context label split Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train This was done by MERT optimization ( ) towards post-edits under the TER target metric . Uses train She evaluates 3,000 German verbs with a token frequency between 10 and 2,000 against the Duden ( ) . Background train The following four components have been identified as the key elements of a question related to patient care ( ) : Background train report on manually analyzing an open-class vocabulary of 35,000 head words for predicate subcategorization information and comparing the results against the subcategorization details in COMLEX . CompareOrContrast train This is roughly an 11 % relative reduction in error rate over and Bods PCFG-reduction reported in Table 1 . CompareOrContrast train Finally , feedback expressions ( head nods and shakes ) are successfully predicted from speech , prosody and eye gaze in interaction with Embodied Communication Agents as well as human communication ( Fujie et al. , 2004 ; ; Morency et al. , 2007 ; Morency et al. , 2009 ) . Background train We use the agreement checker code developed by and evaluate our baseline ( MaltParser using only CORE12 ) , best performing model ( Easy-First Parser using CORE12 + DET+LMM+PERSON+FN * NGR g + p ) , and the gold reference . Uses train The diagnoser , based on b ) , outputs a diagnosis which consists of lists of correct , contradictory and non-mentioned objects and relations from the student 's answer . Extends train The formalization of DLRs provided by defines a formal lexical rule specification language and provides a semantics for that language in two steps : A rewrite system enriches the lexical rule specification into a fully explicit description of the kind shown in Figure 1 . Background train Similar to our previous work ( Chan and Ng , 2005b ) , we used the supervised WSD approach described in ( ) for our experiments , using the naive Bayes algorithm as our classifier . Uses train Similar to ( a ) , our summarization system is , which consists of three key components : an initial sentence pre-selection module to select some important sentence candidates ; the above compression model to generate n-best compressions for each sentence ; and then an ILP summarization method to select the best summary sentences from the multiple compressed sentences . CompareOrContrast train The extraction procedure consists of three steps : First , the bracketing of the trees in the Penn Treebank is corrected and extended based on the approaches of and Collins ( 1997 ) . Background train transition-based dependency parsing framework ( ) using an arc-eager transition strategy and are trained using the perceptron algorithm as in Zhang and Clark ( 2008 ) with a beam size of 8 . Uses train But their importance has grown far beyond machine translation : for instance , transferring annotations between languages ( Yarowsky and Ngai 2001 ; Hwa et al. 2005 ; Ganchev , Gillenwater , and Taskar 2009 ) ; discovery of paraphrases ( ) ; and joint unsupervised POS and parser induction across languages ( Snyder and Barzilay 2008 ) . Motivation train each relevant document is retrieved ( ) . Uses train have previously examined the task of categorizing sentences in medical abstracts using supervised discriminative machine learning techniques . Background train This recognizer incrementally outputs word hypotheses as soon as they are found in the best-scored path in the forward search ( ) using the ISTAR ( Incremental Structure Transmitter And Receiver ) protocol , which conveys word graph information as well as word hypotheses . Background train The recent great advances in speech and language technologies have made it possible to build fully implemented spoken dialogue systems ( ; Allen et al. , 1996 ; Zue et al. , 2000 ; Walker et al. , 2000 ) . Background train We posit that this would not have a significant effect on the results , in particular for MML-based classification techniques , such as Decision Graphs ( ) . Background val showed how the perceptron algorithm can be used to efficiently compute the best parse with DOP1 's subtrees , reporting a 5.1 % relative reduction in error rate over the model in Collins ( 1999 ) on the WSJ . Background train Tateisi et al. also translated LTAG into HPSG ( ) . CompareOrContrast train The system utilizes several large size biological databases including three NCBI databases ( GenPept [ 11 ] , RefSeq , and Entrez GENE [ 13 ] ) , PSD database from Protein Information Resources ( PIR ) [ 14 ] , and Uses train The LM uses the monolingual data and is trained as a five-gram9 using the SRILM-Toolkit ( ) . Uses train There has also been work focused upon determining the political leaning ( e.g. , `` liberal '' vs. `` conservative '' ) of a document or author , where most previously-proposed methods make no direct use of relationships between the documents to be classified ( the `` unlabeled '' texts ) ( ; Efron , 2004 ; Mullen and Malouf , 2006 ) . Background train A number of speech understanding systems have been developed during the past fifteen years ( Barnett et al. 1980 , Dixon and Martin 1979 , Erman et al. 1980 , Haton and Pierrel 1976 , Lea 1980 , , Medress 1980 , Reddy 1976 , Walker 1978 , and Wolf and Woods 1980 ) . CompareOrContrast train Secondly , the cooperative principle of , 1978 ) , under the assumption that referential levels of a writer and a reader are quite similar , implies that the writer should structure the text in a way that makes the construction of his intended model easy for the reader ; and this seems to imply that he should appeal only to the most direct knowledge of the reader . Motivation train In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ Kaplan and Bresnan 1982 ; Bresnan 2001 ; Dalrymple 2001 ] , head-driven phrase structure grammar [ HPSG ] [ Pollard and Sag 1994 ] , tree-adjoining grammar [ TAG ] [ ] , and combinatory categorial grammar [ CCG ] [ Ades and Steedman 1982 ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Background train In this paper , inspired by KNN-SVM ( ) , we propose a local training method , which trains sentence-wise weights instead of a single weight , to address the above two problems . Motivation train 1990 ) , on linguisitic acquisition ( by the use of Part-of-Speech filters hand-crafted by a linguist ) ( Oueslati , 1999 ) or , more frequently , on a combination of the two ( Smadja , 1993 ; , for example ) . CompareOrContrast train report excellent part-of-speech tagging results using a handcrafted approach that is close to OT .3 More speculatively , imagine an OT grammar for stylistic revision of parsed sentences . Background train Thus for instance , ( ; Copestake et al. , 2001 ) describes a Head Driven Phrase Structure Grammar ( HPSG ) which supports the parallel construction of a phrase structure ( or derived ) tree and of a semantic representation and ( Dalrymple , 1999 ) show how to equip Lexical Functional grammar ( LFG ) with a glue semantics . Background train While many linguistic theories state subcategorization requirements in terms of phrase structure ( CFG categories ) , questions the viability and universality of such an approach because of the variety of ways in which grammatical functions may be realized at the language-specific constituent structure level . Background train We are going to make such a comparison with the theories proposed by J. , 1982 ) that represent a more computationally oriented approach to coherence , and those of T.A. van Dijk and W. Kintch ( 1983 ) , who are more interested in addressing psychological and cognitive aspects of discourse coherence . CompareOrContrast train 11 From ( ) , we find that the performance of SAMT system is similar with the method of labeling SCFG rules with POS tags . Motivation train In this paper we focus on the exploitation of the LDOCE grammar coding system ; Alshawi et al. ( 1985 ) and describe further research in Cambridge utilising different types of information available in LDOCE . Background train observed that some annotators were not familiar with the exact definition of semantic relatedness . Motivation train All current approaches to monolingual TE , either syntactically oriented ( Rus et al. , 2005 ) , or applying logical inference ( Tatu and Moldovan , 2005 ) , or adopting transformation-based techniques ( ; Bar-Haim et al. , 2008 ) , incorporate different types of lexical knowledge to support textual inference . Background train Although originally developed as a tool to assist in query formulation , pointed out that PICO frames can be employed to structure IR results for improving precision . Background train There has been some controversy , at least for simple stemmers ( Lovins , 1968 ; Porter , 1980 ) , about the effectiveness of morphological analysis for document retrieval ( ; Krovetz , 1993 ; Hull , 1996 ) . Background train in history-based models ( ) , the probability estimate for each derivation decision di is conditioned on the previous derivation decisions d1 , ... , d , _ 1 , which is called the derivation history at step i . Uses train The reader is referred to for a more detailed discussion of our use of constraint propagation .32 We illustrate the result of constraint propagation with our example grammar . Background train Such tools make it easy to run most current approaches to statistical markup , chunking , normalization , segmentation , alignment , and noisy-channel decoding , ' including classic models for speech recognition ( Pereira and Riley , 1997 ) and machine translation ( ) . Background train Such technologies require significant human input , and are difficult to create and maintain ( ) . Background train We can define PCAT using a probabilistic grammar ( ) . Background val It is these orthographic variations and complex morphological structure that make Arabic language processing challenging ( ; Xu et al. , 2002 ) . Background train The first is the one used in the chunking competition in CoNLL-2000 ( Tjong Kim ) . Uses train There have been many studies on parsing techniques ( Poller and Becker , 1998 ; Flickinger et al. , 2000 ) , ones on disambiguation models ( Chiang , 2000 ; ) , and ones on programming/grammar-development environ - Background train For example , experimented with abstracts and full article texts in the task of automatically generating index term recommendations and discovered that using full article texts yields at most a 7.4 % improvement in F-score . Background train The ten most specific nouns have been produced by comparing our corpus of computing to the French corpus Le Monde , composed of newspaper articles ( ) . Uses train Secondly , we need to investigate techniques for identifying identical documents , virtually identical documents and highly repetitive documents , such as those pioneered by b ) and shingling techniques described by Chakrabarti ( 2002 ) . Future train Following the work of , we implement a linear-chain CRF merging system using the following features : stemmed ( separated ) surface form , part-of-speech14 and frequencies from the training corpus for bigrams/merging of word and word +1 , word as true prefix , word +1 as true suffix , plus frequency comparisons of these . Uses train attempts to improve on the approach of Brent ( 1993 ) by passing raw text through a stochastic tagger and a finite-state parser ( which includes a set of simple rules for subcategorization frame recognition ) in order to extract verbs and the constituents with which they co-occur . Background train From an IR view , a lot of specialized research has already been carried out for medical applications , with emphasis on the lexico-semantic aspects of dederivation and decomposition ( Pacak et al. , 1980 ; Norton and Pacak , 1983 ; Wolff , 1984 ; Wingert , 1985 ; Dujols et al. , 1991 ; ) . Background val Rather than producing a complete analysis of sentences , the alternative is to perform only partial analysis of the syntactic structures in a text ( ; Abney , 1991 ; Greffenstette , 1993 ) . Background train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train Problems such as these have motivated research on more abstract , dependencybased parser evaluation ( e.g. , Lin 1995 ; Carroll , Briscoe , and Sanfilippo 1998 ; Carroll et al. 2002 ; Clark and Hockenmaier 2002 ; King et al. 2003 ; Preiss 2003 ; ; Miyao and Tsujii 2004 ) . Motivation train They are widely used in MT as a way to figure out how to translate input in one language into output in another language ( ) . Background train Further details about the properties of entropy can be found in textbooks on information theory ( e.g. , ) . Background train , for example , discussed the transcripts of a dialogue between people who assemble a piece of garden furniture ( originally recorded by Candy Sidner ) . Background train Here , PV ( A ) represents an ascent direction chosen as follows : For inequality constraints , it is the projected gradient ( ) ; for equality constraints with slack , we use conjugate gradient ( Nocedal and Wright 1999 ) , noting that when A = 0 , the objective is not differentiable . Uses val Some methods are based on likelihood ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) , error rate ( ; Zhao and Chen , 2009 ; Pauls et al. , 2009 ; Galley and Quirk , 2011 ) , margin ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ) and ranking ( Hopkins and May , 2011 ) , and among which minimum error rate training ( MERT ) ( Och , 2003 ) is the most popular one . Motivation train This idea was inspired by , who used a list of arguments surrounding the main verb together with the verb 's subcategorization information and previously processed examples to analyse semantic roles ( case relations ) . Motivation train , `` domain circumscription '' ( cfXXX ) , and their kin . CompareOrContrast train While these approaches have been reasonably successful ( see Mitkov ( 2002 ) ) , speculate that deeper linguistic knowledge needs to be made available to resolvers in order to reach the next level of performance . Background train A parsing experiment shows that an efficient HPSG parser with the obtained grammar achieved a significant speed-up against an existing LTAG parser ( ) . CompareOrContrast train Features were discovered from the actual sentence plan trees that the SPG generated through the feature derivation process described below , in a manner similar to that used by . CompareOrContrast train Such questions are typically answered by designing appropriate priming experiments ( ) or other lexical decision tasks . Background train The system is in the form of an agenda-driven chart-based parser whose foundation is similar to previous formalizations of Chomsky 's Minimalist Program ( ; Harkema , 2000 ; Niyogi , 2001 ) . CompareOrContrast train and Akkerman et al. ( 1985 ) provide a more detailed analysis of the information encoded by the LDOCE grammar codes and discuss their efficacy as a system of linguistic description . Background train The research described below is taking place in the context of three collaborative projects ( Boguraev , 1987 ; Russell et al. , 1986 ; ) to develop a general-purpose , wide coverage morphological and syntactic analyser for English . Background train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( Bikel et al. , 1997 ; Miller et al. , 1998 ; Borthwick , 1999 ; Mikheev et al. , 1999 ; Soon et al. , 2001 ; ; Florian et al. , 2004 ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train Viewed in this way , gradable adjectives are an extreme example of the `` efficiency of language '' ( ) : Far from meaning something concrete like `` larger than 8 cm '' -- a concept that would have very limited applicability -- or even something more general like `` larger than the average N , '' a word like large is applicable across a wide range of different situations . CompareOrContrast train However , since work in this direction has started , a significant progress has also been made in the research on statistical learning of full parsers , both in terms of accuracy and processing time ( Charniak , 1997b ; Charniak , 1997a ; ; Ratnaparkhi , 1997 ) . Background train It is defined on different kinds of textual units , e.g. documents , parts of a document ( e.g. words and their surrounding context ) , words or concepts ( ) .2 Linguistic distance between words is inverse to their semantic similarity or relatedness . Background train However , the greatest increase is in the amount of raw text available to be processed , e.g. the English Gigaword Corpus ( Linguistic Data ) . Background val For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( ) , ( Meyers et al. , 2000 ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( Richardson et al. , 2001 ) , ( Al-Adhaileh & Tang , 1999 ) . Background val introduced the log-linear model for statistical machine translation ( SMT ) , in which translation is considered as the following optimization problem : Background train Representative systems are described in Boisen et al. ( 1989 ) , De Mattia and Giachin ( 1989 ) , Niedermair ( 1989 ) , , and Young ( 1989 ) . Background train For compound splitting , we follow , using linguistic knowledge encoded in a rule-based morphological analyser and then selecting the best analysis based on the geometric mean of word part frequencies . Uses train showed that the use of a POS factor only results in negligible BLEU improvements , but we need access to the POS in our inflection prediction models . Background train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background train We first identified the most informative unigrams and bigrams using the information gain measure ( ) , and then selected only the positive outcome predictors using odds ratio ( Mladenic and Grobelnik 1999 ) . Uses train Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( Neville and Jensen , 2000 ; Lafferty et al. , 2001 ; Getoor et al. , 2002 ; Taskar et al. , 2002 ; ; Taskar et al. , 2004 ; McCallum and Wellner , 2004 ) . Background val It compares favorably to other stemming or root extraction algorithms ( ; Al-Shalabi and Evens , 1998 ; and Houmame , 1999 ) , with a performance of over 97 % for extracting the correct root in web documents , and it addresses the challenge of the Arabic broken plural and hollow verbs . Motivation train We perceive that these results can be extended to other language models that properly embed bilexical context-free grammars , as for instance the more general history-based models used in ( ) and ( Chelba and Jelinek , 1998 ) . Future train Clearly , what it takes for the adjective to be applicable has not been cast in stone , but is open to fiat : the speaker may decide that 8 cm is enough , or the speaker may set the standards higher ( cfXXX , ) . Background train Many investigators ( e.g. ; Elowitz et al. 1976 ; Luce et al. 1983 ; Cahn 1988 ) have suggested that the poor prosody of synthetic speech , in comparison with natural speech , is the primary factor leading to difficulties in the comprehension of fluent synthetic speech . Motivation train The PERSIVAL project , the most comprehensive study of such techniques applied on medical texts to date , leverages patient records to generate personalized summaries in response to physicians ' queries ( McKeown , Elhadad , and Hatzivassiloglou 2003 ; ) . CompareOrContrast train See for further discussion . Background train We gather similar words using a ) , mining similar verbs from a comparable-sized parsed corpus , and collecting similar nouns from a broader 10 GB corpus of English text .4 We also use Keller and Lapata ( 2003 ) 's approach to obtaining web-counts . Uses train Thus , for example , it can acquire a `` script '' such as the one for going to a restaurant as defined in . Background train ones , DIRT ( ) , VerbOcean ( Chklovski and Pantel , 2004 ) , FrameNet ( Baker et al. , 1998 ) , and Wikipedia ( Mehdad et al. , 2010 ; Kouylekov et al. , 2009 ) . Background train There is some literature on procedure acquisition such as the LISP synthesis work described in Biermann et al. ( 1984 ) and the PROLOG synthesis method of . CompareOrContrast train The system uses a domain-specific content planner to produce input to the surface realizer based on the strategy decision , and a FUF/SURGE ( ) generation system to produce the appropriate text . Uses train However , each of these fields requires further decoding and restructuring to provide client programs with easy access to the information they require ( see for further discussion ) . Background train Our results also confirm the insights gained by , who observed that in crossdomain polarity analysis adding more training data is not always beneficial . CompareOrContrast train We consider the Creative Commons model as the most suitable one to let each author choose the rights to reserve ( ) . Uses val It compares favorably to other stemming or root extraction algorithms ( Yates and Neto , 1999 ; Al-Shalabi and Evens , 1998 ; and ) , with a performance of over 97 % for extracting the correct root in web documents , and it addresses the challenge of the Arabic broken plural and hollow verbs . Motivation train In a similar vein , showed that a different feature-topic model improved predictions on a fill-in-the-blank task . Background val As ( ) show , lexical information improves on NP and VP chunking as well . Future val But while Bod 's estimator obtains state-of-the-art results on the WSJ , comparable to Charniak ( 2000 ) and Collins ( 2000 ) , Bonnema et al. 's estimator performs worse and is comparable to . Background train In multi-party discussion people usually mention each other 's name for the purpose of disentanglement ( ) . Background train Fortunately , indirect associations are usually not difficult to identify , because they tend to be weaker than the direct associations on which they are based ( c ) . Background train The work of and Steedman , Osborne , et al. ( 2003 ) suggests that co-training can be helpful for statistical parsing . Background val TNT refers to the HPSG parser ( ) , C++ implementation of the two-phase parsing algorithm that performs filtering with a compiled CFG ( phase 1 ) and then executes feature unification ( phase 2 ) . CompareOrContrast train 1 ° The body of a plan can be an action or sequence of actions , a goal or sequence 9 Moore and Paris also note that `` a generation system must maintain the kinds of information outlined by Grosz and Sidner '' ( , 203 ) . Background val Surveys and articles on the topic include Lamarche and Retord ( 1996 ) , de Groote and Retord ( 1996 ) , and . Background train Another line of research that is correlated with ours is recognition of agreement/disagreement ( Misra and Walker , 2013 ; Yin et al. , 2012 ; Abbott et al. , 2011 ; Andreas et al. , 2012 ; ; Hillard et al. , 2003 ) and classification of stances ( Walker et al. , 2012 ; Somasundaran and Wiebe , 2010 ) in online forums . CompareOrContrast val The standard way to handle this problem is to handcraft a finite set of features which provides a sufficient summary of the unbounded history ( Ratnaparkhi , 1999 ; Collins , 1999 ; ) . CompareOrContrast train Since the arguments can provide useful semantic information , the SRL is crucial to many natural language processing tasks , such as Question and Answering ( Narayanan and Harabagiu 2004 ) , Information Extraction ( ) , and Machine Translation ( Boas 2002 ) . Background train Most approaches rely on VerbNet ( ) and FrameNet ( Baker et al. , 1998 ) to provide associations between verbs and semantic roles , that are then mapped onto the current instance , as shown by the systems competing in semantic role labelling competitions ( Carreras and Marquez , 2004 ; Carreras and Marquez , 2005 ) and also ( Gildea and Jurafsky , 2002 ; Pradhan et al. , 2005 ; Shi and Mihalcea , 2005 ) . Background train In previous work ( Bachenko et al. 1986 ) , we described an experimental text-to-speech system that determined prosodic phrasing for the Olive -- Liberman synthesizer ( ) . Background train Note that this ensures that greater importance is attributed to longer chunks , as is usual in most EBMT systems ( cfXXX Sato and Nagao 1990 ; Veale and Way 1997 ; ) .7 As an example , consider the translation into French of the house collapsed . Background train This approach is taken , for example , in LKB ( Copestake 1992 ) where lexical rules are introduced on a par with phrase structure rules and the parser makes no distinction between lexical and nonlexical rules ( , 31 ) . CompareOrContrast train 27 argue that semi-productivity of lexical rules , which can be understood as a generalization of exceptions to lexical rules , can be integrated with our approach by assigning probabilities to the automaton associated with a particular lexical entry . Background train Results from other systems show that measures of semantic coherence between a student and a system were positively associated with higher learning gain ( ) . Future train These automatic transformations are based on linguistic rules ( ) . Uses train • Only an automatic evaluation was performed , which relied on having model responses ( Berger and Mittal 2000 ; ) . CompareOrContrast train In comparison , the tag set of the Buckwalter Morphological Analyzer ( ) used in the PATB has a core POS set of 44 tags ( CORE44 ) before morphological extension .8 Cross-linguistically , a core set containing around 12 tags is often CompareOrContrast train • use of low level knowledge from the speech recognition phase , • use of high level knowledge about the domain in particular and the dialogue task in general , • a `` continue '' facility and an `` auto-loop '' facility as described by Biermann and Krishnaswamy ( 1976 ) , • a `` conditioning '' facility as described by , • implementation of new types of paraphrasing , • checking a larger environment in the expectation acquisition algorithm when deciding if an incoming sentence is the same or similar to one already seen , and • examining inter-speaker dialogue patterns . Future train present detailed studies on the task of named entity recognition , which discusses and compares different methods on multiple aspects including chunk representation , inference method , utility of non-local features , and integration of external knowledge . Background train mers ( ; Porter , 1980 ) demonstrably improve retrieval performance . Background train Cases like this would be covered if the decision-theoretic property of Pareto optimality ( e.g. , ) was used as the sole criterion : Formally , an object r E C has a Pareto-optimal combination of Values V iff there is no other x E C such that Background train A formula for the test set perplexity ( ) is :13 Background train This method follows a traditional Information Retrieval paradigm ( ) , where a query is represented by the content terms it contains , and the system retrieves from the corpus a set of documents that best match this query . Uses train Table 1 gives the interpretations of eight adjective-noun combinations discussed in and Vendler ( 1968 ) . Uses train The Penn Treebank results reported here for the Markov model approach are at least equivalent to those reported for the Maximum Entropy approach in ( ) . CompareOrContrast train , by comparison , employ 163 distinct predefined frames . Background train Brockmann and Lapata ( 2003 ) have showed that WordNet-based approaches do not always outperform simple frequency-based models , and a number of techniques have been recently proposed which may offer ideas for refining our current unsupervised approach ( Erk , 2007 ; ) . Future train fθ on demand ( ) can pay off here , since only part of fθ may be needed subsequently . ) Background val An approach ( also based on regulation of the succession of rule application ) to the associated problem of spurious ambiguity is given in but again , to our knowledge , there is no predictive relation between incremental combinatory processing and the kind of processing phenomena cited in the introduction . Background train This equivalence is doing essentially the same job as Pereira 's pronoun abstraction schema in . CompareOrContrast train Finally , we experiment with a method for combining phrase tables proposed in ( ; Nakov and Ng , 2012 ) . Uses train Our work is more similar to NLG work that concentrates on structural constraints such as generative poetry ( Greene et al. , 2010 ) ( Colton et al. , 2012 ) ( Jiang and Zhou , 2008 ) or song lyrics ( ) ( Ramakrishnan A et al. , 2009 ) , where specified meter or rhyme schemes are enforced . CompareOrContrast train For the cases where retrieval took place , we used F-score ( van Rijsbergen 1979 ; ) to determine the similarity between the response from the top-ranked document and the real response ( the formulas for F-score and its contributing factors , recall and precision , appear in Section 4.2 ) . Uses val These translations gave rise to a number of automatically constructed linguistic resources : ( 1 ) the original ( source , target ) phrasal translation pairs , ( 2 ) the marker lexicon , ( 3 ) the gen11 Thanks are due to one of the anonymous reviewers for pointing out that our wEBMT system , seeded with input from multiple translation systems , with a postvalidation process via the Web ( amounting to an n-gram target language model ) , in effect forms a multiengine MT system as described by , Frederking et al. ( 1994 ) , and Hogan and Frederking ( 1998 ) . CompareOrContrast val This includes work on generalized expectation ( Mann and McCallum , 2010 ) , posterior regularization ( Ganchev et al. , 2010 ) and constraint driven learning ( Chang et al. , 2007 ; ) . Background val The list , a synthesis of a number of relation lists cited in the literature , has been designed to be general , domainindependent ( a ) . Motivation val Unlike our approach , those of and Hockenmaier , Bierner , and Baldridge ( 2004 ) include a substantial initial correction and clean-up of the Penn-II trees . CompareOrContrast train 2The algorithm was implemented by the the authors , following the description in . Uses train Our motivation for generation of material for language education exists in work such as Sumita et al. ( 2005 ) and , which deal with automatic generation of classic fill in the blank questions . Motivation train In most recent research , NEs ( person , location and organisations ) are extracted from the text and used as a source of evidence to calculate the similarity between documents - see for instance ( ; Chen and Martin , 2007 ; Popescu and Magnini , 2007 ; Kalashnikov et al. , 2007 ) . Background train Many researchers use the GIZA + + software package ( ) as a black box , selecting IBM Model 4 as a compromise between alignment quality and efficiency . Background train Some works abstract perception via the usage of symbolic logic representations ( ; Chen and Mooney , 2011 ; Matuszek et al. , 2012 ; Artzi and Zettlemoyer , 2013 ) , while others choose to employ concepts elicited from psycholinguistic and cognition studies . Background train Over the past decade , researchers at IBM have developed a series of increasingly sophisticated statistical models for machine translation ( Brown et al. , 1988 ; ; Brown et al. , 1993a ) . Background train One area of current interest concerns the left-to-right arrangement of premodifying adjectives within an NP ( e.g. , ; Malouf 2000 ) . Background train Models of translational equivalence that are ignorant of indirect associations have `` a tendency ... to be confused by collocates '' ( ) . Background train Note that although our feature set was drawn primarily from our prior uncertainty detection experiments ( Forbes-Riley and Litman , 2011a ; ) , we have also experimented with other features , including state-of-theart acoustic-prosodic features used in the last Interspeech Challenges ( Schuller et al. , 2010 ; Schuller et al. , 2009b ) and made freely available in the openSMILE Toolkit ( Florian et al. , 2010 ) . Extends train For this mention-pair coreference model φ ( u , v ) , we use the same set of features used in . Uses train Compared to the reranking technique in , who obtained an LP of 89.9 % and an LR of 89.6 % , our results show a 9 % relative error rate reduction . CompareOrContrast train The paper compares and contrasts the training time needed and performance achieved by our modified learner with two other systems : a standard transformation-based learner , and the ICA system ( ) . CompareOrContrast train There has also been work focused upon determining the political leaning ( e.g. , `` liberal '' vs. `` conservative '' ) of a document or author , where most previously-proposed methods make no direct use of relationships between the documents to be classified ( the `` unlabeled '' texts ) ( Laver et al. , 2003 ; ; Mullen and Malouf , 2006 ) . Background train It projects a functional head , voice ( ) , whose specifier is the external argument . Background train Finite state transducers , which can be learned from bilingual corpora , have been proposed for automatic translation ( Amengual et al. , 2000 ) , as have been bilingual stochastic grammars ( ) . Background train This is noticeable for German ( Brants et al. , 2002 ) and Portuguese ( ) , which still have high overall accuracy thanks to very high attachment scores , but much more conspicuous for Czech ( B ¨ ohmov ´ a et al. , 2003 ) , Dutch ( van der Beek et al. , 2002 ) and Slovene ( Dˇzeroski et al. , 2006 ) , where root precision drops more drastically to about 69 % , 71 % and 41 % , respectively , and root recall is also affected negatively . CompareOrContrast train 3 The degree of precision of the measurement ( , Section 1.5 ) determines which objects can be described by the GRE algorithm , since it determines which objects count as having the same size . Background val There have already been several attempts to develop distributed NLP systems for dialogue systems ( ) and speech recognition ( Hacioglu and Pellom , 2003 ) . Background train Table look-up using an explicit translation lexicon is sufficient and preferable for many multilingual NLP applications , including `` crummy '' MT on the World Wide Web ( Church & Hovy , 1993 ) , certain machine-assisted translation tools ( e.g. ( ; Melamed , 1996b ) ) , concordancing for bilingual lexicography ( Catizone et al. , 1993 ; Gale & Church , 1991 ) , computerassisted language learning , corpus linguistics ( Melby . Background val SWIZZLE is a multilingual enhancement of COCKTAIL ( ) , a coreference resolution system that operates on a mixture of heuristics that combine semantic and textual cohesive information ' . Extends train ( Davis and Ogden , 1997 ; ; Hull and ( 3refenstette , 1996 ) . CompareOrContrast train de URL : http://www.sfs.nphil.uni-tuebingen.de/sfb / b4home.html 1 This is , for example , the case for all proposals working with verbal lexical entries that raise the arguments of a verbal complement ( Hinrichs and Nakazawa 1989 ) that also use lexical rules such as the Complement Extraction Lexical Rule ( ) or the Complement Cliticization Lexical Rule ( Miller and Sag 1993 ) to operate on those raised elements . Background train attempt to translate technical terms using word relation matrices , although the resource from which such relations are derived is a pair of nonparallel corpora . Background train The system is in the form of an agenda-driven chart-based parser whose foundation is similar to previous formalizations of Chomsky 's Minimalist Program ( Stabler , 1997 ; ; Niyogi , 2001 ) . CompareOrContrast val Each component will return a confidence measure of the reliability of its prediction , c.f. ( ) . Motivation train In fact , most of the features3 implemented in existing coreference resolution systems rely solely on mention heads ( ) . Background val The problem of handling ill-formed input has been studied by Carbonell and Hayes ( 1983 ) , , Jensen et al. ( 1983 ) , Kwasny and Sondheimer ( 1981 ) , Riesbeck and Schank ( 1976 ) , Thompson ( 1980 ) , Weischedel and Black ( 1980 ) , and Weischedel and Sondheimer ( 1983 ) . CompareOrContrast train In most recent research , NEs ( person , location and organisations ) are extracted from the text and used as a source of evidence to calculate the similarity between documents - see for instance ( Blume , 2005 ; ; Popescu and Magnini , 2007 ; Kalashnikov et al. , 2007 ) . Background train This conception of lexical rules thus can be understood as underlying the computational approach that treats lexical rules as unary phrase structure rules as , for example , adopted in the LKB system ( ) . Background train Typical letter-to-sound rule sets are those described by Ainsworth ( 1973 ) , McIlroy ( 1973 ) , Elovitz et al. ( 1976 ) , Hurmicutt ( 1976 ) , and . Background train describe an efficient algorithm ( of linear complexity in the number of training sentences ) for computing the LDA transform matrix , which entails computing the withinand between-covariance matrices of the classes , and using Singular Value Decomposition ( SVD ) to compute the eigenvectors of the new space . Uses val For better comparison with work of others , we adopt the suggestion made by to evaluate the parsing quality on sentences up to 70 tokens long . Uses val Some examples include text categorization ( ) , base noun phrase chunking ( Ngai and Yarowsky 2000 ) , part-of-speech tagging ( Engelson Dagan 1996 ) , spelling confusion set disambiguation ( Banko and Brill 2001 ) , and word sense disambiguation ( Fujii et al. 1998 ) . Background val Differently , designed a sampler to infer an STSG by fixing the tree structure and exploring the space of alignment . Motivation val They use a Bag of Visual Words ( BoVW ) model ( ) to create a bimodal vocabulary describing documents . Background train W. discussed sentences of the form * This is a chair but you can sit on it . Background train This imbalance foils thresholding strategies , clever as they might be ( ; Wu & Xia , 1994 ; Chen , 1996 ) . Background train The problem of handling ill-formed input has been studied by Carbonell and Hayes ( 1983 ) , Granger ( 1983 ) , Jensen et al. ( 1983 ) , Kwasny and Sondheimer ( 1981 ) , Riesbeck and Schank ( 1976 ) , , Weischedel and Black ( 1980 ) , and Weischedel and Sondheimer ( 1983 ) . CompareOrContrast train Prototypes of Internet search engines for linguists , corpus linguists and lexicographers have been proposed : WebCorp ( Kehoe and Renouf , 2002 ) , KWiCFinder ( Fletcher , 2004a ) and the Linguist 's Search Engine ( ; Resnik and Elkiss , 2003 ) . Background val In most recent research , NEs ( person , location and organisations ) are extracted from the text and used as a source of evidence to calculate the similarity between documents - see for instance ( Blume , 2005 ; Chen and Martin , 2007 ; Popescu and Magnini , 2007 ; ) . Background train More recently , Silberer et al. ( 2013 ) show that visual attribute classifiers , which have been immensely successful in object recognition ( ) , act as excellent substitutes for feature Background train In other methods , lexical resources are specifically tailored to meet the requirements of the domain ( ) or the system ( Gomez , 1998 ) . Background val Such systems extract information from some types of syntactic units ( clauses in ( Fillmore and Atkins , 1998 ; ; Hull and Gomez , 1996 ) ; noun phrases in ( Hull and Gomez , 1996 ; Rosario et al. , 2002 ) ) . Background val The third version ( VOYAGER ) serves as an interface both with a recognizer and with a functioning database back-end ( ) . Uses train We use the non-projective k-best MST algorithm to generate k-best lists ( ) , where k = 8 for the experiments in this paper . Uses train As for work on Arabic ( MSA ) , results have been reported on the PATB ( Kulick , Gabbard , and Marcus 2006 ; Diab 2007 ; Green and Manning 2010 ) , the Prague Dependency Treebank ( PADT ) ( Buchholz and Marsi 2006 ; ) and the CATiB ( Habash and Roth 2009 ) . Background train For example , McKnight and Srinivasan ( 2003 ) describe a machine learning approach to automatically label sentences as belonging to introduction , methods , results , or conclusion using structured abstracts as training data ( see also ) . Background train • Graph transformations for recovering nonprojective structures ( ) . Uses train 's CCM is an unlabeled bracketing model that generates the span of part-of-speech tags that make up each constituent and the pair of tags surrounding each constituent span ( as well as the spans and contexts of each non-constituent ) . Background train This process produces a hierarchical clustering of the word types in the corpus , and these clusterings have been found useful in many applications ( ; Koo et al. , 2008 ; Miller et al. , 2004 ) . Motivation train Opposition ( called `` adversative '' or `` contrary-to-expectation '' by Halliday and Hasan 1976 ; cfXXX also , p. 672 ) . Background train Aside from the extraction of theory-neutral subcategorization lexicons , there has also been work in the automatic construction of lexical resources which comply with the principles of particular linguistic theories such as LTAG , CCG , and HPSG ( ; Xia 1999 ; Hockenmaier , Bierner , and Baldridge 2004 ; Nakanishi , Miyao , and Tsujii 2004 ) . Background train 1 The representation in is even more compact than ours for grammars that are not self-embedding . CompareOrContrast train 29 This improvement of the covariation encoding can also be viewed as an instance of the program transformation technique referred to as deletion of clauses with a finitely failed body ( ) . CompareOrContrast train The reordering models we describe follow our previous work using function word models for translation ( ; Setiawan et al. , 2009 ) . Extends train Although there are other discussions of the paragraph as a central element of discourse ( e.g. Chafe 1979 , , Longacre 1979 , Haberlandt et al. 1980 ) , all of them share a certain limitation in their formal techniques for analyzing paragraph structure . CompareOrContrast train For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( Meyers et al. , 2000 ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( Richardson et al. , 2001 ) , ( ) . Background train In corpus linguistics building such megacorpora is beyond the scope of individual researchers , and they are not easily accessible ( Kennedy , 1998 : 56 ) unless the web is used as a corpus ( ) . Background train Here 11 is an optimization precision , oc is a step size chosen with the strong Wolfe 's rule ( ) . Uses train See ( ) for a discussion . Background train Figure 2 ( a ) shows the frame-based semantic representation for the utterance `` What time is Analyze This playing 2 See ( ) for how MIMIC 's dialoguelevel knowledge is used to override default prosodic assignments for concept-to-speech generation . Background train For Berkeley system , we use the reported results from . CompareOrContrast train Instead , we will adopt the nomenclature of the Automatic Content Extraction program ( ) : we will call the instances of textual references to objects/abstractions mentions , which can be either named ( e.g. John Mayor ) , nominal ( the president ) or pronominal ( she , it ) . Uses train Such a component would serve as the first stage of a clinical question answering system ( ) or summarization system ( McKeown et al. , 2003 ) . Future train In the areas of Natural Language Processing ( NLP ) and computational linguistics , proposals have been made for using the computational Grid for data-intensive NLP and text-mining for eScience ( ; Hughes et al , 2004 ) . Background train comprehensively compares different approaches to complementation within grammatical theory providing a touchstone against which the LDOCE scheme can be evaluated . CompareOrContrast val To prove that our method is effective , we also make a comparison between the performances of our system and , Xue ( 2008 ) . CompareOrContrast val The obtained SCFG is further used in a phrase-based and hierarchical phrase-based system ( ) . CompareOrContrast train 32 In certain cases an extension of the constraint language with named disjunctions or contexted constraints ( Maxwell and Kaplan 1989 ; Eisele and Dorre 1990 ; ) can be used to circumvent constraint propagation . Background train , 2009 , 2010 ) utilized Bayesian methods to learn synchronous context free grammars ( SCFG ) from a parallel corpus . CompareOrContrast val Since the arguments can provide useful semantic information , the SRL is crucial to many natural language processing tasks , such as Question and Answering ( ) , Information Extraction ( Surdeanu et al. 2003 ) , and Machine Translation ( Boas 2002 ) . Background train In addition to the model based upon a dictionary of stems and words , we also experimented with models based upon character n-grams , similar to those used for Chinese segmentation ( ) . CompareOrContrast train Notable early papers on graph-based semisupervised learning include Blum and Chawla ( 2001 ) , Bansal et al. ( 2002 ) , Kondor and Lafferty ( 2002 ) , and . Background train Due to using a global model like CRFs , our previous work in ( Zhao et al. , 2006 ; c ) reported the best results over the evaluated corpora of Bakeoff-2 until now7 . CompareOrContrast val As already mentioned in the literature , see for example ( ) , knowledge about implicit predicates could be potentially useful for a variety of NLP tasks such as language generation , information extraction , question answering or machine translation . Background train We use the Columbia Arabic Treebank ( CATiB ) ( ) . Uses train This problem may be similar to the situation in which current formal grammars allow nonsensical but parsable collections of words ( e.g. , `` colorless green ideas ... '' ) , while before the advent of Chomskyan formalisms , a sentence was defined as the smallest meaningful collection of words ; , p. 546 ) gives 10 definitions of a sentence . Background train Another technique for making better use of unlabeled data is cotraining ( ) , in which two sufficiently different learners help each other learn by labeling training data for one another . Background val For example , modeling CASE in Czech improves Czech parsing ( ) : CASE is relevant , not redundant , and can be predicted with sufficient accuracy . Motivation train Table look-up using an explicit translation lexicon is sufficient and preferable for many multilingual NLP applications , including `` crummy '' MT on the World Wide Web ( Church & Hovy , 1993 ) , certain machine-assisted translation tools ( e.g. ( Macklovitch , 1994 ; Melamed , 1996b ) ) , concordancing for bilingual lexicography ( Catizone et al. , 1993 ; ) , computerassisted language learning , corpus linguistics ( Melby . Background train Following our previous work ( ; Althaus , Karamanis , and Koller 2004 ) , the input to information ordering is an unordered set of informationbearing items represented as CF lists . Extends train There has been some controversy , at least for simple stemmers ( Lovins , 1968 ; Porter , 1980 ) , about the effectiveness of morphological analysis for document retrieval ( Harman , 1991 ; Krovetz , 1993 ; ) . Background train This situation suggests a response-automation approach that follows the document retrieval paradigm ( ) , where a new request is matched with existing response documents ( e-mails ) . Background train All current approaches to monolingual TE , either syntactically oriented ( Rus et al. , 2005 ) , or applying logical inference ( Tatu and ) , or adopting transformation-based techniques ( Kouleykov and Magnini , 2005 ; Bar-Haim et al. , 2008 ) , incorporate different types of lexical knowledge to support textual inference . Background val This is the approach taken by IBM Models 4 + ( Brown et al. 1993b ; Och and Ney 2003 ) , and more recently by the LEAF model ( ) . CompareOrContrast train We have noted that many of these desiderata make complex question answering quite similar to multi-document summarization ( b ) , but these features are also beyond the capabilities of current summarization systems . CompareOrContrast train :472 ) , but these are the only ones which are explicit in the LDOCE coding system . Background train This paper describes an approach for sharing resources in various grammar formalisms such as Feature-Based Lexicalized Tree Adjoining Grammar ( FB-LTAG1 ) ( Vijay-Shanker , 1987 ; Vijay-Shanker and Joshi , 1988 ) and Head-Driven Phrase Structure Grammar ( HPSG ) ( ) by a method of grammar conversion . Background train These constructs correspond as directly as possible to properties of the linguistic structure that express them and are , to as small an extent as possible , dependent on the requirements of contextual resolution ( unlike , say , the metavariables of standard QLFs [ ] , or the labels of UDRS [ Reyle 1996 ] , which are motivated entirely by the mechanisms that operate on them after grammatical processing ) . Background train Prototypes of Internet search engines for linguists , corpus linguists and lexicographers have been proposed : WebCorp ( Kehoe and Renouf , 2002 ) , KWiCFinder ( Fletcher , 2004a ) and the Linguist 's Search Engine ( Kilgarriff , 2003 ; ) . Background val Table look-up using an explicit translation lexicon is sufficient and preferable for many multilingual NLP applications , including `` crummy '' MT on the World Wide Web ( Church & Hovy , 1993 ) , certain machine-assisted translation tools ( e.g. ( Macklovitch , 1994 ; b ) ) , concordancing for bilingual lexicography ( Catizone et al. , 1993 ; Gale & Church , 1991 ) , computerassisted language learning , corpus linguistics ( Melby . Background train An off-the-shelf speech recognition device , a Nippon Electric Corporation DP-200 , was added to an existing natural language processing system , the Natural Language Computer ( NLC ) ( Ballard 1979 , ) . Background train Some efforts have tackled tasks such as automatic image caption generation ( Feng and Lapata , 2010a ; Ordonez et al. , 2011 ) , text illustration ( Joshi et al. , 2006 ) , or automatic location identification of Twitter users ( Eisenstein et al. , 2010 ; ; Roller et al. , 2012 ) . Background train The system was trained on the Penn Treebank ( Marcus et al. , 1993 ) WSJ Sections 221 and tested on Section 23 ( Table 1 ) , same as used by , Collins ( 1997 ) , and Ratnaparkhi ( 1997 ) , and became a common testbed . CompareOrContrast train predefine 163 verbal subcategorization frames , obtained by manually merging the classes exemplified in the COMLEX ( MacLeod , Grishman , and Meyers 1994 ) and ANLT ( Boguraev et al. 1987 ) dictionaries and adding around 30 frames found by manual inspection . Background train Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( ; Lafferty et al. , 2001 ; Getoor et al. , 2002 ; Taskar et al. , 2002 ; Taskar et al. , 2003 ; Taskar et al. , 2004 ; McCallum and Wellner , 2004 ) . Background val give a sufficiently general finite-state framework to allow this : weights may fall in any set K ( instead of R ) . Uses train FBLTAG ( Vijay-Shanker , 1987 ; ) is an extension of the LTAG formalism . Background train There are several variations of such a method ( Ballesteros and Croft , 1998 ; Pirkola , 1998 ; ) . CompareOrContrast train 12 In order to focus on the computational aspects of the covariation approach , in this paper we will not go into a discussion of the full lexical rule specification language introduced in . Background val Many investigators ( e.g. Allen 1976 ; Elowitz et al. 1976 ; Luce et al. 1983 ; ) have suggested that the poor prosody of synthetic speech , in comparison with natural speech , is the primary factor leading to difficulties in the comprehension of fluent synthetic speech . Motivation train This Principle of Finitism is also assumed by , Jackendoff ( 1983 ) , Kamp ( 1981 ) , and implicitly or explicitly by almost all researchers in computational linguistics . CompareOrContrast train We performed translation experiments with an implementation of the IBM-4 translation model ( ) . Uses train In this paper , a flexible annotation schema called Structured String-Tree Correspondence ( SSTC ) ( ) will be introduced to capture a natural language text , its corresponding abstract linguistic representation and the mapping ( correspondence ) between these two . Background val For the development of these lists we used a collection of texts of about 300,000 words derived from the New York Times ( NYT ) corpus that was supplied as training data for the 7th Message Understanding Conference ( MUC-7 ) ( ) . Background train For the full parser , we use the one developed by Michael Collins ( Collins , 1996 ; ) -- one of the most accurate full parsers around . Uses train More sophisticated approaches have been proposed ( ) , including an extension that , in an interesting reversal of our problem , makes use of sentimentpolarity indicators within speech segments ( Galley et al. , 2004 ) . Background train have demonstrated that differential weighting of automatically labeled sections can lead to improved retrieval performance . Background train The one-sided t-test ( ) at significance level 0.05 indicated that the improvement on Trec5C is not statistically significant . Uses train mers ( Lovins , 1968 ; ) demonstrably improve retrieval performance . Background train It is therefore no surprise that early attempts at response automation were knowledge-driven ( ; Watson 1997 ; Delic and Lahaix 1998 ) . Background train The framework was originally developed for the realization of deep-syntactic structures in NLG ( ) . Background train These features are very much desired in the design of an annotation scheme , in particular for the treatment of linguistic phenomena , which are non-standard , e.g. crossed dependencies ( ) . Background train Few approaches to parsing have tried to handle disfluent utterances ( notable exceptions are Core & Schubert , 1999 ; ; Nakatani & Hirschberg , 1994 ; Shriberg , Bear , & Dowding , 1992 ) . Background train 5 The open source Moses ( ) toolkit from www.statmt.org/moses/ . Uses train There have been several efforts aimed at developing a domain-independent method for generating responses from a frame representation of user requests ( ; Chu-Carroll , 1999 ) . Future val The maximum entropy approach ( ) presents a powerful framework for the combination of several knowledge sources . Uses train Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) [ 18 ] , fly FlyBase [ 19 ] , yeast Saccharomyces Genome Database ( SGD ) , rat -- Rat Genome Database ( RGD ) [ 21 ] , worm -- WormBase [ 22 ] , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) [ 24 ] , and Enzyme Nomenclature Database ( ECNUM ) [ 25 , 26 ] . Uses val We then use the program Snob ( Wallace and Boulton 1968 ; ) to cluster these experiences . Uses train However , studies have shown that existing systems for searching MEDLINE ( such as PubMed , the search service provided by the National Library of Medicine ) are often inadequate and unable to supply clinically relevant answers in a timely manner ( Gorman , Ash , and Wykoff 1994 ; ) . Background train Our implementation of the NP-based QA system uses the Empire noun phrase finder , which is described in detail in . Uses train For all the experiments reported in this article , we used the training portion of PATB Part 3 v3 .1 ( ) , converted to the CATiB Treebank format , as mentioned in Section 2.5 . Uses train In addition , we consider several types of lexical features ( LexF ) inspired by previous work on agreement and disagreement ( Galley et al. , 2004 ; ) . Motivation val All communicative head gestures in the videos were found and annotated with ANVIL using a subset of the attributes defined in the MUMIN annotation scheme ( ) . Uses train Specifically , we examine the strength of association between the verb and the noun constituent of a combination ( the target expression or its lexical variants ) as an indirect cue to its idiomaticity , an approach inspired by . Motivation train The current system learns finite state flowcharts whereas typical learning systems usually acquire coefficient values as in Minsky and Papert ( 1969 ) , assertional statements as in Michalski ( 1980 ) , or semantic nets as in . CompareOrContrast val More recently , an alignment selection approach was proposed in ( ) , which computes confidence scores for each link and prunes the links from multiple sets of alignments using a hand-picked threshold . CompareOrContrast train for example discusses a method where a syntactic parse of the text is performed and the context of a word is modeled using dependency triples . Background train A variety of statistical methods were proposed over the recent years for learning to produce a full parse of free-text sentences ( e.g. , Bod ( 1992 ) , Magerman ( 1995 ) , , Ratnaparkhi ( 1997 ) , and Sekine ( 1998 ) ) . Background train The strategies employed when MIMIC has only dialogue initiative are similar to the mixed initiative dialogue strategies employed by many existing spoken dialogue systems ( e.g. , ( Bennacef et at. , 1996 ; ) ) . CompareOrContrast train 7A11 our results are computed with the evalb program following the now-standard criteria in ( ) . Uses train We see no good reason , however , why such text spans should necessarily be sentences , since the majority of tagging paradigms ( e.g. , Hidden Markov Model [ HMM ] [ ] , Brill 's [ Brill 1995a ] , and MaxEnt [ Ratnaparkhi 1996 ] ) do not attempt to parse an entire sentence and operate only in the local window of two to three tokens . CompareOrContrast train results are based on a corpus of movie subtitles ( ) , and are consequently shorter sentences , whereas the En → Es results are based on a corpus of parliamentary proceedings ( Koehn 2005 ) . Uses train The system is implemented based on ( Galley et al. , 2006 ) and ( ) . Uses train Also relevant is work on the general problems of dialog-act tagging ( Stolcke et al. , 2000 ) , citation analysis ( Lehnert et al. , 1990 ) , and computational rhetorical analysis ( ; Teufel and Moens , 2002 ) . Background val For projective parsing , it is significantly faster than exact dynamic programming , at the cost of small amounts of search error , We are interested in extending these ideas to phrase-structure and lattice parsing , and in trying other higher-order features , such as those used in parse reranking ( ; Huang , 2008 ) and history-based parsing ( Nivre and McDonald , 2008 ) . Future val Word alignments are used primarily for extracting minimal translation units for machine translation ( MT ) ( e.g. , phrases [ Koehn , Och , and Marcu 2003 ] and rules [ Galley et al. 2004 ; ] ) as well as for Background train Sridhar et al. ( 2009 ) obtain promising results in dialogue act tagging of the Switchboard-DAMSL corpus using lexical , syntactic and prosodic cues , while examine the relation between particular acoustic and prosodic turn-yielding cues and turn taking in a large corpus of task-oriented dialogues . Background train We have also applied our more general unification grammar acquisition methodology to the TIGER Treebank ( ) and Penn Chinese Treebank ( Xue , Chiou , and Palmer 2002 ) , extracting wide-coverage , probabilistic LFG grammar Uses train We use the structures previously used by , and propose one new structure . Uses train Two exceptions to this generalisation are the Linguistic String Project ( Sager , 1981 ) and the IBM CRITIQUE ( formerly EPISTLE ) Project ( Heidorn et al. , 1982 ; ) ; the former employs a dictionary of approximately 10,000 words , most of which are specialist medical terms , the latter has well over 100,000 entries , gathered from machine readable sources . CompareOrContrast train A number of proposals in the 1990s deliberately limited the extent to which they relied on domain and/or linguistic knowledge and reported promising results in knowledge-poor operational environments ( Dagan and Itai 1990 , 1991 ; Lappin and Leass 1994 ; Nasukawa 1994 ; Kennedy and Boguraev 1996 ; Williams , Harvey , and Preston 1996 ; ; Mitkov 1996 , 1998b ) . Background train Towards this aim , a flexible annotation structure called Structured String-Tree Correspondence ( SSTC ) was introduced in to record the string of terms , its associated representation structure and the mapping between the two , which is expressed by the sub-correspondences recorded as part of a SSTC . Background val We have built an experimental text-to-speech system that uses our analysis of prosody to generate phrase boundaries for the Olive -- Liberman synthesizer ( ) . Uses train In this paper , I present a computational implementation of Distributed Morphology ( ) , a non-lexicalist linguistic theory that erases the distinction between syntactic derivation and morphological derivation . Uses train Some recent GRE algorithms have done away with the separation between content determination and linguistic realization , interleaving the two processes instead ( Stone and Webber 1998 ; ) . CompareOrContrast train Subsequent processing by the natural language and response generation components was done automatically by the computer ( ) . Uses train This is implemented as a cascade of simple strategies , which were briefly described in . Uses val But , obviously , there are other possibilities -- for instance , the discourse representation structures ( DRS 's ) of , which have been used to translate a subset of English into logical formulas , to model text ( identified with a list of sentences ) , to analyze a fragment of English , and to deal with anaphora . CompareOrContrast train Better results would be expected by combining the PCFG-LA parser with discriminative reranking approaches ( Charniak and Johnson , 2005 ; ) for self training . Future train Intermedia is no more developed and nobody of us had the opportunity to try it ( ) . Background val Morphological alterations of a search term have a negative impact on the recall performance of an information retrieval ( IR ) system ( Choueka , 1990 ; J ¨ appinen and Niemist ¨ o , 1988 ; ) , since they preclude a direct match between the search term proper and its morphological variants in the documents to be retrieved . Background train ment ( ; Doran et al. , 2000 ; Makino et al. , 1998 ) . Background train Finally , feedback expressions ( head nods and shakes ) are successfully predicted from speech , prosody and eye gaze in interaction with Embodied Communication Agents as well as human communication ( ; Morency et al. , 2005 ; Morency et al. , 2007 ; Morency et al. , 2009 ) . Background val Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( ; Steyvers , 2010 ; Feng and Lapata , 2010b ; Bruni et al. , 2011 ; Silberer and Lapata , 2012 ; Johns and Jones , 2012 ; Bruni et al. , 2012a ; Bruni et al. , 2012b ; Silberer et al. , 2013 ) . Background train We also made use of the person-name/instance pairs automatically extracted by .2 This data provides counts for pairs such as `` Edwin Moses , hurdler '' and `` William Farley , industrialist . '' Uses train presented an approach for constructing a BKB based on the S-SSTC . Background train However , since work in this direction has started , a significant progress has also been made in the research on statistical learning of full parsers , both in terms of accuracy and processing time ( b ; Charniak , 1997a ; Collins , 1997 ; Ratnaparkhi , 1997 ) . Background train Also , advanced methods often require many training iterations , for example active learning ( Dagan and Engelson ,1995 ) and co-training ( ) . Background train furthered this work by showing that a bimodal topic model , consisting of both text and feature norms , outperformed models using only one modality on the prediction of association norms , word substitution errors , and semantic interference tasks . Extends val It can be shown ( ) that the use of this model with maximum likelihood parameter estimation is justified on information-theoretic grounds when q represents some prior knowledge about the true distribution and when the expected values of f in the training corpus are identical to their true expected values .3 There is no requirement that the components of f represent disjoint or statistically independent events . Motivation train Shortly after the publication of The Sound Pattern of English ( Chomsky and Halle 1968 ) , Kornai points out , `` Johnson ( 1970 ) demonstrated that the context-sensitive machinery of SPE ... [ could ] be replaced by a much simpler one , based on finite-state transducers ( FSTs ) ; the same conclusion was reached independently by Kaplan and Kay , whose work remained an underground classic until it was finally published in . '' Background train 18 In this article , we use a newer version of the corpus by than the one we used in Marton , Habash , and Rambow ( 2011 ) . Uses train b ) and Topkara et al. ( 2006a ) attained the embedding capacity of 0.5 bits per sentence with the syntactic transformation method . Background train Another possibility that often works better is to use Minimum Bayes-Risk ( MBR ) decoding ( ; Liang , Taskar , and Klein 2006 ; Ganchev , and Taskar 2007 ) . Uses train The subcategorization requirements expressed by semantic forms are enforced at f-structure level through completeness and coherence well-formedness conditions on f-structure ( ) : An f-structure is locally complete iff it contains all the governable grammatical functions that its predicate governs . Background train This paper describes an approach for sharing resources in various grammar formalisms such as Feature-Based Lexicalized Tree Adjoining Grammar ( FB-LTAG1 ) ( Vijay-Shanker , 1987 ; ) and Head-Driven Phrase Structure Grammar ( HPSG ) ( Pollard and Sag , 1994 ) by a method of grammar conversion . Background train Following the example of , we will call the autonomous units of a hypertext lexias ( from ` lexicon ' ) , a word coined by Roland Barthes ( 1970 ) . Uses train ( ) . Background train We use the same set of binary features as in previous work on this dataset ( ; Pang and Lee , 2004 ; Zaidan et al. , 2007 ) . Uses train Various feature selection techniques have been used in automatic text categorization ; they include document frequency ( DF ) , information gain ( IG ) ( ) , minimum description length principal ( Lang , 1995 ) , and the X2 statistic . Background train Politically-oriented text Sentiment analysis has specifically been proposed as a key enabling technology in eRulemaking , allowing the automatic analysis of the opinions that people submit ( Shulman et al. , 2005 ; Cardie et al. , 2006 ; ) . Background train observed that dependency parsers tend to do quite poorly when parsing questions due to their limited exposure to them in the news corpora from the PennTreebank . CompareOrContrast val a ) 's similar word list for eat misses these but includes sleep ( ranked 6 ) and sit ( ranked 14 ) , because these have similar subjects to eat . Background train • Learnability ( ) • Text generation ( Hovy 1988 ; Milosavljevic , Tulloch , and Dale 1996 ) • Speech generation ( Rayner and Carter 1997 ) • Localization ( Sch ¨ aler 1996 ) Background train For example , a ` web page ' is more similar to an infinite canvas than a written page ( ) . Background train Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) , fly FlyBase [ 19 ] , yeast Saccharomyces Genome Database ( SGD ) [ 20 ] , rat -- Rat Genome Database ( RGD ) [ 21 ] , worm -- WormBase [ 22 ] , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) [ 24 ] , and Enzyme Nomenclature Database ( ECNUM ) [ 25 , 26 ] . Uses train 2This view of typed feature structures differs from the perspective on typed feature structures as modeling partial information as in ( ) . Background train This approach resembles the work by Grishman et al. ( 1986 ) and on selectional restrictions . CompareOrContrast train This is noticeable for German ( ) and Portuguese ( Afonso et al. , 2002 ) , which still have high overall accuracy thanks to very high attachment scores , but much more conspicuous for Czech ( B ¨ ohmov ´ a et al. , 2003 ) , Dutch ( van der Beek et al. , 2002 ) and Slovene ( Dˇzeroski et al. , 2006 ) , where root precision drops more drastically to about 69 % , 71 % and 41 % , respectively , and root recall is also affected negatively . CompareOrContrast train LTAG ( ) is a grammar formalism that provides syntactic analyses for a sentence by composing elementary trees with two opera - Background train Translations have been generated by the CrowdFlower3 channel to Amazon Mechanical Turk4 ( MTurk ) , adopting the methodology proposed by ( ) . Uses train This work is a continuation of that initiated in ( ) , which reports an overall NB classification correctness of 75.6 % , in cross validation experiments , on a data set that consists of 100 documents for each of 12 categories ( the data set is collected from different Arabic portals ) . Extends train Our previous work ( ) designed an EMbased method to construct unsupervised trees for tree-based translation models . CompareOrContrast train Then , we binarize the English parse trees using the head binarization approach ( ) and use the resulting binary parse trees to build another s2t system . Uses train This is the strongest version of the sorites paradox ( e.g. , ) . Background train We used a standard implementation of IBM Model 4 ( ) and because changing the existing code is not trivial , we could not use the same stopping criterion to avoid overfitting and we are not able to produce precision/recall curves . Uses train Cross-lingual Textual Entailment ( CLTE ) has been proposed by ( ) as an extension of Textual Entailment ( Dagan and Glickman , 2004 ) that consists in deciding , given two texts T and H in different languages , if the meaning of H can be inferred from the meaning of T . Background val Some approaches apply semantic parsing , where words and sentences are mapped to logical structure meaning ( ) . Background train For projective parsing , it is significantly faster than exact dynamic programming , at the cost of small amounts of search error , We are interested in extending these ideas to phrase-structure and lattice parsing , and in trying other higher-order features , such as those used in parse reranking ( Charniak and Johnson , 2005 ; ) and history-based parsing ( Nivre and McDonald , 2008 ) . Future train Prototypes of Internet search engines for linguists , corpus linguists and lexicographers have been proposed : WebCorp ( Kehoe and Renouf , 2002 ) , KWiCFinder ( a ) and the Linguist 's Search Engine ( Kilgarriff , 2003 ; Resnik and Elkiss , 2003 ) . Background train Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) [ 18 ] , fly FlyBase [ 19 ] , yeast Saccharomyces Genome Database ( SGD ) [ 20 ] , rat -- Rat Genome Database ( RGD ) [ 21 ] , worm -- WormBase , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) [ 24 ] , and Enzyme Nomenclature Database ( ECNUM ) [ 25 , 26 ] . Uses train Other factors , such as the role of focus ( Grosz 1977 , 1978 ; Sidner 1983 ) or quantifier scoping ( ) must play a role , too . Background train Since mid-2002 , the Library has been employing software that automatically suggests MeSH headings based on content ( ) . Background train Discriminative approaches ( especially SVMs ) have been shown to be very effective for many supervised classification tasks ; see , for example , ( Joachims , 1998 ; ) . Background val Also , the approach will be undefined if the pair is unobserved on the web . Uses train The automation of help-desk responses has been previously tackled using mainly knowledge-intensive paradigms , such as expert systems ( ) and case-based reasoning ( Watson 1997 ) . CompareOrContrast train We measure this association using pointwise Mutual Information ( MI ) ( ) . Uses train Some methods of semantic relation analysis rely on predefined templates filled with information from processed texts ( ) . Background train There are two corpora normally used for evaluation in a number of text-processing tasks : the Brown corpus ( ) and the Wall Street Journal ( WSJ ) corpus , both part of the Penn Treebank ( Marcus , Marcinkiewicz , and Santorini 1993 ) . Uses train or quotation of messages in emails or postings ( see Mullen and Malouf ( 2006 ) but cfXXX ) . Background train extend LDA to allow for the inference of document and topic distributions in a multimodal corpus . Background val SNoW ( ; Roth , 1998 ) is a multi-class classifier that is specifically tailored for learning in domains in which the potential number of information sources ( features ) taking part in decisions is very large , of which NLP is a principal example . Uses train The search algorithm is the standard Viterbi search ( ) , except that the match involves a network-to-network alignment problem rather than sequence-to-sequence . Uses train Since then this idea has been applied to several tasks , including word sense disambiguation ( Yarowsky 1995 ) and named-entity recognition ( ) . Background train 7 We employed the LIBSVM package ( ) . Uses train coreference performance on perfect mentions ( e.g. , Incorporate the two knowledge sources in a ) ; and for those that do report percoreference resolver . CompareOrContrast val According to , p. 67 ) , these two sentences are incoherent . CompareOrContrast train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train Another dialogue acquisition system has been developed by . CompareOrContrast train Lexical functional grammar ( Kaplan and Bresnan 1982 ; Bresnan 2001 ; ) is a member of the family of constraint-based grammars . Background val 1Our rules are similar to those from . CompareOrContrast train In psycholinguistics , relatedness of words can also be determined through association tests ( Schulte im ) . Background train In this article , we use an in-house system which provides functional gender , number , and rationality features ( ) . Uses train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background train The use of the web as a corpus for teaching and research on language has been proposed a number of times ( Kilgarriff , 2001 ; Robb , 2003 ; Rundell , 2000 ; , 2004b ) and received a special issue of the journal Computational Linguistics ( Kilgarriff and Grefenstette , 2003 ) . Background val Several authors in communication studies have pointed out that head movements are relevant to feedback phenomena ( see for an overview ) . Background train Our method resorts to some translation examples , which is similar as example-based translation or translation memory ( ; He et al. , 2010 ; Ma et al. , 2011 ) . CompareOrContrast train Baseline Systems We choose three publicly available state-of-the-art end-to-end coreference systems as our baselines : Stanford system ( Lee et al. , 2011 ) , Berkeley system ( ) and HOTCoref system ( Bj ¨ orkelund and Kuhn , 2014 ) . CompareOrContrast val A number of speech understanding systems have been developed during the past fifteen years ( Barnett et al. 1980 , Dixon and Martin 1979 , Erman et al. 1980 , Haton and Pierrel 1976 , Lea 1980 , Lowerre and Reddy 1980 , Medress 1980 , Reddy 1976 , Walker 1978 , and ) . CompareOrContrast train As for work on Arabic ( MSA ) , results have been reported on the PATB ( Kulick , Gabbard , and Marcus 2006 ; Diab 2007 ; Green and Manning 2010 ) , the Prague Dependency Treebank ( PADT ) ( Buchholz and Marsi 2006 ; Nivre 2008 ) and the CATiB ( ) . Background train Other representations use the link structure ( ) or generate graph representations of the extracted features ( Kalashnikov et al. , 2007 ) . Background train From an IR view , a lot of specialized research has already been carried out for medical applications , with emphasis on the lexico-semantic aspects of dederivation and decomposition ( Pacak et al. , 1980 ; Norton and Pacak , 1983 ; ; Wingert , 1985 ; Dujols et al. , 1991 ; Baud et al. , 1998 ) . Background train There has been some controversy , at least for simple stemmers ( Lovins , 1968 ; ) , about the effectiveness of morphological analysis for document retrieval ( Harman , 1991 ; Krovetz , 1993 ; Hull , 1996 ) . Background train This revalidates the observation of that phrase structure representations and dependency representations add complimentary value to the learning task . CompareOrContrast train The parsing algorithm used for all languages is the deterministic algorithm first proposed for unlabeled dependency parsing by Nivre ( 2003 ) and extended to labeled dependency parsing by . Uses train The semantic categories of verbs and other words are extracted from the Semantic Knowledge-base of Contemporary Chinese ( ) . Uses train Moreover , in order to determine whether the performances of the predictive criteria are consistent across different learning models within the same domain , we have performed the study on two parsing models : one based on a context-free variant of tree-adjoining grammars ( Joshi , Levy , and Takahashi 1975 ) , the Probabilistic Lexicalized Tree Insertion Grammar ( PLTIG ) formalism ( ; Hwa 1998 ) , and Collins 's Model 2 parser ( 1997 ) . Uses train 6 The Partial-VP Topicalization Lexical Rule proposed by , 10 ) is a linguistic example . Background train We measure the inter annotator agreement using the Fleiss Kappa ( ) measure ( x ) where the agreement lies around 0.79 . Uses train The task we used to compare different generalisation techniques is similar to that used by and Rooth et al. ( 1999 ) . CompareOrContrast val Other milestones of recent research include the deployment of probabilistic and machine learning techniques ( Aone and Bennett 1995 ; Kehler 1997 ; Ge , Hale , and Charniak 1998 ; Cardie and Wagstaff 1999 ; the continuing interest in centering , used either in original or in revised form ( Abracos and Lopes 1994 ; Strube and Hahn 1996 ; ; Tetreault 1999 ) ; and proposals related to the evaluation methodology in anaphora resolution ( Mitkov 1998a , 2001b ) . Background train It is not aimed at handling dependencies , which require heavy use of lexical information ( , for PP attachment ) . CompareOrContrast val This includes work on question answering ( Wang et al. , 2007 ) , sentiment analysis ( ) , MT reordering ( Xu et al. , 2009 ) , and many other tasks . Background train I A more detailed discussion of various aspects of the proposed parser can be found in ( ) . Background val Our experimental design with professional bilingual translators follows our previous work a ) comparing scratch translation to post-edit . Extends val Lexical functional grammar ( Kaplan and Bresnan 1982 ; ; Dalrymple 2001 ) is a member of the family of constraint-based grammars . Background train Thus , the second class of SBD systems employs machine learning techniques such as decision tree classifiers ( ) , neural networks ( Palmer and Hearst 1994 ) , and maximum-entropy modeling ( Reynar and Ratnaparkhi 1997 ) . Background train Previously , a user study ( ) has shown that people are reluctant to type full natural language questions , even after being told that they were using a questionanswering system and that typing complete questions would result in better performance . CompareOrContrast train We will examine the worst-case complexity of interpretation as well as generation to shed some light on the hypothesis that vague descriptions are more difficult to process than others because they involve a comparison between objects ( Beun and Cremers 1998 , ) . Background train In our case , the clustering is performed by the program Snob , which implements mixture modeling combined with model selection based on the Minimum Message Length ( MML ) criterion ( Wallace and Boulton 1968 ; ) . Uses train The third approach to cross-lingual retrieval is to map queries and documents to some intermediate representation , e.g latent semantic indexing ( LSI ) ( Littman et al , 1998 ) , or the General Vector space model ( GVSM ) , ( ) . CompareOrContrast train Many provide graphical user interfaces ( GUI ) for manual annotation ( e.g. General Architecture for Text Engineering ( GATE ) ( ) and the Alembic Workbench ( Day et al. , 1997 ) ) as well as NLP tools and resources that can be manipulated from the GUI . Background train The keypoints are clustered into 5,000 visual codewords ( centroids ) using k-means clustering ( ) , and images are then quantized over the 5,000 codewords . Uses train Another approach for partial parsing was presented by . Background val In knowledge-lean approaches , coreference resolvers employ only morpho-syntactic cues as knowledge sources in the resolution process ( e.g. , Mitkov ( 1998 ) , ) . Background train The application of domain models and deep semantic knowledge to question answering has been explored by a variety of researchers ( e.g. , , Rinaldi et al. 2004 ) , and was also the focus of recent workshops on question answering in restricted domains at ACL 2004 and AAAI 2005 . Background train This result is consistent with other works using this model with these features ( Andrews et al. , 2009 ; ) . CompareOrContrast train Cases of intracategorial synonymy are relatively straigthtforward as several electronic synonym dictionnaries for french are available ( ) . Background train The feature of head word trigger which we apply to the log-linear model is motivated by the trigger-based approach ( ) . Motivation train Other work on modeling the meanings of verbs using video recognition has also begun showing great promise ( Mathe et al. , 2008 ; ) . Background val based parsing algorithms with an arc-factored parameterization ( ) . Uses train No attempt has been made to map any closed class entries from LDOCE , as a 3,000 word lexicon containing most closed class items has been developed independently by one of the groups collaborating with us to develop the general purpose morphological and syntactic analyser ( see the Introduction and ) . Background train In this paper we focus on the exploitation of the LDOCE grammar coding system ; and Alshawi ( 1987 ) describe further research in Cambridge utilising different types of information available in LDOCE . Background val Thus , the second class of SBD systems employs machine learning techniques such as decision tree classifiers ( Riley 1989 ) , neural networks ( ) , and maximum-entropy modeling ( Reynar and Ratnaparkhi 1997 ) . Background train We see no good reason , however , why such text spans should necessarily be sentences , since the majority of tagging paradigms ( e.g. , Hidden Markov Model [ HMM ] [ Kupiec 1992 ] , Brill 's [ a ] , and MaxEnt [ Ratnaparkhi 1996 ] ) do not attempt to parse an entire sentence and operate only in the local window of two to three tokens . CompareOrContrast train Similarly , ( Barzilay and Lee , 2003 ) and ( ) learn sentence level paraphrase templates from a corpus of news articles stemming from different news source . Background train In future work we plan to experiment with richer representations , e.g. including long-range n-grams ( Rosenfeld , 1996 ) , class n-grams ( ) , grammatical features ( Amaya and Benedy , 2001 ) , etc ' . Future val Furthermore , we demonstrate that our results carry over successfully to another parser , the Easy-First Parser ( ) ( Section 6 ) . Uses train For instance , relating `` they '' to `` apples '' in the sentence ( cfXXX p. 195 ; Zadrozny 1987a ) : We bought the boys apples because they were so cheap Background train We follow our previous work ( ) in our feature choices , using a fiveword window that includes the target stem and two words on either side for context ( see also Tetreault and Chodorow , 2008 ) . Extends train Adjectives , more than other categories , are a striking example of regular polysemy since they are able to take on different meanings depending on their context , viz. , the noun or noun class they modify ( see and the references therein ) . Background train According to , LFG assumes the following universally available inventory of grammatical functions : SUBJ ( ect ) , OBJ ( ect ) , OBJe , COMP , XCOMP , OBL ( ique ) e , ADJ ( unct ) , XADJ . Background train But their importance has grown far beyond machine translation : for instance , transferring annotations between languages ( Yarowsky and Ngai 2001 ; Hwa et al. 2005 ; Ganchev , Gillenwater , and Taskar 2009 ) ; discovery of paraphrases ( Bannard and Callison-Burch 2005 ) ; and joint unsupervised POS and parser induction across languages ( ) . Motivation train 4 This interpretation of the signature is sometimes referred to as closed world ( Gerdemann and King 1994 ; ) . Background val Our training examples are similar to the data created for pseudodisambiguation , the usual evaluation task for SP models ( Erk , 2007 ; Keller and Lapata , 2003 ; ) . CompareOrContrast train There have already been several attempts to develop distributed NLP systems for dialogue systems ( Bayer et al. , 2001 ) and speech recognition ( ) . Background train For instance , implementing an efficient version of the MXPOST POS tagger ( ) will simply involve composing and configuring the appropriate text file reading component , with the sequential tagging component , the collection of feature extraction components and the maximum entropy model component . Future train cue word and name the first ( or several ) associated words that come to mind ( e.g. , Nelson et al. ( 2004 ) ) , and feature norms , where subjects are given a cue word and asked to describe typical properties of the cue concept ( e.g. , ) . Background train and Litman and Hirschberg ( 1990 ) also examine the relation between discourse and prosodic phrasing . Background train evaluate 914 Czech verbs against a custom-made gold standard and record a token recall of 88 % . CompareOrContrast train We use two measures from Information Retrieval to determine the quality of an automatically generated response : precision and F-score ( van Rijsbergen 1979 ; ) . Uses val The automation of help-desk responses has been previously tackled using mainly knowledge-intensive paradigms , such as expert systems ( Barr and Tessler 1995 ) and case-based reasoning ( ) . CompareOrContrast train did very encouraging work on the feature calibration of semantic role labeling . Background train This approach , which uses words that appear in the context of terms to formulate hypotheses on their semantic relatedness ( , for example ) , does not specify the relationship itself . Background train And ( ) use clustering and similarity measures to identify similar contexts in a single corpus and extract verbal paraphrases from these contexts . Background train Not having to represent the frame explicitly not only enables the linguist to express only the relevant things , but also allows a more compact representation of lexical rules where explicit framing would require the rules to be split up ( ) . Background train There has also been work focused upon determining the political leaning ( e.g. , `` liberal '' vs. `` conservative '' ) of a document or author , where most previously-proposed methods make no direct use of relationships between the documents to be classified ( the `` unlabeled '' texts ) ( Laver et al. , 2003 ; Efron , 2004 ; ) . Background train Other similar approaches include those of Cicekli and G ¨ uvenir ( 1996 ) , , Carl ( 1999 ) , and Brown ( 2000 ) , inter alia . Background train To sum up , this work has been carried out to automatically classify Arabic documents using the NB algorithm , with the use of a different data set , a different number of categories , and a different root extraction algorithm from those used in ( ) . CompareOrContrast val Our work builds on earlier research on learning to identify dialogues in which the user experienced poor speech recognizer performance ( ) . Extends val CCGBank ( ) is used to train the model . Uses train More recent work on terminology structuring has focussed on formal similarity to develop hypotheses on the semantic relationships between terms : uses derivational morphology ; Grabar and Zweigenbaum ( 2002 ) use , as a starting point , a number of identical characters . Background train Some researchers ( Cucerzan , 2007 ; ) have explored the use of Wikipedia information to improve the disambiguation process . Background train To address this issue , we use a version of the PATB3 training and dev sets manually annotated with functional gender , number , and rationality ( ) .18 This is the first resource providing all three features ( ElixirFm only provides functional number , and to some extent functional gender ) . Uses train One approach to this problem is that taken by the ASCOT project ( Akkerman et al. , 1985 ; ) . Background train Narrative writings or essays are creative works and they generally treat ownership as authorship , even for the most enthusiastic fellows of free culture ( ) . Background train McDonald has even argued for extending the model to a large number of components ( McDonald 1988 ) , and several systems have indeed added an additional component between the planner and the linguistic component ( Meteer 1994 ; Panaget 1994 ; ) . Background train According to current tagger comparisons ( van Halteren et al. , 1998 ; Zavrel and Daelemans , 1999 ) , and according to a comparsion of the results presented here with those in ( ) , the Maximum Entropy framework seems to be the only other approach yielding comparable results to the one presented here . CompareOrContrast train Similar findings have been proposed by that points out V1 and V2 are paired on the basis of their semantic compatibility , which is subject to syntactic constraints . Background train The first lexical substitution method was proposed by . Background train This can be a hazardous affair , since vague expressions tend to be interpreted in different ways by different people ( ) , sometimes in stark contrast with the intention of the speaker/writer ( Berry , Knapp , and Raynor 2002 ) . Background train We see no good reason , however , why such text spans should necessarily be sentences , since the majority of tagging paradigms ( e.g. , Hidden Markov Model [ HMM ] [ Kupiec 1992 ] , Brill 's [ Brill 1995a ] , and MaxEnt [ ] ) do not attempt to parse an entire sentence and operate only in the local window of two to three tokens . CompareOrContrast train ; also reported in Levelt 1989 ) show that greater differences are most likely to be chosen , presumably because they are more striking . Background val Nevertheless , recent results show that knowledge-poor methods perform with amazing accuracy ( cfXXX ( ) , ( Kennedy and Boguraev , 1996 ) ( Kameyama , 1997 ) ) . Background train Notice that it is not possible to use corpus annotation to determine the likelihood of a given property to be chosen , unless we know in advance all of the properties that can be attributed to a given object , as in the case of Jordan 's work on the COCONUT domain ( ) . Background train Some well-known approaches include rule-based models ( Brill and Resnik 1994 ) , backed-off models ( ) , and a maximumentropy model ( Ratnaparkhi 1998 ) . Background train Other psycholing-uistic studies that confirm the validity of paragraph units can be found in Black and Bower ( 1979 ) and . Background train Tetreault 's contribution features comparative evaluation involving the author 's own centering-based pronoun resolution algorithm called the Left-Right Centering algorithm ( LRC ) as well as three other pronoun resolution methods : Hobbs 's naive algorithm ( Hobbs 1978 ) , BFP ( Brennan , Friedman , and Pollard 1987 ) , and Strube 's 5list approach ( ) . Background train investigated three approaches to the automatic generation of response e-mails : text classification , case-based reasoning , and question answering . CompareOrContrast train This contrasts with one of the traditional approaches ( e.g. , ; Watanabe 1995 ) to posing the translation problem , i.e. , the approach in which translation problems are seen in terms of bridging the gap between the most natural monolingual representations underlying the sentences of each language . CompareOrContrast val In practical context , German , English , and Japanese HPSG-based grammars are developed and used in the Verbmobil project ( ) . Background train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( Richardson et al. , 2001 ) , ( Al-Adhaileh & Tang , 1999 ) . Background train While IA is generally thought to be consistent with findings on human language production ( Hermann and Deutsch 1976 ; Levelt 1989 ; Pechmann 1989 ; ) , the hypothesis that incrementality is a good model of human GRE seems unfalsifiable until a preference order is specified for the properties on which it operates . Background train It helps them build complex knowledge bases by combining components : events , entities and modifiers ( ) . Background train Our group has developed a wide-coverage HPSG grammar for Japanese ( ) , which is used in a high-accuracy Japanese dependency analyzer ( Kanayama et al. , 2000 ) . Background train Word frequency counts in internet search engines are inconsistent and unreliable ( ) . Background train Other milestones of recent research include the deployment of probabilistic and machine learning techniques ( Aone and Bennett 1995 ; Kehler 1997 ; Ge , Hale , and Charniak 1998 ; Cardie and Wagstaff 1999 ; the continuing interest in centering , used either in original or in revised form ( Abracos and Lopes 1994 ; Strube and Hahn 1996 ; Hahn and Strube 1997 ; Tetreault 1999 ) ; and proposals related to the evaluation methodology in anaphora resolution ( a , 2001b ) . Background train It is only recently that the web name ambiguity has been approached as a separate problem and defined as an NLP task Web People Search on its own ( Artiles et al. , 2005 ; ) . Background train Actually , if we use LSH technique ( ) in retrieval process , the local method can be easily scaled to a larger training data . Future train Many lexicons , both automatically acquired and manually created , are more fine grained in their approaches to subcategorized clausal arguments , differentiating , for example , between a that-clause and a to + infinitive clause ( ) . Background train Others provide automatic mappings of natural language instructions to executable actions , such as interpreting navigation directions ( ) or robot commands ( Tellex et al. , 2011 ; Matuszek et al. , 2012 ) . Background train Agreement between two annotation sets is calculated here in terms of Cohen 's kappa ( Cohen , 1960 ) 1 and corrected kappa ( ) 2 . Uses train Second , software for utilizing this ontology already exists : MetaMap ( Aronson 2001 ) identifies concepts in free text , and SemRep ( ) extracts relations between the concepts . Background val The implementation has been inspired by experience in extracting information from very large corpora ( Curran and Moens , 2002 ) and performing experiments on maximum entropy sequence tagging ( Curran and Clark , 2003 ; ) . Motivation train Some efforts have tackled tasks such as automatic image caption generation ( Feng and Lapata , 2010a ; ) , text illustration ( Joshi et al. , 2006 ) , or automatic location identification of Twitter users ( Eisenstein et al. , 2010 ; Wing and Baldridge , 2011 ; Roller et al. , 2012 ) . Background train These types of features result in an improvement in both the mention detection and coreference resolution performance , as shown through experiments on the Arabic data . Uses val The use of the web as a corpus for teaching and research on language has been proposed a number of times ( Kilgarriff , 2001 ; ; Rundell , 2000 ; Fletcher , 2001 , 2004b ) and received a special issue of the journal Computational Linguistics ( Kilgarriff and Grefenstette , 2003 ) . Background train But while Bod 's estimator obtains state-of-the-art results on the WSJ , comparable to Charniak ( 2000 ) and , Bonnema et al. 's estimator performs worse and is comparable to Collins ( 1996 ) . Background train The implementation has been inspired by experience in extracting information from very large corpora ( ) and performing experiments on maximum entropy sequence tagging ( Curran and Clark , 2003 ; Clark et al. , 2003 ) . Motivation train Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( Andrews et al. , 2009 ; Steyvers , 2010 ; Feng and Lapata , 2010b ; Bruni et al. , 2011 ; Silberer and Lapata , 2012 ; Johns and Jones , 2012 ; Bruni et al. , 2012a ; Bruni et al. , 2012b ; ) . Background train 14We parse each sentence with the Collins parser ( ) . Uses train Encouraged by the success of chunk-based verb reordering lattices on ArabicEnglish ( ) , we tried to adapt the same approach to the German-English language pair . Motivation train We followed the same experimental procedure as discussed in ( ) for English polymorphemic words . Uses train In our experiments , we employed the well-known classifier SVM `` ght to obtain individual-document classification scores , treating Y as the positive class and using plain unigrams as features .5 Following standard practice in sentiment analysis ( ) , the input to SVM `` ght consisted of normalized presence-of-feature ( rather than frequency-of-feature ) vectors . Uses train The problem of handling ill-formed input has been studied by Carbonell and Hayes ( 1983 ) , Granger ( 1983 ) , Jensen et al. ( 1983 ) , , Riesbeck and Schank ( 1976 ) , Thompson ( 1980 ) , Weischedel and Black ( 1980 ) , and Weischedel and Sondheimer ( 1983 ) . CompareOrContrast train To address this problem , we are currently working on developing a metagrammar in the sense of ( ) . Future train Sentences like 12 , from , are frequently cited . Background train After the PropBank ( ) was built , Xue and Palmer ( 2005 ) and Xue ( 2008 ) have produced more complete and systematic research on Chinese SRL . Background train In order to estimate the parameters of our model , we develop a blocked sampler based on that of to sample parse trees for sentences in the raw training corpus according to their posterior probabilities . Uses train The extraction procedure utilizes a head percolation table as introduced by in combination with a variation of Collins 's ( 1997 ) approach to the differentiation between complement and adjunct . Background train Many other such cases are described in Danlos 's book ( ) . Background train • Before indexing the text , we process it with Textract ( Byrd and Ravin , 1998 ; ) , which performs lemmatization , and discovers proper names and technical terms . Uses train Therefore , we repeated the experiments with POS tags predicted by the MADA toolkit ( ; Habash , Rambow , and Roth 2012 ) 15 ( see Table 2 , 14 Some parsers predict POS tags internally , instead of receiving them as input , but this is not the case in this article . Uses train Experiments on Chinese SRL ( , Xue 2008 ) reassured these findings . Motivation train An example of this is the estimation of maximum entropy models , from simple iterative estimation algorithms used by that converge very slowly , to complex techniques from the optimisation literature that converge much more rapidly ( Malouf , 2002 ) . Background train According to , there are three prevalent approaches for evaluating SR measures : mathematical analysis , applicationspecific evaluation and comparison with human judgments . Background train The choice of learning algorithm for each classifier is motivated by earlier findings showing that discriminative classifiers outperform other machine-learning methods on error correction tasks ( ) . Motivation train A good study comparing document categorization algorithms can be found in ( ) . Background train For HMMs ( footnote 11 ) , Ti is the familiar trellis , and we would like this computation of ti to reduce to the forwardbackward algorithm ( ) . Background val We follow , for compound merging . Uses val Semantic filters can also be used to prevent multiple versions of the same case frame ( ) showing up as complements . Uses val The terms have been identified as the most specific to our corpus by a program developed by and called TER1vloSTAT . Uses train For statistical significance , we use McNemar 's test on non-gold LAS , as implemented by Nilsson and . Uses train The availability of toolkits for this weighted case ( ; van Noord and Gerdemann , 2001 ) promises to unify much of statistical NLP . Background val tionally reconstructed by and Crouch and Putman ( 1994 ) , the context-independent meaning of a sentence is given by one or more QLFs that are built directly from syntactic and semantic rules . CompareOrContrast val We use a standard split of 268 training documents , 68 development documents , and 106 testing documents ( Culotta et al. , 2007 ; ) . Uses train improved on this by marking prepositions with the case they mark ( one of the most important markups in our system ) . CompareOrContrast train But typical OT grammars offer much richer finite-state models of left context ( a ) than provided by the traditional HMM finite-state topologies . Background train Other tools have been designed around particular techniques , such as finite state machines ( Karttunen et al. , 1997 ; ) . Background train This contrasts with the findings described in where significant improvements could be achieved by increasing the number of source languages . CompareOrContrast train The function selects the Value that removes most distractors , but in case of a tie , the least specific contestant is chosen , as long as it is not less specific than the basic-level Value ( i.e. , the most commonly occurring and psychologically most fundamental level , ) . Background train pointed out that distribution plots of judgments for the word pairs used by Rubenstein and Goodenough display an empty horizontal band that could be used to separate related and unrelated pairs . CompareOrContrast train In fact , Reiter has even argued in favor of this approach , claiming that the interactions are sufficiently minor to be ignored ( or at least handled on an ad hoc basis ) ( ) . Background train The method is called targeted self-training as it is similar in vein to self-training ( ) , with the exception that the new parse data is targeted to produce accurate word reorderings . CompareOrContrast train In our previous work ( ) , we started an initial investigation on conversation entailment . Extends train tried to construct a semantic analysis based on `` prepared '' and `` unprepared mind '' . Background train It is also possible to focus on non-compositional compounds , a key point in bilingual applications ( ; Melamed , 1997 ; Lin , 99 ) . Background train This is mainly due to the fact that Arabic is a non-concatenative language ( ) , and that the stem/infix obtained by suppression of infix and prefix add-ons is not the same for words derived from the same origin called the root . Background train ear regression adapted for classification ( ) , which can be described by the following equation : Uses train For example , the forward-backward algorithm ( Baum , 1972 ) trains only Hidden Markov Models , while ( ) trains only stochastic edit distance . Background val Using the tree-cut technique described above , our previous work ( ) extracted systematic polysemy from WordNet . Extends val ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) used maximum likelihood estimation to learn weights for MT. ( Och , 2003 ; Moore and Quirk , 2008 ; Zhao and Chen , 2009 ; ) employed an evaluation metric as a loss function and directly optimized it . CompareOrContrast val 5 Significant bigrams are obtained using the n-gram statistics package NSP ( ) , which offers statistical tests to decide whether to accept or reject the null hypothesis regarding a bigram ( that it is not a collocation ) . Uses val Liu et al. ( 2005 ) , Meral et al. ( 2007 ) , Murphy ( 2001 ) , and Topkara et al. ( 2006a ) all belong to the syntactic transformation category . Background train 2The WePS-1 corpus includes data from the Web03 testbed ( ) which follows similar annotation guidelines , although the number of document per ambiguous name is more variable . Uses train The local training method ( ) is widely employed in computer vision ( Zhang et al. , 2006 ; Cheng et al. , 2010 ) . Background train Recent work ( Banko and Brill , 2001 ; ) has suggested that some tasks will benefit from using significantly more data . Background train Rather than producing a complete analysis of sentences , the alternative is to perform only partial analysis of the syntactic structures in a text ( Harris , 1957 ; ; Greffenstette , 1993 ) . Background train The feasibility of automatically identifying outcome statements in secondary sources has been demonstrated by . CompareOrContrast train The standard approach is to train two models independently and then intersect their predictions ( ) . CompareOrContrast train We use the same method as for generating our multimodal corpora : for each word token in the text corpus , a feature is selected stochastically from the word 's feature distribution , creating a word-feature pair . Uses train MI was also recently used for inference-rule SPs by . Background train Based on this advise ( Moore and ) exclude the latent segmentation variables and opt for a heuristic training procedure . CompareOrContrast train For example , 10 million words of the American National Corpus ( Ide et al. , 2002 ) will have manually corrected POS tags , a tenfold increase over the Penn Treebank ( ) , currently used for training POS taggers . Background train The forward and backward probabilities , p0j and pkn , can be computed using single-source algebraic path for the simpler semiring ( R , + , x , ∗ ) -- or equivalently , by solving a sparse linear system of equations over R , a much-studied problem at O ( n ) space , O ( nm ) time , and faster approximations ( ) . Background train 4To prove ( 1 ) ⇒ ( 3 ) , express f as an FST and apply the well-known Kleene-Sch ¨ utzenberger construction ( ) , taking care to write each regexp in the construction as a constant times a probabilistic regexp . Uses val Second , software for utilizing this ontology already exists : MetaMap ( ) identifies concepts in free text , and SemRep ( Rindflesch and Fiszman 2003 ) extracts relations between the concepts . Background val Our work is more similar to NLG work that concentrates on structural constraints such as generative poetry ( ) ( Colton et al. , 2012 ) ( Jiang and Zhou , 2008 ) or song lyrics ( Wu et al. , 2013 ) ( Ramakrishnan A et al. , 2009 ) , where specified meter or rhyme schemes are enforced . CompareOrContrast train Most DOP models , such as in Bod ( 1993 ) , Goodman ( 1996 ) , , Sima'an ( 2000 ) and Collins & Duffy ( 2002 ) , use a likelihood criterion in defining the best parse tree : they take ( some notion of ) the most likely ( i.e. most probable ) tree as a candidate for the best tree of a sentence . Background train The best performance on the WSJ corpus was achieved by a combination of the SATZ system ( ) with the Alembic system ( Aberdeen et al. 1995 ) : a 0.5 % error rate . CompareOrContrast train The extracted frames are noisy as a result of parser errors and so are filtered using the binomial hypothesis theory ( BHT ) , following . Background train This method allows the efficient retrieval of arbitrary length n-grams ( Nagao and Mori , 94 ; Haruno et al. , 96 ; Ikehaxa et al. , 96 ; ; Russell , 1998 ) . Background train The current system learns finite state flowcharts whereas typical learning systems usually acquire coefficient values as in Minsky and Papert ( 1969 ) , assertional statements as in , or semantic nets as in Winston ( 1975 ) . CompareOrContrast val Our most accurate product model achieves an F score of 92.5 without the use of discriminative reranking and comes close to the best known numbers on this test set ( ) . CompareOrContrast train To demonstrate that this is possible we have implemented a system which constructs dictionary entries for the PATR-II system ( and references therein ) . Uses val Other works ( Kasper et al. , 1995 ; ) convert HPSG grammars into LTAG grammars . CompareOrContrast train The work of demonstrates that faceted queries can be converted into simple filtering constraints to boost precision . Background val Efficient hardware implementation is also possible via chip-level parallelism ( ) . Future train 2We could just as easily use other symmetric `` association '' measures , such as 02 ( Gale & Church , 1991 ) or the Dice coefficient ( ) . CompareOrContrast train have argued that Dale and Reiter 's ( 1995 ) dichotomy between salient and nonsalient objects ( where the objects in the domain are the salient ones ) should be replaced by an account that takes degrees of salience into account : No object can be too unsalient to be referred to , as long as the right properties are available . Background train The acquisition of dialogue as implemented in VNLCE is reminiscent of the program synthesis methodology developed by where program flowcharts were constructed from traces of their behaviors . CompareOrContrast train Character classes , such as punctuation , are defined according to the Unicode Standard ( ) . Uses train Rather than producing a complete analysis of sentences , the alternative is to perform only partial analysis of the syntactic structures in a text ( Harris , 1957 ; Abney , 1991 ; ) . Background train Finally , it has been shown by Groesser ( 1981 ) that the ratio of derived to explicit information necessary for understanding a piece of text is about 8:1 ; furthermore , our reading of the analysis of five paragraphs by strongly suggests that only the most direct or obvious inferences are being made in the process of building a model or constructing a theory of a paragraph . Motivation train In other methods , lexical resources are specifically tailored to meet the requirements of the domain ( Rosario and Hearst , 2001 ) or the system ( ) . Background train The exact form of M ( Si ) need not be discussed at this point ; it could be a conceptual dependence graph ( ) , a deep parse of Si , or some other representation . Background train Some methods are based on likelihood ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) , error rate ( Och , 2003 ; ; Pauls et al. , 2009 ; Galley and Quirk , 2011 ) , margin ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ) and ranking ( Hopkins and May , 2011 ) , and among which minimum error rate training ( MERT ) ( Och , 2003 ) is the most popular one . Background train Similarly , ( ) and ( Shinyanma et al. , 2002 ) learn sentence level paraphrase templates from a corpus of news articles stemming from different news source . Background train This setup is also scalable to a higher number of word pairs ( 350 ) as was shown in . Background train More recent work on terminology structuring has focussed on formal similarity to develop hypotheses on the semantic relationships between terms : Daille ( 2003 ) uses derivational morphology ; use , as a starting point , a number of identical characters . Background train Louwerse et al. ( 2006 ) and Louwerse et al. ( 2007 ) study the relation between eye gaze , facial expression , pauses and dialogue structure in annotated English map-task dialogues ( ) and find correlations between the various modalities both within and across speakers . Background train Some examples include text categorization ( Lewis and Catlett 1994 ) , base noun phrase chunking ( Ngai and Yarowsky 2000 ) , part-of-speech tagging ( Engelson Dagan 1996 ) , spelling confusion set disambiguation ( ) , and word sense disambiguation ( Fujii et al. 1998 ) . Background train It is known that certain cue words and phrases ( ) can serve as explicit indicators of discourse structure . Motivation val The last point may be seen better if we look at some differences between our system and KRYPTON , which also distinguishes between an object theory and background knowledge ( cfXXX ) . CompareOrContrast val One approach to this problem is that taken by the ASCOT project ( ; Akkerman , 1986 ) . Background val Though we could have used a further downstream measure like BLEU , METEOR has also been shown to directly correlate with translation quality ( ) and is simpler to measure . Motivation train For all experiments reported in this section we used the syntactic dependency parser MaltParser v1 .3 ( Nivre 2003 , 2008 ; Kübler , McDonald , and ) , a transition-based parser with an input buffer and a stack , which uses SVM classifiers Uses val We conducted experiments with gold features to assess the potential of these features , and with predicted features , obtained from training a simple maximum likelihood estimation classifier on this resource ( ) .19 The first part of Table 8 shows that the RAT ( rationality ) feature is very relevant ( in gold ) , but suffers from low accuracy ( no gains in machine-predicted input ) . Uses val In our previous work ( ; Salloum and Habash , 2012 ) , we applied our approach to tokenized Arabic and our DA-MSA transfer component used feature transfer rules only . CompareOrContrast train It has already been used to implement a framework for teaching NLP ( ) . Extends train Due to their remarkable ability to incorporate context structure information and long distance reordering into the translation process , tree-based translation models have shown promising progress in improving translation quality ( Liu et al. , 2006 , 2009 ; Quirk et al. , 2005 ; Galley et al. , 2004 , 2006 ; ; Shen et al. , 2008 ; Zhang et al. , 2011b ) . Background train Liu et al. ( 2005 ) , Meral et al. ( 2007 ) , , Murphy and Vogel ( 2007 ) and Topkara et al. ( 2006a ) all belong to the syntactic transformation category . Background train Similarly , the notion of R + M-abduction is spiritually related to the `` abductive inference '' of , the `` diagnosis from first principles '' of Reiter ( 1987 ) , `` explainability '' of Poole ( 1988 ) , and the subset principle of Berwick ( 1986 ) . CompareOrContrast train On the other side , wikis started as collective works where each entry is not owned by a single author e.g. . Background val To address this inconsistency in the correspondence between inflectional features and morphemes , and inspired by Smrž ( 2007 ) , we distinguish between two types of inflectional features : formbased ( a.k.a. surface , or illusory ) features and functional features .6 Most available Arabic NLP tools and resources model morphology using formbased ( `` surface '' ) inflectional features , and do not mark rationality ; this includes the Penn Arabic Treebank ( PATB ) ( ) , the Buckwalter morphological analyzer ( Buckwalter 2004 ) , and tools using them such as the Morphological Analysis and Disambiguation for Arabic ( MADA ) toolkit ( Habash and Rambow 2005 ; Habash , Rambow , and Roth 2012 ) . CompareOrContrast train has built a semantic role classifier exploiting the interdependence of semantic roles . Uses val Others have applied the NLP technologies of near-duplicate detection and topic-based text categorization to politically oriented text ( Yang and Callan , 2005 ; ) . Background val 8 It is based on the dataset of ,9 which consists of 1000 positive and 1000 negative movie reviews , tokenized and divided into 10 folds ( F0 -- F9 ) . Extends train In previous work ( ) , we described an experimental text-to-speech system that determined prosodic phrasing for the Olive -- Liberman synthesizer ( Olive and Liberman 1985 ) . Extends train The UMLS -- the Unified Medical Language System ( UMLS ) has been developed and maintained by National Library of Medicine ( NLM ) . Background train The data used in the experiment was selected from the Penn Treebank Wall Street Journal , and is the same used by . Uses train As shown in this is a well-motivated convention since it avoids splitting up lexical rules to transfer the specifications that must be preserved for different lexical entries . Motivation train The reader may consult recent papers on this subject ( e.g. Moens and Steedman 1987 ; ) to see what a formal interpretation of events in time might look like . Background train Proceedings of EACL '99 example , the ALE parser ( ) presupposes a phrase structure backbone which can be used to determine whether a constraint is to be interpreted bottom-up or topdown . Background train Following Soon et al. ( 2001 ) , we represent use the ACE training data for acquiring our SC clasSCA as a binary value that indicates whether the insifier ; instead , we use the BBN Entity Type Corpus duced SCs of the two NPs involved are the same or ( ) , which consists of not . Uses train Following , we also compare the performance of our system with a system using features based on the Brown clusters of the word types in a document . Uses train use mutual information to identify collocations , a method they claim is reasonably effective for words with a frequency of not less than five . Background train We do this with a first-order HMM part-ofspeech tagger ( Merialdo ) . Uses train Using the bottom-up , dynamic programming technique ( see the appendix for details ) of computing inside probabilities ( ) , we can efficiently compute the probability of the sentence , P ( w | G ) . Uses train As shown in ( ) • The presented research was carried out at the University of Tubingen , Germany , as part of the Sonderforschungsbereich 340 . Background train According to , paragraphs are made up of segments , which in turn are made up of sentences or clauses , which in turn are made up of phrases . Background train Similarly , the notion of R + M-abduction is spiritually related to the `` abductive inference '' of Reggia ( 1985 ) , the `` diagnosis from first principles '' of , `` explainability '' of Poole ( 1988 ) , and the subset principle of Berwick ( 1986 ) . CompareOrContrast train This choice is motivated by an observation we made previously ( a ) : since each post in a sequence is a reply to the preceding post , we could exploit their dependencies by determining their stance labels together .3 As our sequence learner , we employ a maximum entropy Markov model ( MEMM ) ( McCallum et al. , 2000 ) . Extends train Over the past decade , researchers at IBM have developed a series of increasingly sophisticated statistical models for machine translation ( ; Brown et al. , 1990 ; Brown et al. , 1993a ) . Background train This may be because pipelines have many engineering advantages , and in practice the sort of problems pointed out by Danlos and other pipeline critics do not seem to be a major problem in current applied NLG systems ( ) . Background train This approach has its roots in Fillmore 's Case Grammar ( 1968 ) , and serves as the foundation for two current large-scale semantic annotation projects : FrameNet ( Baker et al. , 1998 ) and PropBank ( ) . Background train Two exceptions to this generalisation are the Linguistic String Project ( ) and the IBM CRITIQUE ( formerly EPISTLE ) Project ( Heidorn et al. , 1982 ; Byrd , 1983 ) ; the former employs a dictionary of approximately 10,000 words , most of which are specialist medical terms , the latter has well over 100,000 entries , gathered from machine readable sources . CompareOrContrast val They can also shift the dialogue 2The notion of the initiative in this paper is different from that of the dialogue initiative of . CompareOrContrast train The Google n-gram data was collected by Google Research for statistical language modelling , and has been used for many tasks such as lexical disambiguation ( ) , and contains English n-grams and their observed frequency counts , for counts of at least 40 . Background train Withindocument coreference resolution has been applied to produce summaries of text surrounding occurrences of the name ( Bagga and Baldwin , 1998 ; ) . Background train To represent the semantics of predicative units , we use FrameNet inventory of frames and frame elements ( C. ) . Uses train Similar observation for surface word frequency was also observed by ( Bertram et al. , 2000 ; Bradley , 1980 ; Burani et al. , 1987 ; Burani et al. , 1984 ; Schreuder et al. , 1997 ; Taft 1975 ; ) where it has been claimed that words having low surface frequency tends to decompose . Background train These two sets of data were used for automatic dialogue act classification , which was run in the Weka system ( ) . Uses train • The transition probability a is 0.7 using the EM algorithm ( ) on the TREC4 ad-hoc query set . Uses train obtain promising results in dialogue act tagging of the Switchboard-DAMSL corpus using lexical , syntactic and prosodic cues , while Gravano and Hirschberg ( 2009 ) examine the relation between particular acoustic and prosodic turn-yielding cues and turn taking in a large corpus of task-oriented dialogues . Background val ( ) has found strong correlations between DF , IG and the X2 statistic for a term . Background train Despite this , to date , there has been little work on corpus-based approaches to help-desk response automation ( notable exceptions are Carmel , Shtalhaim , and Soffer 2000 ; ; Bickel and Scheffer 2004 ; Malik , Subramaniam , and Kaushik 2007 ) . Background train To quantify the relative strengths of these transitive inferences , propose to assign a weight to each link . Background train While these approaches have been reasonably successful ( see ) , Kehler et al. ( 2004 ) speculate that deeper linguistic knowledge needs to be made available to resolvers in order to reach the next level of performance . Background train Lisp is not particularly well suited for interfacing to complex , structured objects , and it was not our intention to embark on a major effort involving the development of a formal model of a dictionary ( of the style described in , eg. , ) ; on the other hand a method of access was clearly required , which was flexible enough to support a range of applications intending to make use of the LDOCE tape . Background train We train a 4-gram language model on the Xinhua portion of the English Gigaword corpus using the SRILM Toolkits ( Stolcke , 2002 ) with modified Kneser-Ney smoothing ( ) . Uses val • Learnability ( Zernik and Dyer 1987 ) • Text generation ( ; Milosavljevic , Tulloch , and Dale 1996 ) • Speech generation ( Rayner and Carter 1997 ) • Localization ( Sch ¨ aler 1996 ) Background train Details of the top performing heuristics of COCKTAIL were reported in ( ) . Background train Representative systems are described in , De Mattia and Giachin ( 1989 ) , Niedermair ( 1989 ) , Niemann ( 1990 ) , and Young ( 1989 ) . Background train In addition , note that our Object Raising rule would assign mean to this category incorrectly . CompareOrContrast train The parallel corpus is word-aligned using GIZA + + ( ) . Uses train tions for the remaining 20 % of the instances ; and ( 3 ) train an SVM classifier ( using the LIBSVM package ( ) ) on these 20 % of the instances , where each instance , i , is represented by a set of 31 binary features . Uses val Our algorithm is similar to the approach taken by for inducing PCFG parsers . CompareOrContrast train Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( Neville and Jensen , 2000 ; Lafferty et al. , 2001 ; Getoor et al. , 2002 ; Taskar et al. , 2002 ; Taskar et al. , 2003 ; Taskar et al. , 2004 ; ) . Background val This experiment was again replicated by with 10 subjects . Background train Children use vague adjectives among their first dozens of words ( Peccei 1994 ) and understand some of their intricacies as early as their 24th month ( ) . Background train Other approaches use less deep linguistic resources ( e.g. , POS-tags Stymne ( 2008 ) ) or are ( almost ) knowledge-free ( e.g. , ) . CompareOrContrast train This includes work on question answering ( Wang et al. , 2007 ) , sentiment analysis ( Nakagawa et al. , 2010 ) , MT reordering ( ) , and many other tasks . Background train ( 4 ) NE : We use BBN 's IdentiFinder ( ) , a MUC-style NE recognizer to determine the NE type of NPZ . Uses train has made the first attempt working on the single semantic role level to make further improvement . CompareOrContrast train In particular , since we treat each individual speech within a debate as a single `` document '' , we are considering a version of document-level sentiment-polarity classification , namely , automatically distinguishing between positive and negative documents ( Das and Chen , 2001 ; Pang et al. , 2002 ; ; Dave et al. , 2003 ) . Background train raw length value as a feature , we follow our previous work ( ; Wagner et al. , 2014 ) and create multiple features for length using a decision tree ( J48 ) . Extends train used unification in an SMT system to model some of the CompareOrContrast train Following Ruch et al. ( 2003 ) and , we employed Hidden Markov Models to model the discourse structure of MEDLINE abstracts . Uses train The language grounding problem has received significant attention in recent years , owed in part to the wide availability of data sets ( e.g. Flickr , Von Ahn ( 2006 ) ) , computing power , improved computer vision models ( Oliva and Torralba , 2001 ; Lowe , 2004 ; Farhadi et al. , 2009 ; ) and neurological evidence of ties between the language , perceptual and motor systems in the brain ( Pulverm ¨ uller et al. , 2005 ; Tettamanti et al. , 2005 ; Aziz-Zadeh et al. , 2006 ) . Background train Due to their remarkable ability to incorporate context structure information and long distance reordering into the translation process , tree-based translation models have shown promising progress in improving translation quality ( Liu et al. , 2006 , 2009 ; Quirk et al. , 2005 ; Galley et al. , 2004 , 2006 ; Marcu et al. , 2006 ; ; Zhang et al. , 2011b ) . Background train Previous work has argued that initiative affects the degree of control an agent has in the dialogue interaction ( Whittaker and Stenton , 1988 ; ; Chu-Carroll and Brown , 1998 ) . Background val Many provide graphical user interfaces ( GUI ) for manual annotation ( e.g. General Architecture for Text Engineering ( GATE ) ( Cunningham et al. , 1997 ) and the Alembic Workbench ( ) ) as well as NLP tools and resources that can be manipulated from the GUI . Background train A number of speech understanding systems have been developed during the past fifteen years ( Barnett et al. 1980 , Dixon and Martin 1979 , Erman et al. 1980 , Haton and Pierrel 1976 , Lea 1980 , Lowerre and Reddy 1980 , , Reddy 1976 , Walker 1978 , and Wolf and Woods 1980 ) . CompareOrContrast train It allows the construction of a non-TAL ( Shieber , 1994 ) , ( ) . Background train In addition , the advantages of using linguistically annotated data over raw data are well documented ( Mair , 2005 ; ) . Background val In particular , the `` Semantic Information Retrieval '' project ( SIR ) systematically investigates the use of lexical-semantic relations between words or concepts for improving the performance of information retrieval systems . Motivation val Our re-ranking approach , like the approach to parse re-ranking of , employs a simpler model -- a local semantic role labeling algorithm -- as a first pass to generate a set of n likely complete assignments of labels to all parse tree nodes . CompareOrContrast train For instance , the Alembic workbench ( ) contains a sentence-splitting module that employs over 100 regular-expression rules written in Flex . Background train In contrast , a single statistical model allows one to maintain a single table ( ) . Background train Our experiments are based on the multimodal extension of Latent Dirichlet Allocation developed by . Uses train First , it has been noted that in many natural language applications it is sufficient to use shallow parsing information ; information such as noun phrases ( NPs ) and other syntactic sequences have been found useful in many large-scale language processing applications including information extraction and text summarization ( Grishman , 1995 ; ) . Background train adopted a Bayesian method to infer an STSG by exploring the space of alignments based on parse trees . CompareOrContrast train In corpus linguistics building such megacorpora is beyond the scope of individual researchers , and they are not easily accessible ( : 56 ) unless the web is used as a corpus ( Kilgarriff and Grefenstette , 2003 ) . Background train In addition , there are a number of projects under way to develop substantial lexicons from machine readable sources ( see for details ) . Background train We carried out two parallel experiments with two parsers available for Czech , parser I ( Hajie et al. , 1998 ) and parser II ( ) . Uses val Part of speech taggers typically require input in the format of a single sentence per line ( for example Brill 's tagger ( ) ) and parsers generally aim to produce a tree spanning each sentence . Background train The elimination of redundant nondeterminism is based on Unfold/Fold transformation techniques ( ) .29 The unfolding transformation is also referred to as partial execution , for example , by Pereira and Shieber ( 1987 ) . Uses train Problems such as these have motivated research on more abstract , dependencybased parser evaluation ( e.g. , Lin 1995 ; Carroll , Briscoe , and Sanfilippo 1998 ; Carroll et al. 2002 ; Clark and Hockenmaier 2002 ; ; Preiss 2003 ; Kaplan et al. 2004 ; Miyao and Tsujii 2004 ) . Motivation train This seems to provide additional evidence of b ) 's suggestion that something like a distributional hypothesis of images is plausible . CompareOrContrast train For instance , GATE currently provides a POS tagger , named entity recogniser and gazetteer and ontology editors ( ) . Background train Latent variables we wish to consider are an increased number of word classes ; more flexible regions -- see on learning a state transition diagram for acoustic regions in phone recognition -- and phonological features and syllable boundaries . Background train Since earlier versions of the SNoW based CSCL were used only to identify single phrases ( Punyakanok and Roth , 2001 ; Munoz et al. , 1999 ) and never to identify a collection of several phrases at the same time , as we do here , we also trained and tested it under the exact conditions of CoNLL-2000 ( Tjong Kim ) to compare it to other shallow parsers . Uses train Using the implicit modeling of argument consistency , we follow the same approach as in our previous work ( ) and trained a logistic regression model to predict verb alignment based on the features in Table 1 . Extends train Recent developments in linguistics , and especially on grammatical theory -- for example , Generalised Phrase Structure Grammar ( GPSG ) ( Gazdar et al. , 1985 ) , Lexical Functional Grammar ( LFG ) ( ) -- and on natural language parsing frameworks -- for example , Functional Unification Grammar ( FUG ) ( Kay , 1984a ) , PATR-II ( Shieber , 1984 ) -- make it feasible to consider the implementation of efficient systems for the syntactic analysis of substantial fragments of natural language . Background train We train a 4-gram language model on the Xinhua portion of the English Gigaword corpus using the SRILM Toolkits ( ) with modified Kneser-Ney smoothing ( Chen and Goodman , 1998 ) . Uses train It is only recently that the web name ambiguity has been approached as a separate problem and defined as an NLP task Web People Search on its own ( ; Artiles et al. , 2007 ) . Background train This deficiency is rectified in the verb classification system employed by in the Brandeis verb catalogue . CompareOrContrast train Previous work has argued that initiative affects the degree of control an agent has in the dialogue interaction ( ; Walker and Whittaker , 1990 ; Chu-Carroll and Brown , 1998 ) . Background val 1990 ) , on linguisitic acquisition ( by the use of Part-of-Speech filters hand-crafted by a linguist ) ( ) or , more frequently , on a combination of the two ( Smadja , 1993 ; Kilgarriff and Tugwell , 2001 , for example ) . CompareOrContrast val It is frequently used in tasks like scene identification , and shows that distance in GIST space correlates well with semantic distance in WordNet . Motivation train Because each rule r consists of a target tree fragment frag and a source string str in the model , we follow and decompose the prior probability P0 ( r | N ) into two factors as follows : Uses train To address this inconsistency in the correspondence between inflectional features and morphemes , and inspired by Smrž ( 2007 ) , we distinguish between two types of inflectional features : formbased ( a.k.a. surface , or illusory ) features and functional features .6 Most available Arabic NLP tools and resources model morphology using formbased ( `` surface '' ) inflectional features , and do not mark rationality ; this includes the Penn Arabic Treebank ( PATB ) ( Maamouri et al. 2004 ) , the Buckwalter morphological analyzer ( Buckwalter 2004 ) , and tools using them such as the Morphological Analysis and Disambiguation for Arabic ( MADA ) toolkit ( ; Habash , Rambow , and Roth 2012 ) . CompareOrContrast train In our prior work ( ) , we examined whether techniques used for predicting the helpfulness of product reviews ( Kim et al. , 2006 ) could be tailored to our peer-review domain , where the definition of helpfulness is largely influenced by the educational context of peer review . Extends train Previous sentiment-analysis work in different domains has considered inter-document similarity ( ; Pang and Lee , 2005 ; Goldberg and Zhu , 2006 ) or explicit Background train a ) show how a BoVW model may be easily combined with a distributional vector space model of language using only vector concatenation . Background train The following are the various tag sets we use in this article : ( a ) the core POS tag sets CORE44 and the newly introduced CORE12 ; ( b ) CATiB Treebank tag set ( CATIB6 ) ( ) and its newly introduced extension of CATIBEX created using simple regular expressions on word form , indicating particular morphemes such as the prefix JI Al + or the suffix v ' + wn ; this tag set is the best-performing tag set for Arabic on predicted values as reported in Section 4 ; ( c ) the PATB full tag set with complete morphological tag ( BW ) ( Buckwalter 2004 ) ; and two extensions of the PATB reduced tag set ( PENN POS , a.k.a. RTS , size 24 [ Diab , Hacioglu , and Jurafsky 2004 ] ) , both outperforming it : ( d ) Kulick , Gabbard , and Marcus ( 2006 ) 's tag set ( KULICK ) , size 43 , one of whose most important extensions is the marking of the definite article clitic , and ( e ) Diab and Benajiba 's ( in preparation ) EXTENDED RTS tag set ( ERTS ) , which marks gender , number , and definiteness , size 134 . Uses val The numeral ( whether it is implicit , as in ( 3 ) , or explicit ) can be construed as allowing the reader to draw inferences about the standards employed ( ; DeVault and Stone 2004 ) : ( 3 ) , for example , implies a standard that counts 10 cm as large and 8 cm as not large . Background train Agreement between two annotation sets is calculated here in terms of Cohen 's kappa ( ) 1 and corrected kappa ( Brennan and Prediger , 1981 ) 2 . Uses train On the WSJ corpus our system performed slightly better than the combination of the Alembic and SATZ systems described in ( 0.44 % vs. 0.5 % error rate ) . CompareOrContrast train studied the issue of disambiguation for mono-lingual M. Background train Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( Neville and Jensen , 2000 ; ; Getoor et al. , 2002 ; Taskar et al. , 2002 ; Taskar et al. , 2003 ; Taskar et al. , 2004 ; McCallum and Wellner , 2004 ) . Background train To address this inconsistency in the correspondence between inflectional features and morphemes , and inspired by Smrž ( 2007 ) , we distinguish between two types of inflectional features : formbased ( a.k.a. surface , or illusory ) features and functional features .6 Most available Arabic NLP tools and resources model morphology using formbased ( `` surface '' ) inflectional features , and do not mark rationality ; this includes the Penn Arabic Treebank ( PATB ) ( Maamouri et al. 2004 ) , the Buckwalter morphological analyzer ( ) , and tools using them such as the Morphological Analysis and Disambiguation for Arabic ( MADA ) toolkit ( Habash and Rambow 2005 ; Habash , Rambow , and Roth 2012 ) . CompareOrContrast val Although in this paper we take modus ponens as the main rule of inference , in general one can consider deductive closures with respect to weaker , nonstandard logics , ( cfXXX Levesque 1984 ; ; Patel-Schneider 1985 ) . CompareOrContrast train Riehemann 1993 ; Oliva 1994 ; Frank 1994 ; ; Sanfilippo 1995 ) . CompareOrContrast train Berger et al. 2000 ; Jijkoun and de Rijke 2005 ; ) . CompareOrContrast train We will examine the worst-case complexity of interpretation as well as generation to shed some light on the hypothesis that vague descriptions are more difficult to process than others because they involve a comparison between objects ( , Krahmer and Theune 2002 ) . Background train A number of proposals in the 1990s deliberately limited the extent to which they relied on domain and/or linguistic knowledge and reported promising results in knowledge-poor operational environments ( Dagan and Itai 1990 , 1991 ; Lappin and Leass 1994 ; Nasukawa 1994 ; ; Williams , Harvey , and Preston 1996 ; Baldwin 1997 ; Mitkov 1996 , 1998b ) . Background train From the Meaning-Text Theory ( MTT ) 1 point of view , Natural Language ( NL ) is considered as a correspondence between meanings and texts ( ) . Background train This means that natural language expressions such as `` A is B , '' `` A is the same as B , '' etc. are not directly represented by logical equality ; similarly , `` not '' is often not treated as logical negation ; cfXXX . CompareOrContrast train The full-listing model claims that polymorphic words are represented as a whole in the human mental lexicon ( ; Butterworth , 1983 ) . Background train As for work on Arabic ( MSA ) , results have been reported on the PATB ( Kulick , Gabbard , and Marcus 2006 ; Diab 2007 ; Green and Manning 2010 ) , the Prague Dependency Treebank ( PADT ) ( ; Nivre 2008 ) and the CATiB ( Habash and Roth 2009 ) . Background train However , learning-based resolvers have not been able to benefit from having an SC agreement feature , presumably because the method used to compute the SC of an NP is too simplistic : while the SC of a proper name is computed fairly accurately using a named entity ( NE ) recognizer , many resolvers simply assign to a common noun the first ( i.e. , most frequent ) WordNet sense as its SC ( e.g. , Soon et al. ( 2001 ) , ) . Background train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train The ability to explicitly identify these sections in unstructured text could play an important role in applications such as document summarization ( ) , information retrieval ( Tbahriti et al. , 2005 ) , information extraction ( Mizuta et al. , 2005 ) , and question answering . Background train Many statistical parsers ( Ratnaparkhi , 1999 ; Collins , 1999 ; Charniak , 2001 ) are based on a history-based probability model ( ) , where the probability of each decision in a parse is conditioned on the previous decisions in the parse . Background train For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( Meyers et al. , 2000 ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( ) , ( Sato , 1991 ) , ( Richardson et al. , 2001 ) , ( Al-Adhaileh & Tang , 1999 ) . Background train For instance , ( ) acquire two-argument templates ( inference rules ) from corpora using an extended version of the distributional analysis in which paths in dependency trees that have similar arguments are taken to be close in meaning . Background train Per-state joint normalization ( b , § 8.2 ) is similar but drops the dependence on a . CompareOrContrast train Other studies on the value of disambiguation for cross-lingual IR include Hiemstra and de Jong , 1999 ; . Background train We have not yet made use of TINA 'S probabilities in adjusting the recognizer scores on the fly , but we have been able to incorporate linguistic scores to resort N-best outputs , giving a significant improvement in performance ( ) . Uses train These features are carefully designed to reduce the data sparseness problem and some of them are inspired by previous work ( He et al. , 2008 ; Gimpel and Smith , 2008 ; ; Chiang et al. , 2009 ; Setiawan et al. , 2009 ; Shen et al. , 2009 ; Xiong et al. , 2009 ) : 1 . Motivation train Nevertheless , the full document text is present in most systems , sometimes as the only feature ( Sugiyama and Okumura , 2007 ) and sometimes in combination with others see for instance ( ; Popescu and Magnini , 2007 ) - . Background train More recently , ( ) has performed a good survey of document categorization ; recent works can also be found in ( Joachims , 2002 ) , ( Crammer and Singer , 2003 ) , and ( Lewis et al. , 2004 ) . Background train McDonald has even argued for extending the model to a large number of components ( McDonald 1988 ) , and several systems have indeed added an additional component between the planner and the linguistic component ( ; Panaget 1994 ; Wanner 1994 ) . Background train It is these orthographic variations and complex morphological structure that make Arabic language processing challenging ( Xu et al. , 2001 ; ) . Background train A number of speech understanding systems have been developed during the past fifteen years ( Barnett et al. 1980 , Dixon and Martin 1979 , , Haton and Pierrel 1976 , Lea 1980 , Lowerre and Reddy 1980 , Medress 1980 , Reddy 1976 , Walker 1978 , and Wolf and Woods 1980 ) . CompareOrContrast train It has been more difficult showing that agreement morphology helps parsing , however , with negative results for dependency parsing in several languages ( Eryigit , Nivre , and Oflazer 2008 ; Nivre , Boguslavsky , and Iomdin 2008 ; ) . Motivation train Some efforts have tackled tasks such as automatic image caption generation ( a ; Ordonez et al. , 2011 ) , text illustration ( Joshi et al. , 2006 ) , or automatic location identification of Twitter users ( Eisenstein et al. , 2010 ; Wing and Baldridge , 2011 ; Roller et al. , 2012 ) . Background train This idea was proposed by Krauwer and des Tombe ( 1981 ) , Langendoen and Langsam ( 1987 ) , and Pulman ( 1986 ) , and was rediscovered by Black ( 1989 ) and recently by . Background train A nonprobabilistic approach for DA labeling proposed by Samuel , Carberry , and Vijay-Shanker ( 1998 ) is transformation-based learning ( ) . CompareOrContrast train However , since work in this direction has started , a significant progress has also been made in the research on statistical learning of full parsers , both in terms of accuracy and processing time ( Charniak , 1997b ; Charniak , 1997a ; Collins , 1997 ; ) . Background val Some methods are based on likelihood ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) , error rate ( Och , 2003 ; Zhao and Chen , 2009 ; Pauls et al. , 2009 ; ) , margin ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ) and ranking ( Hopkins and May , 2011 ) , and among which minimum error rate training ( MERT ) ( Och , 2003 ) is the most popular one . Background train The algorithm we implemented is inspired by the work of on word sense disambiguation . Motivation val Various approaches for computing semantic relatedness of words or concepts have been proposed , e.g. dictionary-based ( Lesk , 1986 ) , ontology-based ( Wu and Palmer , 1994 ; Leacock and Chodorow , 1998 ) , information-based ( Resnik , 1995 ; Jiang and Conrath , 1997 ) or distributional ( ) . Background train EM maximizes G ( 0 ) via block-coordinate ascent on a lower bound F ( q , 0 ) using an auxiliary distribution over the latent variables q ( z | x , y ) ( ) : Uses train It has been argued that generating such modifiers is not a trivial decision because it interferes with the planning of both local and global coherence ( in the sense of ( Grosz and Sidner , 1986 ) ) ( a ) . Background train • Only qualitative observations of the responses were reported ( no formal evaluation was performed ) ( ; Roy and Subramaniam 2006 ) . CompareOrContrast train A stops B from doing something ; A disagreees with B on something , 8 % and 12 % ) Note that in our original work ( ) , only development data were used to show some initial observations . CompareOrContrast train We evaluated on the English CCGBank ( Hockenmaier and Steedman , 2007 ) , which is a transformation of the Penn Treebank ( ) ; the CTBCCG ( Tse and Curran , 2010 ) transformation of the Penn Chinese Treebank ( Xue et al. , 2005 ) ; and the CCG-TUT corpus ( Bos et al. , 2009 ) , built from the TUT corpus of Italian text ( Bosco et al. , 2000 ) . Uses train The EM algorithm ( ) can maximize these functions . Uses train Finally , feedback expressions ( head nods and shakes ) are successfully predicted from speech , prosody and eye gaze in interaction with Embodied Communication Agents as well as human communication ( Fujie et al. , 2004 ; Morency et al. , 2005 ; ; Morency et al. , 2009 ) . Background train For these or for a specific domain , basic synonymic dictionaries can be complemented using learning methods based on distributional similarity ( Pereira et al. , 1993 ; ) . Future train For projective parsing , it is significantly faster than exact dynamic programming , at the cost of small amounts of search error , We are interested in extending these ideas to phrase-structure and lattice parsing , and in trying other higher-order features , such as those used in parse reranking ( Charniak and Johnson , 2005 ; Huang , 2008 ) and history-based parsing ( ) . Future train The coreference system system is similar to the Bell tree algorithm as described by ( ) . CompareOrContrast train Other representations use the link structure ( Malin , 2005 ) or generate graph representations of the extracted features ( ) . Background train The research described below is taking place in the context of three collaborative projects ( Boguraev , 1987 ; ; Phillips and Thompson , 1986 ) to develop a general-purpose , wide coverage morphological and syntactic analyser for English . Background train We experiment with four learners commonly employed in language learning : Decision List ( DL ) : We use the DL learner as described in , motivated by its success in the related tasks of word sense disambiguation ( Yarowsky , 1995 ) and NE classification ( Collins and Singer , 1999 ) . Uses train transition-based dependency parsing framework ( Nivre , 2008 ) using an arc-eager transition strategy and are trained using the perceptron algorithm as in with a beam size of 8 . Uses train Using the basic solution proposed by ( ) as a term of comparison , we experiment with different sources of multilingual lexical knowledge to address the following questions : ( 1 ) What is the potential of the existing multilingual lexical resources to approach CLTE ? CompareOrContrast train Acoustic models for HTK is trained with the continuous speech database of the Acoustical Society of Japan ( ) . Uses train Automatic text categorization has been used in search engines , digital library systems , and document management systems ( ) . Background train For our Text modality , we use deWaC , a large German web corpus created by the WaCKy group ( ) containing approximately 1.7 B word tokens . Uses train Unless very high rates of misspellings are to be expected ( this explains the favorable results for trigram indexing in ( ) ) one can not really recommend this method . CompareOrContrast val For english , there is for instance the 15 year old HewlettPackard test suite , a simple text file listing test sentences and grouping them according to linguistics phenomena ( ) ; and more recently , the much more sophisticated TSNLP ( Test Suite for Natural Language Processing ) which includes some 9500 test items for English , French and German , each of them being annotated with syntactic and application related information ( Oepen and Flickinger , 1998 ) . CompareOrContrast val Previous work on Chinese SRL mainly focused on how to transplant the machine learning methods which has been successful with English , such as , Xue and Palmer ( 2005 ) and Xue ( 2008 ) . Background train A similar method is included in PATR-II ( ) and can be used to encode lexical rules as binary relations in the CUF system ( Dorre and Eisele 1991 ; Done and Dorna 1993b ) or the TFS system ( Emele and Zajac 1990 ; Emele 1994 ) . CompareOrContrast train Our training examples are similar to the data created for pseudodisambiguation , the usual evaluation task for SP models ( Erk , 2007 ; ; Rooth et al. , 1999 ) . CompareOrContrast train With the use of computers in storing the explosive amount of biological information , natural language processing ( NLP ) approaches have been explored to make the task of managing information recorded in free text more feasible . Background val A formula q5 of L ( =-RRB- , the language with equality , is weakly R + M-abductible from an object theory T , denoted by T I-R + m 0 , iff there exists a partial theory T e PT ( T ) and a preferred model M E PM ( T ) such that M = 0 , i.e. 0 is true in at least one preferred model of the partial theory T. Note : The notions of strong provability and strong R + M-abduction can be introduced by replacing `` there exists '' by `` all '' in the above definitions ( cfXXX b ) . CompareOrContrast train The first version ( TIMIT ) was developed for the 450 phonetically rich sentences of the TIMIT database ( ) . Uses val A third problem arises with the approach to the semantics of QLFs that this notion of the relationship between QLF and RQLF encourages one to adopt : it is that taken by . CompareOrContrast train In ( ) , I present evidence from Mandarin Chinese that this analysis is on the right track . Extends train Interaction between components is coordinated by the dialogue manager which uses the informationstate approach ( ) . Uses val Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) [ 18 ] , fly FlyBase [ 19 ] , yeast Saccharomyces Genome Database ( SGD ) [ 20 ] , rat -- Rat Genome Database ( RGD ) [ 21 ] , worm -- WormBase [ 22 ] , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) , and Enzyme Nomenclature Database ( ECNUM ) [ 25 , 26 ] . Uses train There are many more distinctions which are conveyed by the conjunction of grammar codes and word qualifiers ( see , for further details ) . Background train The most common way is to divide each half of the bitext into an equal number of segments and to align the segments so that each pair of segments Si and Ti are translations of each other ( Gale & Church , 1991 ; a ) . Background val ) . Future train proposes readjustment rules similar to those of Chomsky and Halle , but he claims that the readjustment of structure is part of the grammar , not part of the performance model . Background train financial news , we created a probabilistic CzechEnglish dictionary by running GIZA + + training ( translation models 1-4 , see ) on the training part of the English-Czech WSJ parallel corpus extended by the parallel corpus of entry/translation pairs from the manual dictionary . Uses train In the field of machine learning research , incremental training has been employed in the work ( ; Shilton et al. , 2005 ) , but there is little work for tuning parameters of statistical machine translation . Background train A similar method is included in PATR-II ( Shieber et al. 1983 ) and can be used to encode lexical rules as binary relations in the CUF system ( Dorre and Eisele 1991 ; Done and Dorna 1993b ) or the TFS system ( Emele and Zajac 1990 ; ) . CompareOrContrast train In a log-linear parameterization , for example , a prior that penalizes feature strengths far from 1 can be used to do feature selection and avoid overfitting ( ) . Uses train We report performance in terms of two metrics : ( 1 ) the Fmeasure score as computed by the commonly-used MUC scorer ( ) , and ( 2 ) the accuracy on the anaphoric references , computed as the fraction of anaphoric references correctly resolved . Uses train A study of the query log of the AllTheWeb and Altavista search sites gives an idea of the relevance of the people search task : 11-17 % of the queries were composed of a person name with additional terms and 4 % were identified as person names ( ) . Background val This was done because purely unsupervised techniques ( e.g. , Baum-Welch [ Baum and Petrie 1966 ] or Brill 's [ b ] ) enable regularities to be induced for word classes which contain many entries , exploiting the fact that individual words that belong to a POS class occur in different ambiguity patterns . CompareOrContrast train Future research should apply the work of and Blunsom and Osborne ( 2008 ) , who marginalize over derivations to find the most probable translation rather than the most probable derivation , to these multi-nonterminal grammars . Future train In informal experiments described elsewhere ( ) , I found that the G2 statistic suggested by Dunning ( 1993 ) slightly outperforms 02 . Background train porating these two KSs into our resolver : they can Following , we select as the aneach be represented as a constraint or as a feature , tecedent of each NP , NPS , the closest preceding NP and they can be applied to the resolver in isolation that is classified as coreferent with NPS . Motivation train Using the section labels , the HMM was trained using the HTK toolkit ( ) , which efficiently performs the forward-backward algorithm and BaumWelch estimation . Uses train In this paper , we use TF-IDF ( a kind of augmented DF ) as a feature selection criterion , in order to ensure results are comparable with those in ( ) . CompareOrContrast train A variety of statistical methods were proposed over the recent years for learning to produce a full parse of free-text sentences ( e.g. , Bod ( 1992 ) , Magerman ( 1995 ) , Collins ( 1997 ) , , and Sekine ( 1998 ) ) . Background train There is a rich literature on organization and lexical access of morphologically complex words where experiments have been conducted mainly for derivational suffixed words of English , Hebrew , Italian , French , Dutch , and few other languages ( Marslen-Wilson et al. , 2008 ; Frost et al. , 1997 ; ; Drews and Zwitserlood , 1995 ) . Background train In the disambiguation of capitalized words , the most widespread method is POS tagging , which achieves about a 3 % error rate on the Brown corpus and a 5 % error rate on the WSJ corpus , as reported in . CompareOrContrast train The problem with this approach is that any threshold is , to some extent , arbitrary , and there is evidence to suggest that , for some tasks , low counts are important ( ) . Motivation train further labeled the SCFG rules with POS tags and unsupervised word classes . CompareOrContrast train Steganography is concerned with hiding information in some cover medium , by manipulating properties of the medium in such a way that the hidden information is not easily detectable by an observer ( ) . Background val We offer a theorem that highlights the broad applicability of these modeling techniques .4 If f ( input , output ) is a weighted regular relation , then the following statements are equivalent : ( 1 ) f is a joint probabilistic relation ; ( 2 ) f can be computed by a Markovian FST that halts with probability 1 ; ( 3 ) f can be expressed as a probabilistic regexp , i.e. , a regexp built up from atomic expressions a : b ( for a E E U -LCB- E -RCB- , b E A U -LCB- E -RCB- ) using concatenation , probabilistic union + p , and probabilistic closure * p. For defining conditional relations , a good regexp language is unknown to us , but they can be defined in several other ways : ( 1 ) via FSTs as in Fig. 1c , ( 2 ) by compilation of weighted rewrite rules ( Mohri and Sproat , 1996 ) , ( 3 ) by compilation of decision trees ( ) , ( 4 ) as a relation that performs contextual left-to-right replacement of input substrings by a smaller conditional relation ( Gerdemann and van Noord , 1999 ) ,5 ( 5 ) by conditionalization of a joint relation as discussed below . Background train Word pairs containing polysemous words are expanded to concept pairs using GermaNet ( ) , the German equivalent to WordNet , as a sense inventory for each word . Uses train Using the initial target U-trees , source sentences and word alignment , we extract minimal GHKM translation rules7 in terms of frontier nodes ( ) . Uses train For example , while it is difficult to induce a grammar with raw text alone , the task is tractable when the syntactic analysis for each sentence is provided as a part of the training data ( ) . Background train Our results are lower than those of full parsers , e.g. , as might be expected since much less structural data , and no lexical data are being used . CompareOrContrast train Our work is more similar to NLG work that concentrates on structural constraints such as generative poetry ( Greene et al. , 2010 ) ( Colton et al. , 2012 ) ( Jiang and Zhou , 2008 ) or song lyrics ( Wu et al. , 2013 ) ( Ramakrishnan ) , where specified meter or rhyme schemes are enforced . CompareOrContrast val Hermann and Deutsch ( 1976 ; also reported in ) show that greater differences are most likely to be chosen , presumably because they are more striking . Background train The result holds for both the MaltParser ( ) and the Easy-First Parser ( Goldberg and Elhadad 2010 ) . Uses train This system has been successfully tested with the development of plug-ins supporting instant messaging , distributed video encoding ( ) , distributed virtual worlds ( Hughes et al. , 2005 ) and digital library management ( Walkerdine and Rayson , 2004 ) . Background train In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ ; Bresnan 2001 ; Dalrymple 2001 ] , head-driven phrase structure grammar [ HPSG ] [ Pollard and Sag 1994 ] , tree-adjoining grammar [ TAG ] [ Joshi 1988 ] , and combinatory categorial grammar [ CCG ] [ Ades and Steedman 1982 ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Uses train We then use Illinois Chunker ( Punyakanok and Roth , 2001 ) 6 to extract more noun phrases from the text and employ Collins head rules ( ) to identify their heads . Uses train argues CV formations in Hindi and Urdu are either morphological or syntactical and their formation take place at the argument structure . Background train feature Cohen 's k corrected k agreement 73.59 98.74 dial act 84.53 98.87 turn 73.52 99.16 Table 2 : Inter-coder agreement on feedback expression annotation Although researchers do not totally agree on how to measure agreement in various types of annotated data and on how to interpret the resulting figures , see , it is usually assumed that Cohen 's kappa figures over 60 are good while those over 75 are excellent ( Fleiss , 1971 ) . Background val Collins 1996 , Charniak 1997 , Collins 1999 and ) . CompareOrContrast train Two formalizations of lexical rules as used by HPSG linguists have been proposed , the meta-level lexical rules ( MLRs ; Calcagno 1995 ; ) and the . Background train We run TreeTagger ( Schmid , 1994 ) for tokenization , and used the Giza + + ( ) to align the tokenized corpora at the word level . Uses train Other milestones of recent research include the deployment of probabilistic and machine learning techniques ( Aone and Bennett 1995 ; Kehler 1997 ; Ge , Hale , and Charniak 1998 ; Cardie and Wagstaff 1999 ; the continuing interest in centering , used either in original or in revised form ( Abracos and Lopes 1994 ; ; Hahn and Strube 1997 ; Tetreault 1999 ) ; and proposals related to the evaluation methodology in anaphora resolution ( Mitkov 1998a , 2001b ) . Background train We then go on to compare the current approach with that of some other theories with similar aims : the `` standard '' version of quasi-logical form implemented in the Core Language Engine , as rationally reconstructed by and Crouch and Pulman ( 1994 ) ; underspecified Discourse Representation Theory ( Reyle 1993 ) ; and the `` glue language '' approach of Dalrymple et al. ( 1996 ) . CompareOrContrast train Note that this ensures that greater importance is attributed to longer chunks , as is usual in most EBMT systems ( cfXXX ; Veale and Way 1997 ; Carl 1999 ) .7 As an example , consider the translation into French of the house collapsed . Background train Using WordNet , annotating the sem feature of an adjective involves first choosing the correct sense for the adjective 2Some descriptions of int modifiers can be found in ( b ) . Background val One common approach is using Machine Translation ( MT ) to translate the queries to the language of the documents or translate documents to the language of the queries ( Gey et al , 1999 ; ) . CompareOrContrast train Adding selectional restrictions ( semantic feature information , ) does not solve the problem , because isolated features offer only part of the background knowledge necessary for reference disambiguation . Background val Software engineering research on Generative Programming ( ) attempts to solve these problems by focusing on the development of configurable elementary components and knowledge to combine these components into complete systems . Background train Both use the evaluation software and triple encoding presented in . Uses val In this Section , we will describe some example cases , which are drawn from the problem of using synchronous formalisms to define translations between languages ( e.g. cases ) . Background train Some examples include text categorization ( Lewis and Catlett 1994 ) , base noun phrase chunking ( ) , part-of-speech tagging ( Engelson Dagan 1996 ) , spelling confusion set disambiguation ( Banko and Brill 2001 ) , and word sense disambiguation ( Fujii et al. 1998 ) . Background train Most coreference resolution work simply mentions it in passing as a module in the pipelined system ( ; Durrett and Klein , 2013 ; Lee et al. , 2011 ; Bj ¨ orkelund and Kuhn , 2014 ) . Background train Because the judges do not evaluate the same cases , we could not employ standard inter-annotator agreement measures ( ) . Uses train Related are also the studies by Rieks op den Akker and Schulz ( 2008 ) and : both achieve promising results in the automatic segmentation of dialogue acts using the annotations in a large multimodal corpus . Background train Due to advances in statistical syntactic parsing techniques ( Collins , 1997 ; ) , attention has recently shifted towards the harder question of analyzing the meaning of natural language sentences . Background train The reader may consult recent papers on this subject ( e.g. ; Webber 1987 ) to see what a formal interpretation of events in time might look like . Background train Because it is also consistent , it will be chosen as a best interpretation of S , ( cfXXX a , 1987b ) . Background val For our experiments we used the standard division of the WSJ ( ) , with sections 2 through 21 for training ( approx . Uses train With respect to this , we apply the different priming and other lexical decision experiments , described in literature ( ; Bentin , S. and Feldman , 1990 ) specifically for derivationally suffixed polymorphemic words and compound verbs of Bangla . Uses train From this description , it should be clear that TM systems do not translate : Indeed , some researchers consider them to be little more than a search-and-replace engine , albeit a rather sophisticated one ( ) . Background train Other milestones of recent research include the deployment of probabilistic and machine learning techniques ( Aone and Bennett 1995 ; ; Ge , Hale , and Charniak 1998 ; Cardie and Wagstaff 1999 ; the continuing interest in centering , used either in original or in revised form ( Abracos and Lopes 1994 ; Strube and Hahn 1996 ; Hahn and Strube 1997 ; Tetreault 1999 ) ; and proposals related to the evaluation methodology in anaphora resolution ( Mitkov 1998a , 2001b ) . Background val Our group has developed a wide-coverage HPSG grammar for Japanese ( Mitsuishi et al. , 1998 ) , which is used in a high-accuracy Japanese dependency analyzer ( ) . Background train There are several variations of such a method ( Ballesteros and Croft , 1998 ; ; Hull 1997 ) . CompareOrContrast val This evaluation set-up is an improvement versus the one we previously reported ( ) , in which fixed partitions were used for training , development , and testing . Extends train compared the performace of NEs versus BoW features . Background train The right-side context of a non-terminal category -- the probability of generating a category to the right of the current constituent 's category -- corresponds directly to the category transitions used for the HMM supertagger of . CompareOrContrast train The starting point for the approach followed here was a dissatisfaction with certain aspects of the theory of quasi-logical form as described in , 1992 ) , and implemented in SRI 's Core Language Engine ( CLE ) . CompareOrContrast train However , the method we are currently using in the ATIS domain ( ) represents our most promising approach to this problem . Future train Three UniRef tables UniRef100 , UniRef90 and UniRef50 ) are available for download : UniRef100 combines identical sequences and sub-fragments into a single UniRef entry ; and UniRef90 and UniRef50 are built by clustering UniRef100 sequences into clusters based on the CD-HIT algorithm such that each cluster is composed of sequences that have at least 90 % or 50 % sequence similarity , respectively , to the representative sequence . Uses train , Charniak 1997 , Collins 1999 and Charniak 2000 ) . CompareOrContrast train To name a few examples , Rohrbach et al. ( 2010 ) and Socher et al. ( 2013 ) show how semantic information from text can be used to improve zero-shot classification ( i.e. , classifying never-before-seen objects ) , and show that verb clusters can be used to improve activity recognition in videos . Background train The computational treatment of lexical rules as covariation in lexical entries was implemented in Prolog by the authors in cooperation with Dieter Martini for the ConTroll system ( Gerdemann and ; Gotz and Meurers 1997a ) . Uses train Arabic has two kinds of plurals : broken plurals and sound plurals ( ; Chen and Gey , 2002 ) . Background train Although there are other discussions of the paragraph as a central element of discourse ( e.g. Chafe 1979 , Halliday and Hasan 1976 , , Haberlandt et al. 1980 ) , all of them share a certain limitation in their formal techniques for analyzing paragraph structure . CompareOrContrast train To prove that our method is effective , we also make a comparison between the performances of our system and Xue and Palmer ( 2005 ) , . CompareOrContrast train But while Bod 's estimator obtains state-of-the-art results on the WSJ , comparable to and Collins ( 2000 ) , Bonnema et al. 's estimator performs worse and is comparable to Collins ( 1996 ) . Background train Our plan is to implement a windowed or moving-average version of BLEU as in ( ) . Future train As rightly pointed out , however , `` Proper nouns and capitalized words are particularly problematic : some capitalized words are proper nouns and some are not . CompareOrContrast train The recent great advances in speech and language technologies have made it possible to build fully implemented spoken dialogue systems ( Aust et al. , 1995 ; Allen et al. , 1996 ; Zue et al. , 2000 ; ) . Background train For instance , part of the ACE Phase 2 also adopted a corpus-based approach to SC deterevaluation involves classifying an NP as PERSON , mination that is investigated as part of the mention ORGANIZATION , GPE ( a geographical-political redetection ( MD ) task ( e.g. , ) . CompareOrContrast train In most cases , the accuracy of parsers degrades when run on out-of-domain data ( Gildea , 2001 ; McClosky et al. , 2006 ; ; Petrov et al. , 2010 ) . Background train Based on this assumption , the problem of identifying mention heads is a sequential phrase identification problem , and we choose to employ the BILOU-representation as it has advantages over traditional BIO-representation , as shown , e.g. in . Motivation val TF is given by TFD , t , and it denotes frequency of term t in document D. IDF is given by IDFt = log ( N/dft ) , where N is the number of documents in the collection , and dft is the number of documents containing the term t. ( ) proposed the combination of TF and IDF as weighting schemes , and it has been shown that their product gave better performance . Motivation train An HPSG grammar consists of lexical entries and ID grammar rules , each of which is described with typed feature structures ( ) . Background train More recently , Burke , Cahill , et al. ( 2004a ) carried out an evaluation of the automatic annotation algorithm against the publicly available PARC 700 Dependency Bank ( ) , a set of 700 randomly selected sentences from Section 23 which have been parsed , converted to dependency format , and manually corrected and extended by human validators . Background train In our work , we gather sets of sentences , and assume ( but do not employ ) existing approaches for their organization ( ; Barzilay , Elhadad , and McKeown 2001 ; Barzilay and McKeown 2005 ) . Background train For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( Meyers et al. , 2000 ) , ( ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( Richardson et al. , 2001 ) , ( Al-Adhaileh & Tang , 1999 ) . Background train This was done because purely unsupervised techniques ( e.g. , Baum-Welch [ ] or Brill 's [ Brill 1995b ] ) enable regularities to be induced for word classes which contain many entries , exploiting the fact that individual words that belong to a POS class occur in different ambiguity patterns . CompareOrContrast train Two formalizations of lexical rules as used by HPSG linguists have been proposed , the meta-level lexical rules ( MLRs ; ; Calcagno and Pollard 1995 ) and the . Background train In terms of treebank data , the primary training corpus is the Penn Wall Street Journal Treebank ( PTB ) ( ) . Uses train The extraction of each PICO element relies to a different extent on an annotated corpus of MEDLINE abstracts , created through an effort led by the first author at the National Library of Medicine ( ) . Uses train Various feature selection techniques have been used in automatic text categorization ; they include document frequency ( DF ) , information gain ( IG ) ( Tzeras and Hartman , 1993 ) , minimum description length principal ( ) , and the X2 statistic . Background val Usually , the classes are from WordNet ( Miller et al. , 1990 ) , although they can also be inferred from clustering ( ) . Background train The account sketched in Section 4 was superimposed on an incremental GRE algorithm , partly because incrementality is well established in this area ( Appelt 1985 ; ) . Background train In most cases , the accuracy of parsers degrades when run on out-of-domain data ( Gildea , 2001 ; ; Blitzer et al. , 2006 ; Petrov et al. , 2010 ) . Background train Now for some important remarks on efficiency : • Computing ti is an instance of the well-known algebraic path problem ( ; Tar an , 1981a ) . Background train take an entirely different approach by showing that one can successfully infer held out feature norms from weighted mixtures based on textual similarity . Background train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; ; Steinhauser et al. , 2007 ) . Background train Previous work in sentence planning in the natural language generation ( NLG ) community uses hand-written rules to approximate the distribution of linguistic phenomena in a corpus ( see ( ) for a recent example with further references ) . Background train We assume that every determiner has its own equivalence , which resolves it as a quantifier : sometimes this can be quite a complicated matter , as with any ( ) , which will resolve in different ways depending on its linguistic context , but here we avoid this complexity ' 6 Separate equivalences might also make it easier to encode determiner-specific preferences , such as that of each for wide scope . Background train Following , such expressions will be called vague descriptions even though , as we shall see , the vagueness of the adjective does not extend to the description as a whole . Uses train Such approaches have been tried recently in restricted cases ( McCallum et al. , 2000 ; Eisner , 2001b ; ) . Background train The standard way to handle this problem is to handcraft a finite set of features which provides a sufficient summary of the unbounded history ( Ratnaparkhi , 1999 ; ; Charniak , 2000 ) . CompareOrContrast train Following , we measure association norm prediction as an average of percentile ranks . Uses val There is a general consensus among theoretical linguists that the proper representation of verbal argument structure is event structure -- representations grounded in a theory of events that decompose semantic roles in terms of primitive predicates representing concepts such as causality and inchoativity ( Dowty , 1979 ; Jackendoff , 1983 ; b ; Rappaport Hovav and Levin , 1998 ) . Background train Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) [ 18 ] , fly FlyBase [ 19 ] , yeast Saccharomyces Genome Database ( SGD ) [ 20 ] , rat -- Rat Genome Database ( RGD ) , worm -- WormBase [ 22 ] , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) [ 24 ] , and Enzyme Nomenclature Database ( ECNUM ) [ 25 , 26 ] . Uses train re-trained the linguistic parsers bilingually based on word alignment . CompareOrContrast train de URL : http://www.sfs.nphil.uni-tuebingen.de/sfb / b4home.html 1 This is , for example , the case for all proposals working with verbal lexical entries that raise the arguments of a verbal complement ( ) that also use lexical rules such as the Complement Extraction Lexical Rule ( Pollard and Sag 1994 ) or the Complement Cliticization Lexical Rule ( Miller and Sag 1993 ) to operate on those raised elements . Background train Much of theoretical linguistics can be formulated in a very natural manner as stating correspondences ( translations ) between layers of representation structures ( ) , such as the relation between syntax and semantic . Background val From an IR view , a lot of specialized research has already been carried out for medical applications , with emphasis on the lexico-semantic aspects of dederivation and decomposition ( Pacak et al. , 1980 ; Norton and Pacak , 1983 ; Wolff , 1984 ; ; Dujols et al. , 1991 ; Baud et al. , 1998 ) . Background train The following are the various tag sets we use in this article : ( a ) the core POS tag sets CORE44 and the newly introduced CORE12 ; ( b ) CATiB Treebank tag set ( CATIB6 ) ( Habash and Roth 2009 ) and its newly introduced extension of CATIBEX created using simple regular expressions on word form , indicating particular morphemes such as the prefix JI Al + or the suffix v ' + wn ; this tag set is the best-performing tag set for Arabic on predicted values as reported in Section 4 ; ( c ) the PATB full tag set with complete morphological tag ( BW ) ( ) ; and two extensions of the PATB reduced tag set ( PENN POS , a.k.a. RTS , size 24 [ Diab , Hacioglu , and Jurafsky 2004 ] ) , both outperforming it : ( d ) Kulick , Gabbard , and Marcus ( 2006 ) 's tag set ( KULICK ) , size 43 , one of whose most important extensions is the marking of the definite article clitic , and ( e ) Diab and Benajiba 's ( in preparation ) EXTENDED RTS tag set ( ERTS ) , which marks gender , number , and definiteness , size 134 . Uses train A similar problem is discussed in the psycholinguistics of interpretation ( ) : Interpretation is widely assumed to proceed incrementally , but vague descriptions resist strict incrementality , since an adjective in a vague description can only be fully interpreted when its comparison set is known . CompareOrContrast train In this paper , we evaluated the role of low-level image features , SURF and GIST , for their compatibility with the multimodal Latent Dirichlet Allocation model of . Uses train Figure 2 illustrates a DSyntS from a meteorological application , MeteoCogent ( Kittredge and Lavoie , 1998 ) , represented using the standard graphical notation and also the RealPro ASCII notation used internally in the framework ( ) . Extends train And argues for `` keeping track of counts of arbitrary fragments within parse trees '' , which has indeed been carried out in Collins and Duffy ( 2002 ) who use exactly the same set of ( all ) tree fragments as proposed in Bod ( 1992 ) . Background train use a tagged parallel corpus to extract translationally equivalent English-Greek clauses on the basis of word occurrence and co-occurrence probabilities . Background train The key linguistic knowledge sources that we use are morphological analysis and generation of German based on SMOR , a morphological analyzer/generator of German ( Schmid et al. , 2004 ) and the BitPar parser , which is a state-of-the-art parser of German ( ) . Uses train A common computational treatment of lexical rules adopted , for example , in the ALE system ( ) consists of computing the transitive closure of the base lexical entries under lexical rule application at compile-time . CompareOrContrast train Manually defined heuristics are used to automatically annotate each tree in the treebank with partially specified HPSG derivation trees : Head/argument/modifier distinctions are made for each node in the tree based on and Collins ( 1997 ) ; Uses train Authors may choose this right with the No-Deriv option of the Creative Commons licences ( ) . Background train Mathematical analysis can assess a measure with respect to some formal properties , e.g. whether a measure is a metric ( ) .4 However , mathematical analysis can not tell us whether a measure closely resembles human judgments or whether it performs best when used in a certain application . Background train The same annotation scheme as in our previous work on anger detection has been applied , see e.g. ( ) . Extends train We also compute GIST vectors ( ) for every image using LearGIST ( Douze et al. , 2009 ) . Uses train In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ Kaplan and Bresnan 1982 ; Bresnan 2001 ; Dalrymple 2001 ] , head-driven phrase structure grammar [ HPSG ] [ Pollard and Sag 1994 ] , tree-adjoining grammar [ TAG ] [ Joshi 1988 ] , and combinatory categorial grammar [ CCG ] [ ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Background train The analysis of the data we have collected indicates that student satisfaction may be affected if the system rephrases student answers using different words ( for example , using better terminology ) but does n't explicitly explain the reason why different terminology is needed ( ) . Future val A logic that provides the formal architecture required by Pollard and Sag ( 1994 ) was defined by , 1994 ) . Background val fields generally follow the pattern of `` introduction '' , `` methods '' , `` results '' , and `` conclusions '' ( SalangerMeyer , 1990 ; ; Or˘asan , 2001 ) . Background val report that an optimal tag set for parsing Czech consists of a basic POS tag plus a CASE feature ( when applicable ) . Background train The elimination of redundant nondeterminism is based on Unfold/Fold transformation techniques ( Tamaki and Sato 1984 ) .29 The unfolding transformation is also referred to as partial execution , for example , by . Background train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background train 9 We do not relate to specific results in their study because it has been brought to our attention that are in the process of rechecking their code for errors , and rerunning their experiments ( personal communication ) . CompareOrContrast val Inspired by ( ) , we split one phrase type into several subsymbols , which contain category information of current constituent 's parent . Motivation train The recent great advances in speech and language technologies have made it possible to build fully implemented spoken dialogue systems ( Aust et al. , 1995 ; Allen et al. , 1996 ; ; Walker et al. , 2000 ) . Background train Moreover , in order to determine whether the performances of the predictive criteria are consistent across different learning models within the same domain , we have performed the study on two parsing models : one based on a context-free variant of tree-adjoining grammars ( Joshi , Levy , and Takahashi 1975 ) , the Probabilistic Lexicalized Tree Insertion Grammar ( PLTIG ) formalism ( Schabes and Waters 1993 ; ) , and Collins 's Model 2 parser ( 1997 ) . Uses train Most of the early work on automatic f-structure annotation ( e.g. , van Genabith , Way , and Sadler 1999 ; ; Sadler , van Genabith , and Way 2000 ) was applied only to small data sets ( fewer than 200 sentences ) and was largely proof of concept . Background train We prepare the training data by splitting compounds in two steps , following the technique of . Uses train Recent work by on the filtering phase of this approach uses linguistic verb classes ( based on Levin [ 1993 ] ) for obtaining more accurate back-off estimates for hypothesis selection . Background train Numerous previous pseudodisambiguation evaluations only include arguments that occur between 30 and 3000 times ( Erk , 2007 ; Keller and Lapata , 2003 ; ) . CompareOrContrast train Some efforts have tackled tasks such as automatic image caption generation ( Feng and Lapata , 2010a ; Ordonez et al. , 2011 ) , text illustration ( Joshi et al. , 2006 ) , or automatic location identification of Twitter users ( Eisenstein et al. , 2010 ; Wing and Baldridge , 2011 ; ) . Background train Default parameters were used , although experimentation with different parameter settings is an important direction for future work ( ; Munson et al. , 2005 ) . Future train The language chosen for semantic representation is a flat semantics along the line of ( Bos , 1995 ; ; Copestake et al. , 2001 ) . CompareOrContrast train For example , and Jokinen et al. ( 2008 ) find that machine learning algorithms can be trained to recognise some of the functions of head movements , while Reidsma et al. ( 2009 ) show that there is a dependence between focus of attention and assignment of dialogue act labels . Background train As a logical postulate it is not very radical ; it is possible within a finitary framework to develop that part of mathematics that is used or has potential applications in natural science , such as mathematical analysis ( cfXXX ) . Background train In a similar vain to Skut and Brants ( 1998 ) and , the method extends an existing flat shallow-parsing method to handle composite structures . Future val An interesting aspect of our generative approach is that we model HMM outputs as Gaussian vectors ( log probabilities of observing entire sentences based on our language models ) , as opposed to sequences of terms , as done in ( ) . CompareOrContrast train • language learning ( Green 1979 ; Mori and Moeser 1983 ; Morgan , Meier , and Newport 1989 ) • monolingual grammar induction ( Juola 1998 ) • grammar optimization ( Juola 1994 ) • insights into universal grammar ( Juola 1998 ) • machine translation ( Juola 1994 , 1997 ; ; Gough , Way , and Hearne 2002 ) Background train In Charniak ( 1996 ) and , it was observed that treebank grammars ( CFGs extracted from treebanks ) are very large and grow with the size of the treebank . Background val We apply our system to the latest version of the XTAG English grammar ( The XTAG Research ) , which is a large-scale FB-LTAG grammar . Uses train Furthermore , the need to answer questions related to patient care at the point of service has been well studied and documented ( Covell , Uman , and Manning 1985 ; Gorman , Ash , and Wykoff 1994 ; , 2005 ) . Background train Typical examples are Bulgarian ( Simov et al. , 2005 ; Simov and Osenova , 2003 ) , Chinese ( Chen et al. , 2003 ) , Danish ( Kromann , 2003 ) , and Swedish ( ) . Background train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background train Worst case , calculating the set corresponding with such a property , of the form size ( x ) = maxm , for example , involves sorting the distractors as to their size , which may amount to O ( n2d ) or O ( nd log nd ) calculations ( depending on the sorting algorithm : cfXXX [ ] Chapter 8 ) . Background train • language learning ( ; Mori and Moeser 1983 ; Morgan , Meier , and Newport 1989 ) • monolingual grammar induction ( Juola 1998 ) • grammar optimization ( Juola 1994 ) • insights into universal grammar ( Juola 1998 ) • machine translation ( Juola 1994 , 1997 ; Veale and Way 1997 ; Gough , Way , and Hearne 2002 ) Background train Virpioja et al. ( 2007 ) , Badr et al. ( 2008 ) , Luong et al. ( 2010 ) , , and others are primarily concerned with using morpheme segmentation in SMT , which is a useful approach for dealing with issues of word-formation . CompareOrContrast train While corpus driven efforts along the PARSEVAL lines ( ) are good at giving some measure of a grammar coverage , they are not suitable for finer grained analysis and in particular , for progress evaluation , regression testing and comparative report generation . Background val Before using the DCA method , we applied a Russian morphological processor ( ) to convert each word in the text to its main form : nominative case singular for nouns and adjectives , infinitive for verbs , etc. . Uses train The result holds for both the MaltParser ( Nivre 2008 ) and the Easy-First Parser ( ) . Uses train Also relevant is work on the general problems of dialog-act tagging ( Stolcke et al. , 2000 ) , citation analysis ( ) , and computational rhetorical analysis ( Marcu , 2000 ; Teufel and Moens , 2002 ) . Background train The language chosen for semantic representation is a flat semantics along the line of ( ; Copestake et al. , 1999 ; Copestake et al. , 2001 ) . CompareOrContrast val In simple terms , P2P is a technology that takes advantage of the resources and services available at the edge of the Internet ( ) . Background train A common way to combine different models consists of selecting the model that is most confident regarding its decision ( ) . CompareOrContrast train Common sense ( as well as the Gricean maxims ; ) suggests that vague descriptions are preferred by speakers over quantitative ones whenever the additional information provided by a quantitative description is irrelevant to the purpose of the communication . Background train A similar method is included in PATR-II ( Shieber et al. 1983 ) and can be used to encode lexical rules as binary relations in the CUF system ( Dorre and Eisele 1991 ; Done and Dorna 1993b ) or the TFS system ( ; Emele 1994 ) . CompareOrContrast train For automatically extracting these surface level mappings we will draw on the approach to learning paraphrases from a corpus that is described in . Future val This description can then be given the standard set-theoretical interpretation of , 1994 ) . ' Background train helped pave the path for cognitive-linguistic multimodal research , showing that Latent Dirichlet Allocation outperformed Latent Semantic Analysis ( Deerwester et al. , 1990 ) in the prediction of association norms . Background train A more flexible approach is used by , where users can specify boundary values for attributes like rainfall , specifying , for example , rain counts as moderate above 7 mm/h , as heavy above 20 mm/h , and so on . Background train Collins and Duffy ( 2002 ) define a kernel over parse trees and apply it to re-ranking the output of a parser , but the resulting feature space is restricted by the need to compute the kernel efficiently , and the results are not as good as Collins ' previous work on re-ranking using a finite set of features ( ) . Background train The shallow parser used is the SNoW-based CSCL parser ( ; Munoz et al. , 1999 ) . Uses train In practice , perceptron-type algorithms are often applied in a batch learning scenario , i.e. , the algorithm is applied for K epochs to a training sample of size T and then used for prediction on an unseen test set ( Freund and Schapire , 1999 ; ) . CompareOrContrast train Table 2 shows the results on identifying all phrases -- chunking in CoNLL2000 ( Tjong Kim ) terminology . Uses train Unfortunately , as shown in ( ) , with the represetation of sentences that we use , linear classifiers can not discriminate real sentences from sentences sampled from a trigram , which is the model we use as a baseline , so here we resort to a non-linear large-margin classifier ( see section 3 for details ) . Motivation train Perhaps some variation of multi-level bulleted lists , appropriately integrated with interface elements for expanding and hiding items , might provide physicians a better overview of the information landscape ; see , for example , . Background train have shown , in the context of base noun identification , that combining sample selection and cotraining can be an effective learning framework for large-scale training . Background train • The regular TBL , as described in section 2 ; • An improved version of TBL , which makes extensive use of indexes to speed up the rules ' update ; • The FastTBL algorithm ; • The ICA algorithm ( ) . CompareOrContrast train For example , Jokinen and Ragni ( 2007 ) and find that machine learning algorithms can be trained to recognise some of the functions of head movements , while Reidsma et al. ( 2009 ) show that there is a dependence between focus of attention and assignment of dialogue act labels . Background train This approach resembles the work by and Hirschman et al. ( 1975 ) on selectional restrictions . CompareOrContrast train Unless it is desired to intentionally filter these out as being outside of the new domain , one can insert some arbitrarily small probability for these arcs , using , for example , an N-gram back-off model ( ) . Background val feature Cohen 's k corrected k agreement 73.59 98.74 dial act 84.53 98.87 turn 73.52 99.16 Table 2 : Inter-coder agreement on feedback expression annotation Although researchers do not totally agree on how to measure agreement in various types of annotated data and on how to interpret the resulting figures , see Artstein and Poesio ( 2008 ) , it is usually assumed that Cohen 's kappa figures over 60 are good while those over 75 are excellent ( ) . Background train To name a few examples , Rohrbach et al. ( 2010 ) and show how semantic information from text can be used to improve zero-shot classification ( i.e. , classifying never-before-seen objects ) , and Motwani and Mooney ( 2012 ) show that verb clusters can be used to improve activity recognition in videos . Background train Empirical evidence has been brought forward that inflectional and/or derivational stemmers augmented by dictionaries indeed perform substantially better than those without access to such lexical repositories ( Krovetz , 1993 ; ; Tzoukermann et al. , 1997 ) . Background train The XTAG group ( ) at the University of Pennsylvania is also developing Korean , Chinese , and Hindi grammars . Background train HOLMES is given the following set of six domainindependent rules , which are similar to the upward monotone rules introduced by ( ) . CompareOrContrast train The lexicon is used to mediate and map between a language-independent domain model and a language-dependent ontology widely used in NLG , the Upper Model ( ) . Background train ImageNet is a large-scale and widely used image database , built on top of WordNet , which maps words into groups of images , called synsets ( ) . Uses train based parsing algorithms with an arc-factored parameterization ( ) . Uses train Two applications that , like help-desk , deal with question -- answer pairs are : summarization of e-mail threads ( Dalli , Xia , and Wilks 2004 ; Shrestha and McKeown 2004 ) , and answer extraction in FAQs ( Frequently Asked Questions ) ( ; CompareOrContrast val The retrieval process relies on the vector space model ( ) , with the cosine measure expressing the similarity between a query and a document . Uses val For right-branching structures , the leftcorner ancestor is the parent , conditioning on which has been found to be beneficial ( ) , as has conditioning on the left-corner child ( Roark and Johnson , 1999 ) . Background train A few others incorporate various measures of inter-document similarity between the texts to be labeled ( Agarwal and Bhattacharyya , 2005 ; ; Goldberg and Zhu , 2006 ) . Background train A few others incorporate various measures of inter-document similarity between the texts to be labeled ( ; Pang and Lee , 2005 ; Goldberg and Zhu , 2006 ) . Background train This is similar to `` one sense per collocation '' idea of . CompareOrContrast train Against the background of a growing interest in multilingual NLP , multilingual anaphora / coreference resolution has gained considerable momentum in recent years ( Aone and McKee 1993 ; Azzam , Humphreys , and Gaizauskas 1998 ; Harabagiu and Maiorano 2000 ; Mitkov and Barbu 2000 ; ; Mitkov and Stys 1997 ; Mitkov , Belguith , and Stys 1998 ) . Background train This system has been successfully tested with the development of plug-ins supporting instant messaging , distributed video encoding ( Hughes and Walkerdine , 2005 ) , distributed virtual worlds ( Hughes et al. , 2005 ) and digital library management ( ) . Background train be found in figure 2 , which is similar with that in . CompareOrContrast train It is inspired by the system described in . Motivation train • A user study was performed , but it was either very small compared to the corpus ( Carmel , Shtalhaim , and Soffer 2000 ; Jijkoun and de Rijke 2005 ) , or the corpus itself was significantly smaller than ours ( Feng et al. 2006 ; ) . CompareOrContrast train Although a number of methods for query-dependent text summarization are beginning to be developed and evaluated in a variety of realistic settings ( ) , we again propose the use of vector space methods from IR , which can be easily extended to the summarization task ( Salton et al. , 1994 ) : CompareOrContrast train Riehemann 1993 ; ; Frank 1994 ; Opalka 1995 ; Sanfilippo 1995 ) . CompareOrContrast train When we run our classifiers on resource-tight environments such as cell-phones , we can use a random feature mixing technique ( ) or a memory-efficient trie implementation based on a succinct data structure ( Jacobson , 1989 ; Delpratt et al. , 2006 ) to reduce required memory usage . Future train These translations gave rise to a number of automatically constructed linguistic resources : ( 1 ) the original ( source , target ) phrasal translation pairs , ( 2 ) the marker lexicon , ( 3 ) the gen11 Thanks are due to one of the anonymous reviewers for pointing out that our wEBMT system , seeded with input from multiple translation systems , with a postvalidation process via the Web ( amounting to an n-gram target language model ) , in effect forms a multiengine MT system as described by Frederking and Nirenburg ( 1994 ) , , and Hogan and Frederking ( 1998 ) . CompareOrContrast train A variety of statistical methods were proposed over the recent years for learning to produce a full parse of free-text sentences ( e.g. , Bod ( 1992 ) , , Collins ( 1997 ) , Ratnaparkhi ( 1997 ) , and Sekine ( 1998 ) ) . Background train Discriminative approaches ( especially SVMs ) have been shown to be very effective for many supervised classification tasks ; see , for example , ( ; Ng and Jordan , 2001 ) . Background train This is the approach taken by IBM Models 4 + ( Brown et al. 1993b ; ) , and more recently by the LEAF model ( Fraser and Marcu 2007 ) . CompareOrContrast train A few others incorporate various measures of inter-document similarity between the texts to be labeled ( Agarwal and Bhattacharyya , 2005 ; Pang and Lee , 2005 ; ) . Background val 11 proposes to unify these two steps by including an update operator in the Background train For instance , when building translation units in EBMT approaches ( Richardson et al. , 2001 ) , ( Aramaki , 2001 ) , ( AlAdhaileh & Tang , 1999 ) , ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( ) , etc. , where S-SSTC can be used to represent the entries of the BKB or when S-SSTC used as an annotation schema to find the translation correspondences ( lexical and structural correspondences ) for transferrules ' extraction from parallel parsed corpus ( Menezes & Richardson , 2001 ) , ( Watanabe et al. , Background train introduce a new method of multimodal integration based on Canonical Correlation Analysis , and performs a systematic comparison between their CCA-based model and others on association norm prediction , held out feature prediction , and word similarity . Background train More details on how the structural divergences described in ( ) can be accounted for using our formalism can be found in ( Nasr et al. , 1998 ) . Background train At present , the system takes into consideration the number of incorrect answers received in response to the current question and the number of uninterpretable answers .1 In addition to a remediation policy , the tutorial planner implements an error recovery policy ( ) . Uses train More recently , show that visual attribute classifiers , which have been immensely successful in object recognition ( Farhadi et al. , 2009 ) , act as excellent substitutes for feature Background train We found the same number using our previous approach ( ) , which is roughly equivalent to our core module . Extends train Similar things hold for multifaceted properties like intelligence ( ) . Background train We have shown elsewhere ( ; Zadrozny 1987a , 1987b ) that natural language programs , such as on-line grammars and dictionaries , can be used as referential levels for commonsense reasoning -- for example , to disambiguate PP attachment . Extends val ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) used maximum likelihood estimation to learn weights for MT. ( Och , 2003 ; Moore and Quirk , 2008 ; ; Galley and Quirk , 2011 ) employed an evaluation metric as a loss function and directly optimized it . CompareOrContrast train There has been some controversy , at least for simple stemmers ( ; Porter , 1980 ) , about the effectiveness of morphological analysis for document retrieval ( Harman , 1991 ; Krovetz , 1993 ; Hull , 1996 ) . Background train reported an intra-subject correlation of r = .85 for 15 subjects judging the similarity of a subset ( 36 ) of the original 65 word pairs . CompareOrContrast train Withindocument coreference resolution has been applied to produce summaries of text surrounding occurrences of the name ( ; Gooi and Allan , 2004 ) . Background train More recently , have proposed the exploitation of TMs at a subsentential level , while Carl , Way , and Sch ¨ aler ( 2002 ) and Sch ¨ aler , Way , and Carl ( 2003 , pages 108 -- 109 ) describe how phrasal lexicons might come to occupy a central place in a future hybrid integrated translation environment . Background train With a minimal set of features and a small number of lexical entries , Niyogi ( 2001 ) has successfully modeled many of the argument alternations described by Levin ( 1993 ) using a style analysis . Background train For some adjectives , including the ones that called evaluative ( as opposed to dimensional ) , this is clearly inadequate . Background train WIT has been implemented in Common Lisp and C on UNIX , and we have built several experimental and demonstration dialogue systems using it , including a meeting room reservation system ( b ) , a video-recording programming system , a schedule management system ( Nakano et al. , 1999a ) , and a weather infomiation system ( Dohsaka et al. , 2000 ) . Extends train The more conservative approach is to try to integrate existing statistical disambiguation schemes for QLFs , either individually or in a `` packed '' structure ( ) , with the resolution process as described here . Future val They proved to be useful in a number of NLP applications such as natural language generation ( Iordanskaja et al. , 1991 ) , multidocument summarization ( ) , automatic evaluation of MT ( Denkowski and Lavie , 2010 ) , and TE ( Dinu and Wang , 2009 ) . Motivation train While many approaches have addressed this problem , our work is most closely related to that of ( Raina et al. , 2005 ; ; Tatu and Moldovan , 2006 ; Braz et al. , 2005 ) , which convert the inputs into logical forms and then attempt to ` prove ' H from T plus a set of axioms . CompareOrContrast train While this is simply irrelevant for general-purpose morphological analyzers , dealing with such phenomena is crucial for any attempt to cope adequately with medical free-texts in an IR setting ( ) . Background train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( Bikel et al. , 1997 ; Miller et al. , 1998 ; Borthwick , 1999 ; Mikheev et al. , 1999 ; Soon et al. , 2001 ; Ng and Cardie , 2002 ; ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train In English , where the base form is morphologically simpler than the other two , this rule could be argued to follow from Gricean principles ( ) . Background val More generally , distributional clustering techniques ( Sch ¨ utze , 1992 ; ) could be applied to extract semantic classes from the corpus itself . Future train The use of the web as a corpus for teaching and research on language has been proposed a number of times ( Kilgarriff , 2001 ; Robb , 2003 ; ; Fletcher , 2001 , 2004b ) and received a special issue of the journal Computational Linguistics ( Kilgarriff and Grefenstette , 2003 ) . Background train replicates the work of Fung and McKeown with different language pairs using the simpler metric of Levenshtein distance . Background val • Only an automatic evaluation was performed , which relied on having model responses ( ; Berger et al. 2000 ) . CompareOrContrast train Typed feature grammars can be used as the basis for implementations of Head-driven Phrase Structure Grammar ( HPSG ; ) as discussed in ( Gotz and Meurers , 1997a ) and ( Meurers and Minnen , 1997 ) . Background train In a ) we identified several systems that resemble ours in that they provide answers to queries . Background train The problem of handling ill-formed input has been studied by Carbonell and Hayes ( 1983 ) , Granger ( 1983 ) , , Kwasny and Sondheimer ( 1981 ) , Riesbeck and Schank ( 1976 ) , Thompson ( 1980 ) , Weischedel and Black ( 1980 ) , and Weischedel and Sondheimer ( 1983 ) . CompareOrContrast val A variety of such lists for many languages are already available ( e.g. , ) . Background train In principle , this might be done by providing the generator with vague input -- in which case no special algorithms are needed -- but suitably contextualized vague input is often not available ( ) . Motivation train Liu et al. ( 2005 ) , Meral et al. ( 2007 ) , Murphy ( 2001 ) , Murphy and Vogel ( 2007 ) and a ) all belong to the syntactic transformation category . Background train We induced a two-class word-to-word model of translational equivalence from 13 million words of the Canadian Hansards , aligned using the method in ( ) . Uses train observes that accomplishments differ from achievements only in terms of event duration , which is often a question of granularity . Background train [ The current system should be distinguished from an earlier voice system ( VNLC , ) , which had no expectation and which handled discrete speech where a 300 millisecond pause must follow each word . ] CompareOrContrast train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train He was a grammarian who analysed Sanskrit ( ) . Background train After much exploration , discovered that it was not practical to annotate PICO entities at the phrase level due to significant unresolvable disagreement and interannotator reliability issues . Background train The necessity of this kind of merging of arguments has been recognized before : call it abductive unification/matching , Hobbs ( 1978 , 1979 ) refers to such operations using the terms knitting or petty conversational implicature . Background train The OntoNotes-5 .0 dataset , which is released for the CoNLL-2012 Shared Task ( ) , contains 3,145 annotated documents . Uses train and , as described below . CompareOrContrast train We collect substring rationales for a sentiment classification task ( ) and use them to obtain significant accuracy improvements for each annotator . Uses train We apply two different priming experiments namely , the cross modal priming and masked priming experiment discussed in ( ; Rastle et al. , 2000 ; Marslen-Wilson et al. , 1994 ; Marslen-Wilson et al. , 2008 ) for Bangla morphologically complex words . Uses val Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( Neville and Jensen , 2000 ; Lafferty et al. , 2001 ; Getoor et al. , 2002 ; Taskar et al. , 2002 ; Taskar et al. , 2003 ; ; McCallum and Wellner , 2004 ) . Background train More sophisticated approaches have been proposed ( Hillard et al. , 2003 ) , including an extension that , in an interesting reversal of our problem , makes use of sentimentpolarity indicators within speech segments ( ) . Background train The expectation parser uses an ATN-like representation for its grammar ( ) . Uses train • Only an automatic evaluation was performed , which relied on having model responses ( Berger and Mittal 2000 ; ) . CompareOrContrast train The error rate on sentence boundaries in the Brown corpus was not significantly worse than the lowest quoted before ( : 0.28 % vs. 0.20 % error rate ) . CompareOrContrast train We follow our previous work ( b ) and restrict bridging to non-coreferential cases . Extends train In our case , the clustering is performed by the program Snob , which implements mixture modeling combined with model selection based on the Minimum Message Length ( MML ) criterion ( ; Wallace 2005 ) . Uses train This is in line with our previous findings from ( ) that candidates with higher power attempt to shift topics less often than others when responding to moderators . CompareOrContrast train As has been previously observed and exploited in the NLP literature ( ; Agarwal and Bhattacharyya , 2005 ; Barzilay and Lapata , 2005 ) , the above optimization function , unlike many others that have been proposed for graph or set partitioning , can be solved exactly in an provably efficient manner via methods for finding minimum cuts in graphs . CompareOrContrast train It is analogous to the step in other translation model induction algorithms that sets all probabilities below a certain threshold to negligible values ( Brown et al. , 1990 ; ; Chen , 1996 ) . Background train It provides a fine grained NE recognition covering 100 different NE types ( ) . Uses train Only a few such corpora exist , including the Hansard English-French corpus and the HKUST EnglishChinese corpus ( ) . Background train Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( Neville and Jensen , 2000 ; Lafferty et al. , 2001 ; Getoor et al. , 2002 ; ; Taskar et al. , 2003 ; Taskar et al. , 2004 ; McCallum and Wellner , 2004 ) . Background train In FAQs , employed a sentence retrieval approach based on a language model where the entire response to an FAQ is considered a sentence , and the questions and answers are embedded in an FAQ document . Background train The statistical significance test is performed by the re-sampling approach ( ) . Uses val With a minimal set of features and a small number of lexical entries , Niyogi ( 2001 ) has successfully modeled many of the argument alternations described by using a Hale and Keyser ( 1993 ) style analysis . Background train Although evaluated on different data sets , this result is consistent with results from previous work ( Gatt and Belz , 2008 ; ) . CompareOrContrast train We used the revised experimental setup ( ) , based on discrete relatedness scores and presentation of word pairs in isolation , that is scalable to the higher number of pairs . Uses val ; Bobrow , 1978 ) consult relatively small lexicons , typically generated by hand . CompareOrContrast train Typed feature grammars can be used as the basis for implementations of Head-driven Phrase Structure Grammar ( ) .3 ( Meurers and Minnen , 1997 ) propose a compilation of lexical rules into TIT definite clauses Background train In the United States , for example , governmental bodies are providing and soliciting political documents via the Internet , with lofty goals in mind : electronic rulemaking ( eRulemaking ) initiatives involving the `` electronic collection , distribution , synthesis , and analysis of public commentary in the regulatory rulemaking process '' , may `` [ alter ] the citizen-government relationship '' ( ) . Background train This will become even more interesting when mappings of our synonym identifiers to a large medical thesaurus ( MeSH , ( ) ) are incorporated into our system . Future train ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) used maximum likelihood estimation to learn weights for MT. ( Och , 2003 ; ; Zhao and Chen , 2009 ; Galley and Quirk , 2011 ) employed an evaluation metric as a loss function and directly optimized it . CompareOrContrast train raw length value as a feature , we follow our previous work ( Rubino et al. , 2013 ; ) and create multiple features for length using a decision tree ( J48 ) . Extends train Most DOP models , such as in Bod ( 1993 ) , Goodman ( 1996 ) , Bonnema et al. ( 1997 ) , Sima'an ( 2000 ) and , use a likelihood criterion in defining the best parse tree : they take ( some notion of ) the most likely ( i.e. most probable ) tree as a candidate for the best tree of a sentence . Background train In some systems such dependencies are learned from labeled examples ( ) . Background train CD for this type of descriptions along the lines of Section 4 is not difficult once relational descriptions are integrated with a standard GRE algorithm ( , Section 8.6.2 ) : Suppose an initial description is generated describing the set of all those dogs that are in sheds over a given size ( say , size 5 ) ; if this description happens to distinguish an individual dog then this legitimizes the use of the noun phrase the dog in the large shed . Background train Michiels proposed rules for doing this for infinitive complement codes ; however there seems to be no principled reason not to extend this approach to computing the underlying relations in other types of VP as well as in cases of NP , AP and PP predication ( see , for further discussion ) . Background val While wikis have spread from a detailed design ( ) , unfortunately blogs have not been designed under a model . Background train In our experiment , we annotated a high number of pairs similar in size to the test sets by Finkelstein ( 2002 ) and . CompareOrContrast val notation of is more sophisticated , and may be considered another possibility . CompareOrContrast train The recent great advances in speech and language technologies have made it possible to build fully implemented spoken dialogue systems ( Aust et al. , 1995 ; ; Zue et al. , 2000 ; Walker et al. , 2000 ) . Background val Most approaches rely on VerbNet ( ) and FrameNet ( Baker et al. , 1998 ) to provide associations between verbs and semantic roles , that are then mapped onto the current instance , as shown by the systems competing in semantic role labelling competitions ( Carreras and Marquez , 2004 ; Carreras and Marquez , 2005 ) and also ( Gildea and Jurafsky , 2002 ; Pradhan et al. , 2005 ; Shi and Mihalcea , 2005 ) . Background train aimed to embed information by exploiting the linguistic phenomenon of presupposition , with the idea that some presuppositional information can be removed without changing the meaning of a sentence . Background train We first identified the most informative unigrams and bigrams using the information gain measure ( Yang and Pedersen 1997 ) , and then selected only the positive outcome predictors using odds ratio ( ) . Uses train Perhaps some variation of multi-level bulleted lists , appropriately integrated with interface elements for expanding and hiding items , might provide physicians a better overview of the information landscape ; see , for example , . Future train Most DOP models , such as in Bod ( 1993 ) , , Bonnema et al. ( 1997 ) , Sima'an ( 2000 ) and Collins & Duffy ( 2002 ) , use a likelihood criterion in defining the best parse tree : they take ( some notion of ) the most likely ( i.e. most probable ) tree as a candidate for the best tree of a sentence . Background train Some previous works ( ; Zhao et al. , 2009 ; Kouylekov et al. , 2009 ) indicate , as main limitations of the mentioned resources , their limited coverage , their low precision , and the fact that they are mostly suitable to capture relations mainly between single words . Background train Nugget F-score has been employed as a metric in the TREC question-answering track since 2003 , to evaluate so-called definition and `` other '' questions ( ) . Background train Manually defined heuristics are used to automatically annotate each tree in the treebank with partially specified HPSG derivation trees : Head/argument/modifier distinctions are made for each node in the tree based on Magerman ( 1994 ) and ; Uses train Regarding future work , there are many research line that may be followed : i ) Capturing more features by employing external knowledge such as ontological , lexical resource or WordNet-based features ( Basili et al. , 2005a ; Basili et al. , 2005b ; Bloehdorn et al. , 2006 ; Bloehdorn and Moschitti , 2007 ) or shallow semantic trees , ( Giuglea and Moschitti , 2004 ; Giuglea and Moschitti , 2006 ; ; Moschitti et al. , 2007 ; Moschitti , 2008 ; Moschitti et al. , 2008 ) . Future train In addition , there has been much work on the application of linguistic and semantic knowledge to information retrieval ; see a ) for a brief overview . Background train However , claims that the log-likelihood chisquared statistic ( G2 ) is more appropriate for corpus-based NLP . Motivation train In a log-linear parameterization , for example , a prior that penalizes feature strengths far from 1 can be used to do feature selection and avoid overfitting ( ) . Uses val This alignment is done on the basis of both length ( Gale and Church ) and a notion of cognateness ( Simard [ 16 ] ) . Uses train Since earlier versions of the SNoW based CSCL were used only to identify single phrases ( Punyakanok and Roth , 2001 ; ) and never to identify a collection of several phrases at the same time , as we do here , we also trained and tested it under the exact conditions of CoNLL-2000 ( Tjong Kim Sang and Buchholz , 2000 ) to compare it to other shallow parsers . Extends train A more detailed discussion of the various available Arabic tag sets can be found in . Background train The high Dirichlet priors are chosen to prevent sparsity in topic distributions , while the other parameters are selected as the best from . Uses val The resulting speech understanding system is called the Voice Natural Language Computer with Expectation ( VNLCE , ) . Background train In our previous papers ( ; Zhang , Blackwood , and Clark 2012 ) , we applied a set of beams to this structure , which makes it similar to the data structure used for phrase-based MT decoding ( Koehn 2010 ) . CompareOrContrast train A number of speech understanding systems have been developed during the past fifteen years ( , Dixon and Martin 1979 , Erman et al. 1980 , Haton and Pierrel 1976 , Lea 1980 , Lowerre and Reddy 1980 , Medress 1980 , Reddy 1976 , Walker 1978 , and Wolf and Woods 1980 ) . CompareOrContrast train Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( Andrews et al. , 2009 ; Steyvers , 2010 ; Feng and Lapata , 2010b ; Bruni et al. , 2011 ; ; Johns and Jones , 2012 ; Bruni et al. , 2012a ; Bruni et al. , 2012b ; Silberer et al. , 2013 ) . Background train In particular , since we treat each individual speech within a debate as a single `` document '' , we are considering a version of document-level sentiment-polarity classification , namely , automatically distinguishing between positive and negative documents ( ; Pang et al. , 2002 ; Turney , 2002 ; Dave et al. , 2003 ) . Background train A number of alignment techniques have been proposed , varying from statistical methods ( Brown et al. , 1991 ; Gale and Church , 1991 ) to lexical methods ( Kay and Roscheisen , 1993 ; ) . Background train ( contains further description and discussion of LDOCE . ) Background train , 1997 ) assumes that words ending in - ed are verbs . CompareOrContrast train For the development of these lists we used a collection of texts of about 300,000 words derived from the New York Times ( NYT ) corpus that was supplied as training data for the 7th Message Understanding Conference ( MUC-7 ) ( ) . Uses train We would also provide a user facility for choosing the right licence for every lexia , following the model of Creative Commons licences ( ) . Uses train The language grounding problem has received significant attention in recent years , owed in part to the wide availability of data sets ( e.g. Flickr , Von Ahn ( 2006 ) ) , computing power , improved computer vision models ( Oliva and Torralba , 2001 ; ; Farhadi et al. , 2009 ; Parikh and Grauman , 2011 ) and neurological evidence of ties between the language , perceptual and motor systems in the brain ( Pulverm ¨ uller et al. , 2005 ; Tettamanti et al. , 2005 ; Aziz-Zadeh et al. , 2006 ) . Background train There has been some controversy , at least for simple stemmers ( Lovins , 1968 ; Porter , 1980 ) , about the effectiveness of morphological analysis for document retrieval ( Harman , 1991 ; ; Hull , 1996 ) . Background train We have yet to import such a constraint into our model , but we plan to do so in the near future using the weighted majority algorithm ( ) . Future train Generally speaking , we find that the personal public diary metaphor behind blogs ( ) may bring to an unsatisfactory representation of the context . Background train Most probabilistic translation model reestimation algorithms published to date are variations on the theme proposed by b ) . Background train A number of proposals in the 1990s deliberately limited the extent to which they relied on domain and/or linguistic knowledge and reported promising results in knowledge-poor operational environments ( Dagan and Itai 1990 , 1991 ; Lappin and Leass 1994 ; ; Kennedy and Boguraev 1996 ; Williams , Harvey , and Preston 1996 ; Baldwin 1997 ; Mitkov 1996 , 1998b ) . Background train The significance testing is performed by paired bootstrap re-sampling ( ) . Uses train The system is implemented based on ( ) and ( Marcu et al. 2006 ) . Uses train IGEN uses standard chart generation techniques ( ) in its base generator to efficiently produce generation candidates . Background train Position , subcat frame , phrase type , first word , last word , subcat frame + , predicate , path , head word and its POS , predicate + head word , predicate + phrase type , path to BA and BEI , verb class 3 , verb class + head word , verb class + phrase type , from . Uses train The psycholinguistic studies of Martin ( 1970 ) , Allen ( 1975 ) , Hillinger et al. ( 1976 ) , Grosjean et al. ( 1979 ) , Dommergues and Grosjean ( 1983 ) , and , responding to the idea of readjusted syntax as the source of prosodic phrasing , show that grammatical structure , even if readjusted , is not in itself a reliable predictor of prosodic phrasing : mismatches between syntax and prosody occur often and systematically , and can be related to specific nonsyntactic factors such as length and word frequency . Background train A number of studies ( e.g. , Hildebrandt , Katz , and Lin 2004 ) have pointed out shortcomings of the original nugget scoring model , although a number of these issues have been recently addressed ( a , 2006b ) . Background train reported that the translation lexicon that our model induced from this tiny bitext accounted for 30 % of the word types with precision between 84 % and 90 % . Background train describe a simple tool which uses fine-grained rules to identify the arguments of verb occurrences in the Penn-II Treebank . Background train The contextual interpreter then uses a reference resolution approach similar to , and an ontology mapping mechanism ( Dzikovska et al. , 2008a ) to produce a domain-specific semantic representation of the student 's output . CompareOrContrast train maintains a survey of this area . Background train The principle of maximum entropy states that when one searches among probability distributions that model the observed data ( evidence ) , the preferred one is the one that maximizes the entropy ( a measure of the uncertainty of the model ) ( ) . Background train In the context of word alignment , use a state-duration HMM in order to model word-to-phrase translations . Background val With respect to the focus on function words , our reordering model is closely related to the UALIGN system ( ) . CompareOrContrast train We could also introduce new variables , e.g. , nonterminal refinements ( ) , or secondary links Mid ( not constrained by TREE/PTREE ) that augment the parse with representations of control , binding , etc. ( Sleator and Temperley , 1993 ; Buch-Kromann , 2006 ) . Future train Selectional Preferences have also been a recent focus of researchers investigating the learning of paraphrases and inference rules ( ; Roberto et al. , 2007 ) . Background train Previously LDA has been successfully used to infer unsupervised joint topic distributions over words and feature norms together ( Andrews et al. , 2009 ; ) . Background train Something like this approach is in fact used in some systems ( e.g. , Elhadad and Robin 1992 ; PenMan 1989 ; a ) . Background train The emphasis on narrativity takes into account the use of blogs as public diaries on the web , that is still the main current interpretation of this literary genre , or metagenre ( ) . Background train For example , it would be helpful to consider strong correspondence between certain English and Chinese words , as in ( ) . Future train The system utilizes several large size biological databases including three NCBI databases ( GenPept [ 11 ] , RefSeq [ 12 ] , and Entrez GENE [ 13 ] ) , PSD database from Protein Information Resources ( PIR ) , and Uses train We take some core ideas from our previous work on mining script information ( ) . Extends train While IA is generally thought to be consistent with findings on human language production ( Hermann and Deutsch 1976 ; Levelt 1989 ; ; Sonnenschein 1982 ) , the hypothesis that incrementality is a good model of human GRE seems unfalsifiable until a preference order is specified for the properties on which it operates . Background train Typical examples are Bulgarian ( ; Simov and Osenova , 2003 ) , Chinese ( Chen et al. , 2003 ) , Danish ( Kromann , 2003 ) , and Swedish ( Nilsson et al. , 2005 ) . Background val • A user study was performed , but it was either very small compared to the corpus ( Carmel , Shtalhaim , and Soffer 2000 ; Jijkoun and de Rijke 2005 ) , or the corpus itself was significantly smaller than ours ( ; Leuski et al. 2006 ) . CompareOrContrast train Our work on the prosodic phrase status of clause final prepositional phrases , which we discuss below , suggests the existence of a discourse-neutral phrasing that depends on syntactic constituency mediated by string adjacency and length of a potential prosodic phrase .3 Such phrasing provides us with a typical phrasing pattern analogous to the typical phrasal stress patterns examined in , which `` are often overwhelmed by the chiaroscuro of highlight and background in discourse , but retain the status of null-hypothesis patterns that emerge when Computational Linguistics Volume 16 , Number 3 , September 1990 157 J. Bachenko and E. Fitzpatrick Discourse-Neutral Prosodic Phrasing in English there is no good reason to take some other option '' ( p. 251 ) . CompareOrContrast train Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( Andrews et al. , 2009 ; ; Feng and Lapata , 2010b ; Bruni et al. , 2011 ; Silberer and Lapata , 2012 ; Johns and Jones , 2012 ; Bruni et al. , 2012a ; Bruni et al. , 2012b ; Silberer et al. , 2013 ) . Background val When objects are compared in terms of several dimensions , these dimensions can be weighed in different ways ( e.g. , ) . Background train Most DOP models , such as in Bod ( 1993 ) , Goodman ( 1996 ) , Bonnema et al. ( 1997 ) , and Collins & Duffy ( 2002 ) , use a likelihood criterion in defining the best parse tree : they take ( some notion of ) the most likely ( i.e. most probable ) tree as a candidate for the best tree of a sentence . Background train While IA is generally thought to be consistent with findings on human language production ( Hermann and Deutsch 1976 ; ; Pechmann 1989 ; Sonnenschein 1982 ) , the hypothesis that incrementality is a good model of human GRE seems unfalsifiable until a preference order is specified for the properties on which it operates . Background train Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) [ 18 ] , fly FlyBase , yeast Saccharomyces Genome Database ( SGD ) [ 20 ] , rat -- Rat Genome Database ( RGD ) [ 21 ] , worm -- WormBase [ 22 ] , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) [ 24 ] , and Enzyme Nomenclature Database ( ECNUM ) [ 25 , 26 ] . Uses train The RenTAL system automatically converts an FB-LTAG grammar into a strongly equivalent HPSG-style grammar ( ) . Background train The features can be easily obtained by modifying the TAT extraction algorithm described in ( ) . Extends train present preliminary work on the automatic extraction of subcategorization frames for Bulgarian from the BulTreeBank ( Simov , Popova , and Osenova 2002 ) . Background train The use of the web as a corpus for teaching and research on language has been proposed a number of times ( Kilgarriff , 2001 ; Robb , 2003 ; Rundell , 2000 ; Fletcher , 2001 , 2004b ) and received a special issue of the journal Computational Linguistics ( ) . Background val ASARES is based on a Machine Learning technique , Inductive Logic Programming ( ILP ) ( ) , which infers general morpho-syntactic patterns from a set of examples ( this set is noted E + hereafter ) and counter-examples ( E − ) of the elements one Background val built a corpus by iteratively searching Google for a small set of seed terms . Background train Expanding on a suggestion of , we classify verbs as Subject Equi , Object Equi , Subject Raising or Object Raising for each sense which has a predicate complement code associated with it . Extends train This Principle of Finitism is also assumed by Johnson-Laird ( 1983 ) , , Kamp ( 1981 ) , and implicitly or explicitly by almost all researchers in computational linguistics . CompareOrContrast train For more information on CATiB , see and Habash , Faraj , and Roth ( 2009 ) . Background train With the exception of ( Fung , 1995b ) , previous methods for automatically constructing statistical translation models begin by looking at word cooccurrence frequencies in bitexts ( ; Kumano & Hirakawa , 1994 ; Fung , 1995a ; Melamed , 1995 ) . Background val Experiments ( Section 5 ) show that forestbased extraction improves BLEU score by over 1 point on a state-of-the-art tree-to-string system ( Liu et al. , 2006 ; ) , which is also 0.5 points better than ( and twice as fast as ) extracting on 30-best parses . Extends train Our approach to extract and classify social events builds on our previous work ( ) , which in turn builds on work from the relation extraction community ( Nguyen et al. , 2009 ) . Extends train Previous sentiment-analysis work in different domains has considered inter-document similarity ( Agarwal and Bhattacharyya , 2005 ; ; Goldberg and Zhu , 2006 ) or explicit Background val For example , speech repairs , particle omission , and fillers can be dealt with in the framework of unification grammar ( ; Nakano and Shimazu , 1999 ) . Future train employed a Bayesian method to learn discontinuous SCFG rules . CompareOrContrast train Table 5 shows our mapping from publication type and MeSH headings to evidence grades based on principles defined in the Strength of Recommendations Taxonomy ( ) . Uses train We use the same splits as . Uses train These results are slightly worse than those obtained in previous studies using the same annotation scheme ( ) , but are still sat - CompareOrContrast train In , this flattening process is not part of the grammar . Background train MEDLINE , the authoritative repository of abstracts from the medical and biomedical primary literature maintained by the National Library of Medicine , provides the clinically relevant sources for answering physicians ' questions , and is commonly used in that capacity ( ; De Groote and Dorsch 2003 ) . Background train There is a rich literature on organization and lexical access of morphologically complex words where experiments have been conducted mainly for derivational suffixed words of English , Hebrew , Italian , French , Dutch , and few other languages ( Marslen-Wilson et al. , 2008 ; ; Grainger , et al. , 1991 ; Drews and Zwitserlood , 1995 ) . Background train He lists , classifies , and discusses various types of inference , by which he means , generally , `` the linguistic-logical notions of consequent and presupposition '' :112 ) have collected convincing evidence of the existence of language chunks -- real structures , not just orthographic conventions -- that are smaller than a discourse , larger than a sentence , generally composed of sentences , and recursive in nature ( like sentences ) . Background val For this research , we used a coreference resolution system ( ( ) ) that implements different sets of heuristics corresponding to various forms of coreference . Uses train The work that is most similar to ours is that of , who introduced the Constraint Driven Learning algorithm ( CODL ) . CompareOrContrast train For example , in ordinary HMM training , xi = E * and represents a completely hidden state sequence ( cfXXX , who allows any regular set ) , while yi is a single string representing a completely observed emission sequence .11 What to optimize ? Background val tried to solve the inflection prediction problem by simply building an SMT system for translating from stems to inflected forms . CompareOrContrast train As shown in ( ) , using this representation , a linear classifier can not distinguish sentences sampled from a trigram and real sentences . Motivation train Various approaches for computing semantic relatedness of words or concepts have been proposed , e.g. dictionary-based ( Lesk , 1986 ) , ontology-based ( Wu and Palmer , 1994 ; ) , information-based ( Resnik , 1995 ; Jiang and Conrath , 1997 ) or distributional ( Weeds and Weir , 2005 ) . Background train This framework , where the `` semantic load '' is spread more evenly throughout the lexicon to lexical categories not typically thought to bear semantic content , is essentially the model advocated by a ) , among many others . Background val also presents a similar method for the extraction of a TAG from the Penn Treebank . Background train To a first approximation , a CURRENT-FOCUS reaches only nodes that are c-commanded ( ) by its generator . Background train A previous work along this line is , which is based on weighted finite-state transducers ( FSTs ) . CompareOrContrast val Most approaches rely on VerbNet ( Kipper et al. , 2000 ) and FrameNet ( Baker et al. , 1998 ) to provide associations between verbs and semantic roles , that are then mapped onto the current instance , as shown by the systems competing in semantic role labelling competitions ( Carreras and Marquez , 2004 ; Carreras and Marquez , 2005 ) and also ( Gildea and Jurafsky , 2002 ; ; Shi and Mihalcea , 2005 ) . Background val describe an efficient algorithm for accomplishing this in which approximations to Pst ( TIS ) are computed in parallel for all ( new ) features ft by holding all weights in the existing model fixed and optimizing only over a8t . Background train As suggested in this can be done by looking up the ranks of each of the four given words ( i.e. the words occurring in a particular word equation ) within the association vector of a translation candidate , and by multiplying these ranks . Motivation val 32 In certain cases an extension of the constraint language with named disjunctions or contexted constraints ( ; Eisele and Dorre 1990 ; Griffith 1996 ) can be used to circumvent constraint propagation . Background train This article represents an extension of our previous work on unsupervised event coreference resolution ( Bejan et al. 2009 ; ) . Extends val These include devices such as interleaving the components ( McDonald 1983 ; Appelt 1983 ) , backtracking on failure ( Appelt 1985 ; Nogier 1989 ) , allowing the linguistic component to interrogate the planner ( ; Sondheimer and Nebel 1986 ) , and Hovy 's notion of restrictive ( i.e. , bottom-up ) planning ( Hovy 1988a , 1988c ) . Background train See , among others , ( ) . Background train In Table 2 , lem refers to the LTAG parser ( ) , ANSI C implementation of the two-phase parsing algorithm that performs the head corner parsing ( van Noord , 1994 ) without features ( phase 1 ) , and then executes feature unification ( phase 2 ) . CompareOrContrast train and Burkett et al. ( 2010 ) focused on joint parsing and alignment . CompareOrContrast train In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ ; Bresnan 2001 ; Dalrymple 2001 ] , head-driven phrase structure grammar [ HPSG ] [ Pollard and Sag 1994 ] , tree-adjoining grammar [ TAG ] [ Joshi 1988 ] , and combinatory categorial grammar [ CCG ] [ Ades and Steedman 1982 ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Background train For example , when books should n't be copied by hand any longer , authors took the advantage and start writing original books and evaluation -- i.e. literary criticism -- unlike in the previous times ( ) . Background train Furthermore , the availability of rich ontological resources , in the form of the Unified Medical Language System ( UMLS ) ( Lindberg et al. , 1993 ) , and the availability of software that leverages this knowledge -- MetaMap ( ) for concept identification and SemRep ( Rindflesch and Fiszman , 2003 ) for relation extraction -- provide a foundation for studying the role of semantics in various tasks . Background train Since the language generation module works in parallel with the language understanding module , utterance generation is possible even while the system is listening to user utterances and that utterance understanding is possible even while it is speaking ( a ) . Background train This idea was expanded to include nouns and their modifiers through verb nominalizations ( ; Quirk et al. , 1985 ) . Background val For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( Meyers et al. , 2000 ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( Richardson et al. , 2001 ) , ( Al-Adhaileh & Tang , 1999 ) . Background train define a kernel over parse trees and apply it to re-ranking the output of a parser , but the resulting feature space is restricted by the need to compute the kernel efficiently , and the results are not as good as Collins ' previous work on re-ranking using a finite set of features ( Collins , 2000 ) . Background train 9 We only use the minimal GHKM rules ( ) here to reduce the complexity of the sampler . Uses val FBLTAG ( ; Vijay-Shanker and Joshi , 1988 ) is an extension of the LTAG formalism . Background train In addition to headwords , dictionary search through the pronunciation field is available ; has merged information from the pronunciation and hyphenation fields , creating an enhanced phonological representation which allows access to entries by broad phonetic class and syllable structure ( Huttenlocher and Zue , 1983 ) . Uses train Research on shallow parsing was inspired by psycholinguistics arguments ( ) that suggest that in many scenarios ( e.g. , conversational ) full parsing is not a realistic strategy for sentence processing and analysis , and was further motivated by several arguments from a natural language engineering viewpoint . Background train Despite this , to date , there has been little work on corpus-based approaches to help-desk response automation ( notable exceptions are Carmel , Shtalhaim , and Soffer 2000 ; Lapalme and Kosseim 2003 ; ; Malik , Subramaniam , and Kaushik 2007 ) . Background train results are based on a corpus of movie subtitles ( Tiedemann 2007 ) , and are consequently shorter sentences , whereas the En → Es results are based on a corpus of parliamentary proceedings ( ) . Uses train A statistical technique which has recently become popular for NLP is Maximum Entropy/Minimum Divergence ( MEMD ) modeling ( ) . Uses train Our baseline coreference system uses the C4 .5 decision tree learner ( ) to acquire a classifier on the training texts for determining whether two NPs are coreferent . Uses train This alignment is obtained by following the same set of rules learned from the development dataset as in ( ) . Uses train Developed Systems Our developed system is built on the work by , using Constrained Latent Left-Linking Model ( CL3M ) as our mention-pair coreference model in the joint framework10 . Uses train The best performance on the Brown corpus , a 0.2 % error rate , was reported by , who trained a decision tree classifier on a 25-million-word corpus . CompareOrContrast train Most approaches rely on VerbNet ( Kipper et al. , 2000 ) and FrameNet ( Baker et al. , 1998 ) to provide associations between verbs and semantic roles , that are then mapped onto the current instance , as shown by the systems competing in semantic role labelling competitions ( Carreras and Marquez , 2004 ; Carreras and Marquez , 2005 ) and also ( Gildea and Jurafsky , 2002 ; Pradhan et al. , 2005 ; ) . Background train These include devices such as interleaving the components ( McDonald 1983 ; ) , backtracking on failure ( Appelt 1985 ; Nogier 1989 ) , allowing the linguistic component to interrogate the planner ( Mann 1983 ; Sondheimer and Nebel 1986 ) , and Hovy 's notion of restrictive ( i.e. , bottom-up ) planning ( Hovy 1988a , 1988c ) . Background val The system was trained on the Penn Treebank ( Marcus et al. , 1993 ) WSJ Sections 221 and tested on Section 23 ( Table 1 ) , same as used by Magerman ( 1995 ) , Collins ( 1997 ) , and , and became a common testbed . CompareOrContrast train was the first scholar who stressed the impact of the digital revolution to the medium of writing . Background train A similar problem is discussed in the psycholinguistics of interpretation ( ) : Interpretation is widely assumed to proceed incrementally , but vague descriptions resist strict incrementality , since an adjective in a vague description can only be fully interpreted when its comparison set is known . Background train Much of theoretical linguistics can be formulated in a very natural manner as stating correspondences ( translations ) between layers of representation structures ( ) . Background train In our previous work ( ) , we applied this method to a small subset of WordNet nouns and showed potential applicability . Extends train Much of the earlier work in anaphora resolution heavily exploited domain and linguistic knowledge ( Sidner 1979 ; Carter 1987 ; ; Carbonell and Brown 1988 ) , which was difficult both to represent and to process , and which required considerable human input . Background train the mention sub-type , which is a sub-category of the mention type ( ) ( e.g. OrgGovernmental , FacilityPath , etc. ) . Uses train For more details on the proprieties of SSTC , see . Background train This has been reported for other languages , too , dependent on the generality of the chosen approach ( J ¨ appinen and Niemist ¨ o , 1988 ; Choueka , 1990 ; ; Ekmekc ¸ ioglu et al. , 1995 ; Hedlund et al. , 2001 ; Pirkola , 2001 ) . Background train Specifically , we used Decision Graphs ( Oliver 1993 ) for Doc-Pred , and SVMs ( ) for Sent-Pred .11 Additionally , we used unigrams for clustering documents and sentences , and unigrams and bigrams for predicting document clusters and sentence clusters ( Sections 3.1.2 and 3.2.2 ) . Uses train To prepare SMT outputs for post-editing , the creators of the corpus used their own WMT10 system ( Potet et al. , 2010 ) , based on the Moses phrase-based decoder ( ) with dense features . Uses val To model o ( Li , S → T ) , o ( Ri , S → T ) , i.e. the reordering of the neighboring phrases of a function word , we employ the orientation model introduced by . Uses val The syntactic structures of the input data are produced by a parser with good coverage and detailed syntactic information , DIPETT ( ) . Uses train In particular , ( ) lists the converses of some 3 500 predicative nouns . Future train This is similar to the `` deletion '' strategy employed by , but we do it directly in the grammar . CompareOrContrast val Although this study falls under the general topic of discourse modeling , our work differs from previous attempts to characterize text in terms of domainindependent rhetorical elements ( McKeown , 1985 ; ) . CompareOrContrast train Many NLP applications require knowledge about semantic relatedness rather than just similarity ( ) . Background train Other work on modeling the meanings of verbs using video recognition has also begun showing great promise ( ; Regneri et al. , 2013 ) . Background train It has been shown ( ) that the subcategorization tendencies of verbs vary across linguistic domains . Motivation train In the system , we extract both the minimal GHKM rules ( ) , and the rules of SPMT Model 1 ( Galley et al. , 2006 ) with phrases up to length L = 5 on the source side . Uses train Note that this ensures that greater importance is attributed to longer chunks , as is usual in most EBMT systems ( cfXXX Sato and Nagao 1990 ; ; Carl 1999 ) .7 As an example , consider the translation into French of the house collapsed . Background train One of the better-known approaches is described in , which suggested that abbreviations first be extracted from a corpus using abbreviation-guessing heuristics akin to those described in Section 6 and then reused in further processing . Background train Unlike the models proposed by b ) , this model is symmetric , because both word bags are generated together from a joint probability distribution . CompareOrContrast val We apply two different priming experiments namely , the cross modal priming and masked priming experiment discussed in ( Forster and Davis , 1984 ; Rastle et al. , 2000 ; ; Marslen-Wilson et al. , 2008 ) for Bangla morphologically complex words . Uses train ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ; ) proposed other optimization objectives by introducing a margin-based and ranking-based indirect loss functions . Background train Our work extends directions taken in systems such as Ariane ( Vauquois and Boitet , 1985 ) , FoG ( Kittredge and Polguere , 1991 ) , JOYCE ( Rambow and ) , and LFS ( Iordanskaja et al. , 1992 ) . Extends train Since the arguments can provide useful semantic information , the SRL is crucial to many natural language processing tasks , such as Question and Answering ( Narayanan and Harabagiu 2004 ) , Information Extraction ( Surdeanu et al. 2003 ) , and Machine Translation ( ) . Background train Such systems extract information from some types of syntactic units ( clauses in ( Fillmore and Atkins , 1998 ; Gildea and Jurafsky , 2002 ; Hull and Gomez , 1996 ) ; noun phrases in ( Hull and Gomez , 1996 ; ) ) . Background val Association Norms ( AN ) is a collection of association norms collected by Schulte im . Uses train Stanford University is developing the English Resource Grammar , an HPSG grammar for English , as a part of the Linguistic Grammars Online ( LinGO ) project ( ) . Background train The changes made were inspired by those described in , page 75 ) . Motivation train The grammar code system used in LDOCE is based quite closely on the descriptive grammatical framework of , 1985 ) . Extends train Our task was made possible by the fact that while far from being a database in the accepted sense of the word , the LDOCE typesetting tape is the only truly computerised dictionary of English ( ) . Background train A detailed description of the kinds of expectation mechanisms appearing in these systems appears in . Background train Using the GHKM algorithm ( ) , we can get two different STSG derivations from the two U-trees based on the fixed word alignment . Uses train The need for information systems to support physicians at the point of care has been well studied ( Covell et al. , 1985 ; Gorman et al. , 1994 ; ) . Background train Task properties Determining whether or not a speaker supports a proposal falls within the realm of sentiment analysis , an extremely active research area devoted to the computational treatment of subjective or opinion-oriented language ( early work includes Wiebe and Rapaport ( 1988 ) , Hearst ( 1992 ) , , and Wiebe ( 1994 ) ; see Esuli ( 2006 ) for an active bibliography ) . Background val Table look-up using an explicit translation lexicon is sufficient and preferable for many multilingual NLP applications , including `` crummy '' MT on the World Wide Web ( Church & Hovy , 1993 ) , certain machine-assisted translation tools ( e.g. ( Macklovitch , 1994 ; Melamed , 1996b ) ) , concordancing for bilingual lexicography ( ; Gale & Church , 1991 ) , computerassisted language learning , corpus linguistics ( Melby . Background val Our HDP extension is also inspired from the Bayesian model proposed by . Motivation train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( Bikel et al. , 1997 ; Miller et al. , 1998 ; ; Mikheev et al. , 1999 ; Soon et al. , 2001 ; Ng and Cardie , 2002 ; Florian et al. , 2004 ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train As has been previously observed and exploited in the NLP literature ( Pang and Lee , 2004 ; ; Barzilay and Lapata , 2005 ) , the above optimization function , unlike many others that have been proposed for graph or set partitioning , can be solved exactly in an provably efficient manner via methods for finding minimum cuts in graphs . CompareOrContrast train To combine the phrasal matching scores obtained at each n-gram level , and optimize their relative weights , we trained a Support Vector Machine classifier , SVMlight ( ) , using each score as a feature . Uses train Against the background of a growing interest in multilingual NLP , multilingual anaphora / coreference resolution has gained considerable momentum in recent years ( Aone and McKee 1993 ; Azzam , Humphreys , and Gaizauskas 1998 ; Harabagiu and Maiorano 2000 ; ; Mitkov 1999 ; Mitkov and Stys 1997 ; Mitkov , Belguith , and Stys 1998 ) . Background train For an overview of systems designed to answer open-domain factoid questions , the TREC QA track overview papers are a good place to start ( ) . Background train Another interesting extension is to broaden the definition of a `` word '' to include multi-word lexical units ( ) . Future train For example , 10 million words of the American National Corpus ( ) will have manually corrected POS tags , a tenfold increase over the Penn Treebank ( Marcus et al. , 1993 ) , currently used for training POS taggers . Background train From an IR view , a lot of specialized research has already been carried out for medical applications , with emphasis on the lexico-semantic aspects of dederivation and decomposition ( Pacak et al. , 1980 ; ; Wolff , 1984 ; Wingert , 1985 ; Dujols et al. , 1991 ; Baud et al. , 1998 ) . Background train This approach has occasionally been taken , as in Kantrowitz and Bates ( 1992 ) and Danlos ( 1987 ) and , at least implicitly , in and Delin et al. ( 1994 ) ; however , under this approach , all of the flexibility and simplicity of modular design is lost . Background val Typical examples are Bulgarian ( Simov et al. , 2005 ; Simov and Osenova , 2003 ) , Chinese ( ) , Danish ( Kromann , 2003 ) , and Swedish ( Nilsson et al. , 2005 ) . Background train In this paper , we use the Constrained Latent Left-Linking Model ( CL3M ) described in in our experiments . Uses train We use a standard split of 268 training documents , 68 development documents , and 106 testing documents ( ; Bengtson and Roth , 2008 ) . Uses train Aside from the extraction of theory-neutral subcategorization lexicons , there has also been work in the automatic construction of lexical resources which comply with the principles of particular linguistic theories such as LTAG , CCG , and HPSG ( Chen and Vijay-Shanker 2000 ; ; Hockenmaier , Bierner , and Baldridge 2004 ; Nakanishi , Miyao , and Tsujii 2004 ) . Background train An alternative representation based on is presented in Selkirk ( 1984 ) , which contends that prosody , including prosodic phrasing , is more properly represented as a grid instead of a tree . CompareOrContrast train For the joint segmentation and POS-tagging task , we present a novel solution using the framework in this article , and show that it gives comparable accuracies to our previous work ( a ) , while being more than an order of magnitude faster . CompareOrContrast val According to the data available from 1990 U.S. Census Bureau , only 90,000 different names are shared by 100 million people ( ) . Background train Since then this idea has been applied to several tasks , including word sense disambiguation ( ) and named-entity recognition ( Cucerzan and Yarowsky 1999 ) . Background val • History-based feature models for predicting the next parser action ( ) . Uses train For each co-occurring pair of word types u and v , these likelihoods are initially set proportional to their co-occurrence frequency ( „ , v ) and inversely proportional to their marginal frequencies n ( u ) and n ( v ) 1 , following ( ) 2 . Uses train However , rather than output this wrong translation directly , we use a post hoc validation and ( if required ) correction process based on . Uses train Most approaches rely on VerbNet ( Kipper et al. , 2000 ) and FrameNet ( Baker et al. , 1998 ) to provide associations between verbs and semantic roles , that are then mapped onto the current instance , as shown by the systems competing in semantic role labelling competitions ( Carreras and Marquez , 2004 ; Carreras and Marquez , 2005 ) and also ( Gildea and Jurafsky , 2002 ; Pradhan et al. , 2005 ; ) . Background val The powerful mechanism of lexical rules ( ) has been used in many natural language processing systems . Background train The version proposed here combines a basic insight from Lewin ( 1990 ) with higher-order unification to give an analysis that has a strong resemblance to that proposed in , 1991 ) , with some differences that are commented on below . CompareOrContrast train presented a perceptron-based algorithm for learning the phrase-translation parameters in a statistical machine translation system . CompareOrContrast train NLG has to do more than select a distinguishing description ( i.e. , one that unambiguously denotes its referent ; ) : The selected expression should also be felicitous . Background train , 1997 ) conducts some small experiments using his METLA system to show the viability of this approach for English − > French and English − > Urdu . Background train This idea of preserving properties can be considered an instance of the well-known frame problem in AT ( ) , and we will therefore refer to the specifications left implicit by the linguist as the frame specification , or simply frame , of a lexical rule . CompareOrContrast train All experiments have been performed using MaltParser ( ) , version 0.4 , which is made available together with the suite of programs used for preand post-processing .1 Uses val One area of current interest concerns the left-to-right arrangement of premodifying adjectives within an NP ( e.g. , Shaw and Hatzivassiloglou 1999 ; ) . Background train Better results would be expected by combining the PCFG-LA parser with discriminative reranking approaches ( ; Huang , 2008 ) for self training . Future train Some researchers ( ; Nguyen and Cao , 2008 ) have explored the use of Wikipedia information to improve the disambiguation process . Background train Others have applied the NLP technologies of near-duplicate detection and topic-based text categorization to politically oriented text ( ; Purpura and Hillard , 2006 ) . Background val We chose to follow and split the sentences evenly to facilitate further comparison . Uses train That is , a document that contains terms al , a2 and a3 may be ranked higher than a document which contains terms al and b.f. However , the second document is more likely to be relevant since correct translations of the query terms are more likely to co-occur ( ) . Background train The last years have seen considerable advances in the field of anaphora resolution , but a number of outstanding issues either remain unsolved or need more attention and , as a consequence , represent major challenges to the further development of the field ( a ) . Future train Using an accumulator passing technique ( ) , we ensure that upon execution of a call to the interaction predicate q_1 a new lexical entry is derived as the result of successive application of a number of lexical rules . Uses train However , most strategies are based on `` internal '' or `` external methods '' ( ) , i.e. methods that rely on the form of terms or on the information gathered from contexts . CompareOrContrast train With the exception of ( Fung , 1995b ) , previous methods for automatically constructing statistical translation models begin by looking at word cooccurrence frequencies in bitexts ( Gale & Church , 1991 ; Kumano & Hirakawa , 1994 ; Fung , 1995a ; ) . Background train A substring in the sentence that corresponds to a node in the representation tree is denoted by assigning the interval of the substring to SNODE of 2 These definitions are based on the discussion in ( ) and Boitet & Zaharin ( 1988 ) . Uses train The Web People Search task , as defined in the first WePS evaluation campaign ( ) , consists of grouping search results for a given name according to the different people that share it . Background train Although this is only true in cases where y occurs in an upward monotone context ( ) , in practice genuine contradictions between y-values sharing a meronym relationship are extremely rare . Motivation train We use the same data setting with , however a bit different from Xue and Palmer ( 2005 ) . Uses train Erk ( 2007 ) compared a number of techniques for creating similar-word sets and found that both the Jaccard coefficient and a ) 's information-theoretic metric work best . Background train We previously showed that incorporating this intuition into a Bayesian prior can help train a CCG supertagger ( ) . Extends val 4 To turn this likelihood into a certainty , one can add a test at the end of the algorithm , which adds a type-related property if none is present yet ( cfXXX , ) . Background train For descriptions of SMT systems see for example ( Germann et al. , 2001 ; ; Tillmann and Ney , 2002 ; Vogel et al. , 2000 ; Wang and Waibel , 1997 ) . Background train Such tools make it easy to run most current approaches to statistical markup , chunking , normalization , segmentation , alignment , and noisy-channel decoding , ' including classic models for speech recognition ( ) and machine translation ( Knight and Al-Onaizan , 1998 ) . Background train It is analogous to the step in other translation model induction algorithms that sets all probabilities below a certain threshold to negligible values ( Brown et al. , 1990 ; Dagan et al. , 1993 ; ) . CompareOrContrast train The ConTroll grammar development system as described in ( b ) implements the above mentioned techniques for compiling an HPSG theory into typed feature grammars . Background train After calculating the raw score of each sentence , we use a modified version of the Adaptive Greedy Algorithm by to penalize redundant sentences in cohesive clusters . Uses train If each word 's translation is treated as a sense tag ( ) , then `` translational '' collocations have the unique property that the collocate and the word sense are one and the same ! Uses train The values of a vector correspond to the presence or absence of each ( lemmatized ) corpus word in the document in question ( after removing stop-words and words with very low frequency ) .4 The predictive model is a Decision Graph ( ) , which , like Snob , is based on the MML principle . Uses train Therefore , in order to be able to incorporate long-range dependencies in our models , we chose to adopt a re-ranking approach ( ) , which selects from likely assignments generated by a model which makes stronger independence assumptions . Uses val However , it is possible to think about constraining linguistic or logical predicates by simulating physical experiences ( cfXXX ) . Background train One , the VOYAGER domain ( ) , answers questions about places of interest in an urban area , in our case , the vicinity of MIT and Harvard University . Uses train Our task is closer to the work of , who looked at the problem of intellectual attribution in scientific texts . CompareOrContrast val In the seminal work by , similarity judgments were obtained from 51 test subjects on 65 noun pairs written on paper cards . Background train ( ; Blunsom et al. , 2008 ) used maximum likelihood estimation to learn weights for MT. ( Och , 2003 ; Moore and Quirk , 2008 ; Zhao and Chen , 2009 ; Galley and Quirk , 2011 ) employed an evaluation metric as a loss function and directly optimized it . CompareOrContrast val , p. 14 ) writes `` it would be perverse not to take as a working assumption that language is a relatively efficient and accurate encoding of the information it conveys . '' CompareOrContrast train See also ( ; Naish , 1986 ) . Background train Michiels ( 1982 ) and provide a more detailed analysis of the information encoded by the LDOCE grammar codes and discuss their efficacy as a system of linguistic description . Background val We parsed the 3 GB AQUAINT corpus ( Voorhees , 2002 ) using Minipar ( b ) , and collected verb-object and verb-subject frequencies , building an empirical MI model from this data . Uses train More specifically , we use LIBSVM ( ) with a quadratic kernel K ( xZ , xj ) = ( - yxT xj + r ) 2 and the built-in one-versus-all strategy for multi-class classification . Uses train It is wasteful to compute ti as suggested earlier , by minimizing ( cxxi ) of o ( yixE ) , since then the real work is done by an c-closure step ( ) that implements the all-pairs version of algebraic path , whereas all we need is the single-source version . Background train positional features that have been employed by highwe can see , the baseline achieves an F-measure of performing resolvers such as 57.0 and a resolution accuracy of 48.4 . CompareOrContrast train Such a component would serve as the first stage of a clinical question answering system ( Demner-Fushman and Lin , 2005 ) or summarization system ( ) . Future train Briscoe and Carroll ( 1997 ) predefine 163 verbal subcategorization frames , obtained by manually merging the classes exemplified in the COMLEX ( MacLeod , Grishman , and Meyers 1994 ) and ANLT ( ) dictionaries and adding around 30 frames found by manual inspection . Background train The idea resurfaced forcefully at several points in the more recent history of linguistic research ( Tesni`ere , 1959 ; ; Fillmore , 1968 ) . Background train It is therefore no surprise that early attempts at response automation were knowledge-driven ( Barr and Tessler 1995 ; Watson 1997 ; ) . Background val For example , the suite of LT tools ( Mikheev et al. , 1999 ; ) perform tokenization , tagging and chunking on XML marked-up text directly . Background train It is only recently that the web name ambiguity has been approached as a separate problem and defined as an NLP task Web People Search on its own ( Artiles et al. , 2005 ; ) . Uses train To retrieve translation examples for a test sentence , ( ) defined a metric based on the combination of edit distance and TF-IDF ( Manning and Sch ¨ utze , 1999 ) as follows : Uses train By using the EM algorithm ( ) , they can guarantee convergence towards the globally optimum parameter set . Background train • use of low level knowledge from the speech recognition phase , • use of high level knowledge about the domain in particular and the dialogue task in general , • a `` continue '' facility and an `` auto-loop '' facility as described by , • a `` conditioning '' facility as described by Fink et al. ( 1985 ) , • implementation of new types of paraphrasing , • checking a larger environment in the expectation acquisition algorithm when deciding if an incoming sentence is the same or similar to one already seen , and • examining inter-speaker dialogue patterns . Future train Previous work has developed various approaches for grounded semantics mainly for the reference resolution task , i.e. , identifying visual objects in the environment given language descriptions ( Dhande , 2003 ; Gorniak and Roy , 2004 ; Tenbrink and Moratz , 2003 ; Siebert and Schlangen , 2008 ; ) . Extends val In most recent research , NEs ( person , location and organisations ) are extracted from the text and used as a source of evidence to calculate the similarity between documents - see for instance ( Blume , 2005 ; Chen and Martin , 2007 ; ; Kalashnikov et al. , 2007 ) . Background train To provide the required configurability in the static version of the code we will use policy templates ( ) , and for the dynamic version we will use configuration classes . Uses train demonstrates a technique for segmenting Arabic text and uses it as a morphological processing step in machine translation . Uses train In informal experiments described elsewhere ( ) , I found that the G2 statistic suggested by Dunning ( 1993 ) slightly outperforms 02 . Extends train • language learning ( Green 1979 ; Mori and Moeser 1983 ; Morgan , Meier , and Newport 1989 ) • monolingual grammar induction ( Juola 1998 ) • grammar optimization ( ) • insights into universal grammar ( Juola 1998 ) • machine translation ( Juola 1994 , 1997 ; Veale and Way 1997 ; Gough , Way , and Hearne 2002 ) Background train A recent study by also investigates the task of training parsers to improve MT reordering . CompareOrContrast train Due to their remarkable ability to incorporate context structure information and long distance reordering into the translation process , tree-based translation models have shown promising progress in improving translation quality ( Liu et al. , 2006 , 2009 ; Quirk et al. , 2005 ; , 2006 ; Marcu et al. , 2006 ; Shen et al. , 2008 ; Zhang et al. , 2011b ) . Background train 15 show that the question of whether the application criterion of lexical rules should be a subsumption or a unification test is an important question deserving of more attention . Background val Latent Dirichlet Allocation ( ) , or LDA , is an unsupervised Bayesian probabilistic model of text documents . Background train Regarding future work , there are many research line that may be followed : i ) Capturing more features by employing external knowledge such as ontological , lexical resource or WordNet-based features ( Basili et al. , 2005a ; Basili et al. , 2005b ; Bloehdorn et al. , 2006 ; Bloehdorn and Moschitti , 2007 ) or shallow semantic trees , ( Giuglea and Moschitti , 2004 ; ; Moschitti and Bejan , 2004 ; Moschitti et al. , 2007 ; Moschitti , 2008 ; Moschitti et al. , 2008 ) . Future train Third , the paradigm of evidence-based medicine ( ) provides a task-based model of the clinical information-seeking process . Background train introduced factored SMT . CompareOrContrast train Both systems are built around from the maximum-entropy technique ( ) . Uses train We chose the adjectives as follows : we first compiled a list of all the polysemous adjectives mentioned in the lexical semantics literature ( Vendler , 1968 ; ) . Uses train Most web-derived corpora have exploited raw text or HTML pages , so efforts have focussed on boilerplate removal and cleanup of these formats with tools like Hyppia-BTE , Tidy and Parcels3 ( ) . Background train Undesirable consequences of this fact have been termed `` label bias '' ( ) . Background train For the full parser , we use the one developed by Michael Collins ( ; Collins , 1997 ) -- one of the most accurate full parsers around . Uses train Some researchers , however , including , train on predicted feature values instead . CompareOrContrast train Their computational significance arises from the issue of their storage in lexical resources like WordNet ( ) and raises the questions like , how to store morphologically complex words , in a lexical resource like WordNet keeping in mind the storage and access efficiency . Background train Fortunately , there exists a compact PCFG-reduction of DOP1 that generates the same trees with the same probabilities , as shown by , 2002 ) . Background train claims that prosodic phrase boundaries will co-occur with grammatical functions such as subject , predicate , modifier , and adjunct . Background train We rephrase the method of as follows : First , we construct the approximating finite automaton according to the unparameterized RTN method above . Uses train 6The analysis is reminiscent of the treatment of coordination in the Collins parser ( ) . CompareOrContrast val The work is carried out in order to assist terminographers in the enrichment of a dictionary on computing that includes collocational information ( ) . Motivation train CornmandTalk ( ) , Circuit Fix-It Shop ( Smith , 1997 ) and TRAINS-96 ( Traum and Allen , 1994 ; Traum and Andersen , 1999 ) are spoken language systems but they interface to simulation or help facilities rather than semi-autonomous agents . Background train The annotation procedure is dependent on locating the head daughter , for which an amended version of is used . Uses val ; Oliva 1994 ; Frank 1994 ; Opalka 1995 ; Sanfilippo 1995 ) . CompareOrContrast train For instance , recently wrote : `` To our knowledge , learning algorithms , although promising , have not ( yet ) reached the level of rule sets developed by humans '' ( p. 520 ) . Background train Much previous work looks at the impact of using source side information ( i.e. , feature functions on the aligned English ) , such as those of Avramidis and Koehn ( 2008 ) , and others . CompareOrContrast train We use an in-house statistical tagger ( based on ( ) ) to tag the text in which the unknown word occurs . Uses train Based on a computational grammar that associates natural language expressions with both a syntactic and a semantic representation , a paraphrastic gram ` As we shall briefly discuss in section 4 , the grammar is developed with the help of a meta-grammar ( ) thus ensuring an additional level of abstraction . Uses train There are very few reported attempts at corpus-based automation of help-desk responses ( Carmel , Shtalhaim , and Soffer 2000 ; ; Bickel and Scheffer 2004 ; Malik , Subramaniam , and Kaushik 2007 ) . CompareOrContrast train Both tasks are performed with a statistical framework : the mention detection system is similar to the one presented in ( ) and the coreference resolution system is similar to the one described in ( Luo et al. , 2004 ) . CompareOrContrast train The disambiguation of person names in Web results is usually compared to two other Natural Language Processing tasks : Word Sense Disambiguation ( WSD ) ( Agirre and Edmonds , 2006 ) and Cross-document Coreference ( CDC ) ( ) . Background train Our own work ( ) extends the first idea to paraphrase fragment extraction on monolingual parallel and comparable corpora . Extends train They proved to be useful in a number of NLP applications such as natural language generation ( ) , multidocument summarization ( McKeown et al. , 2002 ) , automatic evaluation of MT ( Denkowski and Lavie , 2010 ) , and TE ( Dinu and Wang , 2009 ) . Motivation train Current state-of-the-art statistical parsers ( ; Charniak 2000 ) are all trained on large annotated corpora such as the Penn Treebank ( Marcus , Santorini , and Marcinkiewicz 1993 ) . Background train Although the parser only derives projective graphs , the fact that graphs are labeled allows non-projective dependencies to be captured using the pseudoprojective approach of . Background train Riehemann 1993 ; Oliva 1994 ; ; Opalka 1995 ; Sanfilippo 1995 ) . CompareOrContrast train The idea resurfaced forcefully at several points in the more recent history of linguistic research ( Tesni`ere , 1959 ; Gruber , 1965 ; ) . Background train A detailed introduction to the SBD problem can be found in . Background train Task properties Determining whether or not a speaker supports a proposal falls within the realm of sentiment analysis , an extremely active research area devoted to the computational treatment of subjective or opinion-oriented language ( early work includes Wiebe and Rapaport ( 1988 ) , Hearst ( 1992 ) , Sack ( 1994 ) , and ; see Esuli ( 2006 ) for an active bibliography ) . Background train The BEETLE II system architecture is designed to overcome these limitations ( ) . Background train For the evaluation of the results we use the BLEU score ( ) . Uses train There are very few reported attempts at corpus-based automation of help-desk responses ( Carmel , Shtalhaim , and Soffer 2000 ; Lapalme and Kosseim 2003 ; ; Malik , Subramaniam , and Kaushik 2007 ) . CompareOrContrast train For an introduction to maximum entropy modeling and training procedures , the reader is referred to the corresponding literature , for instance ( ) or ( Ratnaparkhi , 1997 ) . Background train This design idea was adopted from TANKA ( b ) . Uses train The problem of handling ill-formed input has been studied by , Granger ( 1983 ) , Jensen et al. ( 1983 ) , Kwasny and Sondheimer ( 1981 ) , Riesbeck and Schank ( 1976 ) , Thompson ( 1980 ) , Weischedel and Black ( 1980 ) , and Weischedel and Sondheimer ( 1983 ) . CompareOrContrast train In addition to its explanatory capacity , this symbolic acquisition technique has obtained good results for other acquisition tasks when compared to existing statistical techniques ( ) . Motivation val The context of a current token ti is clearly one of the most important features in predicting whether ti is a mention or not ( ) . Background train Features using the word context ( left and right tokens ) have been shown to be very helpful in coreference resolution ( ) . Uses train Here , the PET and GR kernel perform similar : this is different from the results of ( ) where GR performed much worse than PET for ACE data . CompareOrContrast val In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ Kaplan and Bresnan 1982 ; Bresnan 2001 ; ] , head-driven phrase structure grammar [ HPSG ] [ Pollard and Sag 1994 ] , tree-adjoining grammar [ TAG ] [ Joshi 1988 ] , and combinatory categorial grammar [ CCG ] [ Ades and Steedman 1982 ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Background train Some works abstract perception via the usage of symbolic logic representations ( Chen et al. , 2010 ; Chen and Mooney , 2011 ; ; Artzi and Zettlemoyer , 2013 ) , while others choose to employ concepts elicited from psycholinguistic and cognition studies . Background val This is where robust syntactic systems like SATZ ( Palmer and Hearst 1997 ) or the POS tagger reported in , which do not heavily rely on word capitalization and are not sensitive to document length , have an advantage . CompareOrContrast train Our work is inspired by the latent left-linking model in and the ILP formulation from Chang et al. ( 2011 ) . Background val For example , proves that Chinese numerals such as wu zhao zhao zhao zhao zhao wu zhao zhao zhao zhao wu zhao zhao zhao wu zhao zhao wu zhao , for the number 5000000000000000005000000000000005000000000005000000005000 , are not context-free , which implies that Chinese is not a context-free language and thus might parse in exponential worst-case time . Background train also note that the applicability of paraphrases is strongly influenced by context . Background train Promising features for classification include part of speech , frequency of co-occurrence , relative word position , and translational entropy ( ) . Future train For descriptions of SMT systems see for example ( ; Och et al. , 1999 ; Tillmann and Ney , 2002 ; Vogel et al. , 2000 ; Wang and Waibel , 1997 ) . Background train like information extraction ( ) and textual entailment ( Berant et al. , 2010 ) . Background train In a final processing stage , we generalize over the marker lexicon following a process found in . Uses val One approach to this problem consists in defining , within the Cut-free atomic-id space , normal form derivations in which the succession of rule application is regulated ( , Hepple 1990 , Hendriks 1993 ) . Background train The SPR uses rules automatically learned from training data , using techniques similar to ( ; Freund et al. , 1998 ) . CompareOrContrast train have previously examined the task of categorizing sentences in medical abstracts using supervised discriminative machine learning techniques . CompareOrContrast train For complementing this database and for converse constructions , the LADL tables ( ) can furthermore be resorted to , which list detailed syntactico-semantic descriptions for 5 000 verbs and 25 000 verbal expressions . Future train These operations are not domain-specific and are similar to those of previous aggregation components ( Rambow and Korelsky ,1992 ; ; Danlos , 2000 ) , although the various MERGE operations are , to our knowledge , novel in this form . Background train Following construction of the marker lexicon , the ( source , target ) chunks are generalized further using a methodology based on to permit a limited form of insertion in the translation process . Uses train ment ( Sarkar and Wintner , 1999 ; ; Makino et al. , 1998 ) . Background train The ability to explicitly identify these sections in unstructured text could play an important role in applications such as document summarization ( Teufel and Moens , 2000 ) , information retrieval ( Tbahriti et al. , 2005 ) , information extraction ( ) , and question answering . Background train Recent developments in linguistics , and especially on grammatical theory -- for example , Generalised Phrase Structure Grammar ( GPSG ) ( Gazdar et al. , 1985 ) , Lexical Functional Grammar ( LFG ) ( Kaplan and Bresnan , 1982 ) -- and on natural language parsing frameworks -- for example , Functional Unification Grammar ( FUG ) ( Kay , 1984a ) , PATR-II ( ) -- make it feasible to consider the implementation of efficient systems for the syntactic analysis of substantial fragments of natural language . Background train The latter question is tackled by applicationspecific evaluation , where a measure is tested within the framework of a certain application , e.g. word sense disambiguation ( ) or malapropism detection ( Budanitsky and Hirst , 2006 ) . Background train argue for application-specific evaluation of similarity measures , because measures are always used for some task . Background train The first direct application of parse forest in translation is our previous work ( ) which translates a packed forest from a parser ; it is also the base system in our experiments ( see below ) . Extends train 4 This interpretation of the signature is sometimes referred to as closed world ( Gerdemann and ; Gerdemann 1995 ) . Background train In their Gaijin system , give a result of 63 % accurate translations obtained for English − > German on a test set of 791 sentences from CorelDRAW manuals . CompareOrContrast train This method can be generalized , inspired by , who derive N-gram probabilities from stochastic context-free grammars . Background train The system was trained on the Penn Treebank ( Marcus et al. , 1993 ) WSJ Sections 221 and tested on Section 23 ( Table 1 ) , same as used by Magerman ( 1995 ) , , and Ratnaparkhi ( 1997 ) , and became a common testbed . CompareOrContrast val The Nash arbitration plan , for example , would allow a doubly graded description whenever the product of the Values for the referent r exceeds that of all distractors ( Nash 1950 ; cfXXX ; Thorisson 1994 , for other plans ) . Background train The paraphrase dictionary that we use was generated for us by Chris Callison-Burch , using the technique described in , which exploits a parallel corpus and methods developed for statistical machine translation . Uses train To name a few examples , and Socher et al. ( 2013 ) show how semantic information from text can be used to improve zero-shot classification ( i.e. , classifying never-before-seen objects ) , and Motwani and Mooney ( 2012 ) show that verb clusters can be used to improve activity recognition in videos . Background val This approach has now gained wide usage , as exemplified by the work of , 1999 ) , Charniak ( 1996 , 1997 ) , Johnson ( 1998 ) , Chiang ( 2000 ) , and many others . Motivation train For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( Meyers et al. , 2000 ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( Sato , 1991 ) , ( ) , ( Al-Adhaileh & Tang , 1999 ) . Background train Increasingly , corpus researchers are tapping the Web to overcome the sparse data problem ( ) . Background val As such it resembles the parser of the grammar development system Attribute Language Engine ( ALE ) of ( ) . CompareOrContrast train de URL : http://www.sfs.nphil.uni-tuebingen.de/sfb / b4home.html 1 This is , for example , the case for all proposals working with verbal lexical entries that raise the arguments of a verbal complement ( Hinrichs and Nakazawa 1989 ) that also use lexical rules such as the Complement Extraction Lexical Rule ( Pollard and Sag 1994 ) or the Complement Cliticization Lexical Rule ( ) to operate on those raised elements . Background train Research that is more similar in goal to that outlined in this paper is Vosse ( ) . CompareOrContrast train pointed out that little attention had been paid in the named-entity recognition field to the discourse properties of proper names . Uses train Table 5 shows our mapping from publication type and MeSH headings to evidence grades based on principles defined in the Strength of Recommendations Taxonomy ( ) . Uses train While we have observed reasonable results with both G2 and Fisher 's exact test , we have not yet discussed how these results compare to the results that can be obtained with a technique commonly used in corpus linguistics based on the mutual information ( MI ) measure ( ) : Uses val Following , we consider an anaphoric reference , NPi , correctly resolved if NPi and its closest antecedent are in the same coreference chain in the resulting partition . Uses train From an IR view , a lot of specialized research has already been carried out for medical applications , with emphasis on the lexico-semantic aspects of dederivation and decomposition ( ; Norton and Pacak , 1983 ; Wolff , 1984 ; Wingert , 1985 ; Dujols et al. , 1991 ; Baud et al. , 1998 ) . Background train The Gsearch system ( ) also selects sentences by syntactic criteria from large on-line text collections . Background train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background train See also the work of , which considers computer-based pronunciation by analogy but does not mention the possible application to text-to-speech synthesis . Background train The resulting list of POS-tagged lemmas is weighted using the SMART ` ltc ' 8 tf.idf-weighting scheme ( ) . Uses train The psycholinguistic studies of Martin ( 1970 ) , , Hillinger et al. ( 1976 ) , Grosjean et al. ( 1979 ) , Dommergues and Grosjean ( 1983 ) , and Gee and Grosjean ( 1983 ) , responding to the idea of readjusted syntax as the source of prosodic phrasing , show that grammatical structure , even if readjusted , is not in itself a reliable predictor of prosodic phrasing : mismatches between syntax and prosody occur often and systematically , and can be related to specific nonsyntactic factors such as length and word frequency . Background train Some methods are based on likelihood ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) , error rate ( ; Zhao and Chen , 2009 ; Pauls et al. , 2009 ; Galley and Quirk , 2011 ) , margin ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ) and ranking ( Hopkins and May , 2011 ) , and among which minimum error rate training ( MERT ) ( Och , 2003 ) is the most popular one . Background val Numerous previous pseudodisambiguation evaluations only include arguments that occur between 30 and 3000 times ( Erk , 2007 ; ; Rooth et al. , 1999 ) . CompareOrContrast train • cross-language information retrieval ( e.g. , McCarley 1999 ) , • multilingual document filtering ( e.g. , Oard 1997 ) , • computer-assisted language learning ( e.g. , ) , • certain machine-assisted translation tools ( e.g. , Macklovitch 1994 ; Melamed 1996a ) , • concordancing for bilingual lexicography ( e.g. , Catizone , Russell , and Warwick 1989 ; Gale and Church 1991 ) , Background train One would think that the type information ti , which is more specific than that 16 A linguistic example based on the signature given by would be a lexical rule deriving predicative signs from nonpredicative ones , i.e. , changing the PRD value of substantive signs from -- to - F , much like the lexical rule for NPs given by Pollard and Sag ( 1994 , p. 360 , fn . Background train • language learning ( Green 1979 ; ; Morgan , Meier , and Newport 1989 ) • monolingual grammar induction ( Juola 1998 ) • grammar optimization ( Juola 1994 ) • insights into universal grammar ( Juola 1998 ) • machine translation ( Juola 1994 , 1997 ; Veale and Way 1997 ; Gough , Way , and Hearne 2002 ) Background train Notable early papers on graph-based semisupervised learning include , Bansal et al. ( 2002 ) , Kondor and Lafferty ( 2002 ) , and Joachims ( 2003 ) . Background train As a result , researchers have re-adopted the once-popular knowledge-rich approach , investigating a variety of semantic knowledge sources for common noun resolution , such as the semantic relations between two NPs ( e.g. , Ji et al. ( 2005 ) ) , their semantic similarity as computed using WordNet ( e.g. , Poesio et al. ( 2004 ) ) or Wikipedia ( Ponzetto and Strube , 2006 ) , and the contextual role played by an NP ( see ) . Background train An example of psycholinguistically oriented research work can be found in . Background val Alternatively , we may think of user-centered comparative studies ( ) . Future train , p. 112 ) , for example , bemoans the fact that his `` theory lacks a world knowledge component , a mental ` encyclopedia , ' which could be invoked to generate inferences ... '' . Background train However , more recent work ( Cahill et al. 2002 ; Cahill , McCarthy , et al. 2004 ) has presented efforts in evolving and scaling up annotation techniques to the Penn-II Treebank ( ) , containing more than 1,000,000 words and 49,000 sentences . Background train The current system learns finite state flowcharts whereas typical learning systems usually acquire coefficient values as in , assertional statements as in Michalski ( 1980 ) , or semantic nets as in Winston ( 1975 ) . CompareOrContrast train It also shows the structural identity to bilingual grammars as used in ( ) . Uses val Relationships between the unlabeled items consider sequential relations between different types of emails ( e.g. , between requests and satisfactions thereof ) to classify messages , and thus also explicitly exploit the structure of conversations . Background train 100000 word stems of German ( ) . Uses train The computational treatment of lexical rules proposed can be seen as an extension to the principled method discussed by Gotz and , 1996 , 1997b ) for encoding the main building block of HPSG grammars -- the implicative constraints -- as a logic program . Extends train Identical to the standard perceptron proof , e.g. , , by inserting in loss-separability for normal separability . Background train The goal of the JAVOX toolkit is to speech-enable traditional desktop applications -- this is similar to the goals of the MELISSA project ( ) . CompareOrContrast train The simplest strategy for ordering adjectives is what call the direct evidence method . Background train The numeral ( whether it is implicit , as in ( 3 ) , or explicit ) can be construed as allowing the reader to draw inferences about the standards employed ( Kyburg and Morreau 2000 ; ) : ( 3 ) , for example , implies a standard that counts 10 cm as large and 8 cm as not large . Background train Such systems extract information from some types of syntactic units ( clauses in ( ; Gildea and Jurafsky , 2002 ; Hull and Gomez , 1996 ) ; noun phrases in ( Hull and Gomez , 1996 ; Rosario et al. , 2002 ) ) . Background val `` petty conversational implicature '' ( ) , or the metarules of Section 5.2 ? Background train Unlike other POS taggers , this POS tagger ( ) was also trained to disambiguate sentence boundaries . Uses train On the other hand , experiments indicate that mental representation and processing of morphologically complex words are not quite language independent ( ) . Background val ASARES has been previously applied to the acquisition of word pairs sharing semantic relations defined in the Generative Lexicon framework ( ) and called qualia relations ( Bouillon et al. , 2001 ) . Background train People are much more likely to consult such evaluative statements than the actual text of a bill or law under discussion , given the dense nature of legislative language and the fact that ( U.S. ) bills often reach several hundred pages in length ( ) . Background train For english , there is for instance the 15 year old HewlettPackard test suite , a simple text file listing test sentences and grouping them according to linguistics phenomena ( Flickinger et al. , 1987 ) ; and more recently , the much more sophisticated TSNLP ( Test Suite for Natural Language Processing ) which includes some 9500 test items for English , French and German , each of them being annotated with syntactic and application related information ( ) . Background train Its significance is reflected both in the growing interest in annotation software for word sense tagging ( ) and in the long-standing use of part-of-speech taggers , parsers and morphological analysers for data from English and many other languages . Background train Each of these Values has equal status , so the notion of a basic-level Value can not play a role ( cfXXX , ) . Background train Previous work on Chinese SRL mainly focused on how to transplant the machine learning methods which has been successful with English , such as Sun and Jurafsky ( 2004 ) , Xue and Palmer ( 2005 ) and . Background val Accuracy is not the best measure to assess segmentation quality , therefore we also conducted experiments using the WindowDiff measure as proposed by . Uses train The system is in the form of an agenda-driven chart-based parser whose foundation is similar to previous formalizations of Chomsky 's Minimalist Program ( Stabler , 1997 ; Harkema , 2000 ; ) . CompareOrContrast train Optimizing for dependency arc length is particularly important as parsers tend to do worse on longer dependencies ( ) and these dependencies are typically the most meaningful for downstream tasks , e.g. , main verb dependencies for tasks Motivation train Whereas dealt only with an English corpus , the current work shows that this methodology is applicable to a wide range of languages and corpora . CompareOrContrast train We also experiment with a CCG parser ( ) , requiring that the contexts surrounding the original phrase and paraphrase are assigned Uses train Their kernel is also very time consuming and in their more general sparse setting it requires O ( mn3 ) time and O ( mn2 ) space , where m and n are the number of nodes of the two trees ( m > = n ) ( ) . Future train Other molecular biology databases We also included several model organism databases or nomenclature databases in the construction of the dictionary , i.e. , mouse Mouse Genome Database ( MGD ) [ 18 ] , fly FlyBase [ 19 ] , yeast Saccharomyces Genome Database ( SGD ) [ 20 ] , rat -- Rat Genome Database ( RGD ) [ 21 ] , worm -- WormBase [ 22 ] , Human Nomenclature Database ( HUGO ) [ 23 ] , Online Mendelian Inheritance in Man ( OMIM ) [ 24 ] , and Enzyme Nomenclature Database ( ECNUM ) . Uses train ones , DIRT ( Lin and Pantel , 2001 ) , VerbOcean ( ) , FrameNet ( Baker et al. , 1998 ) , and Wikipedia ( Mehdad et al. , 2010 ; Kouylekov et al. , 2009 ) . Background train include decision tree learning and Bayesian learning , nearest neighbor learning , and artificial neural networks , early such works may be found in ( ) , ( Creecy and Masand , 1992 ) and ( Wiene and Pedersen , 1995 ) , respectively . Background val In other words AJAX is a web development technique for creating interactive web applications using a combination of XHTML and CSS , Document Object Model ( or DOM ) , the XMLHTTPRequest object ( ) . Background train For these or for a specific domain , basic synonymic dictionaries can be complemented using learning methods based on distributional similarity ( ; Lin , 1998 ) . Future train converted to numerical features using the standard technique of binarization , and we split values of the FEATS field into its atomic components .4 For some languages , we divide the training data into smaller sets , based on some feature s ( normally the CPOS or POS of the next input token ) , which may reduce training times without a significant loss in accuracy ( ) . Background val • language learning ( Green 1979 ; Mori and Moeser 1983 ; Morgan , Meier , and Newport 1989 ) • monolingual grammar induction ( ) • grammar optimization ( Juola 1994 ) • insights into universal grammar ( Juola 1998 ) • machine translation ( Juola 1994 , 1997 ; Veale and Way 1997 ; Gough , Way , and Hearne 2002 ) Background train Linguistic preprocessing of text documents is carried out by re-using smes , an information extraction core system for real-world German text processing ( ) . Uses val Regarding future work , there are many research line that may be followed : i ) Capturing more features by employing external knowledge such as ontological , lexical resource or WordNet-based features ( Basili et al. , 2005a ; Basili et al. , 2005b ; Bloehdorn et al. , 2006 ; Bloehdorn and Moschitti , 2007 ) or shallow semantic trees , ( Giuglea and Moschitti , 2004 ; Giuglea and Moschitti , 2006 ; Moschitti and Bejan , 2004 ; Moschitti et al. , 2007 ; Moschitti , 2008 ; ) . Future train For English − > Urdu , , page 213 ) notes that `` the system learned the original training corpus ... perfectly and could reproduce it without errors '' ; that is , it scored 100 % accuracy when tested against the training corpus . Background train Much previous work looks at the impact of using source side information ( i.e. , feature functions on the aligned English ) , such as those of , Yeniterzi and Oflazer ( 2010 ) and others . CompareOrContrast val Recent work ( ; Curran and Moens , 2002 ) has suggested that some tasks will benefit from using significantly more data . Background train See for a variant of this approximation that constructs finite transducers rather than finite automata . Background train Due to their remarkable ability to incorporate context structure information and long distance reordering into the translation process , tree-based translation models have shown promising progress in improving translation quality ( Liu et al. , 2006 , 2009 ; ; Galley et al. , 2004 , 2006 ; Marcu et al. , 2006 ; Shen et al. , 2008 ; Zhang et al. , 2011b ) . Background val These include devices such as interleaving the components ( McDonald 1983 ; Appelt 1983 ) , backtracking on failure ( Appelt 1985 ; Nogier 1989 ) , allowing the linguistic component to interrogate the planner ( Mann 1983 ; Sondheimer and Nebel 1986 ) , and Hovy 's notion of restrictive ( i.e. , bottom-up ) planning ( a , 1988c ) . Background train The shallow parser used is the SNoW-based CSCL parser ( Punyakanok and Roth , 2001 ; ) . Uses val Other approaches use less deep linguistic resources ( e.g. , POS-tags ) or are ( almost ) knowledge-free ( e.g. , Koehn and Knight ( 2003 ) ) . CompareOrContrast train Some well-known approaches include rule-based models ( ) , backed-off models ( Collins and Brooks 1995 ) , and a maximumentropy model ( Ratnaparkhi 1998 ) . Background train One approach to partial parsing was presented by , who extended a shallow-parsing technique to partial parsing . Background train compared a predictive approach ( statistical translation ) , a retrieval approach based on a language-model , and a hybrid approach which combines statistical chunking and traditional retrieval . CompareOrContrast train Indeed , such rich semantic links can be used to extend indices or reformulate queries ( similar to the work by with WoRDNET relations ) . CompareOrContrast train In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ Kaplan and Bresnan 1982 ; Bresnan 2001 ; Dalrymple 2001 ] , head-driven phrase structure grammar [ HPSG ] [ ] , tree-adjoining grammar [ TAG ] [ Joshi 1988 ] , and combinatory categorial grammar [ CCG ] [ Ades and Steedman 1982 ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Background train Despite these arguments , most applied NLG systems use a pipelined architecture ; indeed , a pipeline was used in every one of the systems surveyed by and Paiva ( 1998 ) . Background val Other similar approaches include those of Cicekli and G ¨ uvenir ( 1996 ) , McTait and Trujillo ( 1999 ) , , and Brown ( 2000 ) , inter alia . Background train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( Bikel et al. , 1997 ; Miller et al. , 1998 ; Borthwick , 1999 ; Mikheev et al. , 1999 ; ; Ng and Cardie , 2002 ; Florian et al. , 2004 ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train In addition to headwords , dictionary search through the pronunciation field is available ; Carter ( 1987 ) has merged information from the pronunciation and hyphenation fields , creating an enhanced phonological representation which allows access to entries by broad phonetic class and syllable structure ( ) . Background train Following Hockenmaier , Bierner , and Baldridge ( 2002 ) , , and Miyao , Ninomiya , and Tsujii ( 2004 ) , we extract a reference lexicon from Sections 02 -- 21 of the WSJ . Uses train The only disambiguation metric that we used in our previous work ( b ) was the shape-based metric , according to which the `` best '' trees are those that are skewed to the right . Extends train The need for information systems to support physicians at the point of care has been well studied ( ; Gorman et al. , 1994 ; Ely et al. , 2005 ) . Background val Every arc always has a definite direction , i.e. arcs are arrows ( ) . Background train In the first experiment , we use an induction algorithm ( a ) based on the expectation-maximization ( EM ) principle that induces parsers for PLTIGs . Uses train To create the baseline system , we use the opensource Joshua 4.0 system ( Ganitkevitch et al. , 2012 ) to build a hierarchical phrase-based ( HPB ) system , and a syntax-augmented MT ( SAMT ) 11 system ( ) respectively . Uses val 6 For Sent-Pred we also experimented with grammatical and sentence-based syntactic features , such as number of syntactic phrases , grammatical mood , and grammatical person ( ) , but the simple binary bag-of-lemmas representation yielded similar results . Uses val At present , the system uses a heuristic matching algorithm to classify relations into the appropriate category , though in the future we may consider a classifier similar to . Future train It is interesting to compare this analysis with that described in Dalrymple , Shieber , and Pereira ( 1991 ) and , 1991 ) . CompareOrContrast train The importance of including nonheadwords has become uncontroversial ( e.g. Collins 1999 ; ; Goodman 1998 ) . Background train Some methods of semantic relation analysis rely on predefined templates filled with information from processed texts ( ) . Background train The example used to illustrate the power of ATNs ( ) , `` John was believed to have been shot , '' also parses correctly , because the [ object ] node following the verb `` believed '' acts as both an absorber and a ( re ) generator . CompareOrContrast train Problems such as these have motivated research on more abstract , dependencybased parser evaluation ( e.g. , Lin 1995 ; Carroll , Briscoe , and Sanfilippo 1998 ; Carroll et al. 2002 ; Clark and Hockenmaier 2002 ; King et al. 2003 ; ; Kaplan et al. 2004 ; Miyao and Tsujii 2004 ) . Motivation train LiLFeS is one of the fastest inference engines for processing feature structure logic , and efficient HPSG parsers have already been built on this system ( Nishida et al. , 1999 ; ) . Background train As in ( ) , we used unsupervised training data which is automatically segmented to discover previously unseen stems . Uses train To address this limitation , our previous work ( ) has initiated an investigation on the problem of conversation entailment . Extends train ∗ A brief version of this work , with some additional material , first appeared as ( a ) . Extends train They also proposed two major categories of meta-learning approaches for recommender systems , merging and ensemble , each subdivided into the more specific subclasses suggested by as follows . Background train Here , I adopt the model proposed by and decompose lexical verbs into verbalizing heads and verbal roots . Uses val This heuristic is called soft union ( ) . Uses train This approach has its roots in Fillmore 's Case Grammar ( 1968 ) , and serves as the foundation for two current large-scale semantic annotation projects : FrameNet ( ) and PropBank ( Kingsbury et al. , 2002 ) . Background train Many investigators ( e.g. Allen 1976 ; Elowitz et al. 1976 ; ; Cahn 1988 ) have suggested that the poor prosody of synthetic speech , in comparison with natural speech , is the primary factor leading to difficulties in the comprehension of fluent synthetic speech . Motivation train The ability to explicitly identify these sections in unstructured text could play an important role in applications such as document summarization ( Teufel and Moens , 2000 ) , information retrieval ( ) , information extraction ( Mizuta et al. , 2005 ) , and question answering . Background train This semantics was constructed ( a , 1987b ) as a formal framework for default and commonsense reasoning . Uses val There is a general consensus among theoretical linguists that the proper representation of verbal argument structure is event structure -- representations grounded in a theory of events that decompose semantic roles in terms of primitive predicates representing concepts such as causality and inchoativity ( ; Jackendoff , 1983 ; Pustejovsky , 1991b ; Rappaport Hovav and Levin , 1998 ) . Background train Similarly , the notion of R + M-abduction is spiritually related to the `` abductive inference '' of Reggia ( 1985 ) , the `` diagnosis from first principles '' of Reiter ( 1987 ) , `` explainability '' of Poole ( 1988 ) , and the subset principle of . CompareOrContrast train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( Bikel et al. , 1997 ; Miller et al. , 1998 ; Borthwick , 1999 ; ; Soon et al. , 2001 ; Ng and Cardie , 2002 ; Florian et al. , 2004 ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train Representative systems are described in Boisen et al. ( 1989 ) , De Mattia and Giachin ( 1989 ) , , Niemann ( 1990 ) , and Young ( 1989 ) . Background train R98 ( , , , , „ ) uses a variant of Kozima 's semantic similarity measure ( ) to compute block similarity . Extends train , Meral et al. ( 2007 ) , Murphy ( 2001 ) , Murphy and Vogel ( 2007 ) and Topkara et al. ( 2006a ) all belong to the syntactic transformation category . Background val We used a publicly available tagger ( ) to tag the words and then used these in the input to the system . Uses train Some works abstract perception via the usage of symbolic logic representations ( Chen et al. , 2010 ; ; Matuszek et al. , 2012 ; Artzi and Zettlemoyer , 2013 ) , while others choose to employ concepts elicited from psycholinguistic and cognition studies . Background train Multilingual lexical databases aligned with the English WordNet ( e.g. MultiWordNet ( ) ) have been created for several languages , with different degrees of coverage . Background train An off-the-shelf speech recognition device , a Nippon Electric Corporation DP-200 , was added to an existing natural language processing system , the Natural Language Computer ( NLC ) ( , Biermann and Ballard 1980 ) . Background train For future work , we might investigate how machine learning algorithms , which are specifically designed for the problem of domain adaptation ( Blitzer et al. , 2007 ; ) , perform in comparison to our approach . Future train But their importance has grown far beyond machine translation : for instance , transferring annotations between languages ( Yarowsky and Ngai 2001 ; ; Ganchev , Gillenwater , and Taskar 2009 ) ; discovery of paraphrases ( Bannard and Callison-Burch 2005 ) ; and joint unsupervised POS and parser induction across languages ( Snyder and Barzilay 2008 ) . Motivation train That is , if the current hypothesis is unable to label a candidate or is uncertain about it , then the candidate might be a good training example ( ) . Background train Other milestones of recent research include the deployment of probabilistic and machine learning techniques ( Aone and Bennett 1995 ; Kehler 1997 ; Ge , Hale , and Charniak 1998 ; Cardie and Wagstaff 1999 ; the continuing interest in centering , used either in original or in revised form ( Abracos and Lopes 1994 ; Strube and Hahn 1996 ; Hahn and Strube 1997 ; ) ; and proposals related to the evaluation methodology in anaphora resolution ( Mitkov 1998a , 2001b ) . Background val For instance , the derived morphological forms are believed to be represented as a whole , whereas the representation of the inflected forms follows the morphemic model ( ) . Background train We perceive that these results can be extended to other language models that properly embed bilexical context-free grammars , as for instance the more general history-based models used in ( Ratnaparkhi , 1997 ) and ( ) . Future train Another paper ( ) describes the detailed analysis on the factor of the difference of parsing performance . Background train The original and the obtained grammar generated exactly the same number of derivation trees in the parsing experiment with 457 sentences from the ATIS corpus ( ) 6 ( the average length is 6.32 words ) . Uses train The use of running tallies and percentages is based on the assumption that these features are likely to produce generalized predictors ( ) . Motivation train The Longman lexicographers have developed a grammar coding system capable of representing in compact form a nontrivial amount of information , usually to be found only in large descriptive grammars of English ( such as ) . Background train 19 The paper by presents additional , more sophisticated models that we do not use in this article . CompareOrContrast train Empirical evidence has been brought forward that inflectional and/or derivational stemmers augmented by dictionaries indeed perform substantially better than those without access to such lexical repositories ( Krovetz , 1993 ; Kraaij and Pohlmann , 1996 ; ) . Background train Although there are other discussions of the paragraph as a central element of discourse ( e.g. Chafe 1979 , Halliday and Hasan 1976 , Longacre 1979 , ) , all of them share a certain limitation in their formal techniques for analyzing paragraph structure . CompareOrContrast train ( Details of how the average-expert model performs can be found in our prior work ( ) . ) Extends train Self-training should also benefit other discriminatively trained parsers with latent annotations ( ) , although training would be much slower compared to using generative models , as in our case . Future train We introduce here a clearly defined and replicable split of the data , so that future investigations can accurately and correctly compare against the results presented here . Uses train In most cases , the accuracy of parsers degrades when run on out-of-domain data ( ; McClosky et al. , 2006 ; Blitzer et al. , 2006 ; Petrov et al. , 2010 ) . Background train relies on morphosyntactic cues in the untagged Brown corpus as indicators of six predefined subcategorization frames . Background train To build the above s2t system , we first use the parse tree , which is generated by parsing the English side of the bilingual data with the Berkeley parser ( ) . Uses val One obvious approach to this problem is to employ parser reranking ( ) . Background train Ideally , to distinguish between raising and equi verbs , a number of syntactic criteria should be employed ( :460 ff . ) Future train The table also presents the closest comparable experimental results reported by .1 McKnight and Srinivasan ( henceforth , M&S ) created a test collection consisting of 37,151 RCTs from approximately 12 million MEDLINE abstracts dated between 1976 and 2001 . CompareOrContrast train How this mismatched perceptual basis affects referential communication in situated dialogue was investigated in our previous work ( ) . Extends val furthered this work by showing that a bimodal topic model , consisting of both text and feature norms , outperformed models using only one modality on the prediction of association norms , word substitution errors , and semantic interference tasks . Background train A cooccurrence based stemmer ( ) was used to stem Spanish words . Uses train The best performance on the WSJ corpus was achieved by a combination of the SATZ system ( Palmer and Hearst 1997 ) with the Alembic system ( ) : a 0.5 % error rate . CompareOrContrast train In a number of proposals , lexical generalizations are captured using lexical underspecification ( ; Krieger and Nerbonne 1992 ; CompareOrContrast val Although a grid may be more descriptively suitable for some aspects of prosody ( for example , use the grid representation for their implementation of stress assignment in compound nominals ) , we are not aware of any evidence for or against a grid representation of discourseneutral phrasing . CompareOrContrast train The bottom panel of table 1 lists the results for the chosen lexicalized model ( SSN-Freq > 200 ) and five recent statistical parsers ( Ratnaparkhi , 1999 ; Collins , 1999 ; ; Collins , 2000 ; Bod , 2001 ) . CompareOrContrast train Other factors , such as the role of focus ( Grosz 1977 , 1978 ; ) or quantifier scoping ( Webber 1983 ) must play a role , too . Background train linguistic in nature , rather than dealing with superficial properties of the text , e.g. the amount of white space between words ( ) . CompareOrContrast train The second version ( RM ) concerns the Resource Management task ( ) that has been popular within the DARPA community in recent years . Uses train Another line of research that is correlated with ours is recognition of agreement/disagreement ( ; Yin et al. , 2012 ; Abbott et al. , 2011 ; Andreas et al. , 2012 ; Galley et al. , 2004 ; Hillard et al. , 2003 ) and classification of stances ( Walker et al. , 2012 ; Somasundaran and Wiebe , 2010 ) in online forums . CompareOrContrast train The studies presented by and Johnson ( 2007 ) differed in the number of states that they used . CompareOrContrast train A number of alignment techniques have been proposed , varying from statistical methods ( ; Gale and Church , 1991 ) to lexical methods ( Kay and Roscheisen , 1993 ; Chen , 1993 ) . Background train We would like to use features that look at wide context on the input side , which is inexpensive ( ) . Future train Much of the earlier work in anaphora resolution heavily exploited domain and linguistic knowledge ( Sidner 1979 ; Carter 1987 ; Rich and LuperFoy 1988 ; ) , which was difficult both to represent and to process , and which required considerable human input . Background train Experiments on Chinese SRL ( Xue and Palmer 2005 , ) reassured these findings . Background train This is then generalized , following a methodology based on , to generate the `` generalized marker lexicon . '' Uses train A number of speech understanding systems have been developed during the past fifteen years ( Barnett et al. 1980 , Dixon and Martin 1979 , Erman et al. 1980 , Haton and Pierrel 1976 , Lea 1980 , Lowerre and Reddy 1980 , Medress 1980 , , Walker 1978 , and Wolf and Woods 1980 ) . CompareOrContrast train Our proposed method is based on the automatically acquired paraphrase dictionary described in , in which the application of paraphrases from the dictionary encodes secret bits . Uses train For example , frequent words are translated less consistently than rare words ( ) . Background train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; ) . Background train These knowledge sources were effectively used to build a state-of-the-art WSD program in one of our prior work ( ) . Extends train Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( Andrews et al. , 2009 ; Steyvers , 2010 ; Feng and Lapata , 2010b ; ; Silberer and Lapata , 2012 ; Johns and Jones , 2012 ; Bruni et al. , 2012a ; Bruni et al. , 2012b ; Silberer et al. , 2013 ) . Background val Due to advances in statistical syntactic parsing techniques ( ; Charniak , 2001 ) , attention has recently shifted towards the harder question of analyzing the meaning of natural language sentences . Background val This is where robust syntactic systems like SATZ ( ) or the POS tagger reported in Mikheev ( 2000 ) , which do not heavily rely on word capitalization and are not sensitive to document length , have an advantage . CompareOrContrast train category relationships from the weak supervision : the tag dictionary and raw corpus ( ; Garrette et al. , 2015 ) .4 This procedure attempts to automatically estimate the frequency of each word/tag combination by dividing the number of raw-corpus occurrences of each word in the dictionary evenly across all of its associated tags . Uses train WIT has been implemented in Common Lisp and C on UNIX , and we have built several experimental and demonstration dialogue systems using it , including a meeting room reservation system ( Nakano et al. , 1999b ) , a video-recording programming system , a schedule management system ( Nakano et al. , 1999a ) , and a weather infomiation system ( ) . Extends train Lexical functional grammar ( ; Bresnan 2001 ; Dalrymple 2001 ) is a member of the family of constraint-based grammars . Background val present an approach to learn previously unknown frames for Czech from the Prague Dependency Bank ( Hajic Background train A companion paper describes the evaluation process and results in further detail ( ) . Extends train SNoW ( Carleson et al. , 1999 ; ) is a multi-class classifier that is specifically tailored for learning in domains in which the potential number of information sources ( features ) taking part in decisions is very large , of which NLP is a principal example . Uses train Hence , enumerating morphological variants in a semi-automatically generated lexicon , such as proposed for French ( ) , turns out to be infeasible , at least for German and related languages . Background train It is analogous to the step in other translation model induction algorithms that sets all probabilities below a certain threshold to negligible values ( ; Dagan et al. , 1993 ; Chen , 1996 ) . CompareOrContrast train They proved to be useful in a number of NLP applications such as natural language generation ( Iordanskaja et al. , 1991 ) , multidocument summarization ( McKeown et al. , 2002 ) , automatic evaluation of MT ( Denkowski and Lavie , 2010 ) , and TE ( ) . Motivation val The results , which partly confirm those obtained on a smaller dataset in , must be seen in light of the fact that our gesture annotation scheme comprises more fine-grained categories than most of the studies mentioned earlier for both head movements and face expressions . CompareOrContrast train We employ the idea of ultraconservative update ( ; Crammer et al. , 2006 ) to propose two incremental methods for local training in Algorithm 2 as follows . Uses train It has also been shown to be useful in joint inference of text with visual attributes obtained using visual classifiers ( ) . Background train reported a correlation of r = .69 . CompareOrContrast train A more subtle example is weighted FSAs that approximate PCFGs ( Nederhof , 2000 ; ) , or to extend the idea , weighted FSTs that approximate joint or conditional synchronous PCFGs built for translation . Background train As has been previously observed and exploited in the NLP literature ( Pang and Lee , 2004 ; Agarwal and Bhattacharyya , 2005 ; ) , the above optimization function , unlike many others that have been proposed for graph or set partitioning , can be solved exactly in an provably efficient manner via methods for finding minimum cuts in graphs . CompareOrContrast train WIT features an incremental understanding method ( b ) that makes it possible to build a robust and real-time system . Uses train This choice is inspired by recent work on learning syntactic categories ( ) , which successfully utilized such language models to represent word window contexts of target words . Motivation train As for work on Arabic ( MSA ) , results have been reported on the PATB ( Kulick , Gabbard , and Marcus 2006 ; ; Green and Manning 2010 ) , the Prague Dependency Treebank ( PADT ) ( Buchholz and Marsi 2006 ; Nivre 2008 ) and the CATiB ( Habash and Roth 2009 ) . Background train Other studies which view lR as a query generation process include Maron and Kuhns , 1960 ; Hiemstra and Kraaij , 1999 ; ; Miller et al , 1999 . CompareOrContrast train In Section 5 , we discuss the difficulties associated with such user studies , and describe a human-based evaluation we conducted for a small subset of the responses generated by our system ( b ) . Uses train considers the second verb V2 as an aspectual complex comparable to the auxiliaries . Background train 2We could just as easily use other symmetric `` association '' measures , such as 02 ( ) or the Dice coefficient ( Smadja , 1992 ) . CompareOrContrast val In this situation , b , 293 ) recommend `` evaluating the expectations using only a single , probable alignment . '' Motivation val The use of the web as a corpus for teaching and research on language has been proposed a number of times ( ; Robb , 2003 ; Rundell , 2000 ; Fletcher , 2001 , 2004b ) and received a special issue of the journal Computational Linguistics ( Kilgarriff and Grefenstette , 2003 ) . Background val • Learnability ( Zernik and Dyer 1987 ) • Text generation ( Hovy 1988 ; Milosavljevic , Tulloch , and Dale 1996 ) • Speech generation ( ) • Localization ( Sch ¨ aler 1996 ) Background train Following our previous work on stance classification ( c ) , we employ three types of features computed based on the frame-semantic parse of each sentence in a post obtained from SEMAFOR ( Das et al. , 2010 ) . Extends train However , the literature on Linguistic Steganography , in which linguistic properties of a text are modified to hide information , is small compared with other media ( ) . Background train This has been reported for other languages , too , dependent on the generality of the chosen approach ( J ¨ appinen and Niemist ¨ o , 1988 ; Choueka , 1990 ; Popovic and Willett , 1992 ; Ekmekc ¸ ioglu et al. , 1995 ; Hedlund et al. , 2001 ; ) . Background train We offer a theorem that highlights the broad applicability of these modeling techniques .4 If f ( input , output ) is a weighted regular relation , then the following statements are equivalent : ( 1 ) f is a joint probabilistic relation ; ( 2 ) f can be computed by a Markovian FST that halts with probability 1 ; ( 3 ) f can be expressed as a probabilistic regexp , i.e. , a regexp built up from atomic expressions a : b ( for a E E U -LCB- E -RCB- , b E A U -LCB- E -RCB- ) using concatenation , probabilistic union + p , and probabilistic closure * p. For defining conditional relations , a good regexp language is unknown to us , but they can be defined in several other ways : ( 1 ) via FSTs as in Fig. 1c , ( 2 ) by compilation of weighted rewrite rules ( ) , ( 3 ) by compilation of decision trees ( Sproat and Riley , 1996 ) , ( 4 ) as a relation that performs contextual left-to-right replacement of input substrings by a smaller conditional relation ( Gerdemann and van Noord , 1999 ) ,5 ( 5 ) by conditionalization of a joint relation as discussed below . Background train We use the CCG parser to analyse the sentence before and after paraphrasing . Uses train For example , consider a relational description ( cfXXX , ) involving a gradable adjective , as in the dog in the large shed . Background train Previous versions of our work , as described in also assume that phrasing is dependent on predicate-argument structure . Extends train like information extraction ( Yates and Etzioni , 2009 ) and textual entailment ( ) . Background train The inclusion of the coreference task in the Sixth and Seventh Message Understanding Conferences ( MUC-6 and MUC-7 ) gave a considerable impetus to the development of coreference resolution algorithms and systems , such as those described in , Gaizauskas and Humphreys ( 1996 ) , and Kameyama ( 1997 ) . Background train Lee et al. ( 2012 ) model entity coreference and event coreference jointly ; consider joint coreference and entity-linking . Background train However , learning-based resolvers have not been able to benefit from having an SC agreement feature , presumably because the method used to compute the SC of an NP is too simplistic : while the SC of a proper name is computed fairly accurately using a named entity ( NE ) recognizer , many resolvers simply assign to a common noun the first ( i.e. , most frequent ) WordNet sense as its SC ( e.g. , , Markert and Nissim ( 2005 ) ) . Background train This includes work on generalized expectation ( Mann and McCallum , 2010 ) , posterior regularization ( Ganchev et al. , 2010 ) and constraint driven learning ( ; Chang et al. , 2010 ) . CompareOrContrast train The final machine is a trigram language model , specifically a Kneser-Ney ( ) based backoff language model . Uses train Hovy has described another text planner that builds similar plans ( b ) . Background train In particular , since we treat each individual speech within a debate as a single `` document '' , we are considering a version of document-level sentiment-polarity classification , namely , automatically distinguishing between positive and negative documents ( Das and Chen , 2001 ; ; Turney , 2002 ; Dave et al. , 2003 ) . Background val transition-based dependency parsing framework ( ) using an arc-eager transition strategy and are trained using the perceptron algorithm as in Zhang and Clark ( 2008 ) with a beam size of 8 . Uses train Just as easily , we can model link types that coincide with entries in an on-line bilingual dictionary separately from those that do not ( cfXXX ) . Uses train As an alternative , we rely on PubMed to retrieve an initial set of hits that we then postprocess in greater detail -- this is the standard pipeline architecture commonly employed in other question-answering systems ( Voorhees and Tice 1999 ; ) . CompareOrContrast train ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) used maximum likelihood estimation to learn weights for MT. ( ; Moore and Quirk , 2008 ; Zhao and Chen , 2009 ; Galley and Quirk , 2011 ) employed an evaluation metric as a loss function and directly optimized it . CompareOrContrast val We also compare the results with the output generated by the statistical translation system GIZA + + / ISI ReWrite Decoder ( AlOnaizan et al. , 1999 ; ; Germann et al. , 2001 ) , trained on the same parallel corpus . CompareOrContrast train Some efforts have tackled tasks such as automatic image caption generation ( Feng and Lapata , 2010a ; Ordonez et al. , 2011 ) , text illustration ( ) , or automatic location identification of Twitter users ( Eisenstein et al. , 2010 ; Wing and Baldridge , 2011 ; Roller et al. , 2012 ) . Background train Since we are not generating from the model , this does not introduce difficulties ( ) . Motivation train The extraction procedure consists of three steps : First , the bracketing of the trees in the Penn Treebank is corrected and extended based on the approaches of Magerman ( 1994 ) and . Background train As they are required to enable test subjects to distinguish between senses , we use artificial glosses composed from synonyms and hypernyms as a surrogate , e.g. for brother : `` brother , male sibling '' vs. `` brother , comrade , friend '' ( ) . Uses train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background val argues that there are cases , albeit exceptional ones , in which constraints on syntactic category are an issue in subcategorization . Motivation train Corpus frequency : ( ) differentiates between misspellings and neologisms ( new words ) in terms of their frequency . Uses train We use an in-house developed hierarchical phrase-based translation ( ) as our baseline system , and we denote it as In-Hiero . Uses train The types of sentences accepted are essentially those accepted by the original NLC grammar , imperative sentences with nested noun groups and conjunctions ( ) . Background train The task we used to compare different generalisation techniques is similar to that used by Pereira et al. ( 1993 ) and . CompareOrContrast train One of the proposed methods to extract paraphrases relies on a pivot-based approach using phrase alignments in a bilingual parallel corpus ( ) . Background val The Nash arbitration plan , for example , would allow a doubly graded description whenever the product of the Values for the referent r exceeds that of all distractors ( ; cfXXX Gorniak and Roy 2003 ; Thorisson 1994 , for other plans ) . Background train extracts word co-occurrence probabilities from unlabelled text collected from a web crawler . Background train Notable early papers on graph-based semisupervised learning include Blum and Chawla ( 2001 ) , , Kondor and Lafferty ( 2002 ) , and Joachims ( 2003 ) . Background val Due to this inherent ambiguity , manual annotations usually distinguish between sure correspondences for unambiguous translations , and possible , for ambiguous translations ( ) . Background train Thus for instance , ( Copestake and Flickinger , 2000 ; Copestake et al. , 2001 ) describes a Head Driven Phrase Structure Grammar ( HPSG ) which supports the parallel construction of a phrase structure ( or derived ) tree and of a semantic representation and ( ) show how to equip Lexical Functional grammar ( LFG ) with a glue semantics . Background train The contextual interpreter then uses a reference resolution approach similar to Byron ( 2002 ) , and an ontology mapping mechanism ( a ) to produce a domain-specific semantic representation of the student 's output . Uses train only the available five relative scopings of the quantifiers are produced ( , 47 ) , but without the need for a free variable constraint -- the HOU algorithm will not produce any solutions in which a previously bound variable becomes free ; • the equivalences are reversible , and thus the above sentences cart be generated from scoped logical forms ; • partial scopings are permitted ( see Reyle [ 19961 ) • scoping can be freely interleaved with other types of reference resolution ; • unscoped or partially scoped forms are available for inference or for generation at every stage . Background train In the transducers produced by the training method described in this paper , the source and target positions are in the set -LCB- -1 , 0,1 -RCB- , though we have also used handcoded transducers ( ) and automatically trained transducers ( Alshawi and Douglas 2000 ) with a larger range of positions . Uses train Similar approaches are being explored for parsing ( Steedman , ; Hwa et al. 2003 ) . Background train In particular , since we treat each individual speech within a debate as a single `` document '' , we are considering a version of document-level sentiment-polarity classification , namely , automatically distinguishing between positive and negative documents ( Das and Chen , 2001 ; Pang et al. , 2002 ; Turney , 2002 ; ) . Background train would be chunked as follows ( Tjong Kim ) : [ NP He ] [ VP reckons ] [ NP the current account deficit ] [ VP will narrow ] [ PP Background train The flexible architecture we have presented enables interesting future research : ( i ) a straightforward improvement is the use of lexical similarity to reduce data sparseness , e.g. ( ; Basili et al. , 2006 ; Bloehdorn et al. , 2006 ) . Background train 5An alternative strategy to step ( 4 ) is to perform a database lookup based on the ambiguous query and summarize the results ( ) , which we leave for future work . Future train Such approaches have been tried recently in restricted cases ( ; Eisner , 2001b ; Lafferty et al. , 2001 ) . Background train Typed feature grammars can be used as the basis for implementations of Head-driven Phrase Structure Grammar ( HPSG ; Pollard and Sag , 1994 ) as discussed in ( a ) and ( Meurers and Minnen , 1997 ) . Extends val It is therefore no surprise that early attempts at response automation were knowledge-driven ( Barr and Tessler 1995 ; ; Delic and Lahaix 1998 ) . Background val A more subtle example is weighted FSAs that approximate PCFGs ( ; Mohri and Nederhof , 2001 ) , or to extend the idea , weighted FSTs that approximate joint or conditional synchronous PCFGs built for translation . Background train ones , DIRT ( Lin and Pantel , 2001 ) , VerbOcean ( Chklovski and Pantel , 2004 ) , FrameNet ( Baker et al. , 1998 ) , and Wikipedia ( ; Kouylekov et al. , 2009 ) . Background train The Ruby on framework permits us to quickly develop web applications without rewriting common functions and classes . Uses train We further add rules for combining with punctuation to the left and right and allow for the merge rule X → X X of . Uses train Politically-oriented text Sentiment analysis has specifically been proposed as a key enabling technology in eRulemaking , allowing the automatic analysis of the opinions that people submit ( Shulman et al. , 2005 ; ; Kwon et al. , 2006 ) . Background train has made some preliminary attempt on the idea of hierarchical semantic Background train pointed out that many relations between words in a text are non-classical ( i.e. other than typical taxonomic relations like synonymy or hypernymy ) and therefore not covered by semantic similarity . Background train Some of the intuitions we associate with this notion have been very well expressed by , pp. 7-8 ) : ... Semantics is constrained by our models of ourselves and our worlds . Background val In 2009 , the second WePS campaign showed similar trends regarding the use of NE features ( ) . Background train The head words can be automatically extracted using a heuristic table lookup in the manner described by . Uses train Reiter describes a pipelined modular approach as a consensus architecture underlying most recent work in generation ( ) . Background train For example , the suite of LT tools ( ; Grover et al. , 2000 ) perform tokenization , tagging and chunking on XML marked-up text directly . Background train Liu et al. ( 2005 ) , , Murphy ( 2001 ) , Murphy and Vogel ( 2007 ) and Topkara et al. ( 2006a ) all belong to the syntactic transformation category . Background train Brockmann and Lapata ( 2003 ) have showed that WordNet-based approaches do not always outperform simple frequency-based models , and a number of techniques have been recently proposed which may offer ideas for refining our current unsupervised approach ( ; Bergsma et al. , 2008 ) . Future val The RenTAL system is implemented in LiLFeS ( ) 2 . Uses train In the future , we hope to evaluate the automatic annotations and extracted lexicon against Propbank ( ) . Future train Each set of translations is stored separately , and for each set the `` marker hypothesis '' ( ) is used to segment the phrasal lexicon into a `` marker lexicon . '' Uses train One way to increase the precision of the mapping process is to impose some linguistic constraints on the sequences such as simple noun-phrase contraints ( Gaussier , 1995 ; ; hua Chen and Chen , 94 ; Fung , 1995 ; Evans and Zhai , 1996 ) . Uses train Although this study falls under the general topic of discourse modeling , our work differs from previous attempts to characterize text in terms of domainindependent rhetorical elements ( ; Marcu and Echihabi , 2002 ) . CompareOrContrast train `` Coherence , '' as outlined above , can be understood as a declarative ( or static ) version of marker passing ( Hirst 1987 ; ) , with one difference : the activation spreads to theories that share a predicate , not through the IS-A hierarchy , and is limited to elementary facts about predicates appearing in the text . CompareOrContrast val Task properties Determining whether or not a speaker supports a proposal falls within the realm of sentiment analysis , an extremely active research area devoted to the computational treatment of subjective or opinion-oriented language ( early work includes , Hearst ( 1992 ) , Sack ( 1994 ) , and Wiebe ( 1994 ) ; see Esuli ( 2006 ) for an active bibliography ) . Background val First , it has been noted that in many natural language applications it is sufficient to use shallow parsing information ; information such as noun phrases ( NPs ) and other syntactic sequences have been found useful in many large-scale language processing applications including information extraction and text summarization ( ; Appelt et al. , 1993 ) . Background train In addition , a fully flexible access system allows the retrieval of dictionary entries on the basis of constraints specifying any combination of phonetic , lexical , syntactic , and semantic information ( ) . Background val developed a way of incorporating standard n-grams into the cache model , using mixtures of language models and also exponentially decaying the weight for the cache prediction depending on the recency of the word 's last Extends train Riehemann 1993 ; Oliva 1994 ; Frank 1994 ; Opalka 1995 ; ) . CompareOrContrast val Machine learning methods should be interchangeable : Transformation-based learning ( TBL ) ( ) and Memory-based learning ( MBL ) ( Daelemans et al. , 2002 ) have been applied to many different problems , so a single interchangeable component should be used to represent each method . Motivation train Recently , several alternative , often quite sophisticated approaches to collective classification have been proposed ( Neville and Jensen , 2000 ; Lafferty et al. , 2001 ; ; Taskar et al. , 2002 ; Taskar et al. , 2003 ; Taskar et al. , 2004 ; McCallum and Wellner , 2004 ) . Background train We tested the classification of verbs into semantic types using a verb list of 139 pre-classified items drawn from the lists published in and Stockwell et al. ( 1973 ) . Uses train The combination of likelihood and prior modeling , HMMs , and Viterbi decoding is fundamentally the same as the standard probabilistic approaches to speech recognition ( Bahl , Jelinek , and Mercer 1983 ) and tagging ( ) . CompareOrContrast train Cross-lingual Textual Entailment ( CLTE ) has been proposed by ( Mehdad et al. , 2010 ) as an extension of Textual Entailment ( ) that consists in deciding , given two texts T and H in different languages , if the meaning of H can be inferred from the meaning of T . Background train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( Bikel et al. , 1997 ; ; Borthwick , 1999 ; Mikheev et al. , 1999 ; Soon et al. , 2001 ; Ng and Cardie , 2002 ; Florian et al. , 2004 ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train Other solutions such as complete caching of the corpora are not typically adopted due to legal concerns over copyright and redistribution of web data , issues considered at length by a ) . Background val This includes work on generalized expectation ( ) , posterior regularization ( Ganchev et al. , 2010 ) and constraint driven learning ( Chang et al. , 2007 ; Chang et al. , 2010 ) . Background train A very similar formulation , for another grammar transformation , is given in . CompareOrContrast train We run GIZA + + ( ) on the training corpus in both directions ( Koehn et al. , 2003 ) to obtain the word alignment for each sentence pair . Uses train Other attempts to address efficiency include the fast Transformation Based Learning ( TBL ) Toolkit ( ) which dramatically speeds up training TBL systems , and the translation of TBL rules into finite state machines for very fast tagging ( Roche and Schabes , 1997 ) . Background train Consequently , fusion has been applied to a wide variety of pattern recognition and decision theoretic problems -- using a plethora of theories , techniques , and tools -- including some applications in computational linguistics ( e.g. , ; van Halteren , Zavrel , and Daelemans 1998 ) and speech technology ( e.g. , Bowles and Damper 1989 ; Romary and Pierre11989 ) . Background train This observation has led some researchers , e.g. , , to claim a direct mapping between the syntactic phrase and the prosodic phrase . Background train These features are carefully designed to reduce the data sparseness problem and some of them are inspired by previous work ( He et al. , 2008 ; ; Marton and Resnik , 2008 ; Chiang et al. , 2009 ; Setiawan et al. , 2009 ; Shen et al. , 2009 ; Xiong et al. , 2009 ) : 1 . Motivation val In our previous work ( ) , conversation entailment is formulated as the following : given a conversation segment D which is represented by a set of clauses D = d1 ∧ ... ∧ dm , and a hypothesis H represented by another set of clauses H = h1 ∧ ... ∧ hn , the prediction on whether D entails H is determined by the product of probabilities that each hypothesis clause hj is entailed from all the conversation segment clauses d1 ... dm as follows . Extends train Notable early papers on graph-based semisupervised learning include Blum and Chawla ( 2001 ) , Bansal et al. ( 2002 ) , , and Joachims ( 2003 ) . Background train In addition , we consider several types of lexical features ( LexF ) inspired by previous work on agreement and disagreement ( ; Misra and Walker , 2013 ) . Motivation train compared two retrieval approaches ( TF.IDF and query expansion ) and two predictive approaches ( statistical translation and latent variable models ) . Background train Others provide automatic mappings of natural language instructions to executable actions , such as interpreting navigation directions ( Chen and Mooney , 2011 ) or robot commands ( Tellex et al. , 2011 ; ) . Background train The dialogue state is represented by a cumulative answer analysis which tracks , over multiple turns , the correct , incorrect , and not-yet-mentioned parts 1Other factors such as student confidence could be considered as well ( ) . Future train As noted above , it is well documented ( ) that subcategorization frames ( and their frequencies ) vary across domains . Motivation train explore a number of related approaches to the extraction of a lexicalized TAG from the Penn-II Treebank with the aim of constructing a statistical model for parsing . Background train At the same time , we believe our method has advantages over the approach developed initially at IBM ( ; Brown et al. 1993 ) for training translation systems automatically . CompareOrContrast train Building on the work of Ruch et al. ( 2003 ) in the same domain , we present a generative approach that attempts to directly model the discourse structure of MEDLINE abstracts using Hidden Markov Models ( HMMs ) ; cfXXX ( ) . Background train The basic Python reflection has already been implemented and used for large scale experiments with POS tagging , using pyMPI ( a message passing interface library for Python ) to coordinate experiments across a cluster of over 100 machines ( Curran and Clark , 2003 ; ) . Background train As an alternative , we rely on PubMed to retrieve an initial set of hits that we then postprocess in greater detail -- this is the standard pipeline architecture commonly employed in other question-answering systems ( ; Hirschman and Gaizauskas 2001 ) . CompareOrContrast train The language grounding problem has received significant attention in recent years , owed in part to the wide availability of data sets ( e.g. Flickr , Von Ahn ( 2006 ) ) , computing power , improved computer vision models ( Oliva and Torralba , 2001 ; Lowe , 2004 ; Farhadi et al. , 2009 ; Parikh and Grauman , 2011 ) and neurological evidence of ties between the language , perceptual and motor systems in the brain ( Pulverm ¨ uller et al. , 2005 ; Tettamanti et al. , 2005 ; ) . Background train Although in this paper we take modus ponens as the main rule of inference , in general one can consider deductive closures with respect to weaker , nonstandard logics , ( cfXXX ; Frisch 1987 ; Patel-Schneider 1985 ) . CompareOrContrast train As points out , given that no situations were envisaged where the information from the tape would be altered once installed in secondary storage , this simple and convenComputational Linguistics , Volume 13 , Numbers 3-4 , July-December 1987 205 Bran Boguraev and Ted Briscoe Large Lexicons for Natural Language Processing tional access strategy is perfectly adequate . Background train Other factors , such as the role of focus ( , 1978 ; Sidner 1983 ) or quantifier scoping ( Webber 1983 ) must play a role , too . Background train This paper describes an approach for sharing resources in various grammar formalisms such as Feature-Based Lexicalized Tree Adjoining Grammar ( FB-LTAG1 ) ( ; Vijay-Shanker and Joshi , 1988 ) and Head-Driven Phrase Structure Grammar ( HPSG ) ( Pollard and Sag , 1994 ) by a method of grammar conversion . Background train This is because the binary structure has been verified to be very effective for tree-based translation ( ; Zhang et al. , 2011a ) . Motivation train There is a general consensus among theoretical linguists that the proper representation of verbal argument structure is event structure -- representations grounded in a theory of events that decompose semantic roles in terms of primitive predicates representing concepts such as causality and inchoativity ( Dowty , 1979 ; ; Pustejovsky , 1991b ; Rappaport Hovav and Levin , 1998 ) . Background train measure the standard intrinsic parser metrics unlabeled attachment score ( UAS ) and labeled attachment score ( LAS ) ( ) . Uses train The EDR has close ties to the named entity recognition ( NER ) and coreference resolution tasks , which have been the focus of several recent investigations ( ; Miller et al. , 1998 ; Borthwick , 1999 ; Mikheev et al. , 1999 ; Soon et al. , 2001 ; Ng and Cardie , 2002 ; Florian et al. , 2004 ) , and have been at the center of evaluations such as : MUC-6 , MUC-7 , and the CoNLL '02 and CoNLL '03 shared tasks . Background train Accordingly , we convert examples such as ( 27 ) into their generalized equivalents , as in ( 28 ) : ( 28 ) good man : bon homme That is , where substitutes variables for various words in his templates , we replace certain lexical items with their marker tag . CompareOrContrast train How it is done is beyond the scope of this paper but is explained in detail in . Background train These tools use a highly optimised GIS implementation and provide sophisticated Gaussian smoothing ( ) . Uses train Also relevant is work on the general problems of dialog-act tagging ( Stolcke et al. , 2000 ) , citation analysis ( Lehnert et al. , 1990 ) , and computational rhetorical analysis ( Marcu , 2000 ; ) . Background train The inference rules that were necessary to convert one list of properties into another do not sit comfortably within the received NLG pipeline model ( e.g. , ) . Background train has developed an agenda-driven chart parser for the feature-driven formalism described above ; please refer to his paper for a description of the parsing algorithm . Extends val An example of psycholinguistically oriented research work can be found in . Background train In addition to a referring function , noun phrases ( NP ) can also serve communicative goals such as providing new information about the referent and expressing the speaker 's emotional attitude towards the referent ( Appelt , 1985 ; ) . Background train Some well-known approaches include rule-based models ( Brill and Resnik 1994 ) , backed-off models ( Collins and Brooks 1995 ) , and a maximumentropy model ( ) . Background train There is some literature on procedure acquisition such as the LISP synthesis work described in and the PROLOG synthesis method of Shapiro ( 1982 ) . CompareOrContrast val Against the background of a growing interest in multilingual NLP , multilingual anaphora / coreference resolution has gained considerable momentum in recent years ( Aone and McKee 1993 ; Azzam , Humphreys , and Gaizauskas 1998 ; ; Mitkov and Barbu 2000 ; Mitkov 1999 ; Mitkov and Stys 1997 ; Mitkov , Belguith , and Stys 1998 ) . Background train The list of semantic relations with which we work is based on extensive literature study ( a ) . Uses train The research described below is taking place in the context of three collaborative projects ( ; Russell et al. , 1986 ; Phillips and Thompson , 1986 ) to develop a general-purpose , wide coverage morphological and syntactic analyser for English . Background train In most cases , the accuracy of parsers degrades when run on out-of-domain data ( Gildea , 2001 ; McClosky et al. , 2006 ; Blitzer et al. , 2006 ; ) . Background train This paper presents experiments with generative content models for analyzing the discourse structure of medical abstracts , which has been confirmed to follow the four-section pattern discussed above ( ) . Background train Following , one approach for achieving this objective consists of applying supervised learning , where a winning method is selected for each case in the training set , all the training cases are labeled accordingly , and then the system is trained to predict a winner for unseen cases . CompareOrContrast train Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background val Due to their remarkable ability to incorporate context structure information and long distance reordering into the translation process , tree-based translation models have shown promising progress in improving translation quality ( , 2009 ; Quirk et al. , 2005 ; Galley et al. , 2004 , 2006 ; Marcu et al. , 2006 ; Shen et al. , 2008 ; Zhang et al. , 2011b ) . Background train 11 reports that non-projective and pseudo-projective algorithms outperform the `` eager '' projective algorithm in MaltParser , but our training data did not contain any non-projective dependencies . CompareOrContrast train In particular , boosting ( Schapire , 1999 ; ) offers the possibility of achieving high accuracy from a collection of classifiers which individually perform quite poorly . Future train For MT the most commonly used heuristic is called grow diagonal final ( ) . CompareOrContrast train ECM-F is an entity-constrained mention Fmeasure ( cfXXX ( ) for how ECM-F is computed ) , and ACE-Value is the official ACE evaluation metric . Uses val In order to obtain semantic representations of each word , we apply our previous strategy ( ) . Extends train annotated a larger set of word pairs ( 353 ) , too . Background train A more flexible approach is used by , where users can specify boundary values for attributes like rainfall , specifying , for example , rain counts as moderate above 7 mm/h , as heavy above 20 mm/h , and so on . Background train Furthermore , a number of performance features , largely based on the PARADISE dialogue evaluation scheme ( ) , were automatically logged , derived , or manually annotated . Uses train propose to generalize the direct evidence method so that it can apply to unseen pairs of adjectives by computing the transitive closure of the ordering relation . Background train We evaluated our translations with IBM 's BLEU evaluation metric ( ) , using the same evaluation method and reference retranslations that were used for evaluation at HLT Workshop 2002 at CLSP ( Haji 6 et al. , 2002 ) . Uses train Other attempts to address efficiency include the fast Transformation Based Learning ( TBL ) Toolkit ( Ngai and Florian , 2001 ) which dramatically speeds up training TBL systems , and the translation of TBL rules into finite state machines for very fast tagging ( ) . Background train PR is closely related to the work of , 2008 ) , who concurrently developed the idea of using penalties based on posterior expectations of features to guide semi-supervised learning . Background train The best results on most of our data were obtained using Hidden Naive Bayes ( HNB ) ( ) . Uses train de URL : http://www.sfs.nphil.uni-tuebingen.de/sfb / b4home.html 1 This is , for example , the case for all proposals working with verbal lexical entries that raise the arguments of a verbal complement ( Hinrichs and Nakazawa 1989 ) that also use lexical rules such as the Complement Extraction Lexical Rule ( ) or the Complement Cliticization Lexical Rule ( Miller and Sag 1993 ) to operate on those raised elements . Background train A number of proposals in the 1990s deliberately limited the extent to which they relied on domain and/or linguistic knowledge and reported promising results in knowledge-poor operational environments ( Dagan and Itai 1990 , 1991 ; Lappin and Leass 1994 ; Nasukawa 1994 ; Kennedy and Boguraev 1996 ; Williams , Harvey , and Preston 1996 ; Baldwin 1997 ; , 1998b ) . Background train ASARES has been previously applied to the acquisition of word pairs sharing semantic relations defined in the Generative Lexicon framework ( Pustejovsky , 1995 ) and called qualia relations ( ) . Background train mlSystem ruleFeats + atomFeats We augment mlSystem ruleFeats with more features from our previous work ( Markert et al. , 2012 ; a ; Hou et al. , 2013b ) on bridging anaphora recognition and antecedent selection . Extends train The description of the EAGLE workbench for linguistic engineering ( ) mentions a case normalization module that uses a heuristic in which a capitalized word in an ambiguous position should be rewritten without capitalization if it is found lower-cased in the same document . CompareOrContrast val An exception is , who experimented with determining the political orientation of websites essentially by classifying the concatenation of all the documents found on that site . Background val Until now , translation models have been evaluated either subjectively ( e.g. White and O'Connell 1993 ) or using relative metrics , such as perplexity with respect to other models ( b ) . CompareOrContrast val The three preprocessing steps ( tokenization , POS-tagging , lemmatization ) are performed using TreeTagger ( ) . Uses train Many statistical parsers ( Ratnaparkhi , 1999 ; ; Charniak , 2001 ) are based on a history-based probability model ( Black et al. , 1993 ) , where the probability of each decision in a parse is conditioned on the previous decisions in the parse . Background train In informal experiments described elsewhere ( Melamed 1995 ) , I found that the G2 statistic suggested by slightly outperforms 02 . Background val For right-branching structures , the leftcorner ancestor is the parent , conditioning on which has been found to be beneficial ( Johnson , 1998 ) , as has conditioning on the left-corner child ( ) . Background train This strategy is certainly the right one to start out with , since anaphora is always the more typical direction of reference in English prose ( , p. 329 ) . Motivation train This indicates that parse trees are usually not the optimal choice for training tree-based translation models ( ) . Background val In and Krotov et al. ( 1998 ) , it was observed that treebank grammars ( CFGs extracted from treebanks ) are very large and grow with the size of the treebank . Background val For example , the forward-backward algorithm ( ) trains only Hidden Markov Models , while ( Ristad and Yianilos , 1996 ) trains only stochastic edit distance . Background train ment ( Sarkar and Wintner , 1999 ; Doran et al. , 2000 ; ) . Background train argues that , aside from missing domain-specific complementation trends , dictionaries produced by hand will tend to lag behind real language use because of their static nature . Motivation train Finally , the Natural Language Toolkit ( NLTK ) is a package of NLP components implemented in Python ( ) . Background train Politically-oriented text Sentiment analysis has specifically been proposed as a key enabling technology in eRulemaking , allowing the automatic analysis of the opinions that people submit ( ; Cardie et al. , 2006 ; Kwon et al. , 2006 ) . Background train Burkett and Klein ( 2008 ) and focused on joint parsing and alignment . CompareOrContrast train But the general outlines are reasonably clear , and we can adapt some of the UDRS ( ) work to our own framework . Uses train KUbler , McDonald , and describe a `` typical '' MaltParser model configuration of attributes and features .13 Starting with it , in a series of initial controlled experiments , we settled on using buf [ 0-1 ] + stk [ 0-1 ] for word-forms , and buf [ 0-3 ] + stk [ 0-2 ] for POS tags . Uses train For this evaluation , we randomly selected 50 abstracts with disorders indexed as the main topic from abstracts retrieved using PubMed on the five clinical questions described in . Uses train The need for information systems to support physicians at the point of care has been well studied ( Covell et al. , 1985 ; ; Ely et al. , 2005 ) . Background train This approach is taken in computational syntactic grammars ( e.g. ) ; the number of unlikely parses is severely reduced whenever possible , but no attempt is made to define only the so-called grammatical strings of a language . Background train Previous work on Chinese SRL mainly focused on how to transplant the machine learning methods which has been successful with English , such as Sun and Jurafsky ( 2004 ) , and Xue ( 2008 ) . Background train For example , the interaction of lexical rules is explored at run-time , even though the possible interaction can be determined at compile-time given the information available in the lexical rules and the base lexical entries .2 Based on the research results reported in , 1996 ) , we propose a new computational treatment of lexical rules that overcomes these shortcomings and results in a more efficient processing of lexical rules as used in HPSG . Motivation train `` Coherence , '' as outlined above , can be understood as a declarative ( or static ) version of marker passing ( ; Charniak 1983 ) , with one difference : the activation spreads to theories that share a predicate , not through the IS-A hierarchy , and is limited to elementary facts about predicates appearing in the text . CompareOrContrast train On small data sets all of the Bayesian estimators strongly outperform EM ( and , to a lesser extent , VB ) with respect to all of our evaluation measures , confirming the results reported in . CompareOrContrast train In addition , the advantages of using linguistically annotated data over raw data are well documented ( ; Granger and Rayson , 1998 ) . Background train For the A * algorithm ( ) as applied to speech recognition , the actual path score is typically augmented with an estimated score for the unseen portion . Uses train Robust natural language understanding in Atlas-Andes is provided by Rosé 's CARMEL system ( Rosé 2000 ) ; it uses the spelling correction algorithm devised by . Uses train The framework represents a generalization of several predecessor NLG systems based on Meaning-Text Theory : FoG ( Kittredge and Polguere , 1991 ) , LFS ( Iordanskaja et al. , 1992 ) , and JOYCE ( Rambow and ) . Extends train Nevertheless , , page 23 ) observes that `` a slightly more general mapping , where two adjacent terminal symbols can be merged into a single lexical item ( for example , a word and its case-marking ) , can capture this sort of result quite handily . '' Background train Other similar approaches include those of Cicekli and G ¨ uvenir ( 1996 ) , McTait and Trujillo ( 1999 ) , Carl ( 1999 ) , and , inter alia . Background train ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ; ) proposed other optimization objectives by introducing a margin-based and ranking-based indirect loss functions . CompareOrContrast train utilized a transformation-based method to learn a sequence of monolingual tree transformations for translation . CompareOrContrast train Baseline language model : For P0 we used a trigram with modified kneser-ney smoothing [ Chen and ] , which is still considered one of the best smoothing methods for n-gram language models . Uses train Therefore , we preprocess Ontonote-5 .0 to derive mention heads using Collins head rules ( ) with gold constituency parsing information and gold named entity information . Uses train Our most accurate single grammar achieves an F score of 91.6 on the WSJ test set , rivaling discriminative reranking approaches ( ) and products of latent variable grammars ( Petrov , 2010 ) , despite being a single generative PCFG . CompareOrContrast train We have presented an ensemble approach to word sense disambiguation ( ) where multiple Naive Bayesian classifiers , each based on co -- occurrence features from varying sized windows of context , is shown to perform well on the widely studied nouns interest and line . Background train Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; ; Aleven et al. , 2001 ; Buckley and Wolska , 2007 ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background train We shall see this in the next example : two sentences , regarded as a fragment of paragraph , are a variation on a theme by . Extends train Nevertheless , the full document text is present in most systems , sometimes as the only feature ( ) and sometimes in combination with others see for instance ( Chen and Martin , 2007 ; Popescu and Magnini , 2007 ) - . Background train The recognizer for these systems is the SUMMIT system ( ) , which uses a segmental-based framework and includes an auditory model in the front-end processing . Uses train Some recent GRE algorithms have done away with the separation between content determination and linguistic realization , interleaving the two processes instead ( ; Krahmer and Theune 2002 ) . CompareOrContrast train To sample from our proposal distribution , we use a blocked Gibbs sampler based on the one proposed by Goodman ( 1998 ) and used by that samples entire parse trees . Uses train Semantic construction proceeds from the derived tree ( ) rather than -- as is more common in TAG -- from the derivation tree . Background train It allows the construction of a non-TAL ( ) , ( Harbusch & Poller , 2000 ) . Background val Following previous work ( e.g. , and Ponzetto and Strube ( 2006 ) ) , we generate training instances as follows : a positive instance is created for each anaphoric NP , NPj , and its closest antecedent , NPi ; and a negative instance is created for NPj paired with each of the intervening NPs , NPi +1 , NPi +2 , ... , NPj_1 . Uses train TF-IDF ( term frequency-inverse document frequency ) is one of the widely used feature selection techniques in information retrieval ( ) . Background train The gap mechanism resembles the Hold register idea of ATNs ( ) and the treatment of bounded domination metavariables in lexical functional grammars ( LFGs ) ( Bresnan 1982 , p. 235 ff . ) CompareOrContrast train To solve these scaling issues , we implement Online Variational Bayesian Inference ( Hoffman et al. , 2010 ; ) for our models . Uses val Typical examples are Bulgarian ( Simov et al. , 2005 ; ) , Chinese ( Chen et al. , 2003 ) , Danish ( Kromann , 2003 ) , and Swedish ( Nilsson et al. , 2005 ) . Background train This appeared to solve the problem , and the results presented later for the average degree of generalisation do not show an over-generalisation compared with those given in . CompareOrContrast train These observations and this line of reasoning has not escaped the attention of theoretical linguists : propose that argument structure is , in fact , encoded syntactically . Background train inter-document references in the form of hyperlinks ( ) . Background train While IA is generally thought to be consistent with findings on human language production ( ; Levelt 1989 ; Pechmann 1989 ; Sonnenschein 1982 ) , the hypothesis that incrementality is a good model of human GRE seems unfalsifiable until a preference order is specified for the properties on which it operates . Background train Arabic has two kinds of plurals : broken plurals and sound plurals ( Wightwick and Gaafar , 1998 ; ) . Background val Secondly , as ( ) show , marginalizing out the different segmentations during decoding leads to improved performance . Future train Nivre ( 2008 ) reports experiments on Arabic parsing using his MaltParser ( ) , trained on the PADT . Background train More recently , ( Sebastiani , 2002 ) has performed a good survey of document categorization ; recent works can also be found in ( ) , ( Crammer and Singer , 2003 ) , and ( Lewis et al. , 2004 ) . Background train The grammar conversion from LTAG to HPSG ( ) is the core portion of the RenTAL system . Background train This includes work on question answering ( ) , sentiment analysis ( Nakagawa et al. , 2010 ) , MT reordering ( Xu et al. , 2009 ) , and many other tasks . Background val Our knowledge extractors rely extensively on MetaMap ( ) , a system for identifying segments of text that correspond to concepts in the UMLS Metathesaurus . Uses train We study the cases where a 9Recall that even the system , built on the world 's largest corpus , achieves only 34 % recall ( Table 1 ) ( with only 48 % of positives and 27 % of all pairs previously observed , but see Footnote 5 ) . CompareOrContrast train Inspired by ( Blunsom et al. , 2009 ) and ( ) , we define P ( str | frag ) as follows : where csw is the number of words in the source string . Motivation train Inspired by ( ) and ( Cohn and Blunsom , 2009 ) , we define P ( str | frag ) as follows : where csw is the number of words in the source string . Motivation train The TNT POS tagger ( ) has also been designed to train and run very quickly , tagging between 30,000 and 60,000 words per second . Background train Some previous works ( Bannard and Callison-Burch , 2005 ; ; Kouylekov et al. , 2009 ) indicate , as main limitations of the mentioned resources , their limited coverage , their low precision , and the fact that they are mostly suitable to capture relations mainly between single words . Background val Similarly , report that the use of a subset of Spanish morphological features ( number for adjectives , determiners , nouns , pronouns , and verbs ; and mode for verbs ) outperforms other combinations . Background train To solve these scaling issues , we implement Online Variational Bayesian Inference ( ; Hoffman et al. , 2012 ) for our models . Uses train Other studies which view lR as a query generation process include ; Hiemstra and Kraaij , 1999 ; Ponte and Croft , 1998 ; Miller et al , 1999 . CompareOrContrast train As a result , researchers have re-adopted the once-popular knowledge-rich approach , investigating a variety of semantic knowledge sources for common noun resolution , such as the semantic relations between two NPs ( e.g. , Ji et al. ( 2005 ) ) , their semantic similarity as computed using WordNet ( e.g. , Poesio et al. ( 2004 ) ) or Wikipedia ( ) , and the contextual role played by an NP ( see Bean and Riloff ( 2004 ) ) . Background train There have been many studies on parsing techniques ( ; Flickinger et al. , 2000 ) , ones on disambiguation models ( Chiang , 2000 ; Kanayama et al. , 2000 ) , and ones on programming/grammar-development environ - Background train An example of this is the estimation of maximum entropy models , from simple iterative estimation algorithms used by Ratnaparkhi ( 1998 ) that converge very slowly , to complex techniques from the optimisation literature that converge much more rapidly ( ) . Background train For example , such schema can serve as a mean to represent translation examples , or find structural correspondences for the purpose of transfer grammar learning ( Menezes & Richardson , 2001 ) , ( Aramaki et al. , 2001 ) , ( Watanabe et al. , 2000 ) , ( Meyers et al. , 2000 ) , ( Matsumoto et al. , 1993 ) , ( kaji et al. , 1992 ) , and example-base machine translation EBMT3 ( Sato & Nagao , 1990 ) , ( ) , ( Richardson et al. , 2001 ) , ( Al-Adhaileh & Tang , 1999 ) . Background train Later works , such as a ) , Bolshakov ( 2004 ) , Taskiran et al. ( 2006 ) and Topkara et al. ( 2006b ) , further made use of part-ofspeech taggers and electronic dictionaries , such as WordNet and VerbNet , to increase the robustness of the method . Background train results are based on a corpus of movie subtitles ( Tiedemann 2007 ) , and are consequently shorter sentences , whereas the En → Es results are based on a corpus of parliamentary proceedings ( ) . Uses train We work with a semi-technical text on meteorological phenomena ( ) , meant for primary school students . Uses train This Principle of Finitism is also assumed by Johnson-Laird ( 1983 ) , Jackendoff ( 1983 ) , , and implicitly or explicitly by almost all researchers in computational linguistics . CompareOrContrast train The candidate feature templates include : Voice from . Uses train Over the past decade , researchers at IBM have developed a series of increasingly sophisticated statistical models for machine translation ( Brown et al. , 1988 ; Brown et al. , 1990 ; a ) . Background train The PICO framework ( ) for capturing well-formulated clinical queries ( described in Section 2 ) can serve as the basis of a knowledge representation that bridges the needs of clinicians and analytical capabilities of a system . Background train The Chinese PropBank has labeled the predicateargument structures of sentences from the Chinese TreeBank ( ) . Uses val Furthermore , manually selected word pairs are often biased towards highly related pairs ( ) , because human annotators tend to select only highly related pairs connected by relations they are aware of . Background train Disjunctive feature descriptions are also possible ; WIT incorporates an efficient method for handling disjunctions ( ) . Uses train Our strategy is based on the approach presented by . Uses train Indeed , contrary to the more classical statistical methods ( Mutual Information , Loglike ... , see below ) used for collocation acquisition ( see ( ) for a review ) , these patterns allow : Background train A further complication is that different speakers can regard very different values as prototypical , making it difficult to assess which of two objects is greener even on one dimension ( , pages 10 -- 12 ) . Background train The M step then treats c as fixed , observed data and adjusts 0 until the predicted vector of total feature counts equals c , using Improved Iterative Scaling ( Della ; Chen and Uses train Both kinds of annotation were carried out using ANVIL ( ) . Uses train The article classifier is a discriminative model that draws on the state-of-the-art approach described in . Uses train The language grounding problem has received significant attention in recent years , owed in part to the wide availability of data sets ( e.g. Flickr , Von Ahn ( 2006 ) ) , computing power , improved computer vision models ( ; Lowe , 2004 ; Farhadi et al. , 2009 ; Parikh and Grauman , 2011 ) and neurological evidence of ties between the language , perceptual and motor systems in the brain ( Pulverm ¨ uller et al. , 2005 ; Tettamanti et al. , 2005 ; Aziz-Zadeh et al. , 2006 ) . Background val , p. 294 ) studied , among other simple text normalization techniques , the effect of case normalization for different words and showed that `` sometimes case variants refer to the same thing ( hurricane and Hurricane ) , sometimes they refer to different things ( continental and Continental ) and sometimes they do n't refer to much of anything ( e.g. , anytime and Anytime ) . '' Background train We follow the notation convention of . Uses train Later works , such as Atallah et al. ( 2001a ) , Bolshakov ( 2004 ) , Taskiran et al. ( 2006 ) and b ) , further made use of part-ofspeech taggers and electronic dictionaries , such as WordNet and VerbNet , to increase the robustness of the method . Background train did not report inter-subject correlation for their larger dataset . CompareOrContrast train This is a similar conclusion to our previous work in . CompareOrContrast train We then use the program Snob ( ; Wallace 2005 ) to cluster these experiences . Uses train The priorities are used for disambiguating interpretation in the incremental understanding method ( b ) . Uses train A number of alignment techniques have been proposed , varying from statistical methods ( Brown et al. , 1991 ; ) to lexical methods ( Kay and Roscheisen , 1993 ; Chen , 1993 ) . Background train present an illustrative first-order fragment along these lines and are able to supply a coherent formal semantics for the CLF-QLFs themselves , using a technique essentially equivalent to supervaluations : a QLF is true iff all its possible RQLFs are , false iff they are all false , and undefined otherwise . Background val We use the open-source Moses toolkit ( ) to build a phrase-based SMT system trained on mostly MSA data ( 64M words on the Arabic side ) obtained from several LDC corpora including some limited DA data . Uses train This approach is taken , for example , in LKB ( ) where lexical rules are introduced on a par with phrase structure rules and the parser makes no distinction between lexical and nonlexical rules ( Copestake 1993 , 31 ) . CompareOrContrast train This confirms that although Kozima 's approach ( ) is computationally expensive , it does produce more precise segmentation . CompareOrContrast train recently described a hybrid method for finding abbreviations and their definitions . Background train More specifically , the notion of the phrasal lexicon ( used first by ) has been used successfully in a number of areas : Background train run a finite-state NP parser on a POS-tagged corpus to calculate the relative frequency of the same six subcategorization verb classes . Background train This section , which elaborates on preliminary results reported in , describes extraction algorithms for population , problems , interventions , outcomes , and the strength of evidence . Extends train It would seem therefore that the iteration of the PT operation to form a closure is needed ( cfXXX b ) . CompareOrContrast train asked subjects to identify the target of a vague description in a visual scene . Background train Our most accurate single grammar achieves an F score of 91.6 on the WSJ test set , rivaling discriminative reranking approaches ( Charniak and Johnson , 2005 ) and products of latent variable grammars ( ) , despite being a single generative PCFG . CompareOrContrast train Our recovery policy is modeled on the TargetedHelp ( ) policy used in task-oriented dialogue . Extends train It has been argued that , in an incremental approach , gradable properties should be given a low preference ranking because they are difficult to process ( ) . CompareOrContrast train ` See ( ) for a discussion of the appropriateness of TIG for HPSG and a comparison with other feature logic approaches designed for HPSG . Background train Specifically , we used Decision Graphs ( ) for Doc-Pred , and SVMs ( Vapnik 1998 ) for Sent-Pred .11 Additionally , we used unigrams for clustering documents and sentences , and unigrams and bigrams for predicting document clusters and sentence clusters ( Sections 3.1.2 and 3.2.2 ) . Uses val There are many plausible representations , such as pairs of trees from synchronous tree adjoining grammars ( Abeille et al. 1990 ; Shieber 1994 ; ) , lexical conceptual structures ( Dorr 1992 ) and WordNet synsets ( Fellbaum 1998 ; Vossen 1998 ) . Background train ones , DIRT ( Lin and Pantel , 2001 ) , VerbOcean ( Chklovski and Pantel , 2004 ) , FrameNet ( ) , and Wikipedia ( Mehdad et al. , 2010 ; Kouylekov et al. , 2009 ) . Background train In the latter case , we can also take care of transferring the value of z. However , as discussed by , creating several instances of lexical rules can be avoided . Motivation train It maximizes the probability of getting the entire DA sequence correct , but it does not necessarily find the DA sequence that has the most DA labels correct ( ) . Background train A substring in the sentence that corresponds to a node in the representation tree is denoted by assigning the interval of the substring to SNODE of 2 These definitions are based on the discussion in ( Tang , 1994 ) and . Uses train We found that the oldest system ( ) yielded the best prototypes , and that using these prototypes gave state-of-the-art performance on WSJ , as well as improvements on nearly all of the non-English corpora . Background train Other definitions of predicates may be found in ( ) . Background train For the sake of completeness , we report in this section also the results obtained adopting the `` basic solution '' proposed by ( ) . CompareOrContrast train The representations used by Danlos ( 2000 ) , Gardent and Webber ( 1998 ) , or are similar , but do not ( always ) explicitly represent the clause combining operations as labeled nodes . Background train Since earlier versions of the SNoW based CSCL were used only to identify single phrases ( ; Munoz et al. , 1999 ) and never to identify a collection of several phrases at the same time , as we do here , we also trained and tested it under the exact conditions of CoNLL-2000 ( Tjong Kim Sang and Buchholz , 2000 ) to compare it to other shallow parsers . Extends train If differences in meaning between senses are very fine-grained , distinguishing between them is hard even for humans ( ) .6 Pairs containing such words are not suitable for evaluation . Background train The application of domain models and deep semantic knowledge to question answering has been explored by a variety of researchers ( e.g. , Jacquemart and Zweigenbaum 2003 , ) , and was also the focus of recent workshops on question answering in restricted domains at ACL 2004 and AAAI 2005 . Background train Griffiths et al. ( 2007 ) helped pave the path for cognitive-linguistic multimodal research , showing that Latent Dirichlet Allocation outperformed Latent Semantic Analysis ( ) in the prediction of association norms . Background train Although not the first to employ a generative approach to directly model content , the seminal work of is a noteworthy point of reference and comparison . CompareOrContrast train Others include selectional preferences , transitivity ( ) , mutual exclusion , symmetry , etc. . Background train • cross-language information retrieval ( e.g. , ) , • multilingual document filtering ( e.g. , Oard 1997 ) , • computer-assisted language learning ( e.g. , Nerbonne et al. 1997 ) , • certain machine-assisted translation tools ( e.g. , Macklovitch 1994 ; Melamed 1996a ) , • concordancing for bilingual lexicography ( e.g. , Catizone , Russell , and Warwick 1989 ; Gale and Church 1991 ) , Background val For example , ( ) discusses the evaluation of two different text categorization strategies with several variations of their feature spaces . Background train As stated before , the experiments are run in the ACE '04 framework ( ) where the system will identify mentions and will label them ( cfXXX Section 4 ) with a type ( person , organization , etc ) , a sub-type ( OrgCommercial , OrgGovernmental , etc ) , a mention level ( named , nominal , etc ) , and a class ( specific , generic , etc ) . Uses train Thus , the second class of SBD systems employs machine learning techniques such as decision tree classifiers ( Riley 1989 ) , neural networks ( Palmer and Hearst 1994 ) , and maximum-entropy modeling ( ) . Background train or quotation of messages in emails or postings ( see but cfXXX Agrawal et al. ( 2003 ) ) . Background train The first work to do this with topic models is b ) . Background train The language chosen for semantic representation is a flat semantics along the line of ( Bos , 1995 ; Copestake et al. , 1999 ; ) . CompareOrContrast train Tetreault 's contribution features comparative evaluation involving the author 's own centering-based pronoun resolution algorithm called the Left-Right Centering algorithm ( LRC ) as well as three other pronoun resolution methods : Hobbs 's naive algorithm ( ) , BFP ( Brennan , Friedman , and Pollard 1987 ) , and Strube 's 5list approach ( Strube 1998 ) . Background train For future work , we might investigate how machine learning algorithms , which are specifically designed for the problem of domain adaptation ( ; Jiang and Zhai , 2007 ) , perform in comparison to our approach . Future train The X2 statistic is performing at least as well as G2 , throwing doubt on the claim by that the G2 statistic is better suited for use in corpus-based NLP . CompareOrContrast train Provided with the candidate fragment elements , we previously ( ) used a chunker3 to finalize the output fragments , in order to follow the linguistic definition of a ( para - ) phrase . Extends train There is a rich literature on organization and lexical access of morphologically complex words where experiments have been conducted mainly for derivational suffixed words of English , Hebrew , Italian , French , Dutch , and few other languages ( Marslen-Wilson et al. , 2008 ; Frost et al. , 1997 ; Grainger , et al. , 1991 ; ) . Background train This method of incorporating dictionary information seems simpler than the method proposed by Brown et al. for their models ( b ) . CompareOrContrast train One important example is the constituentcontext model ( CCM ) of , which was specifically designed to capture the linguistic observation made by Radford ( 1988 ) that there are regularities to the contexts in which constituents appear . Background train Japanese ( ) , despite a very high accuracy , is different in that attachment score drops from 98 % to 85 % , as we go from length 1 to 2 , which may have something to do with the data consisting of transcribed speech with very short utterances . CompareOrContrast train 7 We ignore the rare `` false idafa '' construction ( , p. 102 ) . Background train Various approaches for computing semantic relatedness of words or concepts have been proposed , e.g. dictionary-based ( Lesk , 1986 ) , ontology-based ( ; Leacock and Chodorow , 1998 ) , information-based ( Resnik , 1995 ; Jiang and Conrath , 1997 ) or distributional ( Weeds and Weir , 2005 ) . Background train Some methods are based on likelihood ( Och and Ney , 2002 ; Blunsom et al. , 2008 ) , error rate ( Och , 2003 ; Zhao and Chen , 2009 ; Pauls et al. , 2009 ; Galley and Quirk , 2011 ) , margin ( Watanabe et al. , 2007 ; Chiang et al. , 2008 ) and ranking ( ) , and among which minimum error rate training ( MERT ) ( Och , 2003 ) is the most popular one . Background train We follow in allowing a small set of generic , linguistically-plausible unary and binary grammar rules . Uses train ( 7 ) NEIGHBOR : Research in lexical semantics suggests that the SC of an NP can be inferred from its distributionally similar NPs ( see a ) ) . Motivation val Discrepancies in length throw constituents off balance , and so prosodic phrasing will cross constituent boundaries in order to give the phrases similar lengths ; this is the case in Chickens were eating II the remaining green vegetables , where the subject-predicate boundary finds no prosodic correspondent .4 The most explicit version of this approach is the analysis presented in ( henceforth G&G ) . CompareOrContrast train Our approach to the problem is more compatible with the empirical evidence we presented in our prior work ( ) where we analyzed the output of Chinese to English machine translation and found that there is no correlation between sentence length and MT quality . CompareOrContrast train For all experiments reported in this section we used the syntactic dependency parser MaltParser v1 .3 ( , 2008 ; Kübler , McDonald , and Nivre 2009 ) , a transition-based parser with an input buffer and a stack , which uses SVM classifiers Uses train Against the background of a growing interest in multilingual NLP , multilingual anaphora / coreference resolution has gained considerable momentum in recent years ( Aone and McKee 1993 ; Azzam , Humphreys , and Gaizauskas 1998 ; Harabagiu and Maiorano 2000 ; Mitkov and Barbu 2000 ; Mitkov 1999 ; ; Mitkov , Belguith , and Stys 1998 ) . Background train We use the same data setting with Xue ( 2008 ) , however a bit different from . CompareOrContrast train They proved to be useful in a number of NLP applications such as natural language generation ( Iordanskaja et al. , 1991 ) , multidocument summarization ( McKeown et al. , 2002 ) , automatic evaluation of MT ( ) , and TE ( Dinu and Wang , 2009 ) . Motivation train Moreover , a sandbox is a temporary view of a document itself i.e. a sandbox can not cause a change in the history ( ) . Background train Regarding future work , there are many research line that may be followed : i ) Capturing more features by employing external knowledge such as ontological , lexical resource or WordNet-based features ( a ; Basili et al. , 2005b ; Bloehdorn et al. , 2006 ; Bloehdorn and Moschitti , 2007 ) or shallow semantic trees , ( Giuglea and Moschitti , 2004 ; Giuglea and Moschitti , 2006 ; Moschitti and Bejan , 2004 ; Moschitti et al. , 2007 ; Moschitti , 2008 ; Moschitti et al. , 2008 ) . Future val Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( Andrews et al. , 2009 ; Steyvers , 2010 ; Feng and Lapata , 2010b ; Bruni et al. , 2011 ; Silberer and Lapata , 2012 ; ; Bruni et al. , 2012a ; Bruni et al. , 2012b ; Silberer et al. , 2013 ) . Background train Future research should apply the work of Blunsom et al. ( 2008 ) and , who marginalize over derivations to find the most probable translation rather than the most probable derivation , to these multi-nonterminal grammars . Future train We have since improved the interface by incorporating a capability in the recognizer to propose additional solutions in turn once the first one fails to parse ( ) To produce these `` N-best '' alternatives , we make use of a standard A * search algorithm ( Hart 1968 , Jelinek 1976 ) . Uses train Background train OT therefore holds out the promise of simplifying grammars , by factoring all complex phenomena into simple surface-level constraints that partially mask one another .1 Whether this is always possible under an appropriate definition of `` simple constraints '' ( e.g. , b ) is of course an empirical question . Background train Consider , for example , the lexical rule in Figure 2 , which encodes a passive lexical rule like the one presented by Pollard and Sag ( 1987 , 215 ) in terms of the setup of , ch . Background train This result is consistent with other works using this model with these features ( ; Silberer and Lapata , 2012 ) . CompareOrContrast train reported a correlation of r = .9026.10 The results are not directly comparable , because he only used noun-noun pairs , words instead of concepts , a much smaller dataset , and measured semantic similarity instead of semantic relatedness . CompareOrContrast test Similar observation for surface word frequency was also observed by ( Bertram et al. , 2000 ; ; Burani et al. , 1987 ; Burani et al. , 1984 ; Schreuder et al. , 1997 ; Taft 1975 ; Taft , 2004 ) where it has been claimed that words having low surface frequency tends to decompose . Background test But their importance has grown far beyond machine translation : for instance , transferring annotations between languages ( ; Hwa et al. 2005 ; Ganchev , Gillenwater , and Taskar 2009 ) ; discovery of paraphrases ( Bannard and Callison-Burch 2005 ) ; and joint unsupervised POS and parser induction across languages ( Snyder and Barzilay 2008 ) . Motivation test Previous sentiment-analysis work in different domains has considered inter-document similarity ( Agarwal and Bhattacharyya , 2005 ; Pang and Lee , 2005 ; ) or explicit Background test However , the method we are currently using in the ATIS domain ( ) represents our most promising approach to this problem . Uses test Henceforth the collaborative traits of blogs and wikis ( ) emphasize annotation , comment , and strong editing . Background test The ICA system ( ) aims to reduce the training time by introducing independence assumptions on the training samples that dramatically reduce the training time with the possible downside of sacrificing performance . Background test To this end , several toolkits for building spoken dialogue systems have been developed ( Barnett and Singh , 1997 ; ) . Background test Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; Ratnaparkhi , 1997 ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; ; Tjong Kim Sang and Buchholz , 2000 ) . Background test Task properties Determining whether or not a speaker supports a proposal falls within the realm of sentiment analysis , an extremely active research area devoted to the computational treatment of subjective or opinion-oriented language ( early work includes Wiebe and Rapaport ( 1988 ) , , Sack ( 1994 ) , and Wiebe ( 1994 ) ; see Esuli ( 2006 ) for an active bibliography ) . Background test Various approaches for computing semantic relatedness of words or concepts have been proposed , e.g. dictionary-based ( Lesk , 1986 ) , ontology-based ( Wu and Palmer , 1994 ; Leacock and Chodorow , 1998 ) , information-based ( ; Jiang and Conrath , 1997 ) or distributional ( Weeds and Weir , 2005 ) . Background test Both tasks are performed with a statistical framework : the mention detection system is similar to the one presented in ( Florian et al. , 2004 ) and the coreference resolution system is similar to the one described in ( ) . CompareOrContrast test The advantage of tuning similarity to the application of interest has been shown previously by . CompareOrContrast test Although there are other discussions of the paragraph as a central element of discourse ( e.g. , Halliday and Hasan 1976 , Longacre 1979 , Haberlandt et al. 1980 ) , all of them share a certain limitation in their formal techniques for analyzing paragraph structure . CompareOrContrast test Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers ( Collins , 1997 ; Charniak , 1997a ; Charniak , 1997b ; ) , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ( Church , 1988 ; Ramshaw and Marcus , 1995 ; Argamon et al. , 1998 ; Cardie and Pierce , 1998 ; Munoz et al. , 1999 ; Punyakanok and Roth , 2001 ; Buchholz et al. , 1999 ; Tjong Kim Sang and Buchholz , 2000 ) . Background test We experiment with four learners commonly employed in language learning : Decision List ( DL ) : We use the DL learner as described in Collins and Singer ( 1999 ) , motivated by its success in the related tasks of word sense disambiguation ( ) and NE classification ( Collins and Singer , 1999 ) . Motivation test A central technique is to define a joint relation as a noisy-channel model , by composing a joint relation with a cascade of one or more conditional relations as in Fig. 1 ( ; Knight and Graehl , 1998 ) . Background test We use the same set of binary features as in previous work on this dataset ( Pang et al. , 2002 ; ; Zaidan et al. , 2007 ) . Uses test Our classification framework , directly inspired by , integrates both perspectives , optimizing its labeling of speech segments based on both individual speech-segment classification scores and preferences for groups of speech segments to receive the same label . Uses test As for work on Arabic ( MSA ) , results have been reported on the PATB ( Kulick , Gabbard , and Marcus 2006 ; Diab 2007 ; ) , the Prague Dependency Treebank ( PADT ) ( Buchholz and Marsi 2006 ; Nivre 2008 ) and the CATiB ( Habash and Roth 2009 ) . Background test For instance , report that the SATZ system ( decision tree variant ) was trained on a set of about 800 labeled periods , which corresponds to a corpus of about 16,000 words . CompareOrContrast test One possible direction is to consider linguistically motivated approaches , such as the extraction of syntactic phrase tables as proposed by ( ) . Future test Later works , such as Atallah et al. ( 2001a ) , , Taskiran et al. ( 2006 ) and Topkara et al. ( 2006b ) , further made use of part-ofspeech taggers and electronic dictionaries , such as WordNet and VerbNet , to increase the robustness of the method . Background test A number of speech understanding systems have been developed during the past fifteen years ( Barnett et al. 1980 , Dixon and Martin 1979 , Erman et al. 1980 , , Lea 1980 , Lowerre and Reddy 1980 , Medress 1980 , Reddy 1976 , Walker 1978 , and Wolf and Woods 1980 ) . CompareOrContrast test The bottom panel of table 1 lists the results for the chosen lexicalized model ( SSN-Freq > 200 ) and five recent statistical parsers ( Ratnaparkhi , 1999 ; Collins , 1999 ; Charniak , 2000 ; ; Bod , 2001 ) . CompareOrContrast test The basic Python reflection has already been implemented and used for large scale experiments with POS tagging , using pyMPI ( a message passing interface library for Python ) to coordinate experiments across a cluster of over 100 machines ( ; Clark et al. , 2003 ) . Background test This imbalance foils thresholding strategies , clever as they might be ( Gale & Church , 1991 ; Wu & Xia , 1994 ; ) . Background test Training was done on the Penn Treebank ( ) Wall Street Journal data , sections 02-21 . Uses test We performed Latent Semantic Analysis ( LSA ) over Wikipedia using the jLSI tool ( ) to measure the relatedness between words in the dataset . Uses test For example , our previous work ( ; Nakov and Ng , 2012 ) experimented with various techniques for combining a small bi-text for a resource-poor language ( Indonesian or Spanish , pretending that Spanish is resource-poor ) with a much larger bi-text for a related resource-rich language ( Malay or Portuguese ) ; the target language of all bi-texts was English . CompareOrContrast test Various approaches for computing semantic relatedness of words or concepts have been proposed , e.g. dictionary-based ( ) , ontology-based ( Wu and Palmer , 1994 ; Leacock and Chodorow , 1998 ) , information-based ( Resnik , 1995 ; Jiang and Conrath , 1997 ) or distributional ( Weeds and Weir , 2005 ) . Background test Another line of research approaches grounded language knowledge by augmenting distributional approaches of word meaning with perceptual information ( Andrews et al. , 2009 ; Steyvers , 2010 ; Feng and Lapata , 2010b ; Bruni et al. , 2011 ; Silberer and Lapata , 2012 ; Johns and Jones , 2012 ; a ; Bruni et al. , 2012b ; Silberer et al. , 2013 ) . Background test replicated the experiment of Rubenstein and Goodenough with the original 65 word pairs translated into German . Background test One approach to this more general problem , taken by the ` Nitrogen ' generator ( a ; Langkilde and Knight , 1998b ) , takes advantage of standard statistical techniques by generating a lattice of all possible strings given a semantic representation as input and selecting the most likely output using a bigram language model . Uses test where mk is one mention in entity e , and the basic model building block PL ( L = 1 | e , mk , m ) is an exponential or maximum entropy model ( ) . Uses test 13 We also employed sequence-based measures using the ROUGE tool set ( ) , with similar results to those obtained with the word-by-word measures . Uses test Second , using continuous distributions allows us to leverage a variety of tools ( e.g. , LDA ) that have been shown to be successful in other fields , such as speech recognition ( ) . Background test In this section , we validate the contribution of key tag sets and morphological features -- and combinations thereof -- using a different parser : the Easy-First Parser ( ) . Uses test The typical solution to the redundancy problem is to group verbs according to their argument realization patterns ( ) , possibly arranged in an inheritance hierarchy . CompareOrContrast test Later , , 1982 ) proposed a knowledge base in which information about language and the world would be encoded , and he emphasized the need for using `` salience '' in choosing facts from this knowledge base . Background test Another technique is automatic discovery of translations from parallel or non-parallel corpora ( ) . Background test ASARES is presented in detail in ( ) . Uses test Opposition ( called `` adversative '' or `` contrary-to-expectation '' by ; cfXXX also Quirk et al. 1972 , p. 672 ) . Background test A number of applications have relied on distributional analysis ( ) in order to build classes of semantically related terms . Background test Previous work with MaltParser in Russian , Turkish , and Hindi showed gains with CASE but not with agreement features ( Eryigit , Nivre , and Oflazer 2008 ; Nivre , Boguslavsky , and Iomdin 2008 ; ) . CompareOrContrast test Consider , for example , the lexical rule in Figure 2 , which encodes a passive lexical rule like the one presented by , 215 ) in terms of the setup of Pollard and Sag ( 1994 , ch . CompareOrContrast test Two applications that , like help-desk , deal with question -- answer pairs are : summarization of e-mail threads ( Dalli , Xia , and Wilks 2004 ; ) , and answer extraction in FAQs ( Frequently Asked Questions ) ( Berger and Mittal 2000 ; CompareOrContrast test The language grounding problem has received significant attention in recent years , owed in part to the wide availability of data sets ( e.g. Flickr , Von Ahn ( 2006 ) ) , computing power , improved computer vision models ( Oliva and Torralba , 2001 ; Lowe , 2004 ; ; Parikh and Grauman , 2011 ) and neurological evidence of ties between the language , perceptual and motor systems in the brain ( Pulverm ¨ uller et al. , 2005 ; Tettamanti et al. , 2005 ; Aziz-Zadeh et al. , 2006 ) . Background test In addition , we find that the Bayesian SCFG grammar can not even significantly outperform the heuristic SCFG grammar ( ) 5 . CompareOrContrast test There are several grammars developed in the FB-LTAG formalism , including the XTAG English grammar , a large-scale grammar for English ( The XTAG Research ) . Background test Although the approach may have potential , the shifting of complex accounting into the unification algorithm is at variance with the findings of , who report large speed-ups from the elimination of disjunction processing during unification . CompareOrContrast test For the task of unsupervised dependency parsing , add a constraint of the form `` the average length of dependencies should be X '' to capture the locality of syntax ( at least half of the dependencies are between adjacent words ) , using a scheme they call structural annealing . Background test The speech and language processing architecture is based on that of the SRI CommandTalk system ( ; Stent et a. , 1999 ) . Uses test Second , in line with the findings of ( ) , the results obtained over the MT-derived corpus are equal to those we achieve over the original RTE3 dataset ( i.e. 63.50 % ) . CompareOrContrast test Therefore , inter-subject correlation is lower than the results obtained by . CompareOrContrast test There is a general consensus among theoretical linguists that the proper representation of verbal argument structure is event structure -- representations grounded in a theory of events that decompose semantic roles in terms of primitive predicates representing concepts such as causality and inchoativity ( Dowty , 1979 ; Jackendoff , 1983 ; Pustejovsky , 1991b ; Rappaport ) . Background test For example , some similar measures have been used in stylistic experiments in information retrieval on the basis of a robust parser built for information retrieval purposes ( ) . Background test The resulting training procedure is analogous to the one presented in ( Brown et al. , 1993 ) and ( ) . CompareOrContrast test successfully parses , or until a quitting criterion is reached , such as an upper bound on N. Whereas in the loosely coupled system the parser acts as a filter only on completed candidate solutions ( ) , the tightly coupled system allows the parser to discard partial theories that have no way of continuing . Uses test substituted the non-terminal X in hierarchical phrase-based model by extended syntactic categories . CompareOrContrast test Much of the earlier work in anaphora resolution heavily exploited domain and linguistic knowledge ( ; Carter 1987 ; Rich and LuperFoy 1988 ; Carbonell and Brown 1988 ) , which was difficult both to represent and to process , and which required considerable human input . Background test The paradigm is `` write many , read many '' ( ) . Background test The Praat tool was used ( ) . Uses test 2 The reader is asked to focus on any reasonable size measurement , for example , the maximal horizontal or vertical distance , or some combination of dimensions ( ; also Section 8.1 of the present article ) . Background test The implementation has been inspired by experience in extracting information from very large corpora ( Curran and Moens , 2002 ) and performing experiments on maximum entropy sequence tagging ( ; Clark et al. , 2003 ) . Motivation test Default parameters were used , although experimentation with different parameter settings is an important direction for future work ( Daelemans and Hoste , 2002 ; ) . Future test Our work is inspired by the latent left-linking model in and the ILP formulation from Chang et al. ( 2011 ) . Uses test Furthermore , the availability of rich ontological resources , in the form of the Unified Medical Language System ( UMLS ) ( Lindberg et al. , 1993 ) , and the availability of software that leverages this knowledge -- MetaMap ( Aronson , 2001 ) for concept identification and SemRep ( ) for relation extraction -- provide a foundation for studying the role of semantics in various tasks . Background test The names given to the components vary ; they have been called `` strategic '' and `` tactical '' components ( e.g. , McKeown 1985 ; Thompson 1977 ; Danlos 1987 ) 1 , `` planning '' and `` realization '' ( e.g. , McDonald 1983 ; a ) , or simply `` what to say '' versus `` how to say it '' ( e.g. , Danlos 1987 ; Reithinger 1990 ) . Background test Over the last decade there has been a lot of interest in developing tutorial dialogue systems that understand student explanations ( Jordan et al. , 2006 ; Graesser et al. , 1999 ; Aleven et al. , 2001 ; ; Nielsen et al. , 2008 ; VanLehn et al. , 2007 ) , because high percentages of selfexplanation and student contentful talk are known to be correlated with better learning in humanhuman tutoring ( Chi et al. , 1994 ; Litman et al. , 2009 ; Purandare and Litman , 2008 ; Steinhauser et al. , 2007 ) . Background test We use the TRIPS dialogue parser ( ) to parse the utterances . Uses test In order to address these limitations in a practical way , we conducted a small user study where we asked four judges ( graduate students from the Faculty of Information Technology at Monash University ) to assess the responses generated by our system ( a ) . Uses test The understanding module utilizes ISSS ( Incremental Significant-utterance Sequence Search ) ( b ) , which is an integrated parsing and discourse processing method . Uses test We applied our system to the XTAG English grammar ( The XTAG Research ) 3 , which is a large-scale FB-LTAG grammar for English . Uses test After the extraction , pruning techniques ( ) can be applied to increase the precision of the extracted paraphrases . Background test In this paper , we extend two classes of model adaptation methods ( i.e. , model interpolation and error-driven learning ) , which have been well studied in statistical language modeling for speech and natural language applications ( e.g. , ; Bellegarda , 2004 ; Gao et al. , 2006 ) , to ranking models for Web search applications . Background test GATE goes beyond earlier systems by using a component-based infrastructure ( ) which the GUI is built on top of . Background test Since sentences can refer to events described by other sentences , we may need also a quotation operator ; describes how first order logic can be augmented with such an operator . Background test The system uses a knowledge base implemented in the KM representation language ( Clark and Porter , 1999 ; ) to represent the state of the world . Uses test A possible future direction would be to compare the query string to retrieved results using a method similar to that of . Future test description-level lexical rules ( DLRs ; ) .5 2.2.1 Meta-Level Lexical Rules . Background test All EBMT systems , from the initial proposal by to the recent collection of Carl and Way ( 2003 ) , are premised on the availability of subsentential alignments derived from the input bitext . Background test The necessity of this kind of merging of arguments has been recognized before : Charniak and McDermott ( 1985 ) call it abductive unification/matching , , 1979 ) refers to such operations using the terms knitting or petty conversational implicature . Background test In a number of proposals , lexical generalizations are captured using lexical underspecification ( Kathol 1994 ; ; CompareOrContrast test These keywords are potentially useful features because some of them are subclasses of the ACE SCs shown in the left column of Table 1 , while others appear to be correlated with these ACE SCs .2 ( 6 ) INDUCED CLASS : Since the first-sense heuristic used in the previous feature may not be accurate in capturing the SC of an NP , we employ a corpusbased method for inducing SCs that is motivated by research in lexical semantics ( e.g. , ) . Motivation test Other psycholing-uistic studies that confirm the validity of paragraph units can be found in and Haberlandt et al. ( 1980 ) . Background test The bottom panel of table 1 lists the results for the chosen lexicalized model ( SSN-Freq > 200 ) and five recent statistical parsers ( Ratnaparkhi , 1999 ; Collins , 1999 ; Charniak , 2000 ; Collins , 2000 ; ) . CompareOrContrast test Nevertheless , the full document text is present in most systems , sometimes as the only feature ( Sugiyama and Okumura , 2007 ) and sometimes in combination with others see for instance ( Chen and Martin , 2007 ; ) - . Background test In a similar vain to and Buchholz et al. ( 1999 ) , the method extends an existing flat shallow-parsing method to handle composite structures . Future test As a result , researchers have re-adopted the once-popular knowledge-rich approach , investigating a variety of semantic knowledge sources for common noun resolution , such as the semantic relations between two NPs ( e.g. , ) , their semantic similarity as computed using WordNet ( e.g. , Poesio et al. ( 2004 ) ) or Wikipedia ( Ponzetto and Strube , 2006 ) , and the contextual role played by an NP ( see Bean and Riloff ( 2004 ) ) . Background test We built a two-stage baseline system , using the perceptron segmentation model from our previous work ( ) and the perceptron POS tagging model from Collins ( 2002 ) . Extends test Note that although our current system uses MeSH headings assigned by human indexers , manually assigned terms can be replaced with automatic processing if needed ( ) . Future test Furthermore , medical terminology is characterized by a typical mix of Latin and Greek roots with the corresponding host language ( e.g. , German ) , often referred to as neo-classical compounding ( ) . Background test Previously ( ) , we assessed the importance of various implicit argument feature groups by conducting feature ablation tests . Extends test To model d ( FWi − 1 , S → T ) , d ( FWi +1 , S → T ) , i.e. whether Li , S → T and Ri , S → T extend beyond the neighboring function word phrase pairs , we utilize the pairwise dominance model of . Uses test For instance , , p. 8 ) says that the sentence `` Reagan thinks bananas , '' which is otherwise strange , is in fact acceptable if it occurs as an answer to the question `` What is Kissinger 's favorite fruit ? '' Motivation test Semantic Role labeling ( SRL ) was first defined in . Background test AJAX function lets the communication works asyncronously between a client and a server through a set of messages based on HTTP protocol and XML ( ) . Background test The inclusion of the coreference task in the Sixth and Seventh Message Understanding Conferences ( MUC-6 and MUC-7 ) gave a considerable impetus to the development of coreference resolution algorithms and systems , such as those described in Baldwin et al. ( 1995 ) , Gaizauskas and Humphreys ( 1996 ) , and . Background test The most detailed evaluation of link tokens to date was performed by ( ) , who trained Brown et al. 's Model 2 on 74 million words of the Canadian Hansards . CompareOrContrast test Log-linear models have proved successful in a wide variety of applications , and are the inspiration behind one of the best current statistical parsers ( ) . CompareOrContrast test While we have observed reasonable results with both G2 and Fisher 's exact test , we have not yet discussed how these results compare to the results that can be obtained with a technique commonly used in corpus linguistics based on the mutual information ( MI ) measure ( ) : Background test Morphological alterations of a search term have a negative impact on the recall performance of an information retrieval ( IR ) system ( ; J ¨ appinen and Niemist ¨ o , 1988 ; Kraaij and Pohlmann , 1996 ) , since they preclude a direct match between the search term proper and its morphological variants in the documents to be retrieved . Background test For shuffling paraphrases , french alternations are partially described in ( ) and a resource is available which describes alternation and the mapping verbs/alternations for roughly 1 700 verbs . Background test A more recent approach , advocated by Rappaport Hovav and Levin ( 1998 ) , describes a basic set of event templates corresponding to Vendler 's event classes ( ) : ( 3 ) a. [ x ACT ] ( activity ) b. [ x ] ( state ) c. [ BECOME [ x ] ] ( achievement ) d. [ x CAUSE [ BECOME [ x ] ] ] ( accomplishment ) Background test combines lexical and dependency mappings to form his generalizations . Background test Thus for instance , ( Copestake and Flickinger , 2000 ; ) describes a Head Driven Phrase Structure Grammar ( HPSG ) which supports the parallel construction of a phrase structure ( or derived ) tree and of a semantic representation and ( Dalrymple , 1999 ) show how to equip Lexical Functional grammar ( LFG ) with a glue semantics . Background test The reordering models we describe follow our previous work using function word models for translation ( Setiawan et al. , 2007 ; ) . Extends test And Collins ( 2000 ) argues for `` keeping track of counts of arbitrary fragments within parse trees '' , which has indeed been carried out in who use exactly the same set of ( all ) tree fragments as proposed in Bod ( 1992 ) . Motivation test In our work , we gather sets of sentences , and assume ( but do not employ ) existing approaches for their organization ( Goldstein et al. 2000 ; Barzilay , Elhadad , and McKeown 2001 ; ) . Background test criteria and data used in our experiments are based on the work of . Uses test We present experiments on the two standard coreference resolution datasets , ACE-2004 ( NIST , 2004 ) and OntoNotes-5 .0 ( ) . Uses test • Only qualitative observations of the responses were reported ( no formal evaluation was performed ) ( Lapalme and Kosseim 2003 ; ) . CompareOrContrast test And subderivations headed by A1 with external nonterminals only at the leaves , internal nonterminals elsewhere , have probability 1/a1 ( ) . Background test • Support vector machines for mapping histories to parser actions ( ) . Uses test , 1998 ) developed a polynomial time PCFG-reduction of DOP1 whose size is linear in the size of the training set , thus converting the exponential number of subtrees to a compact grammar . Background test avoids enumerating the various senses for adjectives like fast by exploiting the semantics of the nouns they modify . Background test have conducted a study on dependency parsing for 21 languages using features that encode whether the values for certain attributes are equal or not for a node and its governor . Background test Such approaches have been tried recently in restricted cases ( McCallum et al. , 2000 ; b ; Lafferty et al. , 2001 ) . Background test The relation between discourse and prosodic phrasing has been examined in some detail by , who argues that each noun phrase in an utterance constitutes a separate prosodic phrase unless it is destressed because of reference to previous discourse . Background test By contrast , Turkish ( Oflazer et al. , 2003 ; ) exhibits high root accuracy but consistently low attachment scores ( about 88 % for length 1 and 68 % for length 2 ) . CompareOrContrast test The candidate examples that lead to the most disagreements among the different learners are considered to have the highest TUV ( Cohn , Atlas , and Ladner 1994 ; ) . Background test Subsequently , we extracted the bilingual phrase table from the aligned corpora using the Moses toolkit ( ) . Uses test Representative systems are described in Boisen et al. ( 1989 ) , De Mattia and Giachin ( 1989 ) , Niedermair ( 1989 ) , Niemann ( 1990 ) , and . Background test Our rules for phonological word formation are adopted , for the most part , from G & G , , and the account of monosyllabic destressing in Selkirk ( 1984 ) . Uses test As a generalization , notes that lexicons such as COMLEX tend to demonstrate high precision but low recall . Background test Such systems extract information from some types of syntactic units ( clauses in ( Fillmore and Atkins , 1998 ; Gildea and Jurafsky , 2002 ; ) ; noun phrases in ( Hull and Gomez , 1996 ; Rosario et al. , 2002 ) ) . Background test Various approaches for computing semantic relatedness of words or concepts have been proposed , e.g. dictionary-based ( Lesk , 1986 ) , ontology-based ( Wu and Palmer , 1994 ; Leacock and Chodorow , 1998 ) , information-based ( Resnik , 1995 ; ) or distributional ( Weeds and Weir , 2005 ) . Background test Besides WordNet , the RTE literature documents the use of a variety of lexical information sources ( Bentivogli et al. , 2010 ; ) . Background test The question answering system developed by belongs to the merging category of approaches , where the output of an individual method can be used as input to a different method ( this corresponds to Burke 's cascade sub-category ) . CompareOrContrast test More recently , ( Sebastiani , 2002 ) has performed a good survey of document categorization ; recent works can also be found in ( Joachims , 2002 ) , ( ) , and ( Lewis et al. , 2004 ) . Background test Discriminant analysis has been employed by researchers in automatic text genre detection ( b ; Karlgren and Cutting 1994 ) since it offers a simple and robust solution despite the fact that it presupposes normal distributions of the discriminating variables . Background test This model has previously been shown to provide excellent performance on multiple tasks , including prediction of association norms , word substitution errors , semantic inferences , and word similarity ( ; Silberer and Lapata , 2012 ) . Extends test In other words , existing treatments of gradables in GRE fail to take the `` efficiency of language '' into account ( ; see our Section 2 ) . Background test Word alignments are used primarily for extracting minimal translation units for machine translation ( MT ) ( e.g. , phrases [ Koehn , Och , and Marcu 2003 ] and rules [ ; Chiang et al. 2005 ] ) as well as for Background test Following , the IR system ranks documents according to the probability that a document D is relevant given the query Q , P ( D is R IQ ) . Uses test In modern syntactic theories ( e.g. , lexical-functional grammar [ LFG ] [ Kaplan and Bresnan 1982 ; ; Dalrymple 2001 ] , head-driven phrase structure grammar [ HPSG ] [ Pollard and Sag 1994 ] , tree-adjoining grammar [ TAG ] [ Joshi 1988 ] , and combinatory categorial grammar [ CCG ] [ Ades and Steedman 1982 ] ) , the lexicon is the central repository for much morphological , syntactic , and semantic information . Background test We have shown elsewhere ( Jensen and Binot 1988 ; a , 1987b ) that natural language programs , such as on-line grammars and dictionaries , can be used as referential levels for commonsense reasoning -- for example , to disambiguate PP attachment . Extends test Thus rather than a single training procedure , we can actually partition the examples by predicate , and train a 1For a fixed verb , MI is proportional to 's conditional probability scores for pseudodisambiguation of ( v , n , n ′ ) triples : Pr ( v | n ) = Pr ( v , n ) / Pr ( n ) , which was shown to be a better measure of association than co-occurrence frequency f ( v , n ) . Motivation test