---
language:
- en
pretty_name: MOTIF
license: cc-by-nc-4.0
task_categories:
- image-to-text
- text-to-image
tags:
- multimodal
- image-text-retrieval
- image-text-alignment
- complex-word-identification
- language-learning
- l2-reading
- readability
- datasets
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train.parquet
dataset_info:
  features:
  - name: image
    dtype: image
  - name: id
    dtype: string
  - name: context
    dtype: string
  - name: focus
    dtype: string
  - name: image_id
    dtype: string
---

# MOTIF

MOTIF (MultimOdal ConTextualized Images For Language Learners) is a multimodal dataset introduced in the LREC 2022 paper [MOTIF: Contextualized Images for Complex Words to Improve Human Reading](https://aclanthology.org/2022.lrec-1.263/). The dataset pairs simplified English reading contexts, complex focus words, and contextualized images intended to support second-language reading comprehension.

The uploaded archive contains 1,125 examples from the L2Corpus release. Each example has one reading context, one target focus word, and one annotated image.

## Dataset Structure

This Hugging Face repository is organized as a Parquet image dataset:

```text
data/
  train.parquet
```

The Parquet file embeds each annotated PNG image in an `image` column and keeps the original text metadata alongside it.

## Fields

- `image`: annotated image associated with the focus word and context.
- `id`: original example identifier.
- `context`: reading context sentence.
- `focus`: complex word or target focus word in the context.
- `image_id`: original image identifier without the `.png` suffix.

## Dataset Statistics

| Item | Count |
| --- | ---: |
| examples | 1,125 |
| images | 1,125 |
| unique focus words | 277 |

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("shanewang/MOTIF")
example = dataset["train"][0]

image = example["image"]
context = example["context"]
focus = example["focus"]
```

## Intended Use

MOTIF is intended for research on multimodal language learning, contextualized image retrieval, image-text alignment, complex word identification, and reading support for L2 learners. The dataset can be used to evaluate whether images are contextually suitable for explaining complex words in simplified English reading contexts.

## License

The uploaded archive did not include a standalone license file. The associated LREC 2022 paper is distributed under CC-BY-NC-4.0, so this dataset card uses `cc-by-nc-4.0` as the conservative release license metadata. Please verify the intended data and image licensing before commercial reuse.

## Citation

```bibtex
@inproceedings{wang-etal-2022-motif,
    title = "{MOTIF}: Contextualized Images for Complex Words to Improve Human Reading",
    author = "Wang, Xintong and Schneider, Florian and Alacam, {\"O}zge and Chaudhury, Prateek and Biemann, Chris",
    booktitle = "Proceedings of the Thirteenth Language Resources and Evaluation Conference",
    month = jun,
    year = "2022",
    address = "Marseille, France",
    publisher = "European Language Resources Association",
    url = "https://aclanthology.org/2022.lrec-1.263/",
    pages = "2468--2477"
}
```