Ruth-Ann/jampatoisnli

Dataset Card for [Dataset Name]

Dataset Summary

JamPatoisNLI provides the first dataset for natural language inference in a creole language, Jamaican Patois. Many of the most-spoken low-resource languages are creoles. These languages commonly have a lexicon derived from a major world language and a distinctive grammar reflecting the languages of the original speakers and the process of language birth by creolization. This gives them a distinctive place in exploring the effectiveness of transfer from large monolingual or multilingual pretrained models.

Supported Tasks and Leaderboards

Natural language inference

Languages

Jamaican Patois

Data Fields

premise, hypothesis, label

Data Splits

Train: 250 Val: 200 Test: 200

Data set creation + Annotations

Premise collection: 97% of examples from Twitter; remaining pulled from literature and online cultural website

Hypothesis construction: For each premise, hypothesis written by native speaker (our first author) so that pair’s classification would be E, N or C

Label validation: Random sample of 100 sentence pairs double annotated by fluent speakers

Social Impact of Dataset

JamPatoisNLI is a low-resource language dataset in an English-based Creole spoken in the Caribbean, Jamaican Patois. The creation of the dataset contributes to expanding the scope of NLP research to under-explored languages across the world.

Dataset Curators

@ruth-ann

Citation Information

@misc{https://doi.org/10.48550/arxiv.2212.03419, doi = {10.48550/ARXIV.2212.03419},

url = {https://arxiv.org/abs/2212.03419},

author = {Armstrong, Ruth-Ann and Hewitt, John and Manning, Christopher},

keywords = {Computation and Language (cs.CL), Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences, I.2.7},

title = {JamPatoisNLI: A Jamaican Patois Natural Language Inference Dataset},

publisher = {arXiv},

year = {2022},

copyright = {arXiv.org perpetual, non-exclusive license} }

Contributions

Thanks to Prof. Christopher Manning and John Hewitt for their contributions, guidance, facilitation and support related to the creation of this dataset.