Natural Language Inference (NLI) is an NLP task that requires recognising whether there is a logical entailment or contradiction between two natural language statements, or the lack thereof.
The Cross-lingual Natural Language Inference Corpus (XNLI) already provides NLI benchmarking data in 15 other languages. The myXNLI corpus extends XNLI with Myanmar (Burmese) language.
For myXNLI, we human-translated all 7,500 sentence pairs from XNLI English dev and test sets into Myanmar. The NLI and Genre labels from English dev and test sets are also reused for the Myanmar datasets.
The dataset also includes the NLI training data in Myanmar which is created by machine-translating the MultiNLI training data from English into Myanmar. Similar to XNLI, we also reuse the existing NLI and Genre labels for English training data for the Myanmar version.
In addition to the NLI dev, test and training datasets, we also appended Myanmar translations to the XNLI 15-language parallel corpus, to create a 16-language parallel corpus.
Myanmar XNLI: Building a Dataset and Exploring Low-resource Approaches to Natural Language Inference with Myanmar https://www.researchsquare.com/article/rs-4329843/
- Myanmar NLI Test Dataset - 5010 records (tsv)
- Myanmar NLI Validation Dataset - 1490 records (tsv)
- Myannmar NLI Training Data - 392,702 records (tsv.gz)
- Parallel Corpus in 16 Languages (ar, bg, de, el, en, es, fr, hi, my, ru, sw, th, tr, ur, vi, zh) (tsv)
- HuggingFace dataset
This dataset is licensed under Creative Commons Attribution-NonCommercial
| Sentence-1 (Premise) | Sentence-2 (Hypothesis) | Label | Genre |
|---|---|---|---|
| သင် ဒီမှာနေစရာ မလိုပါဘူး။ | မင်း ထွက်သွားနိုင်တယ်။ | Entailment | face-to-face |
| You don’t have to stay there. | You can leave. | ||
| သင် ဒီမှာနေစရာ မလိုပါဘူး။ | မင်း သွားချင်ရင် အိမ်ကို သွားနိုင်တယ်။ | Neutral | face-to-face |
| You don’t have to stay there. | You can go home if you want to. | ||
| သင် ဒီမှာနေစရာ မလိုပါဘူး။ | မင်း အဲ့ဒီနေရာအတိအကျမှာ နေဖို့လိုတယ်။ | Contradiction | face-to-face |
| You don’t have to stay there. | You need to stay in that exact spot! |
Under translation folder, there are 100 files, each containing 100 blocks. Each block has a block number, an English sentence and a placeholder for Myanmar Translation. An example entry in a translation file is described below.
114
We were watching something on TV.
ကျွန်တော်တို့ တီဗီမှာ တခုခု ကြည့်နေခဲ့သည်။
# REVIEW
# This is a comment explaining details about the problem.
The first line makes a reference to the line number of the English sentence in the XNLI corpus.
The second line contains the actual English sentence to be translated.
The third line is reserved for Myanmar translation of the English sentence.
Additional and optional lines for human translator notes are also allowed with a hash prefix (#). This is useful for flagging translations that require review or documenting any observations made during translation.
Lastly, each entry in the file is separated by a blank line followed by another entry.
The translation revision was carried out in a private git repo, but final revised translation files have been imported into this (myXNLI) repo.
Each phase of myXNLI dataset development is contributed by the following volunteers.
- Aung Kyaw Htet
- Aye Mya Hlaing
- Hsu Myat Mo
- Win Pa Pa
- Yi Mon Shwe Sin
- Aye Nyein Mon
- Ei Myat Myat Noe
- Hay Mar Soe Naing
- Hnin Nandar Zaw
- Myint Myint Wai
- Wai Lai Lai Phyu
- Yadanar Oo
- Zaw Mee
- Aung Kyaw Htet
- Htoo Htet Aung
- Junie Soe
- Thar Htet
- Thein Aung Tan
- Thidar Nwe
- Thiha Kyaw Zaw
- Yair Pike
- Yi Sandi Soe
- 2 Freelancers with the financial support from Macquarie University
- Htet Cho
- Lin Thurein Tun
- Thein Than Phyo
- Zay Ye Htut