This repository contains the MoNLI dataset, with the original splits used in the paper.
The files are in JSONL format, where each line is a dictionary with the following structure:
'sentence1'(str): premise sentence'sentence2'(str): hypothesis sentence'gold_label'(str): either "entailment" or "neutral"'sentence1_lex'(str): the target lexical item in the premise'sentence2_lex'(str): the target lexical item in the hypothesis'depth'(int): the distance between'sentence1_lex'and'sentence2_lex'in the WordNet graph
import json
def load_split(filename):
dataset = []
with open(filename) as f:
for line in f:
d = json.loads(line)
dataset.append(d)
return dataset
@inproceedings{geiger-etal-2020-neural,
address = {Online},
author = {Geiger, Atticus and Richardson, Kyle and Potts, Christopher},
booktitle = {Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP},
doi = {10.18653/v1/2020.blackboxnlp-1.16},
month = nov,
pages = {163--173},
publisher = {Association for Computational Linguistics},
title = {Neural Natural Language Inference Models Partially Embed Theories of Lexical Entailment and Negation},
url = {https://www.aclweb.org/anthology/2020.blackboxnlp-1.16},
year = {2020}}