Probing Neural Network Comprehension of Natural Language Arguments

Probing Neural Network Comprehension of Natural Language Arguments. Niven, T. & Kao, H. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4658–4664, Stroudsburg, PA, USA, 2019. Association for Computational Linguistics.

Paper doi abstract bibtex

We are surprised to find that BERT's peak performance of 77% on the Argument Reasoning Comprehension Task reaches just three points below the average untrained human baseline. However, we show that this result is entirely accounted for by exploitation of spurious statistical cues in the dataset. We analyze the nature of these cues and demonstrate that a range of models all exploit them. This analysis informs the construction of an adversarial dataset on which all models achieve random accuracy. Our adversarial dataset provides a more robust assessment of argument comprehension and should be adopted as the standard in future work.

@inproceedings{Niven2019,
abstract = {We are surprised to find that BERT's peak performance of 77% on the Argument Reasoning Comprehension Task reaches just three points below the average untrained human baseline. However, we show that this result is entirely accounted for by exploitation of spurious statistical cues in the dataset. We analyze the nature of these cues and demonstrate that a range of models all exploit them. This analysis informs the construction of an adversarial dataset on which all models achieve random accuracy. Our adversarial dataset provides a more robust assessment of argument comprehension and should be adopted as the standard in future work.},
address = {Stroudsburg, PA, USA},
author = {Niven, Timothy and Kao, Hung-Yu},
booktitle = {Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},
doi = {10.18653/v1/P19-1459},
file = {:Users/shanest/Documents/Library/Niven, Kao/Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics/Niven, Kao - 2019 - Probing Neural Network Comprehension of Natural Language Arguments.pdf:pdf},
keywords = {method: adversarial data,phenomenon: argumentation},
pages = {4658--4664},
publisher = {Association for Computational Linguistics},
title = {{Probing Neural Network Comprehension of Natural Language Arguments}},
url = {https://www.aclweb.org/anthology/P19-1459},
year = {2019}
}

Downloads: 0

{"_id":"3WBbBXEiivXAxcWe8","bibbaseid":"niven-kao-probingneuralnetworkcomprehensionofnaturallanguagearguments-2019","authorIDs":[],"author_short":["Niven, T.","Kao, H."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"We are surprised to find that BERT's peak performance of 77% on the Argument Reasoning Comprehension Task reaches just three points below the average untrained human baseline. However, we show that this result is entirely accounted for by exploitation of spurious statistical cues in the dataset. We analyze the nature of these cues and demonstrate that a range of models all exploit them. This analysis informs the construction of an adversarial dataset on which all models achieve random accuracy. Our adversarial dataset provides a more robust assessment of argument comprehension and should be adopted as the standard in future work.","address":"Stroudsburg, PA, USA","author":[{"propositions":[],"lastnames":["Niven"],"firstnames":["Timothy"],"suffixes":[]},{"propositions":[],"lastnames":["Kao"],"firstnames":["Hung-Yu"],"suffixes":[]}],"booktitle":"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics","doi":"10.18653/v1/P19-1459","file":":Users/shanest/Documents/Library/Niven, Kao/Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics/Niven, Kao - 2019 - Probing Neural Network Comprehension of Natural Language Arguments.pdf:pdf","keywords":"method: adversarial data,phenomenon: argumentation","pages":"4658–4664","publisher":"Association for Computational Linguistics","title":"Probing Neural Network Comprehension of Natural Language Arguments","url":"https://www.aclweb.org/anthology/P19-1459","year":"2019","bibtex":"@inproceedings{Niven2019,\nabstract = {We are surprised to find that BERT's peak performance of 77% on the Argument Reasoning Comprehension Task reaches just three points below the average untrained human baseline. However, we show that this result is entirely accounted for by exploitation of spurious statistical cues in the dataset. We analyze the nature of these cues and demonstrate that a range of models all exploit them. This analysis informs the construction of an adversarial dataset on which all models achieve random accuracy. Our adversarial dataset provides a more robust assessment of argument comprehension and should be adopted as the standard in future work.},\naddress = {Stroudsburg, PA, USA},\nauthor = {Niven, Timothy and Kao, Hung-Yu},\nbooktitle = {Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},\ndoi = {10.18653/v1/P19-1459},\nfile = {:Users/shanest/Documents/Library/Niven, Kao/Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics/Niven, Kao - 2019 - Probing Neural Network Comprehension of Natural Language Arguments.pdf:pdf},\nkeywords = {method: adversarial data,phenomenon: argumentation},\npages = {4658--4664},\npublisher = {Association for Computational Linguistics},\ntitle = {{Probing Neural Network Comprehension of Natural Language Arguments}},\nurl = {https://www.aclweb.org/anthology/P19-1459},\nyear = {2019}\n}\n","author_short":["Niven, T.","Kao, H."],"key":"Niven2019","id":"Niven2019","bibbaseid":"niven-kao-probingneuralnetworkcomprehensionofnaturallanguagearguments-2019","role":"author","urls":{"Paper":"https://www.aclweb.org/anthology/P19-1459"},"keyword":["method: adversarial data","phenomenon: argumentation"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","biburl":"https://www.shane.st/teaching/575/win20/MachineLearning-interpretability.bib","creationDate":"2020-01-05T04:04:02.893Z","downloads":0,"keywords":["method: adversarial data","phenomenon: argumentation"],"search_terms":["probing","neural","network","comprehension","natural","language","arguments","niven","kao"],"title":"Probing Neural Network Comprehension of Natural Language Arguments","year":2019,"dataSources":["okYcdTpf4JJ2zkj7A","C5FtkvWWggFfMJTFX","znj7izS5PeehdLR3G"]}