Probing neural network comprehension of natural language arguments

Timothy Niven; Hung Yu Kao

Probing neural network comprehension of natural language arguments

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

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.

Original language	English
Title of host publication	ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference
Publisher	Association for Computational Linguistics (ACL)
Pages	4658-4664
Number of pages	7
ISBN (Electronic)	9781950737482
Publication status	Published - 2020
Event	57th Annual Meeting of the Association for Computational Linguistics, ACL 2019 - Florence, Italy Duration: 2019 Jul 28 → 2019 Aug 2

Publication series

Name	ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference

Conference

Conference	57th Annual Meeting of the Association for Computational Linguistics, ACL 2019
Country/Territory	Italy
City	Florence
Period	19-07-28 → 19-08-02

All Science Journal Classification (ASJC) codes

Language and Linguistics
Computer Science(all)
Linguistics and Language

Cite this

Niven, T., & Kao, H. Y. (2020). Probing neural network comprehension of natural language arguments. In ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference (pp. 4658-4664). (ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference). Association for Computational Linguistics (ACL).

Niven, Timothy ; Kao, Hung Yu. / Probing neural network comprehension of natural language arguments. ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference. Association for Computational Linguistics (ACL), 2020. pp. 4658-4664 (ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference).

@inproceedings{afdc14f81f994491bf160599ddb4b509,

title = "Probing neural network comprehension of natural language arguments",

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.",

author = "Timothy Niven and Kao, {Hung Yu}",

note = "Publisher Copyright: {\textcopyright} 2019 Association for Computational Linguistics; 57th Annual Meeting of the Association for Computational Linguistics, ACL 2019 ; Conference date: 28-07-2019 Through 02-08-2019",

year = "2020",

language = "English",

series = "ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference",

publisher = "Association for Computational Linguistics (ACL)",

pages = "4658--4664",

booktitle = "ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference",

}

Niven, T & Kao, HY 2020, Probing neural network comprehension of natural language arguments. in ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference. ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference, Association for Computational Linguistics (ACL), pp. 4658-4664, 57th Annual Meeting of the Association for Computational Linguistics, ACL 2019, Florence, Italy, 19-07-28.

Probing neural network comprehension of natural language arguments. / Niven, Timothy; Kao, Hung Yu.

ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference. Association for Computational Linguistics (ACL), 2020. p. 4658-4664 (ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Probing neural network comprehension of natural language arguments

AU - Niven, Timothy

AU - Kao, Hung Yu

PY - 2020

Y1 - 2020

N2 - 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.

AB - 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.

UR - http://www.scopus.com/inward/record.url?scp=85084006698&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85084006698&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:85084006698

T3 - ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference

SP - 4658

EP - 4664

BT - ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference

PB - Association for Computational Linguistics (ACL)

T2 - 57th Annual Meeting of the Association for Computational Linguistics, ACL 2019

Y2 - 28 July 2019 through 2 August 2019

ER -

Niven T, Kao HY. Probing neural network comprehension of natural language arguments. In ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference. Association for Computational Linguistics (ACL). 2020. p. 4658-4664. (ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Proceedings of the Conference).

Probing neural network comprehension of natural language arguments

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Other files and links

Fingerprint

Cite this