Explaining Contextualization in Language Models using Visual Analytics.

Rita Sevastjanova; Aikaterini-Lida Kalouli; Christin Beck; Hanna Schäfer; Mennatallah El-Assady

doi:10.18653/v1/2021.acl-long.39

Explaining Contextualization in Language Models using Visual Analytics.

Rita Sevastjanova, Aikaterini-Lida Kalouli, Christin Beck, Hanna Schäfer, Mennatallah El-Assady

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

Abstract

Despite the success of contextualized language models on various NLP tasks, it is still unclear what these models really learn. In this paper, we contribute to the current efforts of explaining such models by exploring the continuum between function and content words with respect to contextualization in BERT, based on linguistically-informed insights. In particular, we utilize scoring and visual analytics techniques: we use an existing similarity-based score to measure contextualization and integrate it into a novel visual analytics technique, presenting the model's layers simultaneously and highlighting intra-layer properties and inter-layer differences. We show that contextualization is neither driven by polysemy nor by pure context variation. We also provide insights on why BERT fails to model words in the middle of the functionality continuum.

Original language	English
Title of host publication	Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
Editors	Chengqing Zong, Fei Xia, Wenjie Li, Roberto Navigli
Publisher	Association for Computational Linguistics
Pages	464-476
Number of pages	13
ISBN (Electronic)	9781954085527
ISBN (Print)	9781954085527
DOIs	https://doi.org/10.18653/v1/2021.acl-long.39
Publication status	Published - 2021

Bibliographical note

Funding Information:
We thank the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for funding within project BU 1806/10-2 “Questions Visualized” of the FOR2111 and project D02 “Evaluation Metrics for Visual Analytics in Linguistics” (Project ID: 251654672 – TRR 161).

Publisher Copyright:
© 2021 Association for Computational Linguistics

Access to Document

10.18653/v1/2021.acl-long.39Licence: CC BY

2021.acl-long.39Final published version, 1.79 MBLicence: CC BY

Cite this

Sevastjanova, R., Kalouli, A.-L., Beck, C., Schäfer, H., & El-Assady, M. (2021). Explaining Contextualization in Language Models using Visual Analytics. In C. Zong, F. Xia, W. Li, & R. Navigli (Eds.), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) (pp. 464-476). Association for Computational Linguistics. https://doi.org/10.18653/v1/2021.acl-long.39

Sevastjanova, Rita ; Kalouli, Aikaterini-Lida ; Beck, Christin et al. / Explaining Contextualization in Language Models using Visual Analytics. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). editor / Chengqing Zong ; Fei Xia ; Wenjie Li ; Roberto Navigli. Association for Computational Linguistics, 2021. pp. 464-476

@inproceedings{96109759dd274597bc6755bc61be5d34,

title = "Explaining Contextualization in Language Models using Visual Analytics.",

abstract = "Despite the success of contextualized language models on various NLP tasks, it is still unclear what these models really learn. In this paper, we contribute to the current efforts of explaining such models by exploring the continuum between function and content words with respect to contextualization in BERT, based on linguistically-informed insights. In particular, we utilize scoring and visual analytics techniques: we use an existing similarity-based score to measure contextualization and integrate it into a novel visual analytics technique, presenting the model's layers simultaneously and highlighting intra-layer properties and inter-layer differences. We show that contextualization is neither driven by polysemy nor by pure context variation. We also provide insights on why BERT fails to model words in the middle of the functionality continuum.",

author = "Rita Sevastjanova and Aikaterini-Lida Kalouli and Christin Beck and Hanna Sch{\"a}fer and Mennatallah El-Assady",

note = "Funding Information: We thank the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for funding within project BU 1806/10-2 “Questions Visualized” of the FOR2111 and project D02 “Evaluation Metrics for Visual Analytics in Linguistics” (Project ID: 251654672 – TRR 161). Publisher Copyright: {\textcopyright} 2021 Association for Computational Linguistics",

year = "2021",

doi = "10.18653/v1/2021.acl-long.39",

language = "English",

isbn = "9781954085527",

pages = "464--476",

editor = "Chengqing Zong and Fei Xia and Wenjie Li and Roberto Navigli",

booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",

publisher = "Association for Computational Linguistics",

address = "United States",

}

Sevastjanova, R, Kalouli, A-L, Beck, C, Schäfer, H & El-Assady, M 2021, Explaining Contextualization in Language Models using Visual Analytics. in C Zong, F Xia, W Li & R Navigli (eds), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Association for Computational Linguistics, pp. 464-476. https://doi.org/10.18653/v1/2021.acl-long.39

Explaining Contextualization in Language Models using Visual Analytics. / Sevastjanova, Rita; Kalouli, Aikaterini-Lida; Beck, Christin et al.
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). ed. / Chengqing Zong; Fei Xia; Wenjie Li; Roberto Navigli. Association for Computational Linguistics, 2021. p. 464-476.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Explaining Contextualization in Language Models using Visual Analytics.

AU - Sevastjanova, Rita

AU - Kalouli, Aikaterini-Lida

AU - Beck, Christin

AU - Schäfer, Hanna

AU - El-Assady, Mennatallah

N1 - Funding Information: We thank the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for funding within project BU 1806/10-2 “Questions Visualized” of the FOR2111 and project D02 “Evaluation Metrics for Visual Analytics in Linguistics” (Project ID: 251654672 – TRR 161). Publisher Copyright: © 2021 Association for Computational Linguistics

PY - 2021

Y1 - 2021

N2 - Despite the success of contextualized language models on various NLP tasks, it is still unclear what these models really learn. In this paper, we contribute to the current efforts of explaining such models by exploring the continuum between function and content words with respect to contextualization in BERT, based on linguistically-informed insights. In particular, we utilize scoring and visual analytics techniques: we use an existing similarity-based score to measure contextualization and integrate it into a novel visual analytics technique, presenting the model's layers simultaneously and highlighting intra-layer properties and inter-layer differences. We show that contextualization is neither driven by polysemy nor by pure context variation. We also provide insights on why BERT fails to model words in the middle of the functionality continuum.

AB - Despite the success of contextualized language models on various NLP tasks, it is still unclear what these models really learn. In this paper, we contribute to the current efforts of explaining such models by exploring the continuum between function and content words with respect to contextualization in BERT, based on linguistically-informed insights. In particular, we utilize scoring and visual analytics techniques: we use an existing similarity-based score to measure contextualization and integrate it into a novel visual analytics technique, presenting the model's layers simultaneously and highlighting intra-layer properties and inter-layer differences. We show that contextualization is neither driven by polysemy nor by pure context variation. We also provide insights on why BERT fails to model words in the middle of the functionality continuum.

UR - https://www.scopus.com/pages/publications/85118923186

U2 - 10.18653/v1/2021.acl-long.39

DO - 10.18653/v1/2021.acl-long.39

M3 - Conference contribution

SN - 9781954085527

SP - 464

EP - 476

BT - Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)

A2 - Zong, Chengqing

A2 - Xia, Fei

A2 - Li, Wenjie

A2 - Navigli, Roberto

PB - Association for Computational Linguistics

ER -

Sevastjanova R, Kalouli AL, Beck C, Schäfer H, El-Assady M. Explaining Contextualization in Language Models using Visual Analytics. In Zong C, Xia F, Li W, Navigli R, editors, Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Association for Computational Linguistics. 2021. p. 464-476 doi: 10.18653/v1/2021.acl-long.39

Explaining Contextualization in Language Models using Visual Analytics.

Abstract

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this