Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging

Aarne Talman; Hande Celikkanat; Sami Virpioja; Markus Heinonen; Jörg Tiedemann

Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging

Aarne Talman, Hande Celikkanat, Sami Virpioja, Markus Heinonen, Jörg Tiedemann

University of Helsinki

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

24 Downloads (Pure)

Abstract

This paper introduces Bayesian uncertainty modeling using Stochastic Weight Averaging-Gaussian (SWAG) in Natural Language Understanding (NLU) tasks. We apply the approach to standard tasks in natural language inference (NLI) and demonstrate the effectiveness of the method in terms of prediction accuracy and correlation with human annotation disagreements. We argue that the uncertainty representations in SWAG better reflect subjective interpretation and the natural variation that is also present in human language understanding. The results reveal the importance of uncertainty modeling, an often neglected aspect of neural language modeling, in NLU tasks.

Original language	English
Title of host publication	Proceedings of the 24th Nordic Conference on Computational Linguistics
Publisher	University of Tartu Library
Pages	358-365
ISBN (Electronic)	978-9916-21-999-7
Publication status	Published - 2023
MoE publication type	A4 Conference publication
Event	Nordic Conference on Computational Linguistics - Tórshavn, Faroe Islands Duration: 22 May 2023 → 24 May 2023 Conference number: 24

Conference

Conference	Nordic Conference on Computational Linguistics
Abbreviated title	NoDaLiDa
Country/Territory	Faroe Islands
City	Tórshavn
Period	22/05/2023 → 24/05/2023

Access to Document

Uncertainty-Aware Natural Language Inference with Stochastic Weight AveragingFinal published version, 368 KBLicence: CC BY

https://aclanthology.org/2023.nodalida-1.37/Licence: CC BY

Cite this

@inproceedings{45e6f42314cd4aeea38bb3f88b5ea40a,

title = "Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging",

abstract = "This paper introduces Bayesian uncertainty modeling using Stochastic Weight Averaging-Gaussian (SWAG) in Natural Language Understanding (NLU) tasks. We apply the approach to standard tasks in natural language inference (NLI) and demonstrate the effectiveness of the method in terms of prediction accuracy and correlation with human annotation disagreements. We argue that the uncertainty representations in SWAG better reflect subjective interpretation and the natural variation that is also present in human language understanding. The results reveal the importance of uncertainty modeling, an often neglected aspect of neural language modeling, in NLU tasks.",

author = "Aarne Talman and Hande Celikkanat and Sami Virpioja and Markus Heinonen and J{\"o}rg Tiedemann",

note = "| openaire: EC/H2020/771113/EU//FoTran ; Nordic Conference on Computational Linguistics, NoDaLiDa ; Conference date: 22-05-2023 Through 24-05-2023",

year = "2023",

language = "English",

pages = "358--365",

booktitle = "Proceedings of the 24th Nordic Conference on Computational Linguistics",

publisher = "University of Tartu Library",

address = "Estonia",

}

Talman, A, Celikkanat, H, Virpioja, S, Heinonen, M & Tiedemann, J 2023, Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging. in Proceedings of the 24th Nordic Conference on Computational Linguistics. University of Tartu Library, pp. 358-365, Nordic Conference on Computational Linguistics, Tórshavn, Faroe Islands, 22/05/2023. <https://aclanthology.org/2023.nodalida-1.37/>

Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging. / Talman, Aarne; Celikkanat, Hande; Virpioja, Sami et al.
Proceedings of the 24th Nordic Conference on Computational Linguistics. University of Tartu Library, 2023. p. 358-365.

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging

AU - Talman, Aarne

AU - Celikkanat, Hande

AU - Virpioja, Sami

AU - Heinonen, Markus

AU - Tiedemann, Jörg

N1 - Conference code: 24

PY - 2023

Y1 - 2023

N2 - This paper introduces Bayesian uncertainty modeling using Stochastic Weight Averaging-Gaussian (SWAG) in Natural Language Understanding (NLU) tasks. We apply the approach to standard tasks in natural language inference (NLI) and demonstrate the effectiveness of the method in terms of prediction accuracy and correlation with human annotation disagreements. We argue that the uncertainty representations in SWAG better reflect subjective interpretation and the natural variation that is also present in human language understanding. The results reveal the importance of uncertainty modeling, an often neglected aspect of neural language modeling, in NLU tasks.

AB - This paper introduces Bayesian uncertainty modeling using Stochastic Weight Averaging-Gaussian (SWAG) in Natural Language Understanding (NLU) tasks. We apply the approach to standard tasks in natural language inference (NLI) and demonstrate the effectiveness of the method in terms of prediction accuracy and correlation with human annotation disagreements. We argue that the uncertainty representations in SWAG better reflect subjective interpretation and the natural variation that is also present in human language understanding. The results reveal the importance of uncertainty modeling, an often neglected aspect of neural language modeling, in NLU tasks.

M3 - Conference article in proceedings

SP - 358

EP - 365

BT - Proceedings of the 24th Nordic Conference on Computational Linguistics

PB - University of Tartu Library

T2 - Nordic Conference on Computational Linguistics

Y2 - 22 May 2023 through 24 May 2023

ER -

Uncertainty-Aware Natural Language Inference with Stochastic Weight Averaging

Abstract

Conference

Access to Document

Fingerprint

Cite this