@inproceedings{lee-etal-2023-local,
title = "Local Temperature Beam Search: Avoid Neural Text {D}e{G}eneration via Enhanced Calibration",
author = "Lee, Dongkyu and
Kim, Gyeonghun and
Han, Janghoon and
Hong, Taesuk and
Kim, Yi-Reun and
Choi, Stanley Jungkyu and
Zhang, Nevin L.",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings-acl.628/",
doi = "10.18653/v1/2023.findings-acl.628",
pages = "9903--9915",
abstract = "Previous studies have constantly observed that a language model repeats itself, creating repetitions in an output sequence. To cope with the issue, stochastic decoding schemes have been the de facto approaches; the strategies add randomness in inference, hence avoiding the {\textquotedblleft}self-loop{\textquotedblright}. However, the remedy comes at the cost of sacrificing output quality due to the randomness involved. In this work, we introduce a deterministic decoding scheme, local temperature beam search. This inference algorithm is an embarrassingly simple variant of beam search, yet it reduces repetition, whose level is superior to that of a sampling-based decoding algorithm, while maintaining the level of coherence as in beam search. Our idea is rooted in the concept of model calibration; we view a repetition as a casualty from overconfidence in a model. Therefore, our work mitigates the miscalibration present in the course of inference with a post-calibration approach applied in beam-specific manner. Our inference scheme is validated on text completion tasks, in which the repetition problem is seen most clearly, and is exhaustively compared with existing inference schemes."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lee-etal-2023-local">
<titleInfo>
<title>Local Temperature Beam Search: Avoid Neural Text DeGeneration via Enhanced Calibration</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dongkyu</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gyeonghun</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Janghoon</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Taesuk</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yi-Reun</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stanley</namePart>
<namePart type="given">Jungkyu</namePart>
<namePart type="family">Choi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nevin</namePart>
<namePart type="given">L</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2023</title>
</titleInfo>
<name type="personal">
<namePart type="given">Anna</namePart>
<namePart type="family">Rogers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jordan</namePart>
<namePart type="family">Boyd-Graber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naoaki</namePart>
<namePart type="family">Okazaki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Previous studies have constantly observed that a language model repeats itself, creating repetitions in an output sequence. To cope with the issue, stochastic decoding schemes have been the de facto approaches; the strategies add randomness in inference, hence avoiding the “self-loop”. However, the remedy comes at the cost of sacrificing output quality due to the randomness involved. In this work, we introduce a deterministic decoding scheme, local temperature beam search. This inference algorithm is an embarrassingly simple variant of beam search, yet it reduces repetition, whose level is superior to that of a sampling-based decoding algorithm, while maintaining the level of coherence as in beam search. Our idea is rooted in the concept of model calibration; we view a repetition as a casualty from overconfidence in a model. Therefore, our work mitigates the miscalibration present in the course of inference with a post-calibration approach applied in beam-specific manner. Our inference scheme is validated on text completion tasks, in which the repetition problem is seen most clearly, and is exhaustively compared with existing inference schemes.</abstract>
<identifier type="citekey">lee-etal-2023-local</identifier>
<identifier type="doi">10.18653/v1/2023.findings-acl.628</identifier>
<location>
<url>https://aclanthology.org/2023.findings-acl.628/</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>9903</start>
<end>9915</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Local Temperature Beam Search: Avoid Neural Text DeGeneration via Enhanced Calibration
%A Lee, Dongkyu
%A Kim, Gyeonghun
%A Han, Janghoon
%A Hong, Taesuk
%A Kim, Yi-Reun
%A Choi, Stanley Jungkyu
%A Zhang, Nevin L.
%Y Rogers, Anna
%Y Boyd-Graber, Jordan
%Y Okazaki, Naoaki
%S Findings of the Association for Computational Linguistics: ACL 2023
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F lee-etal-2023-local
%X Previous studies have constantly observed that a language model repeats itself, creating repetitions in an output sequence. To cope with the issue, stochastic decoding schemes have been the de facto approaches; the strategies add randomness in inference, hence avoiding the “self-loop”. However, the remedy comes at the cost of sacrificing output quality due to the randomness involved. In this work, we introduce a deterministic decoding scheme, local temperature beam search. This inference algorithm is an embarrassingly simple variant of beam search, yet it reduces repetition, whose level is superior to that of a sampling-based decoding algorithm, while maintaining the level of coherence as in beam search. Our idea is rooted in the concept of model calibration; we view a repetition as a casualty from overconfidence in a model. Therefore, our work mitigates the miscalibration present in the course of inference with a post-calibration approach applied in beam-specific manner. Our inference scheme is validated on text completion tasks, in which the repetition problem is seen most clearly, and is exhaustively compared with existing inference schemes.
%R 10.18653/v1/2023.findings-acl.628
%U https://aclanthology.org/2023.findings-acl.628/
%U https://doi.org/10.18653/v1/2023.findings-acl.628
%P 9903-9915
Markdown (Informal)
[Local Temperature Beam Search: Avoid Neural Text DeGeneration via Enhanced Calibration](https://aclanthology.org/2023.findings-acl.628/) (Lee et al., Findings 2023)
ACL