Centroid-Based Efficient Minimum Bayes Risk Decoding

Centroid-Based Efficient Minimum Bayes Risk Decoding

Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation. However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Deguchi, Hiroyuki, Sakai, Yusuke, Kamigaito, Hidetaka, Watanabe, Taro, Tanaka, Hideki, Utiyama, Masao
Format: Artikel
Sprache:eng
Schlagworte:
Computer Science - Computation and Language
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Deguchi, Hiroyuki
Sakai, Yusuke
Kamigaito, Hidetaka
Watanabe, Taro
Tanaka, Hideki
Utiyama, Masao
description Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation. However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference translations. We propose centroid-based MBR (CBMBR) decoding to improve the speed of MBR decoding. Our method clusters the reference translations in the feature space, and then calculates the score using the centroids of each cluster. The experimental results show that our CBMBR not only improved the decoding speed of the expected score calculation 5.7 times, but also outperformed vanilla MBR decoding in translation quality by up to 0.5 COMET in the WMT'22 En$\leftrightarrow$Ja, En$\leftrightarrow$De, En$\leftrightarrow$Zh, and WMT'23 En$\leftrightarrow$Ja translation tasks.
doi_str_mv 10.48550/arxiv.2402.11197
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2402_11197</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2402_11197</sourcerecordid><originalsourceid>FETCH-LOGICAL-a677-c84b67bdd7fdb48874bdc17e030e950362197a4009a51ed0bf2875c4dcbaa85c3</originalsourceid><addsrcrecordid>eNotzkFuwjAQhWFvWFS0B-gKXyDpOLFjZ1lSWipRISH20dhjoxEkoKStyu0LtKsn_YunT4hHBbl2xsATDj_8nRcailwpVds7oZvYfw5HpmyOYyS5SIkDX5r84J67r07O8RxHueFxL19iOBL3u3sxSXgY48P_TsX2dbFtltlq_fbePK8yrKzNgtO-sp7IJvLaOas9BWUjlBBrA2VVXASoAWo0KhL4VDhrgqbgEZ0J5VTM_m5v7PY0cIfDub3y2xu__AURnz8J</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Centroid-Based Efficient Minimum Bayes Risk Decoding</title><source>arXiv.org</source><creator>Deguchi, Hiroyuki ; Sakai, Yusuke ; Kamigaito, Hidetaka ; Watanabe, Taro ; Tanaka, Hideki ; Utiyama, Masao</creator><creatorcontrib>Deguchi, Hiroyuki ; Sakai, Yusuke ; Kamigaito, Hidetaka ; Watanabe, Taro ; Tanaka, Hideki ; Utiyama, Masao</creatorcontrib><description>Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation. However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference translations. We propose centroid-based MBR (CBMBR) decoding to improve the speed of MBR decoding. Our method clusters the reference translations in the feature space, and then calculates the score using the centroids of each cluster. The experimental results show that our CBMBR not only improved the decoding speed of the expected score calculation 5.7 times, but also outperformed vanilla MBR decoding in translation quality by up to 0.5 COMET in the WMT'22 En$\leftrightarrow$Ja, En$\leftrightarrow$De, En$\leftrightarrow$Zh, and WMT'23 En$\leftrightarrow$Ja translation tasks.</description><identifier>DOI: 10.48550/arxiv.2402.11197</identifier><language>eng</language><subject>Computer Science - Computation and Language</subject><creationdate>2024-02</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2402.11197$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2402.11197$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Deguchi, Hiroyuki</creatorcontrib><creatorcontrib>Sakai, Yusuke</creatorcontrib><creatorcontrib>Kamigaito, Hidetaka</creatorcontrib><creatorcontrib>Watanabe, Taro</creatorcontrib><creatorcontrib>Tanaka, Hideki</creatorcontrib><creatorcontrib>Utiyama, Masao</creatorcontrib><title>Centroid-Based Efficient Minimum Bayes Risk Decoding</title><description>Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation. However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference translations. We propose centroid-based MBR (CBMBR) decoding to improve the speed of MBR decoding. Our method clusters the reference translations in the feature space, and then calculates the score using the centroids of each cluster. The experimental results show that our CBMBR not only improved the decoding speed of the expected score calculation 5.7 times, but also outperformed vanilla MBR decoding in translation quality by up to 0.5 COMET in the WMT'22 En$\leftrightarrow$Ja, En$\leftrightarrow$De, En$\leftrightarrow$Zh, and WMT'23 En$\leftrightarrow$Ja translation tasks.</description><subject>Computer Science - Computation and Language</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotzkFuwjAQhWFvWFS0B-gKXyDpOLFjZ1lSWipRISH20dhjoxEkoKStyu0LtKsn_YunT4hHBbl2xsATDj_8nRcailwpVds7oZvYfw5HpmyOYyS5SIkDX5r84J67r07O8RxHueFxL19iOBL3u3sxSXgY48P_TsX2dbFtltlq_fbePK8yrKzNgtO-sp7IJvLaOas9BWUjlBBrA2VVXASoAWo0KhL4VDhrgqbgEZ0J5VTM_m5v7PY0cIfDub3y2xu__AURnz8J</recordid><startdate>20240217</startdate><enddate>20240217</enddate><creator>Deguchi, Hiroyuki</creator><creator>Sakai, Yusuke</creator><creator>Kamigaito, Hidetaka</creator><creator>Watanabe, Taro</creator><creator>Tanaka, Hideki</creator><creator>Utiyama, Masao</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20240217</creationdate><title>Centroid-Based Efficient Minimum Bayes Risk Decoding</title><author>Deguchi, Hiroyuki ; Sakai, Yusuke ; Kamigaito, Hidetaka ; Watanabe, Taro ; Tanaka, Hideki ; Utiyama, Masao</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a677-c84b67bdd7fdb48874bdc17e030e950362197a4009a51ed0bf2875c4dcbaa85c3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Computation and Language</topic><toplevel>online_resources</toplevel><creatorcontrib>Deguchi, Hiroyuki</creatorcontrib><creatorcontrib>Sakai, Yusuke</creatorcontrib><creatorcontrib>Kamigaito, Hidetaka</creatorcontrib><creatorcontrib>Watanabe, Taro</creatorcontrib><creatorcontrib>Tanaka, Hideki</creatorcontrib><creatorcontrib>Utiyama, Masao</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Deguchi, Hiroyuki</au><au>Sakai, Yusuke</au><au>Kamigaito, Hidetaka</au><au>Watanabe, Taro</au><au>Tanaka, Hideki</au><au>Utiyama, Masao</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Centroid-Based Efficient Minimum Bayes Risk Decoding</atitle><date>2024-02-17</date><risdate>2024</risdate><abstract>Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation. However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference translations. We propose centroid-based MBR (CBMBR) decoding to improve the speed of MBR decoding. Our method clusters the reference translations in the feature space, and then calculates the score using the centroids of each cluster. The experimental results show that our CBMBR not only improved the decoding speed of the expected score calculation 5.7 times, but also outperformed vanilla MBR decoding in translation quality by up to 0.5 COMET in the WMT'22 En$\leftrightarrow$Ja, En$\leftrightarrow$De, En$\leftrightarrow$Zh, and WMT'23 En$\leftrightarrow$Ja translation tasks.</abstract><doi>10.48550/arxiv.2402.11197</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2402.11197
ispartof
issn
language eng
recordid cdi_arxiv_primary_2402_11197
source arXiv.org
subjects Computer Science - Computation and Language
title Centroid-Based Efficient Minimum Bayes Risk Decoding
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-09T18%3A19%3A20IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Centroid-Based%20Efficient%20Minimum%20Bayes%20Risk%20Decoding&rft.au=Deguchi,%20Hiroyuki&rft.date=2024-02-17&rft_id=info:doi/10.48550/arxiv.2402.11197&rft_dat=%3Carxiv_GOX%3E2402_11197%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true