@inproceedings{casas-etal-2020-combining,
title = "Combining Subword Representations into Word-level Representations in the Transformer Architecture",
author = "Casas, Noe and
Costa-juss{\`a}, Marta R. and
Fonollosa, Jos{\'e} A. R.",
editor = "Rijhwani, Shruti and
Liu, Jiangming and
Wang, Yizhong and
Dror, Rotem",
booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop",
month = jul,
year = "2020",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2020.acl-srw.10",
doi = "10.18653/v1/2020.acl-srw.10",
pages = "66--71",
abstract = "In Neural Machine Translation, using word-level tokens leads to degradation in translation quality. The dominant approaches use subword-level tokens, but this increases the length of the sequences and makes it difficult to profit from word-level information such as POS tags or semantic dependencies. We propose a modification to the Transformer model to combine subword-level representations into word-level ones in the first layers of the encoder, reducing the effective length of the sequences in the following layers and providing a natural point to incorporate extra word-level information. Our experiments show that this approach maintains the translation quality with respect to the normal Transformer model when no extra word-level information is injected and that it is superior to the currently dominant method for incorporating word-level source language information to models based on subword-level vocabularies.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="casas-etal-2020-combining">
<titleInfo>
<title>Combining Subword Representations into Word-level Representations in the Transformer Architecture</title>
</titleInfo>
<name type="personal">
<namePart type="given">Noe</namePart>
<namePart type="family">Casas</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marta</namePart>
<namePart type="given">R</namePart>
<namePart type="family">Costa-jussà</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">José</namePart>
<namePart type="given">A</namePart>
<namePart type="given">R</namePart>
<namePart type="family">Fonollosa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2020-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shruti</namePart>
<namePart type="family">Rijhwani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiangming</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yizhong</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rotem</namePart>
<namePart type="family">Dror</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Online</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>In Neural Machine Translation, using word-level tokens leads to degradation in translation quality. The dominant approaches use subword-level tokens, but this increases the length of the sequences and makes it difficult to profit from word-level information such as POS tags or semantic dependencies. We propose a modification to the Transformer model to combine subword-level representations into word-level ones in the first layers of the encoder, reducing the effective length of the sequences in the following layers and providing a natural point to incorporate extra word-level information. Our experiments show that this approach maintains the translation quality with respect to the normal Transformer model when no extra word-level information is injected and that it is superior to the currently dominant method for incorporating word-level source language information to models based on subword-level vocabularies.</abstract>
<identifier type="citekey">casas-etal-2020-combining</identifier>
<identifier type="doi">10.18653/v1/2020.acl-srw.10</identifier>
<location>
<url>https://aclanthology.org/2020.acl-srw.10</url>
</location>
<part>
<date>2020-07</date>
<extent unit="page">
<start>66</start>
<end>71</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Combining Subword Representations into Word-level Representations in the Transformer Architecture
%A Casas, Noe
%A Costa-jussà, Marta R.
%A Fonollosa, José A. R.
%Y Rijhwani, Shruti
%Y Liu, Jiangming
%Y Wang, Yizhong
%Y Dror, Rotem
%S Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop
%D 2020
%8 July
%I Association for Computational Linguistics
%C Online
%F casas-etal-2020-combining
%X In Neural Machine Translation, using word-level tokens leads to degradation in translation quality. The dominant approaches use subword-level tokens, but this increases the length of the sequences and makes it difficult to profit from word-level information such as POS tags or semantic dependencies. We propose a modification to the Transformer model to combine subword-level representations into word-level ones in the first layers of the encoder, reducing the effective length of the sequences in the following layers and providing a natural point to incorporate extra word-level information. Our experiments show that this approach maintains the translation quality with respect to the normal Transformer model when no extra word-level information is injected and that it is superior to the currently dominant method for incorporating word-level source language information to models based on subword-level vocabularies.
%R 10.18653/v1/2020.acl-srw.10
%U https://aclanthology.org/2020.acl-srw.10
%U https://doi.org/10.18653/v1/2020.acl-srw.10
%P 66-71
Markdown (Informal)
[Combining Subword Representations into Word-level Representations in the Transformer Architecture](https://aclanthology.org/2020.acl-srw.10) (Casas et al., ACL 2020)
ACL