@inproceedings{stiff-etal-2020-self,
title = "How Self-Attention Improves Rare Class Performance in a Question-Answering Dialogue Agent",
author = "Stiff, Adam and
Song, Qi and
Fosler-Lussier, Eric",
editor = "Pietquin, Olivier and
Muresan, Smaranda and
Chen, Vivian and
Kennington, Casey and
Vandyke, David and
Dethlefs, Nina and
Inoue, Koji and
Ekstedt, Erik and
Ultes, Stefan",
booktitle = "Proceedings of the 21th Annual Meeting of the Special Interest Group on Discourse and Dialogue",
month = jul,
year = "2020",
address = "1st virtual meeting",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2020.sigdial-1.24",
doi = "10.18653/v1/2020.sigdial-1.24",
pages = "196--202",
abstract = "Contextualized language modeling using deep Transformer networks has been applied to a variety of natural language processing tasks with remarkable success. However, we find that these models are not a panacea for a question-answering dialogue agent corpus task, which has hundreds of classes in a long-tailed frequency distribution, with only thousands of data points. Instead, we find substantial improvements in recall and accuracy on rare classes from a simple one-layer RNN with multi-headed self-attention and static word embeddings as inputs. While much research has used attention weights to illustrate what input is important for a task, the complexities of our dialogue corpus offer a unique opportunity to examine how the model represents what it attends to, and we offer a detailed analysis of how that contributes to improved performance on rare classes. A particularly interesting phenomenon we observe is that the model picks up implicit meanings by splitting different aspects of the semantics of a single word across multiple attention heads.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="stiff-etal-2020-self">
<titleInfo>
<title>How Self-Attention Improves Rare Class Performance in a Question-Answering Dialogue Agent</title>
</titleInfo>
<name type="personal">
<namePart type="given">Adam</namePart>
<namePart type="family">Stiff</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Song</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Eric</namePart>
<namePart type="family">Fosler-Lussier</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 21th Annual Meeting of the Special Interest Group on Discourse and Dialogue</title>
</titleInfo>
<name type="personal">
<namePart type="given">Olivier</namePart>
<namePart type="family">Pietquin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Smaranda</namePart>
<namePart type="family">Muresan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vivian</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Casey</namePart>
<namePart type="family">Kennington</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Vandyke</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nina</namePart>
<namePart type="family">Dethlefs</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Koji</namePart>
<namePart type="family">Inoue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Erik</namePart>
<namePart type="family">Ekstedt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stefan</namePart>
<namePart type="family">Ultes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">1st virtual meeting</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Contextualized language modeling using deep Transformer networks has been applied to a variety of natural language processing tasks with remarkable success. However, we find that these models are not a panacea for a question-answering dialogue agent corpus task, which has hundreds of classes in a long-tailed frequency distribution, with only thousands of data points. Instead, we find substantial improvements in recall and accuracy on rare classes from a simple one-layer RNN with multi-headed self-attention and static word embeddings as inputs. While much research has used attention weights to illustrate what input is important for a task, the complexities of our dialogue corpus offer a unique opportunity to examine how the model represents what it attends to, and we offer a detailed analysis of how that contributes to improved performance on rare classes. A particularly interesting phenomenon we observe is that the model picks up implicit meanings by splitting different aspects of the semantics of a single word across multiple attention heads.</abstract>
<identifier type="citekey">stiff-etal-2020-self</identifier>
<identifier type="doi">10.18653/v1/2020.sigdial-1.24</identifier>
<location>
<url>https://aclanthology.org/2020.sigdial-1.24</url>
</location>
<part>
<date>2020-07</date>
<extent unit="page">
<start>196</start>
<end>202</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T How Self-Attention Improves Rare Class Performance in a Question-Answering Dialogue Agent
%A Stiff, Adam
%A Song, Qi
%A Fosler-Lussier, Eric
%Y Pietquin, Olivier
%Y Muresan, Smaranda
%Y Chen, Vivian
%Y Kennington, Casey
%Y Vandyke, David
%Y Dethlefs, Nina
%Y Inoue, Koji
%Y Ekstedt, Erik
%Y Ultes, Stefan
%S Proceedings of the 21th Annual Meeting of the Special Interest Group on Discourse and Dialogue
%D 2020
%8 July
%I Association for Computational Linguistics
%C 1st virtual meeting
%F stiff-etal-2020-self
%X Contextualized language modeling using deep Transformer networks has been applied to a variety of natural language processing tasks with remarkable success. However, we find that these models are not a panacea for a question-answering dialogue agent corpus task, which has hundreds of classes in a long-tailed frequency distribution, with only thousands of data points. Instead, we find substantial improvements in recall and accuracy on rare classes from a simple one-layer RNN with multi-headed self-attention and static word embeddings as inputs. While much research has used attention weights to illustrate what input is important for a task, the complexities of our dialogue corpus offer a unique opportunity to examine how the model represents what it attends to, and we offer a detailed analysis of how that contributes to improved performance on rare classes. A particularly interesting phenomenon we observe is that the model picks up implicit meanings by splitting different aspects of the semantics of a single word across multiple attention heads.
%R 10.18653/v1/2020.sigdial-1.24
%U https://aclanthology.org/2020.sigdial-1.24
%U https://doi.org/10.18653/v1/2020.sigdial-1.24
%P 196-202
Markdown (Informal)
[How Self-Attention Improves Rare Class Performance in a Question-Answering Dialogue Agent](https://aclanthology.org/2020.sigdial-1.24) (Stiff et al., SIGDIAL 2020)
ACL