<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="qu-etal-2021-explore">
    <titleInfo>
        <title>Explore Better Relative Position Embeddings from Encoding Perspective for Transformer Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Anlin</namePart>
        <namePart type="family">Qu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jianwei</namePart>
        <namePart type="family">Niu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Shasha</namePart>
        <namePart type="family">Mo</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2021-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Online and Punta Cana, Dominican Republic</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Relative position embedding (RPE) is a successful method to explicitly and efficaciously encode position information into Transformer models. In this paper, we investigate the potential problems in Shaw-RPE and XL-RPE, which are the most representative and prevalent RPEs, and propose two novel RPEs called Low-level Fine-grained High-level Coarse-grained (LFHC) RPE and Gaussian Cumulative Distribution Function (GCDF) RPE. LFHC-RPE is an improvement of Shaw-RPE, which enhances the perception ability at medium and long relative positions. GCDF-RPE utilizes the excellent properties of the Gaussian function to amend the prior encoding mechanism in XL-RPE. Experimental results on nine authoritative datasets demonstrate the effectiveness of our methods empirically. Furthermore, GCDF-RPE achieves the best overall performance among five different RPEs.</abstract>
    <identifier type="citekey">qu-etal-2021-explore</identifier>
    <identifier type="doi">10.18653/v1/2021.emnlp-main.237</identifier>
    <location>
        <url>https://aclanthology.org/2021.emnlp-main.237</url>
    </location>
    <part>
        <date>2021-11</date>
        <extent unit="page">
            <start>2989</start>
            <end>2997</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Explore Better Relative Position Embeddings from Encoding Perspective for Transformer Models
%A Qu, Anlin
%A Niu, Jianwei
%A Mo, Shasha
%S Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
%D 2021
%8 November
%I Association for Computational Linguistics
%C Online and Punta Cana, Dominican Republic
%F qu-etal-2021-explore
%X Relative position embedding (RPE) is a successful method to explicitly and efficaciously encode position information into Transformer models. In this paper, we investigate the potential problems in Shaw-RPE and XL-RPE, which are the most representative and prevalent RPEs, and propose two novel RPEs called Low-level Fine-grained High-level Coarse-grained (LFHC) RPE and Gaussian Cumulative Distribution Function (GCDF) RPE. LFHC-RPE is an improvement of Shaw-RPE, which enhances the perception ability at medium and long relative positions. GCDF-RPE utilizes the excellent properties of the Gaussian function to amend the prior encoding mechanism in XL-RPE. Experimental results on nine authoritative datasets demonstrate the effectiveness of our methods empirically. Furthermore, GCDF-RPE achieves the best overall performance among five different RPEs.
%R 10.18653/v1/2021.emnlp-main.237
%U https://aclanthology.org/2021.emnlp-main.237
%U https://doi.org/10.18653/v1/2021.emnlp-main.237
%P 2989-2997

Markdown (Informal)

[Explore Better Relative Position Embeddings from Encoding Perspective for Transformer Models](https://aclanthology.org/2021.emnlp-main.237) (Qu et al., EMNLP 2021)

• Explore Better Relative Position Embeddings from Encoding Perspective for Transformer Models (Qu et al., EMNLP 2021)

ACL

• Anlin Qu, Jianwei Niu, and Shasha Mo. 2021. Explore Better Relative Position Embeddings from Encoding Perspective for Transformer Models. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2989–2997, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.