Reranking with multiple features for better transliteration

Yan Song, Chunyu Kit, Hai Zhao

    Research output: Chapters, Conference Papers, Creative and Literary WorksRGC 32 - Refereed conference paper (with host publication)peer-review

    6 Citations (Scopus)
    9 Downloads (CityUHK Scholars)

    Abstract

    Effective transliteration of proper names via grapheme conversion needs to find transliteration patterns in training data, and then generate optimized candidates for testing samples accordingly. However, the top-1 accuracy for the generated candidates cannot be good if the right one is not ranked at the top. To tackle this issue, we propose to rerank the output candidates for a better order using the averaged perceptron with multiple features. This paper describes our recent work in this direction for our participation in NEWS2010 transliteration evaluation. The official results confirm its effectiveness in English-Chinese bidirectional transliteration. © 2010 Association for Computational Linguistics
    Original languageEnglish
    Title of host publicationProceedings of the 2010 Named Entities Workshop
    EditorsA Kumaran, Haizhou Li
    PublisherAssociation for Computational Linguistics
    Pages62-65
    ISBN (Print)978-1-932432-78-7, 1-932432-78-7
    Publication statusPublished - Jul 2010
    Event48th Annual Meeting of the Association for Computational Linguistics, ACL 2010 - Uppsala, Sweden
    Duration: 11 Jul 201016 Jul 2010

    Publication series

    NameProceedings of the Annual Meeting of the Association for Computational Linguistics
    ISSN (Print)0736-587X

    Conference

    Conference48th Annual Meeting of the Association for Computational Linguistics, ACL 2010
    PlaceSweden
    CityUppsala
    Period11/07/1016/07/10

    Research Keywords

    • transliteration
    • proper name
    • rerank
    • averaged perceptron

    Publisher's Copyright Statement

    • This full text is made available under CC-BY-NC-SA 3.0. https://creativecommons.org/licenses/by-nc-sa/3.0/

    Fingerprint

    Dive into the research topics of 'Reranking with multiple features for better transliteration'. Together they form a unique fingerprint.

    Cite this