複数のパーザを利用した統計的部分係り受け解析  [in Japanese] Committe-based Decision Making in Probabilistic Partial Parsing  [in Japanese]

    • 乾 孝司 INUI TAKASHI
    • 奈良先端科学技術大学院大学情報科学研究科 Graduate School of Information Science, Nara Institute of Science and Technology
    • 乾 健太郎 INUI KENTARO
    • 九州工業大学大学院情報工学研究科 Graduate School of Computer Science and Systems Engineering, Kyushu Institute of Technology

Abstract

我々はこれまでに, 信頼のおける部分だけを出力し被覆率を犠牲とする代償として正解率を向上させる統計的部分解析手法の調査を進めてきた.本稿では, さらにこの考えに委員会方式という概念を統合した枠組みを提案し, その評価を行った.委員会方式とは, 複数の解析器(委員)の出力解を組み合わせることにより解析精度の向上をはかる手法である.ここでは, 各委員から得られる解析結果に基づき委員会で多数決により統計的部分解析を行うために, 従来の基本的な委員会方式に3つの拡張を施した:(1)解析器(委員)が推定した係り受け確率を票の重みと見なして重みつきの票を投じる確率的投票, (2)委員間での票の重みの信頼性を標準化する重み標準化, (3)各係り文節に対して2位以下の係り先候補にも重みつきの票を投じる多重投票.既存の5つの統計的解析器を用いて, 京大コーパスを対象データとする解析実験を行った.その結果, 委員の組合せによって精度変化には多少の揺れがあるものの, 総合的には提案した枠組みおよび3つの拡張が解析精度の向上に有効に作用する見通しを得た.

In this paper, we explored two new direction for the next step beyond the state of the art of statistical parsing: probabilistic partial parsing and committee-based decision making. Probabilistic partial parsing makes only as an output partial parse tree that is probabilistically highly reliable. Committee-based decision making is to combine several outputs from different systems (parsers) to make a better decision. Aiming at this coupling, we present a general framework which have three extensions against orginal basic framework to committee-based decision making. (1)probabilistic voting: a commitee accepts probabilistically parameterized votes as its input. (2)weight standardization: a commitee provides a means for standardizing original votes to guarantee reliability of them. (3)multiple voting: a committee allows a committee member to vote not only to the best-scored candidate but also to all other potential candidates. From the result of our experiments on the Kyoto japanese corpus, we show that our presented framework have some contributions.

Journal

Transactions of Information Processing Society of Japan   [List of Volumes]

Transactions of Information Processing Society of Japan 42(12), 3160-3172, 2001-12-15  [Table of Contents]

Information Processing Society of Japan (IPSJ)

References:  23

You must have a user ID to see the references.If you already have a user ID, please click "Login" to access the info.New users can click "Sign Up" to register for an user ID.

Cited by:  4

You must have a user ID to see the cited references.If you already have a user ID, please click "Login" to access the info.New users can click "Sign Up" to register for an user ID.

Preview

Preview

Codes

  • NII Article ID (NAID) :
    110002726119
  • NII NACSIS-CAT ID (NCID) :
    AN00116647
  • Text Lang :
    JPN
  • Article Type :
    Journal Article
  • ISSN :
    03875806
  • NDL Article ID :
    6010115
  • NDL Source Classification :
    ZM13(科学技術--科学技術一般--データ処理・計算機)
  • NDL Call No. :
    Z14-741
  • Databases :
    CJP  CJPref  NDL  NII-ELS 

Export