Journal of Natural Language Processing
Online ISSN : 2185-8314
Print ISSN : 1340-7619
ISSN-L : 1340-7619
Volume 15, Issue 4
Displaying 1-5 of 5 articles from this issue
  • [in Japanese]
    2008 Volume 15 Issue 4 Pages 1-2
    Published: September 10, 2008
    Released on J-STAGE: March 01, 2011
    JOURNAL FREE ACCESS
    Download PDF (229K)
  • TETSUYA ISHIKAWA, AKIRA KITAUCHI, OTOYA SHIROTSUKA
    2008 Volume 15 Issue 4 Pages 3-18
    Published: September 10, 2008
    Released on J-STAGE: March 01, 2011
    JOURNAL FREE ACCESS
    Our goal of this study is to contribute to the progress in historical science by developing a system for building a historical ontology from historical materials and making it available to the public.We digitize all the books of “Meiji-mae Nippon Kagaku-shi” (Pre-modern Japanese History of Science and Technology) published by Nippon Gakushiin (The Japan Academy), and extract the attribution and the works of scientists and engineers from the books to build a database of person information in pre-modern Japanese history.We extract the names of persons, positions, places, and books as the attribution and the works of persons by pattern matching.The experimental results show that the F-measures for the names of persons, positions, and places are over 0.8.
    Download PDF (4733K)
  • Masato Hagiwara, Yasuhiro Ogawa, Katsuhiko Toyama
    2008 Volume 15 Issue 4 Pages 19-42
    Published: September 10, 2008
    Released on J-STAGE: March 01, 2011
    JOURNAL FREE ACCESS
    Distributional similarity is a widely adopted concept to compute lexical semantic relatedness of words.Whereas the calculation is based on the distributional hypothesis and utilizes contextual clues of words, little attention has been paid to what kind of contextual information is effective for the purpose.As one of the ways to extend contextual information, we pay attention to the use of indirect dependency, where two or more words are related via several contiguous dependency relations.We have investigated the effect of indirect dependency using automatic synonym acquisition task, and shown that the performance can be improved by using indirect dependency in addition to normal direct dependency.We have also verified its effectiveness under various experimental settings including weight functions, similarity measures, and context representations, and shown that context representations which incorporate richer syntactic information are more effective.
    Download PDF (5030K)
  • TORU HIRANO, YOSHIHIRO MATSUO, GENICHIRO KIKUI
    2008 Volume 15 Issue 4 Pages 43-58
    Published: September 10, 2008
    Released on J-STAGE: March 01, 2011
    JOURNAL FREE ACCESS
    This paper proposes a supervised learning method for detecting a semantic relationbetween a given pair of named entities, which may be located in different sentences.he method employs newly introduced contextual features based on Salient Referent List as well as conventional syntactic and word-based features.These features are organized as a tree structure and are fed into a boosting-based classification algorithm.Experimental results show the proposed method outperformed prior methods, andincreased precision and recall by 11.3% and 14.2%.
    Download PDF (2791K)
  • Towards Construction of Ontology of Adjectives from a real data
    KYOKO KANZAKI, QING MA, EIKO YAMAMOTO, TAMOTSU SHIRADO, HITOSHI ISAHAR ...
    2008 Volume 15 Issue 4 Pages 59-88
    Published: September 10, 2008
    Released on J-STAGE: March 01, 2011
    JOURNAL FREE ACCESS
    The method of organizing word meanings is a crucial issue with lexical databases.Weare aiming to extract the semantic structure of concepts of adjectives from corporaautomatically.The first step to achieving this is to obtain the concepts of adjec-tives from corpora, for which we used abstract nouns.We constructed linguistic databy extracting semantic relations between abstract nouns and adjectives from corpusdata.This paper describes how to hierarchically organize abstract concepts of adjec-tives mainly using the Complementary Similarity Measure (CSM) which calculatesinclusion relations (hypernym/hyponym relations) between words.To estimate hy-pernym/hyponym relations between words, we compared three hierarchical structuresof abstract concepts of adjectives: according to CSM, CSM with frequency (Freq), and an alternative similarity measure based on coefficient overlap.We evaluated au-tomatically generated concept hierarchies of adjectives with those in EDR, and foundthat 43% of those automatically generated were better than EDR.
    Download PDF (3187K)
feedback
Top