Image estimation of words based on adjective Co-occurrences

Kouhei Shimizu, Masafumi Hagiwara

研究成果: Article査読

抄録

In natural language, words convey various impressions such as "kurai (dark)-afajrai (bright)" or "kitanai (dirty)-utsukushii (beautiful)." Impressions play an important role in inferring the speaker's intentions or feelings. Systems that use natural language to support conceptualization or retrieve information by teaching the computer the impressions of words are growing significantly. Therefore, in this paper, the authors propose a technique for estimating the impressions or images that words convey. The proposed technique consists of two processes: similarity measurement and spatial positioning. Similarity measurement uses the enormous amount of text on the Internet to obtain similarities by measuring the co-occurrence frequency of words. Co-occurrence conditions, which are limited to the joint co-occurrences of adjectives or adjectival nouns, are applied to image estimation. Spatial positioning positions words in space by quantifying the similarities that were obtained according to images. The authors used a simulation of an electrical circuit for quantification. The results of evaluation experiments verified that the proposed system could estimate the images of words with high precision. It was also apparent that the proposed system could estimate the images with sufficient precision even for words that were difficult to analyze because of the insufficient number of samples in an existing corpus, including new words and colloquial words or slang.

本文言語English
ページ(範囲)14-24
ページ数11
ジャーナルSystems and Computers in Japan
38
12
DOI
出版ステータスPublished - 2007 11 15

ASJC Scopus subject areas

  • 理論的コンピュータサイエンス
  • 情報システム
  • ハードウェアとアーキテクチャ
  • 計算理論と計算数学

フィンガープリント

「Image estimation of words based on adjective Co-occurrences」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル