BARRS, Keith
Department Hiroshima shudo University The Faculty of Humanities and Human Sciences Position Professor |
|
Language | English |
Publication Date | 2020/02 |
Type | Articles |
Title | Deriving Lists of Collocates from Web Corpora : Issues and Implications of Different Association Measures |
Contribution Type | Single-Authored Publication |
Journal | 広島修大論集 |
Journal Type | Japan |
Publisher | 広島修道大学 |
Volume, Issue, Pages | 60(2),pp.71-78 |
Number of pages | 8 |
Details | This research investigates the issues and implications of using different association measures to derive a list of collocates for words in a web corpus. A case study was carried out on the search word 'zero' in the enTenTen12 corpus of web-based English, comparing the list of its collocates ranked by raw frequency with lists ranked by the traditional asso- ciation measures of MI, t-score and log-likelihood, and then further comparing these with a list ranked by the more modern logDice statistic. It was found that the logDice measure was by far the most effective of the five at building up a semantic profile of the search word, carrying the implication that the corpus analyst needs to be aware of the various issues involved in the application of different association measures to the analyses of web corpora. |