Evaluating Lexicon Coverage for Cross-Language Information Retrieval
Gina-Anne Levow, Douglas W. Oard · 2000
Cross-language retrieval systems use queries expressed in one natural language to retrieve documents that may be written in a different language. Term-by-term translation techniques based on bilingual lexicons are now widely used for cross-language retrieval, but little is known about the way in which retrieval effectiveness varies with dictionary coverage. This paper compares three types of coverage measures using two relatively large Chinese-English dictionaries. An average precision measure calculated using twenty Chinese queries to search the English portion of the Topic Detection and Tracking collection provides the baseline against which the coverage measures are compared. The results indicate that lexicon size is not a suitable coverage measure for this task. In contrast, we find that two other measures, joint "by token" coverage and joint "IDFweighted, by token coverage" both successfully predict retrieval performance for different (non-merged) lexical resources. 1 Introductio...