Exploring the Visual Annotatability of Query Concepts for Interactive Cross-Language Information Retrieval

In interactive CLIR, (Cross-Language Information Retrieval), it is crucial to provide users with useful clues for signifying the word senses that a translated query term can have. Among the possible means, visual clues can be effectively employed, as they are intuitive and language-neutral to some extent. This paper therefore examined the possibility of Web images as an intuitive and effective clue for the purpose of signifying word senses. We designed an experiment to collect human assessments of the relevance of Web images. Through statistical analyses applied to the assessment data, this paper shows: (1) the semantic class of a word sense together with familiarity is a good indicator for predicting the applicability of Web images as a word sense clue; (2) Web biases should be considered when gathering Web images, particularly for terms used as entity names.