Visualization of Senses and Complex Concepts by Web Images

Abstract

Images, deemed language-independent to some extent, can be utilized as an effective medium for representing/disambiguating user's search intent in interactive information access systems, such as a cross-language information retrieval system. If a target query concept (e.g. beaver) could be decomposed into a set of salient concept-feature pairs (e.g. beaver builds dams; beaver chews on woods, etc.), the depiction of each concept-feature pair may greatly help manifest the user's search intent. Given this motivation, we conducted an investigation on the Web-imageability of complex concepts (concept-feature pairs), which the psycholinguistic semantic feature database developed by McRae and his colleagues provides. More specifically, this paper reports on the human assessment results on how the images acquired from the Web by means of an image search engine, given a concept-feature pair, can adequately deliver the meaning. The results clearly show that physical motions and feeding behaviors performed by animate beings were more adequately depicted in the acquired Web images. This paper also discusses possible utility of visual images in interactive information access systems, and further argues mutual grounding of visual and linguistic information.