Combining Words and Compound Terms for Monolingual and Cross-Language Information Retrieval

Most existing systems of Information retrieval (IR) use single words as index to represent the contents of documents and queries. One of the consequences is the low recall level. In this paper, we propose to integrate compound terms as additional indexing units because terms are more precise representation units than words. Terms are recognized through the use of a terminology database and an automatic term extraction tool, which is based on syntactic templates and statistical analysis. In this paper, we first show that the use of compound terms is greatly beneficial to monolingual IR. Then compound terms are incorporated in statistical translation models trained on a large set of parallel texts. Our experiments on cross-language information retrieval (CLIR) show that such a translation model leads to a much better CLIR effectiveness when compound terms are integrated.