Corpus Analysis for Indexing: when corpus-based terminology makes a difference

This paper describes preliminary work in corpus-based indexing of a sizeable specialized Web portal, with (comparable, but not parallel) information in Portuguese and English. The interdisciplinary work involved illustrates the urgent need to create greater cooperation between information retrieval and corpus-based terminology. The aim of the work described is twofold: to provide a case for terminology-based search engine deployment while improving the availability of the information in a specific Website, and to suggest further measures in order to successfully marry IR and corpus-based terminology. After presenting the Corpógrafo, a fairly mature Web-based environment for terminology work which includes information extraction capabilities such as named entity recognition and semiautomatic harvesting of definitions and semantic relations, we describe the practical task we wanted to deal with, namely, improving Linguateca’s search system over its website by adding properly chosen index terms. We begin by a general introduction to the portal, with a first analysis of the current state and limitations of the search engine, Busca, together with a study of its users, based on the logs. Then we explain the assumptions underlying our attempt to improve the efficiency and usefulness of dedicated search engines with a terminology-inspired methodology. Finally, we describe the empirical work, providing a quantitative description of the terminological units detected from a qualitative and quantitative point of view, and suggesting how the work can be employed to give Busca more intelligent search capabilities. 1 We finish the paper by emphasizing the need to build bridges between IR and terminology in the training phase of terminologists. 1