Extending, trimming and fusing WordNet for technical documents

Abstract

This paper describes a tool for the automatic
\nextension and trimming of a multilingual
\nWordNet database for cross-lingual retrieval
\nand multilingual ontology building in
\nintranets and domain-specific document
\ncollections. Hierarchies, built from
\nautomatically extracted terms and combined
\nwith the WordNet relations, are trimmed
\nwith a disambiguation method based on the
\ndocument salience of the words in the
\nglosses. The disambiguation is tested in a
\ncross-lingual retrieval task, showing
\nconsiderable improvement (7%-11%). The
\ncondensed hierarchies can be used as
\nbrowse-interfaces to the documents
\ncomplementary to retrieval.