Corpus-based bilingual terminology extraction in the power engineering domain

Abstract

Abstract This paper presents the resources and tools used to extract and evaluate bilingual, English-Serbian terminology in the power engineering domain. The resources consist of existing general and domain lexica, and a domain parallel corpus; tools include term extractors for both languages and a tool for aligning the segments belonging to corpus sentences. The system was tested by varying a match function that establishes the presence of an extracted term in an aligned segment (a chunk), ranging from very loose to strict. The evaluation of results showed that the precision of English term extraction was 92%, Serbian term extraction 86%, while the precision of bilingual pair extraction was 72% based on the strictest match function. The result of extraction was 2,684 correct bilingual pairs that enhanced the terminology database and can further be used to support the search of the power engineering aligned collection stored in a digital library.