Comparing the Value of Latent Semantic Analysis on two English-to-Indonesian lexical mapping tasks
Eliza Margaretha, Hendra Manurung · 2008
This paper describes an experiment that attempts to automatically map English words and concepts, derived from the Princeton WordNet, to their Indonesian analogues appearing in a widely-used Indonesian dictionary, using Latent Semantic Analysis (LSA). A bilingual semantic model is derived from an English-Indonesian parallel corpus. Given a particular word or concept, the semantic model is then used to identify its neighbours in a highdimensional semantic space. Results from various experiments indicate that for bilingual word mapping, LSA is consistently outperformed by the basic vector space model, i.e. where the full-rank word-document matrix is applied. We speculate that this is due to the fact that the ‘smoothing ’ effect LSA has on the worddocument matrix, whilst very useful for revealing implicit semantic patterns, blurs the cooccurrence information that is necessary for establishing word translations.