Computing Semantic Clusters by Semantic Mirroring and Spectral Graph Partitioning

Wednesday, October 12, 2011 - 11:00am - 12:00pm
Lind 305
Lars Eldén (Linköping University)
A great deal of linguistic knowledge is encoded implicitly in bilingual resources such as parallel texts and dictionaries. Semantic mirroring is a linguistic technique, where one performs two-way translations using a bilingual lexicon. The translations involving a set of words in a source language can be seen to constitute an undirected graph, where the vertices are words in the source language and the edges the translations via words in the target language. The connectedness of the graph holds information about the different meanings of words that occur in the translations. Spectral graph partitioning is used to cluster the words according to different senses. Results using a lexicon of Swedish and English adjectives are reported.