Paper: Cross-lingual WSD for Translation Extraction from Comparable Corpora

ACL ID W13-2501
Title Cross-lingual WSD for Translation Extraction from Comparable Corpora
Venue Building and Using Comparable Corpora
Session
Year 2013
Authors

We propose a data-driven approach to en- hance translation extraction from compa- rable corpora. Instead of resorting to an external dictionary, we translate source vector features by using a cross-lingual Word Sense Disambiguation method. The candidate senses for a feature correspond to sense clusters of its translations in a parallel corpus and the context used for disambiguation consists of the vector that contains the feature. The translations found in the disambiguation output con- vey the sense of the features in the source vector, while the use of translation clusters permits to expand their translation with several variants. As a consequence, the translated vectors are less noisy and richer, and allow for the extraction of higher qual- ity lexicons compared to simpler methods.