Finding translations for low-frequency words in comparable corpora

Abstract. The paper proposes a method to improve the extraction of lowfrequency translation equivalents from comparable corpora. Prior to performing the mapping between vector spaces of different languages, the method models context vectors of rare words using their distributional similarity to words of the same language to predict unseen co-occurrences as well as to smooth rare, unreliable ones. Our evaluation shows that the proposed method delivers a consistent and significant improvement on the conventional approach to this task.

• Type: Journal
• Year: 2006
• Where: MT
• Authors: Viktor Pekar, Ruslan Mitkov, Dimitar Blagoev, Andrea Mulloni
• Keywords: Comparable Corpora, Lowfrequency Translation Equivalents, MT 2006, Vector Spaces