Previous work concerned with the identification of word translations from text collections has been either based on parallel or on comparable corpora of the respective languages. In the case of comparable corpora basic dictionaries have been necessary to form a bridge between the languages under consideration. We present here a novel approach to identify word translations from a single monolingual corpus without necessarily requiring dictionaries, although, as will be shown, a dictionary can still be useful for improving the results. Our approach is based on the observation that for various reasons monolingual corpora typically contain many foreign words (for example citations). Relying on standard newsticker texts, we will show that their cooccurrence- based associations can be successfully used to identify word translations.
展开▼
