| Abstract: |
The present paper focuses on the translation relation: it is argued that the time-honoured practice of bilingual lexicography and automatic bilingual dictionary extraction, a field of natural language processing (NLP), have definitively something to learn from one another. On the one hand, automatic bilingual lexicon extraction provides a consistent account of the fundamental notion of the translation relation, which-in spite of the centuries of scientific investigation-remained a fragmented and somewhat obscure concept of lexicography. On the other hand, lexicographical insights concerning the translation relation are particularly important from the perspective of NLP, as they offer strong arguments for a particular technique among the competing extraction methods. More precisely, it is argued that conditional probability as estimated from parallel corpora via word alignment is suitable to model the translation relation because of its gradual and asymmetric nature. Furthermore, it will be shown that this approach turns out to be suitable to extract reversible encoding dictionaries automatically. |