Translations as a Semantic Knowledge Source

Theories of meaning are sometimes used to throw light on the phenomenon of translation. We argue that light can fruitfully be thrown in the opposite direction: we can use translations to get a handle on meaning. More specifically, we will motivate and present a method for the automatic extraction of wordnet-type information from translational data, and review some results. The basic insight behind the method is that much information about the semantic relations among the words in a language resides in the way in which the sets of their possible translations into some other language overlap. Therefore, if we take the translational relation between two languages as a theoretical primitive, languages can serve as each other’s “semantic mirrors”.