Principes d’analyse automatique des proverbes

We present software tools capable of identifying a proverb automatically in a text and of generating a translation for it. They include a library of finite state automata administered by the INTEX system (developed by M. Silberztein 1993). We have constructed a series of automata and transducers relating to certain syntactic classes of French and Italian proverbs. We differentiate between the automata containing attested proverbs and automata that we call "predictive", particular lexical and syntactic forms that are inserted by analogy with the actual forms retrieved, which could prove useful in the automatic analysis of texts. Wherever we know the equivalent form in the other language, we establish a transducer. We present a detailed model of the analyser, made up of a network of automata, which takes account of the formulations that link the proverb to the discourse: "as the proverb says, according to folk wisdom, as grandmother used to say, etc". These phrases can sometimes be inserted into the proverb itself, as a kind of interpolation, and can present an obstacle for automatic recognition process.