Multilingual Aspects of Multiword Lexical Units

As most of the machine-readable dictionaries contain clearly insufficient information about multiword lexical units, there is a constant need to extend and tune specialized lexical databases to account for new expressions. In this paper, we present a system exclusively based on statistics that massively extracts from unrestricted text corpora contiguous and noncontiguous rigid multiword lexical units. For that purpose, a new association measure called the Mutual Expectation is conjugated with a new acquisition process based on an algorithm of local maxima. The system has been applied to a Portuguese, French, English and Italian parallel corpus and has evidenced that multiword lexical units embody a great deal of cross-language regularities.