Lexical stress assignment as a problem of probabilistic inference