Spoken language interpretation: On the use of dynamic Bayesian networks for semantic composition