A procedure is described for possible use in automatic identification of the language being spoken. The method is to assume that the gross linguistic classes of a language are characterized by certain statistical constraints, namely, they are probabilistic functions of a Markov chain; and that these gross classes can be identified automatically with usable accuracy. The present discussion is in the form of a feasibility experiment and deals only with artificial data (data not derived from acoustic signals). The results indicate that statistical models, as described, can discriminate amongst languages.