In this paper some preliminary results are presented concerning a study performed upon self-organized linguistic models for a multilingual speech processing system based on large vocabularies as part of the Esprit project 860 "Linguistic Analysis of the European languages". The model considered in this case consists of P.O.S. transition matrices, automatically derived from learning texts, for seven European languages, namely Dutch, English, French, German, Greek, Italian and Spanish. Several parameters have been considered and experimented for evaluating the power of the different linguistic models which can be obtained using different PCS sets.