Languages are supposed to differ in the status of certain sequences. Thus, in Hebrew, [ts] is a single consonant, but in English, it is two consonants. On the other hand, [tʃ] is a single consonant in English but a cluster of two consonants in Hebrew. What information do the learners use to figure out which representation is right for their languages? Our computational model is based on the hypothesis that statistical distributions of these sequences and their parts offer a clue.