I just ran across this protein-filled piece on phonology written by Professor Edward Vajda (editor of Word) for his Linguistics 201 class at Western Washington University. It contains all sorts of gems (including, for those of you who took part in the discussion of “nucular” a few days ago, this: “Metathesis rule reorders the segments that are present: ask/aks; nuclear, ‘nucular’… These are examples of a rule randomly applied”); I was particularly struck by this:
A more striking example of a morphological constraint on phonetic distribution is to be found in Cherokee. Cherokee has a sound [m] that contrasts with other sounds to create changes in meaning: ama ‘salt’; ada ‘baby bird’; ana ‘strawberry’; ata ‘young girl’. However, the sound [m] appears in only about 10 morphemes: ugama ‘soup’; kamama ‘butterfly’; gugama ‘cucumber.’ Although most of these words seem to be foreign borrowings, no new words using [m] seem to be entering the language. Nor do new words containing [m] seem be made in Cherokee on any regular basis. Thus, the sound [m], which definitely would be considered a phoneme in the phoneme theory of phonology, is highly restricted in its distribution, at least as far as concerns the present state of Cherokee. The restriction is random: the sound [m] only appears in a small collection of words with no specific meaning in common. Yet the restriction on the distribution of [m] is morphological rather than phonological: [m] is restricted to a specific and limited set of words.
An even more extreme example is to be found in Quileute, a Native American language from the Olympic Peninsula of Washington State. The sound [g] appears in only one word in the entire language: hága’y ‘frog.’ Thus, this sound, which is in contrastive distribution with other phonemes, is entirely restricted in function to being able to contribute to the makeup of a single phoneme, the word for ‘frog.’ It is even possible to say that [g] in Quileute has a specific function: to contribute to the morpheme meaning ‘frog.’