The phoneme can be roughly defined as a minimal unit of sound that can be used to distinguish words in a language. The question of how to theoretically define the phoneme dominated linguistics from the late 19th century until about 1960. In 1959, Morris Halle first published his famous argument against the “classical phoneme” in his monograph The Sound Pattern of Russian. Halle and Chomsky reiterated that argument in The Sound Pattern of English, which had been widely circulated in the 1960s in various manuscript forms before it was published in 1968. The argument, now familiar to most linguists, had to do with an asymmetry in the Russian phoneme inventory. Halle had noticed that Russian regressive voice assimilation produced both phonemic and allophonic outputs. That is, it could produce a phonemic output /z/ from an underlying /s/ or an allophonic variant [ɣ] from an underlying /x/. If one were to posit a significant level of classical phonemic representation in a phonological derivation, then it would be necessary to apply regressive assimilation twice–above and below the level of phonemic representation. Hence, classical phonemic representation had to be ruled out as a significant level of phonological representation. This argument basically signaled the death of the phoneme in early generative phonological theory. As a consequence, the input to phonological derivations became the level of systematic phonemics, which captured generalizations about morphophonological alternations, a considerably more abstract level of representation than in classical phonemic theories.
So much for history as we know it. Modern phonology does not completely ignore the distinction between morphophonological alternations and allophonic variation, but there is no consensus on how to handle it. The concept of the phoneme has become almost an embarrassment, and some phonologists even openly question whether it should be taught in an introduction to linguistics. Over a century after Baudouin de Courtenay first defined phonemic theory in terms of a basic dichotomy between two types of phonetic alternations, that dichotomy is being swept under the rug as a kind of “inconvenient truth” for phonological theory. What I want to do here is explain what Baudouin’s original insight was and why we need it back. In the process, I want to make an appeal that will probably fall on a lot of deaf ears–that linguists stop calling morphophonological analysis phonology. We need to restore phonological theory to its rightful place in linguistic theory, and the best way to do that is not to confuse it with morphophonology, an entirely different area of grammatical description.
Baudouin de Courtenay invented the concept of alternational phonology. His best known work (because a version was published in German) was An Attempt at a Theory of Phonetic Alternations. It was first published in English in Edward Stankiewicz’s 1972 A Baudouin de Courtenay Anthology, but the sheer weight of material in that work tends to overwhelm the simple, fundamental insight that Baudouin came to distinguish as the physiophonetic/psychophonetic alternational dichotomy. To take one of the many examples from his work, consider the Polish word for frost: mróz (nom. sg) in contrast to mroza (gen. sg.). The phonetic representations would be [mrus] and [‘mroza], respectively. This word pair contains one psychophonetic alternation [u] vs. [o] and one physiophonetic alternation [s] vs. [z]. The difference, according to Baudouin, was that the [u/o] alternation represented two distinct phonemes of Polish and the [s/z] alternation represented a single phoneme. That is, the Polish speaker perceived and pronounced two distinct vocalic sounds, but a single fricative sound. The single phoneme at the end of mróz was /z/, which happened to correspond with Polish spelling. Baudouin did not have a process-oriented theory of language to work with, but there can be no doubt that he saw the generalization that modern generative linguists capture as a rule of obstruent devoicing in the codas of Polish syllables. Notice a very important fact about this example. Baudouin de Courtenay started out with a theory of the phoneme that incorporated so-called phonemic neutralization. By the same token, he still recognized alternations between different phonemes. Baudouin also gave examples of what would be called allophonic variation by later phonemicists, but he did not make a big deal of the difference between allophony and phonemic neutralization. Hence, Morris Halle’s argument against so-called classical phonemic representation never even touched Baudouin’s dichotomy. If anything, it exonerated the original dichotomy, which had fallen into disfavor during the structuralist period.
N.S. Trubetzkoy was later to rename physiophonetics as phonology and psychophonetics as morphophonology. He did not have much to say about morphophonology in Principles of Phonology, although he did refer to it as a neglected “branch of grammar” in Appendix II. Thoughts on Morphophonology. In other words, Trubetzkoy, who was working with Baudouin’s original alternational dichotomy, did not want to confuse phonology and morphophonology. The original definition of phonology encompassed just those phonetic alternations that had been called physiophonetics. In addition to the Prague School of phonology, however, Baudouin’s theory gave rise to two additional schools of Russian phonology: the Leningrad school and the Moscow school. Western scholars knew very little about the two Soviet schools and their rivalry during the Cold War period, and that ignorance continues almost unabated today outside of the community of those who specialize in the field of Slavic linguistics. I will not dwell on that difference here, although it has many ramifications for the history of phonology and how we lost the phoneme. (I don’t blame it all on generativists. Baudouin’s student Scherba and Western structuralism had a hand in it.)
One final note before I conclude: Edward Sapir. Sapir was the revered American linguist who inspired our original concept of the psychological phoneme. To my knowledge, Sapir never even mentioned Baudouin de Courtenay in his writings, but he was using Baudouin’s theory of the phoneme, which had been around for decades before he began his work. Sapir is said to have kept up a lively correspondence with Trubetzkoy, although that correspondence seems not to have survived. But consider this passage from page 62 of Language:
…It is important to bear in mind that a linguistic phenomenon cannot be looked upon as illustrating a definite “process” unless it has an inherent functional value. The consonantal change in English, for instance, of book-s and bag-s (s in the former, z in the latter) is of no functional significance. It is a purely external, mechanical change induced by the presence of a preceding voiceless consonant, k, in the former case, of a voiced consonant, g, in the latter. This mechanical alternation is objectively the same as that between the noun house and the verb to house. In the latter case, however, it has an important grammatical function, that of transforming a noun into a verb. The two alternations belong, then, to entirely different psychological categories. Only the latter is a true illustration of consonantal modification as a grammatical process.
This could not be a clearer example of Baudouin’s physiophonetic/psychophonetic dichotomy. In this case, the same phonetic alternation–[s/z]–falls into different categories, depending on something he called its functional value. Sapir did not confuse phonology and morphophonology.
Now, I pointed out earlier that Morris Halle’s famous argument against a classical phonemic level of representation in phonological derivations did not even touch Baudouin de Courtenay’s concept of phonemic representation. His argument was a vindication of the psychological phoneme, not an argument against having phonemic representation. Halle was aware of Sapir’s earlier tolerance of phonemic neutralization, and he named his book, Sound Pattern of Russian, after Sapir’s “Sound Patterns in Language” (which famously appeared in the first issue of Language, the LSA’s journal). Similarly, Chomsky and Halle’s 1968 classic The Sound Pattern of English (SPE) extolled Sapir and his precedence and influence on their work. Nevertheless, they mistakenly conflated Sapir’s phonemic representation with their level of the systematic phoneme. They were not justified in doing that. If Sapir had been a generativist, he would still have emerged from Halle’s argument with three levels of representation (morphonemic, phonemic, phonetic), not the two levels that SPE ended up with. And Baudouin would not be rolling over in his grave every time a generative linguist confused physiophonetic and psychophonetic alternations. If generative theory is going to toss out the bathwater of phonemic neutralization, it should at least save linguistic theory’s original baby–the phoneme. (I love cliches. :-))
Finally, I would just like to say that this is not all I have to say about phonemes. I have not really covered what went wrong with Scherba and how his fundamental shift away from Baudouin’s insight led to the classical phoneme in the West. I have not explained why we ought not to have a three-level phonology, which I believe that we should not. Nor have I pointed out why I think Baudouin’s insight is not really compatible with generative grammar, which I believe it is not. In fact, this is only the tip of the iceberg of what I would like to say. But it is enough just to make the simple point that generative phonology has lost its way. It is time to bring Baudouin de Courtenay back and have a serious look at what started phonology out as a field of study in the first place. If generative linguists still want to have a three-level phonology…well, that’s a subject for a later rant.