Subject: Do phonemes = sounds? From: Robert Port <port(at)CS.INDIANA.EDU> Date: Tue, 7 Jun 2005 13:54:10 -0500I can't resist responding to this issue. I have spent my career trying to figure out what words are made of. I have finally come to the conclusion that: BOTH PHONEMES AND PHONES ARE INTUITIVELY PERSUASIVE PRIMARILY BECAUSE OF OUR LIFELONG EXPERIENCE WITH ALPHABETS. Speech sounds are very short (15-20 per second), and the relevant motor gestures are mostly invisible (tongue, glottis, velum, etc). And those of us in the European cultural tradition learn to use letters beginning as young as 2. Letters are a great engineering solution to preserving language in graphic form and our education system assures that we all become proficient at thinking about speech in letter-like terms from an early age. But phones and phonemes inherit many graphic properties from letters: * SERIAL ORDER (no temporal patterns allowed), * NONOVERLAPPING (hence the artificial `coarticulation problem'), * PERFECTLY CONTRASTIVE FROM ONE ANOTHER (no near contrasts or partial contrasts) * STATIC (diphthongs, glides and affricates present awkward inconsistencies). In the late 19th C, de Courtenay proposed the notion of the `phoneme' which was very quickly adopted by phoneticians and linguists and treated as a great discovery about human language. In fact, all that happened is that scientists began to think seriously about the psychological representation of language (which had been largely ignored earlier) and thought `Maybe we have something in our heads that represents words the way letters do. There must be something analogous to letters to keep words distinct from each other. There must be PHONEMES!' The phoneme was not discovered, it was just postulated by analogy with alphabetical written language. So how ARE words `represented' in the head?: By gestures and gesture components of various sizes - from feature size to syllable size to the size of whole phrases - WHATEVER STATISTICAL REGULARITIES SPEAKERS HAPPEN TO PICK UP as they learn how TO talk. Each language has its own conventional solution to keeping utterances distinct. Of course, languages have a phonological system, something a little like an `inventory' of possible sound contrasts in various positions. This system should be thought of as a social institution that children learn to adapt their speech habits to. But these inventories are very different from a cognitive spelling system. Eg, they always have many uncertainties, places where you cannot tell which phonemes (letters) to employ: what is the V in `beer'? (same as bead or same as bid?), what is the stop in `stow'? Same as tow or same as doe? What is the pstop butter? Same a butt or same as Bud? Or neither? (But the data show two slightly different flaps are used in ladder and latter!) If language were REALLY spelled with a phonological alphabet, then these uncertainties could not happen. I have a couple mspts on my website addressing these issues. Obviously there are a great many empirical implications of this iconoclastic hypothesis: implications about how speech can be recognized (by humans or machine), for how languages are learned, how they change over time, differences between the linguistic intuitions of alphabet-literates vs. nonalphabet-literates, constraints on phonological language games (like pig-latin), etc etc. Basically ALL the data come out just as this hypothesis predicts. I claim, in fact: NO DATA WHATEVER, (aside from the powerful intuitions of lay people and linguists educated in the alphabetic tradition) SUPPORT A SEGMENTAL (that is, C and V) DESCRIPTION OF LANGUAGE! I would like to hear the evidence if anyone wishes to challenge this claim. I have a couple papers on my website presenting these arguments in more detail. http://www.cs.indiana.edu/~port/pubs.html Bob Port ( ( ( O ) ) ) ( ( ( O ) ) ) ( ( ( O ) ) ) Lingstcs/Comp Sci/Cogntv Sci ROBERT F. PORT 330 Memorial Hall, Indiana University Bloomington, Indiana 47405 812-855-9217 Fx 812-855-5363 http://www.cs.indiana.edu/~port