phonemes and phonological knowledge

Mark William Post mpost at DARKWING.UOREGON.EDU
Thu Apr 3 20:44:50 UTC 2003

Dear Suzette (and others),

How I Learned to Stop Worrying and Love the Phoneme (despite the
apparent fact that it doesn't exist)

The basic question seems to be this: what do language users know about
the sounds of language, and how is that knowledge implemented in
communication? There are two very well-supported facts that I see
emerging from thousands of years of research into this question, which
appear at first to be contradictory: 1) language users organize
phonetic material into contrastive categories, which are in most cases
readily comparable across languages 2) the precise content of these
categories - as they are realized in speech - is infinitely variable.

Another way of putting this: language after language agrees on roughly
the same ways of categorizing phonetic content - the high, front,
vowel of language x sounds to a speaker of language y more or less
like the high, front, vowel of language y. At the same time, the
implementation of the categories of language x in actual discourse -
in terms of both sheer phonetic detail and the ways it behaves or is
reshaped in certain contexts - is never - *ever* - precisely the same
as in language y, even given highly controlled experimental
conditions. The origins of these rough categories, or why some
phonetic material seems pretty ripe for recruitment as a contrastive
category, have been shown in many cases to originate in biological and
physical facts to which any language user is subject. And yet it is
clear that what is apparently stored and actually implemented from
language user to language user, and from language to language -
exceeds anything these universal factors alone can explain.

Phonemes seem to me to be theoretical constructs which have been
designed to describe the first fact very well. They do a terrible job
of handling the second fact, often with the result that fact 2 is
simply ignored. They are *useful* in so far as they can often if not
always handle the first fact. But they are virtually *useless* to
anybody interested in accounting for the second, i.e., to anybody
intersted in accurately accounting for a language user's phonological

Does this mean phonemes don't exist, i.e. don't have cognitive
reality? Well, if you design your notion of phoneme (and/or feature)
to accurately describe a language user's phonological knowledge (as is
generally done), you fall short of accounting for the facts - language
users evidently store and implement a great deal more phonological
knowledge than you're capable of capturing with your notion of
phonemes. It would seem probabilistic mathematical models come closer,
although I don't think there's yet a clear idea of how to structure

And yet for transcription, shorthand, representation of the rough
behavior of speech sounds, there is simply no substitute. This is the
use-value of phonemes. They are handy and accessible, and do a
reasonable job of letting us represent our data. But this usefulness
is limited to certain domains, and do not in fact satisfy the demands
of representing phonological knowledge. And as for what to tell people
outside the ivory tower, I'd tell them pretty much that.

For further info, and said in a clearer fashion than I could ever hope
to, see Pierrehumbert (1999) What People Know about Sounds of
Language. Studies in the Linguistic Sciences 29, no. 2 (1999 Fall): p.
and Ohala J. (1974) Phonetic Explanation in Phonology Papers from the
Regional Meetings, Chicago Linguistic Society, Special Issue, Apr,

Thanks for your time,

Mark Post

More information about the Funknet mailing list