question about phonemes

Spike Gildea spike at DARKWING.UOREGON.EDU
Thu Apr 3 22:51:28 UTC 2003

Sorry for any confusion, folks -- Geoff and Dan didn't see my message
because I had replied privately to Suzette.  I didn't really want to
make time to participate fully in a discussion right now, and I had
this silly idea that I could throw out a quick thought off the top of
my head and then avoid a longer discussion of anything that might
turn out to be more controversial than I had realized.  Here is the
message I sent to Suzette, followed by a brief explanation of my
assertion that the notion of "phoneme" is no longer widely accepted
-- an assertion which appears not to be true after all!

>Date: Wed, 2 Apr 2003 14:35:40 -0800
>To: Suzette Haden Elgin <ocls at MADISONCOUNTY.NET>
>From: Spike Gildea <spike at>
>Subject: Re: question about phonemes
>Hi Susan --
>I just wanted to stick in a couple cents in favor of not asking the
>question (at least not in this way).  In general, superlatives in
>Linguistics (most, fewest) have always struck me as artificial, so I
>would advocate just giving a range, just like you do when talking
>about how many languages there are.  And since most linguists no
>longer believe in the cognitive reality of the notion 'phoneme', we
>are unlikely to ever meaningfully resolve even the question of how
>many is the "right" number for any individual language, much less
>which languages have the most and which the least.
>To be a bit more concrete about your specific question, the issue of
>how many phonemes a language has is not a question of fact, it is a
>question of analysis, and the outcome depends crucially on your
>criteria for identifying phonemes.  Places where problems come up
>include whether or not to count contrasts that are arguably
>attributable to suprasegmental features overlaid on more compact
>feature bundles (cf. Dan's  decision not to count tone contrasts in
>Pirahã as phonemic, but to isolate out tongue position and lip
>rounding as the only 'phonemic' distinctions) and whether to count
>contours as individual segments or as sequences (cf. Claire's caveat
>about whether to count "click plus accompaniment" as a single
>phoneme, and similar issues in 'fortis/lenis' contrasts, diphthongs,
>affricates, prenasalized stops, etc.).  Different analysts could
>count between 3 and 64 phonemes in a vowel system with a simple i,
>u, a inventory, but additional contrasts in nasalization, vowel
>length and a four-tone system (3*2*2*4).
>Since I'm teaching intro this term, I expect the issue to come up
>again (it often does in that context), but I'm not sure if there's
>any context where I'd like to entertain the question seriously.
>>April 2, 2003
>>Could someone(s) on the list give me their opinion(s) on the maximum number
>>of phonemes in human languages? Long long ago I was taught that it was
>>roughly 70; in recent years I've seen claims that it's roughly 150. I keep
>>seeing different totals in different sources, and there's a lot of space
>>between 70 and 150.  (It makes me wonder if it's analogous to deciding "how
>>many languages exist" and showing a range from 5000 to 10,000 based on how
>>one defines "language" and "dialect.") And -- if 150 is near the mark -- is
>>that a rare extreme?
>>Thanks for your help. I realize that it's not a profound question [at least
>>I don't _think_  it is, but I'm not a phonologist and may be wrong about
>>that], but I'm not satisfied with the answers that I'm finding.
>>Suzette Haden Elgin

First, I retract my glib assertion about what "most linguists"
believe -- obviously I haven't polled a representative sample, and
there's not enough riding on the answer to make it worth the work.

Second, a quick stab at the substantive question: Do phonemes exist?
Depends on what you mean by phoneme, and what you mean by exist.  I
imagine that everyone agrees on the cognitive reality of contrast,
and it is clear that phonemes provide one way to model that contrast.
But unless we are working with different, updated definitions, the
notion of 'phoneme' that I am familiar with is embedded in phonemic
theory, where it entails the idea of indivisible, atomic units, which
are then discrete building blocks of morphemes.  Both the phoneme and
the theory that launched it certainly seem to have been rejected many
times in the theoretical literature, both by classical generative
theory and subsequent theories that still state phonological
generalizations in terms of discrete features (which may coalesce
into more or less autonomous bundles, depending on the analyst and
the language), and also from cognitive theories like the one
represented in Joan Bybee's recent work, that see phonemic behaviors
as schemata that are constructed by generalizing over a body of
stored, phonetically rich word-level exemplars.  Since the theory
that originally gave meaning to the term phoneme is clearly not still
practiced, what is the theoretical status of the notion in more
current theories that still want to make a place for it?  Perhaps the
notion of phoneme is both more current and more coherent than I have
imagined from the subset of the literature that I have read, and I'm
interested in learning more from people who have spent more time
researching this issue.  I am looking forward to reading Geoff's
textbook when it comes out, and I'd appreciate a chance to look at
his list of references sooner (privately, if nobody else wonders
about these questions).

Some more thoughts, partly triggered by Geoff's and Dan's postings...
If the phoneme is seen as a taxonomic unit, then the question is
whether the unit operates as simply a way to conveniently represent
one's data, or whether there is some deeper theoretical/cognitive
significance given to this taxonomic unit.  Obviously, a taxonomic
unit like a phoneme is useful for things like orthographies, without
which it is difficult to even begin to study much of the stuff that
both functionalists and historical linguists have traditionally spent
most of our time on.  And there are certainly many patterns in data
that can be modeled well with reference to the notion phoneme.  It is
less obvious that these patterns cannot be modeled well *without*
reference to the notion phoneme, or that there is a clear need for
this taxonomic unit in order to understand the processing of sound in
linguistic cognition.

To take one of the arguments that Dan cites from LP as supporting the
taxonomic phoneme, evidence that sound change operates one segment at
a time undermines the unity of the notion natural class.  But this
same argument is taken one step farther by Joan Bybee to argue
against the taxonomic phoneme, since evidence that sound change also
operates one *word* at a time undermines the unity of the notion
segment.  Joan has advanced this argument in several publications,
and it is only one of many "frequency effects" that have been
documented (see, e.g., the studies in Bybee and Hopper 2001,
Frequency and the Emergence of Linguistic Structure.  She and others
have been working towards a theory that gets "phoneme effects" as a
secondary phenomenon, without assigning a primary cognitive role to
the notion phoneme.

So can we be confident that phonemes exist?  As a label for a
distributional pattern in data, of course.  As a contrastive
taxonomic unit in writing systems and grammars, of course.  As
something that can at least be constructed for use in tasks like
reading, again, no doubt.  As a primary cognitive unit that
structures our perception and production of speech, I think there's
plenty of room for skepticism (obviously -- I'm the guy who thought
most lingists were skeptics).

And to return to the original question, even if we agree that
phonemes exist, that is not the same thing as saying that we can
always determine precisely how many of them we should posit for any
given language, with no room for debate.  I really wonder whether we
have a set of universally agreed-upon criteria that theoreticians
from various camps and descriptive linguists (who generally are the
ones that provide the data for typological/theoretical
generalizations) all rely on in making the analytical decision of how
many of the contrasts that speakers attend to should be counted as
"phonemic".  Pretty much every language presents at least one
problematic issue, whether with autosegmental features, with
contours, or with consistent "subphonemic" variation, and thus
presents analysts with non-automatic choices about precisely where to
draw the phonemic line.  So even if we grant that pretty much all
linguists use -- and maybe most also believe in some cognitive
reality to -- some notion "phoneme", I remain dubious about the
validity of any precise answer to the question of which language has
"the most" or "the least" phonemes, and how many that might be.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the Funknet mailing list