I believe there was some important research published in recent decades which brought a fundamental change to the way linguists think about phonemes.

Or is it that the concept of the phoneme has always been contentious despite its seeming wide acceptance to those of us who only dabble in linguistics?

I seem to recall reading that a certain linguist studied a certain language and in the end decided that how we thought of phonemes just didn't apply to this language and that we instead had to look at things another way.

Does my recollection have any truth to it? Who was the linguist and what is the groundbreaking paper or name of the competing theory?

I know the question is pretty vague so please help me improve it. Answers telling me I'm barking up the wrong tree are absolutely fine.

  • 2
    The title to the question is pretty presumptive..or at least I've never heard of any such demise. '...research that claims to undermine the concept...' maybe?
    – Mitch
    Sep 16, 2011 at 13:48
  • 1
    @hippietrail, Languagehat is indeed an excellent blog, but the audience tends to have a theoretical perspective that is somewhat (for lack of a better word) iconoclastic. So of all the groups of internet linguists, the LH readers would be the most likely to already be familiar with such a line of research.
    – Aaron
    Sep 16, 2011 at 18:12
  • 1
    Not sure if it's relevant or not, but your question reminds me of this discussion on the existence of syllables.
    – grautur
    Sep 16, 2011 at 19:18
  • 2
    I don't have enough time to write a proper response, but I will say that the "classical" phoneme as a contrastive set of allophones has been discredited almost from the moment of its inception. Both the early concept of "archiphoneme" and the generative phonological idea of underlying and surface segments go outside of the classical phoneme, and OT goes further afield yet. The basis for all three of those schemes is easily found in any linguistics textbook. Sep 17, 2011 at 0:32
  • 1
    @JSBᾶngs: Well like I said in my question please help me improve it. Rather than close it I think you should submit an answer saying basically what you've said in your comments. I'm happy to reword my question. Just closing it on these grounds makes it seems like I should've been able to predict the answer, in which case I wouldn't have needed to ask. Maybe something like "What is the competing theory to the phoneme?" or "Is the concept of the phoneme universally accepted?" Sep 17, 2011 at 14:28

8 Answers 8


A part of the question thusfar unattended is whether the concept of phoneme has always been contentious. It has.

The position defended in Sapir's (1925, 1933) classic papers, that the phoneme is not just a unit of analysis, but has some psychological status, has always had its critics. There are two major facets of this claim that were challenged: that mental processes of language users involved a segment-sized atomic unit, and that this segment-sized unit was a primitive, called the phoneme.

Sapir's position is challenged by Twaddell (1935: sec.2), who is suspicious of the whole enterprise of inferring mental entities. A fine quotation from that work is given here:

Such a [mentalist] definition [of the phoneme] is invalid because (1) we have no right to guess about the linguistic workings of an inaccessible 'mind', and (2) we can secure no advantage from such guesses. The linguistic processes of the 'mind' as such are quite simply unobservable; and introspection about linguistic processes is notoriously a fire in a wooden stove. Our only information about the 'mind' is derived from the behavior of the individual whom it inhabits. To interpret that behavior in terms of 'mind' is to commit the logical fallacy of 'explaining' a fact of unknown cause by giving that unknown cause a name, and then citing the named x as the cause of the fact. 'Mind' is indeed a summation of such x's, unknown causes of human behavior.

A similar criticism, this time by a British phonetician, is found in Butlin (1937), who maintains (reflecting the then popular instrumentalist philosophy of science) that the phoneme is a "convenient fiction." Abercrombie (1991: 27ff.) [whose ideas were developed much earlier than the citation date suggests] cites Butlin's paper, disputing the status of segment-sized atomic units in language.

Eventual attacks on the phoneme from the side of generative linguists are anticipated by Martinet (1949), who argues that while most phonological concepts should be regarded as fictions, it is necessary to select at least one having status in reality, so that the theory could be grounded in facts describable in the languages of other branches of science. The concept he selects as "...la seule unité pour laquelle nous postulions une existence réele..." (1949: 46) [the only unit for which we postulate actual existence] is the feature rather than the phoneme. Martinet, however, appears to favor a non-mentalist conception of the feature.

Beginning with Halle (1954), American phonologists in what would become the generative tradition started to argue that the segment has psychologically real status, but that the phoneme is epiphenomenal, its observed properties actually being due to featural processes.

A very nice historical introduction to phonology prior to the 1970's is Fischer-Jørgensen (1975).

  • There is something really important that many researchers forget (although they shouldn't). As Rebecca Frumkina put it, science doesn't deal with natural objects. In other words, things we study do not exist in that form in the outer world; they are constructs of our research.
    – Alex B.
    Feb 25, 2012 at 19:09
  • That is why it is methodologically wrong to say that the phoneme is or isn't psychologically real. Nothing we study is. The question we should ask is whether the phoneme is a useful tool for analysis.
    – Alex B.
    Feb 25, 2012 at 19:19
  • 1
    @AlexB. the position you are putting forth is often not adopted because researchers forget about it, but because it leads to many difficult issues if taken seriously (not to say that it is wrong, but its truth is not exactly self-evident). If none of the units of linguistic analysis can be grounded in reality, then there is no way of verifying the correctness or even near-correctness of an analysis. We could say that a phoneme is not real, but epiphenomenal, but saying that is implicitly claiming that there is some real unit driving the effect.
    – user483
    Feb 25, 2012 at 20:05
  • 1
    "verifying the correctness of an analysis" - I don't think it's methodologically correct to say that a certain analysis is either correct or incorrect. imho a better way to put it is to say analysis X covers/describes more data or makes better predictions.
    – Alex B.
    Jul 15, 2014 at 23:05
  • 1
    @AlexB. you had me waiting by the computer for two years for your response! I do agree it's complicated about how to prefer analyses. you might find works by Larry Laudan on the philosophy of science interesting.
    – user483
    Jul 15, 2014 at 23:10

"Phoneme" is a theoretical construct which probably can't go away. Especially in languages with writing, most speakers can be asked "which sound did you make here?" in a word, and provide an answer; the answer often roughly corresponds to some canonical choice of representative sound for that phoneme. I believe speakers of languages which use pictographic representations are often also able to do this; I'm not sure what the situation is like with speakers of languages without a widely accepted writing system.

"Phone," on the other hand, seems like it could be more contentious. For instance, in Uighur /o/ and /u/ are separate phonemes; if asked, speakers will readily clarify which one they used; but the actual acoustics of their articulations in context easily bleed into each other.

  • 1
    I had heard that in fact phonemes don't come naturally to people but have to be learned. I believe they are a bit easier to learn for people literate in a language with a fairly phonetic alphabet. Apparently it's the syllable which we naturally handle. Maybe there's a good question we can ask about that. Sep 18, 2011 at 18:32
  • 4
    Yes, "Literacy Training and Speech Segmentation" and "The Ability to Manipulate Speech Sounds Depends on Knowing Alphabetic Writing" in Cognition 24 are good examples of that. However, there's a great deal of evidence that the phoneme exists in our mind but without alphabetic literacy we lack a consciousness of them. "The Psychological Reality of Phonemes" in [here][amazon.com/Selected-Writings-Language-Culture-Personality/dp/… has a great explanation of it.
    – Nate Glenn
    Sep 18, 2011 at 18:54

There is not much contention about the concept of a phoneme, but there are several researchers who think it is more a function of literacy than reality.

Suffice to say, we all use phonemes in discussing languages, but when you look at speech, you do not see discrete units. You see lots of variation that can be made discrete-seeming through speech processing.

Even if we were to say "phonemes do not exist as primary things" (i.e., in the same way as we can say "chairs exist"), we can still approximate the concept of a phoneme using things that do exist like pitch, intensity, formants, noise, times, etc.

  • When you say "speech processing" this way do you only mean with a computer or also in the brain? Sep 21, 2011 at 19:52
  • Either. We can definitely do it with a computer, and we at least have some discretization of language in the brain. Sep 21, 2011 at 22:16
  • 1
    And to add: our approximated concept of a phoneme will still be an abstraction, maybe some averaging of spectral characteristics and duration. Sep 21, 2011 at 22:26
  • That was Pike's point. The full title of his seminal 1947 book was Phonemics: a technique for reducing languages to writing.
    – jlawler
    Apr 10, 2013 at 23:21

A small addition to the above answers: I believe psycholinguistics in general is perhaps the field where the opposition to the notion of phoneme is the strongest. I once read a summary of a number of psychological studies which were supposed to prove that phonemes don't really exist in our minds and what speakers typically consider to be phonemes are in fact letters. I can't remember very well but I'm almost sure it was a paper by Robert F. Port. I can also remember seeing a few phonological theories which attempted to do away with the phoneme completely.

I don't want to evaluate all these ideas here. I think they might be indeed true but I surely can't prove it. This way or another, even despite its notorious indefinability, phoneme is a very handy concept in linguistic research and I guess most linguists use it many times each day, whether they believe in its psychological reality or not.

I certainly believe in the case you mention. There are probably many cases where it is very difficult to determine whether a certain sound is a phoneme or not, see e.g. Slavonic [y] (as in Russian ‹ы›) or Turkic [ḱ]. Nevertheless, on the whole, phoneme is a handy terminological tool and still widely used — just like we talk about nouns and verbs in English even though separating one from the other is often impossible.

  • "I once read a summary of a number of psychological studies which were supposed to prove that phonemes don't really exist in our minds and what speakers typically consider to be phonemes are in fact letters." Can you provide a link or a reference to this article/summary?
    – Alenanno
    Feb 24, 2012 at 23:45
  • I can't remember very well but I'm almost sure it was a paper by Robert F. Port, most probably this one: cs.indiana.edu/~port/HDphonol/Port.graphical.basis.phones.pdf.
    – kamil-s
    Feb 25, 2012 at 0:33
  • Rich phonology proposed by Port and Bybee (mainly) doesn't deal away with phonemes, but it reinterprets them a s categories encompassing multiple exemplars. A phoneme in Rich Phonology is not some abstract representation, but the how we organize multiple exemplar, rich representations of the actual sounds.
    – MGN
    Jun 1, 2013 at 10:30

I remember reading in a book Eskimo-Aleut languages that the author disagreed with the concept of phonemes and allophones. He said (about Greenlandic; my translation from the German):

Under certain conditions only [f x χ ɬ], under others, only [v ɣ ʁ l]. These conditions are the quantity. [Summary of the remainder of the paragraph: the former only occur long, the latter only short. -PN]

There are linguists who would not simply describe this fact and leave it at that. In a functionalistic analysis, they would combine those sounds to four units called phonemes: they would combine [f] and [v], [x] and [ɣ], [χ] and [ʁ], and [ɬ] and [l]; and one would come up with labels for the phonemes such as /f x χ ɬ/ or /v ɣ ʁ l/. The situation where one phoneme corresponds to a number of different sounds is then called allophony. The really interesting question, however, is what the point of the concept of allophony is. There are some notes on this in Holst (2001: 38, 48f.), but they are not particularly numerous.

Functionalists start from the presupposition that languages are composed of units of sound whose differences distinguish meaning. For example, the German words Paß [pas] and Baß [bas] have two separate units in their respective onsets which determine the meaning difference between the two words. There are, however, cases where differences of sound cannot make a difference in meaning, because the language system [what I would call the phonology, PN] regulates which sound appears where. For example, in German, the sound [x] only appears after back vowels and [ç] appears in other positions. This is a complementary distribution. The situation with Greenlandic [f] and [v], etc., is another example. When functionalists encounter such cases, they deny that those sounds are individual units of the respective language, say that this is a case of allophony, and combine the two (or more) sounds to a single phoneme. This course of action is, however, absurd. Afundamental trait of science is that it is empirically founded: observations are made and then conclusions are drawn. If one observes, for example, that German [x] and [ç] are in complementary distribution, then one may and should state this, but one may not reinterpret the data. The basic assumption of functionalism, which was stated at the beginning of this paragraph, is incorrect: languages are simply not composed of units of sounds, the distinctions between which always cause differences in meaning. One encounters the truth of this statement empirically again and again.

It has to be stated clearly: the concept of allophony is pure nonsense. Such combining of sounds are not legitimate in a scientific treatment of language and should therefore be refrained from.

(Jan Henrik Holst, Einführung in die eskimo-aleutischen Sprachen, Helmut Buske Verlag, Hamburg 2005. ISBN 3-87548-386-3. Pages 39f. Notes in [italics and square brackets] are mine.)

  • 4
    The passage you quote is very interesting. The author is claiming that the concept of multiple phones being allophones of a single phoneme is "absurd" and "pure nonsense", but I don't see that he's making any argument to support this idea. He shows that some contrasts (German [p] vs. [b]) can distinguish words while others (German [x] vs. [ç]) can't. What he doesn't do is show why either 1) sounds like [x] and [ç] can't be described as belonging to some higher-level structure (a phoneme) or 2) pairs like [x]/[ç] and [p]/[b] shouldn't be treated differently.
    – Joe
    Oct 5, 2011 at 5:10
  • It probably is absurd to him. It may be absurd for many others. For still others it makes all kinds of sense. There's a lot of variation. There is no reason to suppose that everybody internally structures the speech sounds they hear and produce in exactly the same way, any more than we would suppose that everybody internally structures the sexual phenomena they observe and produce in exactly the same way. I.e, phonemes are one way we can do it, but not the only one.
    – jlawler
    Apr 10, 2013 at 23:27
  • @Joe, I agree that the author didn't say why they couldn't, as you say, belong to some higher level structure; He merely points out that it should be empirically founded. If one understands allophones, as I do, to be the sum of sounds one would accept to be the same phoneme, then I'd be on the fence about [x]/[ç] because the complimentary distribution is distinct in my dialect; That's not what makes them alophones (imho, and I agree so far with the author), but other dialects apply them differently, and, in one view, I have to accept those; In my view, I may still reject them as wrong.
    – vectory
    Oct 29, 2019 at 16:42

The one point that doesn't seem to have been mentioned is the distinction between segmental and non-segmental approaches to phonology going back at least to the 1960s. So while something phoneme-like may indeed be useful for some kinds of linguistic analysis, the psychological reality (as pointed out) or practical utility (for speech recognition, for instance), has long been in question.

Approaches such as autosegmental phonology or exemplar phonology all seem to be dancing around the same issue of phoneme being at the same time so attractive and so limiting as an analytic concept.


Edward Sapir's classic paper The Psychological Reality of Phonemes argues that the phoneme (as opposed to the phone, an object properly in the domain of phonetics) is necessary to understand phonological systems. More recently, probably beginning with Chomsky & Halle's The Sound Pattern of English, phonemes were `decomposed' into bundles of features that can be the target of phonological rules. From the wiki page:

Chomsky and Halle represent speech sounds as bundles of plus-or-minus valued features (e.g. vocalic, high, back, anterior, nasal, etc..) The phonological component of each lexical entry is considered to be a linear sequence of these feature bundles. A number of context-sensitive rules transform the underlying form of a sequence of words into the final phonetic form that is uttered by the speaker. These rules are allowed access to the tree structure that the syntax is said to output. This access allows rules that apply, for example, only at the end of a word, or only at the end of a noun phrase.

As far as I'm aware (though I am no phonologist), there is not much of a debate within rule-based phonology about the status of the phoneme (or, the feature bundle it `names') as a theoretical necessity. Outside of rule-based phonology, I'm not sure the issue is treated.


Lots of linguists have disliked phonemes.

Who was the linguist you have in mind? It could be Noam Chomsky, if you go clear back to 1964 and his paper Current Issues in Linguistic Theory. Chomsky argued that phonemic theory could be characterized by certain assumptions about the relationships of allophones and phonemes, which assumptions were vulnerable to factual arguments. Defenders of phonemics naturally attacked the assumptions as not actually characterizing phonemics at all.

Generative phonologists (followers, to at least some extent, of Halle and Chomsky's theories) who had a descriptive interest in phonology have tended to use the term phoneme to mean segment of underlying phonological representation, which is probably very, very different from what a phonemicist would have understood a phoneme to be like.

  • Is there an auditory agnosia where a single allophone fails to be perceived?
    – amI
    Sep 12, 2017 at 17:29
  • @amI, I don't know.
    – Greg Lee
    Sep 12, 2017 at 19:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.