LINGUIST List 10.1778

Tue Nov 23 1999

Disc: New: What exactly are allophones?

Editor for this issue: Karen Milligan <>


  1. Martin Salzmann, What exactly are Allophones?

Message 1: What exactly are Allophones?

Date: Sun, 21 Nov 1999 23:20:10 +0100
From: Martin Salzmann <>
Subject: What exactly are Allophones?

Editor's Note:
Martin Salzmann's summary for his query (both reprinted below)
contained such a wide range of responses that the moderators felt 
this topic might prove to be a good one for a discussion.

 The Query

Dear all,

 I'm currently trying to teach the basics of phonology to my students
 beginning with the classical distinction between phonemes and

 Now to my great surprise, I've encountered a problem that has never come
 to my attention before: In the classical structuralist sense, phonemes
 belong to the domain of "langue", i.e. the phonological system, while
 (allo)phones belong to the domain of "parole", they are the actual
 phonetic realizations of a phoneme. No problem so far.
 A phoneme like /i:/ (any other vocalic phoneme could serve as example as
 well) in English is realized as [i:] - but this phonetic notation is of
 course an abstraction since every [i:] that is uttered is somewhat
 different - so we'd have to say that there is an infinite number of
 allophones of the phoneme /a/. Usually one only says that there is one
 allophone - probably because all the realizations are considered in some
 way similar.
 But what about complementarily distributed allophones? Take for
 instance the English voiceless plosives which depending on the
 environment are realized either as aspirated or non aspirated. These are
 called allophones but they are again an abstraction, each of the
 allophones, e.g. [p +asp] and [p -asp] can be realized in an infinite
 number of different ways. Now what should be called an allophone? If the
 term were restricted to the actual phonetic realization (as in classical
 structuralism), we'd have to find a new term for the "abstract
 So do we end up with tree levels instead of two? Or to put the question
 differently: Which phenomena belong to phonology, which to phonetics?
 Since complementary distribution is an abstract regular pattern not
 solely due to physiological necessity, i'd have to belong to the domain
 of phonology in my opinion. But is there a possibility to express this
 in classical structural phonology or is the theory just simply flawed or
 is my argumentation faulty?

 I believe, the picture would be different in generative phonology. If
 I'm not completely mistaken, the three-way contrast indicated above is
 in fact represented by the classical derivational model: There are the
 URs (roughly comparable with the phonemes) and the surface
 representations. As far as I can tell, these surface representations are
 not phonetic entities, but still feature bundles (different from the URs
 only in being fully specified and possibly having undergone some rules).
 This is how the are in the phonological component (probably
 corresponding to what I called "abstract allophones" above). What in the
 classical structuralist sense is called a phone, conceived as a physical
 entity, would then be the result of the interaction of the phonological
 component with the sensory system.

 There is a third problem for which I've been unable to find a satisfying
 solution. Quite often, a distinction is made between complementary
 distribution and Coarticulation, the distinguishing factor being
 physiological (in)evitability. In a major textbook like Spencer
 Phonology (1996), the different pronunciations of /k/ in <key> and
 <car>, the first palatal, the second velar, are considered an instance 
 of coarticulation because of a physiological inevitability. Now this
 inevitability is more properly called a physiological inevitability of
 the native speakers of English since the two sounds can be contrastive
 in a number of languages, for instance Turkish. But if there is no
 universal physiological necessity to pronounce the sounds this way,
 couldn't one regard the two realizations as complementarily distributed
 allophones or surface representations?
 In German phonology, when speaking of complementary distribution, the
 example always adduced is the distribution of the voiceless velar and
 palatal fricatives; There occurrence depends on the value for the
 feature [back] of the preceding vowel (i.e. velar after a back vowel,
 palatal after a front vowel). For a native speaker of standard high
 German, this distribution is about as physiologically inevitable as the
 key/car distribution for a native speaker of English, but no one has
 ever spoken of coarticulation in this case. What might be the reasons?
 Am I just simply wrong? Do these examples represent truly different

 The Responses

I'm not an expert in phonology, but I do teach it, and the story I
usually tell my students is that there are 3 levels: the phoneme, the
allophone, and the phone. I have, in fact, seen this perspective
taken in several textbooks. A phone is a physical event, an allophone
is a category of physical events, and a phoneme is a category of
allophones. In fact, I don't see any way around looking at it this
way. Of course, a phone is really still an abstraction, because
speech is not segmented.

This does bring up deeper questions about the distinction between
phonetics and phonology, which I have never seen resolved to my
satisfaction. However, one could, if one wanted to gloss over things,
give the problem of how phones are categorized into allophones to
phonetics (the search for acoustic cues to segment identities is part
of this task) and give the problem of how allophones are grouped into
phonemes to phonology. Such a division would only convince beginning
students, of course, but it is a sort of ideal that almost reflects

As for the distinction between complementary distribution and
coarticulation, I think that you've confused the taxonomy a bit.
Coarticulation is merely one process that leads to complementary
distribution. That is, the two /k/ allophones in English are an
example of complementary distribution. The source of this
distribution is coarticulation affects. Furthermore, coarticulation
affects of this sort are not a physiological inevitability, as you
point out.

The book you refer to may be trying to make a distinction between the
minor, though predictable, differences in pronunciation between
different occurences of the *same allophone*, and the larger
differences recognized as allophonic variation (differences which
distinguish allophones *from each other*). If so, coarticulation is
not the word to use to describe the minor differences. I have never
heard a term for this notion, because there is clearly a gradient.
Whether we recognize two physical occurences as two occurences of the
same allophone, or as occurrences of different allophones is not a
question that I think has a definite answer. That is, the border
between phonetics and phonology is not clearly defined. Perhaps work
in phonetics will one day make things clearer.


Bob Knippen
Dept. of English
Texas A&M University

As far as I know, pure Saussurean theory does not address phonological
variation at this point; the theory of phonemes which led to
distinctive feature theory came from Boas and Sapir (cf. Boas's paper
"On Alternating Sounds" and Sapir's paper "The Sound System of
Language")--that Jakobson was influenced by Boas is clear in his
writing on Boas (to be found in his collected writings). Boas
explained that the sounds of American languages were not random, that
there was a system; Sapir realized that the sounds of the language
were organized systematically with respect to one another and that
phonemes were therefore psychologically real for speakers in a way
that simple phones are not; that is, one can physically make sounds
not present in one's own language but it is difficult to place them
into one's own speech system; also, it is difficult for most speakers
to hear allophonic variation in their own language. If I were to put
it into structuralist terms, then, the rule for deriving allophones
which stand in complementary distribution would be a part of langue,
that is the language system at the type level. One could make a case
for this in Saussurean terms by referring to the concepts of linearity
and relative motivation (that is, the relationship of phonemes to each
other is determined by pure difference (i.e. distinctive features),
but the presence of a phoneme in a syntagm is realized in parole
through the syntagmatic rules through which a phoneme, standing next
to other phonemes in a linear relation, is realized in
speech--however, this is NOT what Saussure argues. For Saussure, the
relation between type-level langue and token-level parole production
is problematic--he does not give a cogent theory of speech production
and the relationship of langue to the ideal speech community [masse
parlante] is particularly problematic. The theory of the phoneme
which later developed into generative phonology did not develop with
Saussure; rather recognition of a phonemic system as psychologically
real for speakers came through Sapir in the 1920s, after Saussure's
time; Jakobson united this idea of a phonological system with the
Saussurean idea of value as produced through negative difference,
creating with Halle the theory of generative phonology (different
still from articulatory phonetics--although the connection of the
phonological system with the body as realized in phonetic utterances
is important for some theorists).

In short, YOUR argumentation is not faulty; rather the problem is with
the lack of interface between type and token in Saussurean
theory--this problem runs throughout in terms of language change but
also in terms of the relation between phonetic realizations of
rule-based behavior, such as one sees in phonology. This is a problem
with roots in Western semiotic thinking since Aristotle and Plato.
But in generative phonology, the insight into distinctive features of
the phonological system did come from the Saussurean concept of value;
that is that nothing has value except in terms of how it is different
from other possibilities within a syntagm. However, that view has
been critiqued in terms of the lexicon (Jakobson: Six lectures de son
et sens) but it appears to have validity for the phonological system
at the level of speaker awareness of the system (that is, what
speakers recognize as phonemes and where they have trouble
distinguishing complementary distribution in their hearing). More
cognitively-based approaches, such as the current Optimality Theory,
appear to critique this view but are not as approachable in terms of
speaker awareness--whether they are valid as representations of actual
cognitive phenomena is a current or future debate, I'm sure.

John Thiels
Ph.D. student
Department of Anthropology
Brandeis University
Waltham, MA
Interesting posting on Linguist. The skepticism you
exhibit towards the accepted distinction appeals to
me. I'd say you're on the right track when you
suggest that the original distinction between
"phoneme" and "allophone" is flawed. As you point,
out every realization is unique, so how can there be
classes, allophones? Without going into detail, I'd
just suggest you look at two works for a different
view of the task of phonology:

William Diver. "Theory" in _Meaning as Explanation:
Advances in Linguistic Sign Theory._ Ellen
Contini-Morava & Barbara Sussman Goldberg (eds.).
Mouton de Gruyter, 1995.

Yishai Tobin. _Phonology as Human Behavior._ Duke
University Press. Recent date.

Joseph Davis
City College of New York
Hola Martin:
en primer lugar, mis excusas por no contestarte en ingl�s ya que no s�
escribirlo; espero que puedas comprender el espa�ol. De todos modos, mi
nota es muy breve. Te voy a dar una referencia bibliogr�fica que a m� me
servido de mucho en mis clases sobre estos temas. Se trata de un trabajo
Coseriu, Eugenio (1962): "Forma y Sustancia en los Sonidos del
en Teor�a del lenguaje y ling��stica general. Cinco Estudios. Madrid,
Lo puedes encontrar tambi�n en Revista de la Facultad de Humanidades y
Ciencias, 12, 143-217.
Tambi�n aparece en una edici�n independiente en Montevideo en 1954.
Espero que me hayas entendido y que te sirva de ayuda.
Un saludo,

Javier Sim�n Casas
Departamento de Ling��stica General e Hisp�nica
Universidad de Zaragoza
Dear Martin.
I think that the problems you present are very common.

Jonh Laver (1994) distinguishes between the linguistic sound (the
sound, what you hear) and the phone. de Saussure also (1916) speaks
an "acoustic image" (psicological) and the sound (physical) and says
the acoustic image and the sound mustn't be confounded. The phono is

Coseriu (1952/1973) thinks that the sound is in the 'parole'; the phone
in the 'norm'; and the phoneme is in the 'system'. I think that this way
classifying sounds in structural phonology is better than saying that
phones are in 'parole' and phonemes in 'langue'.

Francisco Dubert Garc�a
Departamento de Filolox�a Galega
Universidade de Santiago de Compostela
Santiago de Compostela
 Your question on Linguist left me somewhat puzzled. I was under
impression that allophones were purely in the domaine of phonology and
co-articulation under the domaine of phonetics. Allophones and
co-articulation being both context motivated but the first one at the
phonological level (i.e. how a certain pheneme has to be realised under
specific phonological environment) and co-articulation at the phonetic
level (i.e. how certain allophones are realised under specific phonetic,
articulatory, environment).
 In your message, you ask wether there are 3 levels of
realisation instead of the usual 2. This somewhat gave me food for
and digestion kept me awake for a little time. Maybe there is a language

specific phonetics? i.e. the fine tuning of surface realisation will
an effect on the surrounding segments, effect that may not be universal.
Montreal French, there is a phenomena of affrication of dental stops
followed by front high vowel, /i,y/. In European French, this phenomena
not present, at least one does not hear it. When a sonograme analysis is

run on both dialects, affrication is present in both, though at
"strength". Laurent Santerre, deceased phonetician at Universite de
Montreal, once told me that it would be a perception problem since
accousticaly both dialects affricate their dental stops in that
environment. The specific aperture of Montreal French will reinforce the

level of affrication, making it more audible for speakers of that
He noticed the same distinction in the diphtongisation of long vowels in

both dialects.
 The velar-palatal difference between the 2 realisatons of /k/ in

English is, as you mentioned, considered as a co-articulation phenomena
while in Turkish it is considered as a dictinctive feature. I would be
inclined to think, from the affrication example in French, that in both
languages there is a co-articualtion phenomena of palatalisation in
of front vowels and of velarisation in front of back vowels in both
languages. The difference being distinctive in Turkish, it is not
but would show on a sonograme. The dinstinctiveness of the feature would

(and this is completly conjectural) weaken the perception of the
co-articulation as the aperture of European French vowels weakens the
perception, or stength, of affrication of dental stops.
 Now, should there be 3 levels (2 abstract and one concrete) of
sound production? I would say yes. Phonology (phonemes and allophones)
being abstract objects, one being the output of the other, and
co-articulation being the acoustic output of allophones. This is a
risky proposition since allophones have nearly always been tought as the

actual surface realisation of phonemes, but, as you wrote, no speaker
produce the exact same realisation of a phoneme. It was argued that the
differences were trivial and that they should not be taken into account
since they do not influence the perception, i.e. thay pass through the
percptual filters of speakers of a given language.
 On the other hand, the phonology of a language is nothing more
the statistical average of the phonology of the same language native
speakers, I would say that the relation between allophones and
co-articulation motivated difference in actual surface realisation is
same as the difference between the phonology of a language and the
individual phonology of the speakers. One is and average to be expected
>from any individual speaker and the other being the actual individual
realisation of the said phonology. Allophones are the average, expected
realisation of a phoneme in the language while co-articulation is the
actualindividual realistion.
 To comme back to the palatal-velar influence in the realisation
/k/ in English versus the dinstictiveness of those features un Turkish,
would say that both [k+velar] and [k+pal] in English sould be considered
allophones as far as perception is concerned while they are phonemes in
Turkish. As far as acoustics is concerned, I would say that they are
subject to the co-articulation phenomena (if it is demonstrated that
sonogrames in both English ant Turkish show palatalisation in a similar


Alain Theriault
Ph.D. candidate
Universite de Montreal
I am a linguistics and speech pathology student. I don't have a
"professional" opinion on your problem, but I've certainly thought about
a lot. I'm afraid that after studying phonetics, the only conclusion I
make is that phonologists just don't consider the reality of speech
They came up with a system of phonemes and allophones where one is
to be abstract and the other "real," but as you said there are tons of
allophones, not just a few. For example, an acoustic phonetician (can't

remember which) suggested there might be over 100 allophones of /s/ in
English alone. These depend on context of course, which raises your
about coarticulation.
I don't see any difference between comp distribution and
coarticulation. If
coarticulation is language specific, which you show with the Turkish
then it has to be allophonic variation. I think the reason allophones
about was maybe an attempt to account for the realities of speech, but
certainly doesn't come anywhere close to doing this. In terms of
I'm not sure there is a problem though. If you teach coarticulation as
a phonological process, it follows naturally that allophones are
Maybe any sort of assimilation, dissimilation, deletion, etc. is a type
language specific coarticulation - after all, speech is just one long
of coarticulated sounds.

Cori Kropf
Dear Martin,

Phonemes and allophones can be clearly distinguished by taking into
account two criteria: distribution and functional contrast, cf. John
Lyons (1981): Language and
Linguistics, Cambridge: CUP, pp. 85ff.
That is to say, allophones are phonetically similar (whatever this may
be, though) realizations of the abstract unit of a phoneme, technically
in the same way as allomorphs and
word-forms are concrete realizations of abstract morphemes and lexemes
respectively. With regard to phonology, phonemes can be identified on
the basis of the aforesaid
two criteria.
If phonetically similar sounds don't occur in the same context
(complementary distribution: e.g. light [l] vs. dark [l]), they can be
referred to as two allophones of one phoneme:
there is no minimal pair in existence, in which the distinction between
these two allophones leads to a functional contrast (in meaning).
On the other hand, if - even phonetically similar - sounds occur in the
same context and are able to create a functional contrast, i.e. fulfill
a distinctive function (e.g. voiced /b/
vs. voiceless /p/ in /bit/ vs. /pit/), these sounds are two different
phonemes in the language-system under issue.
However, I think, your problem lies within the concept of phonetic
similarity. For example, [t] and the regional variant of a glottal stop
in words such as <butter> can as well be
regarded as two allophones in free variation, because they don't fulfill
a distinctive function, but it goes without saying that the two lack
phonetic similarity since both place and
manner of articulation differ.
Nevertheless, phonemes can always be identified by the criterion of
functional contrast, whereas allophones as realizations of phonemes can
always be found either in
complementary distribution or in free variation without functional
contrast. I guess the best way is to ignore, to a certain extent, the
traditional belief that allophones must be
phonetically similar. Concentrating on this criterion, my students in
the foundation courses also tend to be confused.

Yours sincerely

J. Mukherjee

Englisches Seminar der Rheinischen
Friedrich-Wilhelms-Universit�t Bonn
Regina-Pacis-Weg 5
D - 53113 Bonn
In my opinion, most of the answer to your question involves a
meta-rumination, a meta-reflection,
on what a theory of phonology (like a theory of many laboratory-based
sciences) is, rather than
a position particular to one brand or another of structuralist or
generative phonology -- though
it may be hard to untangle what is particular to (say) Hockett-ian
structuralist phonology and
what is Hockett's meta-rumination about how science works.

Every science has to deal with the fact that there's a smooth (hence
infinite) range of
laboratory descriptions of the events that are being observed. But due
kinds of restrictions, both theoretical and practical (and the line
is often
hard to draw), we limit our "representation" of the data -- for example,
limit it to
N significant digits; we often say we do that due to limitations of the
being used to conduct observations.

We -- still not just linguists, but anybody doing laboratory science --
another step
frequently, and collapse into larger categories observations that we
are different
(that we measure as different), based on explicit or implicit
This gives us a more symbolic representation of the observations. And
us to what a linguist might call the narrowest phonetic representation.
Some linguists,
both contemporary but also from earlier periods in the century, would
begin to
feel uncomfortable at this point, and say that as far as speakers are
there is no justification for assuming that this level of representation

to anything in the head; most linguists would shrug and say, oh probably
and we can't get anywhere without making that assumption.

This question is not one of levels, at least not of linguistic levels;
part of the
meta-reflections we share with other sciences. Linguistic levels are
theory-internal, and in all cases that I"m aware of, deal with
symbolic representations.

A lot of phoneticians are uncomfortable with any symbolic system, any
system at all, which leaves us rather high and dry when trying to work
out a
way to interface with phonologists' theories. At the other end of the
is SPE's theory of phonetics, designed specifically to meet the needs
of phonologists, not phoneticians.

Questions of allophony are by their very nature questions of phonology,
phoneticians can say nothing about them until there is at least some
of working agreement about how phonetics and phonology interact, and
divide up responsibilities. On the other hand, terms like coarticulation

are fundamentally phonetic in nature, and have no particular status in
phonology (except as convenient labels).

No-one, I think, is attracted by the notion that phonetics is universal
phonology is language-particular -- all recognize that there is
univeral and particular in both. I myself think that the notion of
phonetically-motivated, or articulatorily-motivated, is unpalatable
and unappealing, and often vacuous, but I'm probably in a minority
on that.

This just touches the surface, but maybe this is helpful.
Best, John Goldsmith
Dear Mr Salzmann!
I saw your posting about allophones at the Linguist List. If my reply
would be of any help to you, I'll be very glad.
The problem is seen from the point of view of the phonological school
which taught me -
St-Petersburg school of phonology.

Allophone does not belong to parole, but to langue. Phoneme and
allophone are both abstractions, they are correlative as general and
particular. There cannot be an infinite number of allophones. The
number of allophones for each phoneme of a certain language is
limited. The number of phones is infinite.
There are certainly three levels: phone - allophone - phoneme. I
think that phoneme belongs to phonology, and allophone - to phonetics.

I don't see how coarticulation and complementary distribution can be
opposed, they are too different phenomena. But we can say that
coarticulation causes a type of compl. distribution, forming
combinative allophones (there are two types of allophones: caused by
position and by combination - i.e. coarticulation).
And both phenomena that you show (key/car and back/front vowel-
palatal/velar consonant) can be called complementary distributed
allophones, caused by coarticulation.

- -
Lena Pigrova (
Dear Martin,

I appreciated your discussion of the problems of the notion of
allophony. I addressed some of these problems in my thesis,
Pronunciation modeling in speech synthesis. Traditionally, people have
ascribed symbols to certain allophones in certain languages. For
example, it is common to use a flap symbol in "phonetic" (as opposed to
phonological) transcriptions of American English. However, it becomes
clear that symbols are most appropriate for phonological analyses, and
their usefulness for allophony is questionable. In chapter 4, I
describe an experiment comparing gradient and discrete aspects of
postlexical variation. You can download my thesis from

Corey Miller
Nuance Communications

>But what about complementarily distributed allophones? Take for
>instance the English voiceless plosives which depending on the
>environment are realized either as aspirated or non aspirated. These
>called allophones but they are again an abstraction, each of the
>allophones, e.g. [p +asp] and [p -asp] can be realized in an infinite
>number of different ways. Now what should be called an allophone? If
>term were restricted to the actual phonetic realization (as in
>structuralism), we'd have to find a new term for the "abstract
>So do we end up with tree levels instead of two? Or to put the question

>differently: Which phenomena belong to phonology, which to phonetics?
>Since complementary distribution is an abstract regular pattern not
>solely due to physiological necessity, i'd have to belong to the domain

>of phonology in my opinion. But is there a possibility to express this
>in classical structural phonology or is the theory just simply flawed
>is my argumentation faulty?

I confess I don't see where the problem lies here. Allophones are,
speaking, the realizations of phonemes. As such, and as you underlined
for /a:/, their number, for one given phoneme, is virtually infinite,
in the case of 'complementary distribution'. One might speak about
classes' of allophones (e.g. +asp. and -asp. plosives in English) ; I
guess, however, that this would be an artefact, partly due to the use of

IPA symbols, and that phonetic data reveal a gradient, rather than
categories, within allophonic variation.

>I believe, the picture would be different in generative phonology. If
>I'm not completely mistaken, the three-way contrast indicated above is
>in fact represented by the classical derivational model: There are the
>URs (roughly comparable with the phonemes) and the surface
>representations. As far as I can tell, these surface representations
>not phonetic entities, but still feature bundles (different from the
>only in being fully specified and possibly having undergone some
>This is how the are in the phonological component (probably
>corresponding to what I called "abstract allophones" above). What in
>classical structuralist sense is called a phone, conceived as a
>entity, would then be the result of the interaction of the phonological

>component with the sensory system.

In fact, the three-way contrast established by classical generative
phonology is not similar to the one you suggest, for two reasons.
the URs are morpho-phonemes, not phonemes in the structuralist sense.
example, the /k/ of electriC and the /s/ of electriCity are distinct
phonemes but two 'reflexes' of the same UR in SPE-based frameworks.
Secondly, there is not a specific level in generative phonology
corresponding to the phonemic level of earlier theories. Rather,
phonology could be said to posit a n-way contrast according to the
(n) of rules that apply to a given UR.

>There is a third problem for which I've been unable to find a
>solution. Quite often, a distinction is made between complementary
>distribution and Coarticulation, the distinguishing factor being
>physiological (in)evitability. In a major textbook like Spencer
>Phonology (1996), the different pronunciations of /k/ in <key> and
><car>, the first palatal, the second velar, are considered an instance
>of coarticulation because of a physiological inevitability. Now this
>inevitability is more properly called a physiological inevitability of
>the native speakers of English since the two sounds can be contrastive
>in a number of languages, for instance Turkish. But if there is no
>universal physiological necessity to pronounce the sounds this way,
>couldn't one regard the two realizations as complementarily distributed

>allophones or surface representations?
>In German phonology, when speaking of complementary distribution, the
>example always adduced is the distribution of the voiceless velar and
>palatal fricatives; There occurrence depends on the value for the
>feature [back] of the preceding vowel (i.e. velar after a back vowel,
>palatal after a front vowel). For a native speaker of standard high
>German, this distribution is about as physiologically inevitable as the

>key/car distribution for a native speaker of English, but no one has
>ever spoken of coarticulation in this case. What might be the reasons?
>Am I just simply wrong? Represent these examples truly different

Well, I always told my students that 'complementary distribution' and
'coarticulation' (generally) represent the same phenomena but a
point of view. On pure phonological grounds, you speak about
distribution' ; this is a mere statement of distributional facts. Now,
you want to 'explain' such distributions on phonetic grounds, the notion
'coarticulation' is generally unavoidable (as is the case for <key> and
<car> as well as for the palatal allophone of German /k/). Note that
(too simple) distinction between 'phonetics' and 'phonology' is only
within a linear framework. In autosegmental models, for example,
coarticulation is directly represented insofar as the palatality of the
in <key> can be viewed as belonging to the phoneme /i/, which will be
linked to two slots. In this theory, 'complementary distribution'
explicitly appears as a segmental *effect* of coarticulation.

Best regards.

 Joaquim Brandao de Carvalho 
 Departement de linguistique 
 Faculte des Sciences Humaines et Sociales - Sorbonne 
 Universite Rene Descartes - Paris V 

Hello Martin,

your query raises a real interesting point; namely, that of the
level(s) of
abstraction of our linguistic (and generally "scientific") description.
IPA, I guess, tried to distinguish between langue and parole with
respect to
the phonological system, and perhaps were not aware that their
are as much of an abstraction as are their phonemes, albeit on a
level/nature. I remember discussions back when I was a grad student at
(i.e., among others with Peter Ladefoged and Vicki Fromkin) about this
question of "abstraction" in the IPA alphabet. (Note the label!) They

were also called "phonetic symbols"; IOW, at some level, at least, we
quite aware of our "abstraction". This may be another case of a method
(technique) of analysis (data handling), once accepted for convenience
acquires the status of "fact", and is no longer questioned by the
peoople in
the discipline.

With all the work that's been done on categorical perception, one can
understand the IPA's abstractions and the need for them.

Lieber Martin Salzmann,

zu Ihrer Linguist-List-Anfrage kurz folgendes:

ad 1) Das von Ihnen dargestellte Problem scheint mir in der Tat immer
dann zu
bestehen, wenn man die Begriffe 'Phonem' und 'Allophon' vor dem
Hintergrund des
Saussureschen Begriffspaars 'Langue' - 'Parole' zu beschreiben versucht
insofern ist dieses Problem durchaus real und zeigt eine
Unzul�nglichkeit der
'Langue-Parole'-Dichotomie. In der deutschsprachigen Romanistik hat ein
Vorschlag von E. Coseriu viel Beachtung gefunden, der die Saussure'sche
Dichotomie zu einer Trias 'System - (usuelle) Norm - Rede (parole)'
hat (etwa E. Coseriu, Sprachkompetenz, T�bingen: Francke 1988). Ihre
'abstrakten Allophone' lie�en sich in diesem Rahmen der Ebene der Norm,
konkreten Realisierungen der Ebene der Rede (Parole) zuordnen. Wenn in
einf�hrenden linguistischen Darstellungen Allophone einfach der Ebene
'Parole' zugeordnet werden, begreift man Allophone offensichtlich
als 'Types' und l��t ihre Realisierung als 'tokens' au�er acht - ob dies
durch eine zuvor gegebene Definition des Begriffs der 'Parole' gedeckt
ist oder
nicht. Im Bereich der Morphologie stellen sich im �brigen ganz �hnliche
dar�ber hinaus noch weitere Probleme, wenn Morpheme der Langue und
der Parole zugewiesen werden. Ich habe damit verbundene Probleme einmal
in einem
einf�hrenden Kurs thematisiert, was bei den Studierenden zu ernormen
gef�hrt hat. Ich er�rterte seitdem die Begriffe der strukturalistischen
Phonologie und Morphologie in der Regel ohne expliziten Bezug auf das
Begriffspaar 'Langue - Parole'.

ad 2): Mit Ihrer Einsch�tzung der generativen Phonologie stimme ich
�berein. Das Problem stellt sich hier nicht.

ad 3): Diese Problem scheint mir nicht ganz einfach zu l�sen. Ich wei�
genau, was Spencer (1996) zu engl. /k/ in <key> und <car> schreibt (ich
habe das
Buch leider nicht bei mir zu Hause, sondern in meinem B�ro in der Uni
und kann
deshalb dort auch nicht noch einmal nachschlagen). Letztendlich handelt
es sich
sowohl bei dem engl. Beispiel wie bei dem von Ihnen angef�hrten klass.
Lehrbuch-Beispiel um ein Assimilationsph�nomen. Die Frage, die sich
stellt (und
die m�glicherweise durchaus kontrovers zu beantworten ist), ist, ob
Ph�nomene den gleichen Stellenwert habe. Die Realisierung des dt.
Phonems /x/
als velarer oder als palataler Frikativ ist auf jeden Fall Bestandteil
sprachspezifischer phonologischer Regeln des Standardhochdeutschen. Ich
hier in keinem Falle von physiologischer Notwendigkeit oder
sprechen. Der engl. Fall - entsprechendes gilt mutatis mutandis auch f�r
Deutsche, die romanischen Sprachen und sicherlich zahlreiche weitere
Sprachen -
ist m�glicherweise generellerer Natur, so da� man hier schon eher von
physiologischer Koartikulation sprechen und dementsprechend das Ph�nomen
'low level phonetic rules' (und nicht mehr phonologische Regeln)
k�nnte - �hnlich etwa der (akustisch nachweisbaren) phonetischen
Nasalierung von
Vokalen im Kontext von Nasalkonsonanten (wie in dt. <Mann> oder engl.
<man>), im
Unterschied zu allophonischer Nasalierung, wie z.B. im Portugiesischen
unter abstrakten Analysen von NV auch im Franzs�ischen). Es gibt
jedoch Arbeiten, in denen die unterschiedlichen Realisierungen von /k/
vorderen und hinteren Vokalen als (komplement�r distribuierte) Allophone

aufgefa�t und damit wohl der phonologischen Komponente der Grammatik
werden. In der Tat k�nnen solche Ausgangslagen zu Phonologisierungen und

Restrukturierungen lexikalischer Repr�sentationen f�hren: cf. etwa lat.
/k/ vor
vorderen Vokalen, das im Italienischen, Spanischen usw. zu einer
Affrikate wird, w�hrend lat. /k/ vor hinteren Vokalen als velarer
erhalten bleibt. Weitere Entwicklungen haben dann dazu gef�hrt, da� /k/
/tsch/ (Pardon, keine phonetische Zeichen in diesem Editor!)
Minimalpaare bilden
(cf. it. <chi> /ki/ 'wer' und <ci> /tschi/ 'dort'). Frage also: Sind
unterschiedliche Realisierungen von /k/ in den o.g. englischen
Beispielen oder
in dt. <Kirche> und <Kugel> in der heutigen Synchronie das Resultat
phonologischer oder phonetischer Regeln? Spielt der
artikulatorische/auditive/akustische Abstand zwischen den Realisierungen
f�r die
 Entscheidung der Frage eine Rolle (Ist /k/ vor vorderem Vokal noch
velar oder
bereits ein palataler Okklusiv, i.e. [c]?). Ist die Frage �berhaupt
zu beantworten? Gibt es Entscheidungskriterien?

Andreas Gather

Dr. Andreas Gather
Ruhr-Universit�t Bochum
Romanisches Seminar
GB 8/133
Universit�tsstr. 150
D-44780 Bochum
Email: ODER
You have raised a number of interesting questions in your posting, and I

thought I'd give you my thoughts on the matter.

First, I'm not sure all structuralists would have accepted your claim
phonemes are part of langue and allophones part of parole. Of course,
there was considerable variation among structuralists--first of all,
between American and European, second, in Europe between Britain and the

continent (i.e. between say, Jones and Trubetzkoy) and in the US between

the Sapirians and the Bloofieldians. For Sapir allophones were probably

part of parole, while phonemes were mental percepts (similar to what
Baudouin thought), but for Bloomfield the langue/parole distinction was
meaningless and phonemes were just classifications of sounds into
boxes. Current generative theory (at least some of it) would
between phonological rules and phonetic implementation rules, although
normally allow for language-specific instances of the latter as well.

The German case you bring up is interesting because it is claimed, at
for Standard German that the alternation is actually not automatic after

all--certain morpheme boundaries block it. A typical example is the
contrast `tauchen' : `Pfauchen', which supposedly has a velar in the
former and a palatal in the latter. I could dig up some references to
issue if you're interested.

There's much more that could be said about these questions--Steven
deals with some of it in his history of phonological theory, and others,

myself included, have other opinions. I'd be glad to discuss it further
you're interested.

Geoff Nathan

Geoffrey S. Nathan
Southern Illinois University at Carbondale
Carbondale, IL, 62901-4517
Phone: (618) 453-3421 (Office)
 (618) 549-0106 (Home)


A topic which I often hear mentioned in the recent Laboratory Phonology
(or at conferences) is whether there is any difference between
non-contrastive phonetic differences which are traditionally considered
allophonic alternations and non-contrastive phonetic differences which
usually not mentioned at all or relegated to "phonetic implementation."
This seems to be the same topic you are bringing up.

First, as you point out, a distinction of variation which is
physiologically necessary vs. not doesn't seem very useful. Kingston
Diehl (Language, 1994) argue that much obviously phonetic variation is
language specific, and therefore learned rather than physiologically
caused. (They also cite various references on this.)

I'm not sure classic generative phonology deals with this problem any
better than structuralism does. It seems obvious that structuralism
also require some sort of "phonetic implementation" (whatever that is),
although we know now that that can't be universal. But in generative
phonology, how do we decide what's worthy of being part of the phonology

and what's (language specific) phonetic implementation? As you point
physiological necessity isn't a good criterion. Whether the difference
can be distinctive in some other language probably isn't a good
either, since a wide variety of differences are distinctive in _some_
language, and this has nothing to do with the system of the language
analyzed. (For example, English /u/ is similar to a front rounded vowel

in the environment /tut/ because the alveolars raise the F2. /u/ vs.
is distinctive in many languages. But if this means that English /u/
an allophone [y], then the vowel /u/ must have quite a variety of
allophones, conditioned by place of the preceding and following
consonant.) Whether the variation is language specific or not doesn't
seem to be a good criterion either.

In the end, there are examples of clearly phonetic variation in
which may be language specific. However, these same types of variation
may be phonologized in a given language, so one often finds similar
processes which look more or less phonological. Some current work in OT

seems to put all such variation (and perhaps all variation of any sort)
into the grammar. I think it's an open question as to whether a
distinction between "real" allophonic variation and "lower-level
variation is useful, and how such a distinction should be included in
phonological theory.

Natasha Warner


Natasha Warner Ph.: 31-(0)24-3521372
Max Planck Institute for Psycholinguistics FAX: 31-(0)24-3521213
PB 310 Email:
NL-6500 AH Nijmegen
the Netherlands
Dear Martin,

Allophone does seem to have been an ill-defined notion.

A Greek native speaker once told me he had never realized that palatal
chi and velar chi were different sounds until he had read it in a Greek
language course for foreigners.
I am tempted to conclude that realizations of the same phoneme are
labeled as different allophones when they sound different to a linguist
who's not a native speaker. That's an exaggerated conclusion, but guess

there's some truth in it.
By the way, the Modern Greek chi does not only have a front and a back
realization (before front and back vowel respectively), but also
intermediate realizations (before various consonants - there's a paper
by Mirambel on this). I think the existence of such intermediaries can
be a criterion on where we have one or two phonemes. However, there
might be the contrary example of the velar and palatal fricatives in
Standard High German, which can be argued to be one and the same phoneme

(though the word "Frauchen" makes the issue controversial).

In a search for a definition, one should return to the writings of those

who first introduced the notion. It doesn't seem to be Trubetzkoy. The
notion is used by Martinet but I don't know if he's the first. I confess

I don't know if it is used by e.g. Jakobson or Chomsky.
Insofar as I remember, Martinet did not provide a definition but
proceeded with examples. With such an approach, it is possible to pick
up just two realizations of one phoneme (among the infinity of
realizations) and to say: these two sounds are not different phonemes,
they are allophones of one another (= realizations of the same phoneme).

Of course the examples (languages, phonemes and phones) are so selected
as to be cases of sizeable phonetic differences. This is perhaps a
human-science approach, as opposed to the exact-science approach which
wants proper definitions as you have thems in mathematics. So to speak,
only when the phonetic difference is real big does one bother to take
the word "allophone" from its shelf.
Insofar as I remember (again), the word "allophones" does not refer to
all differences that are possible in "parole", but only to such phonetic

variations that are determined by the phonological context (i.e.
ignoring svariations due to regional or social background, speech
situation, and chance).

Theoretical definition: allophone is a synonym of realization (more
exactly: two phones are allophones of one another = they are
realizations of the same phoneme). That might not be a very useful
notion, i.e. the word realization and phone are enough to do the job.
Practical definition, used for language teaching: only those allophonic
differences are considered that are useful for a good pronunciation of
the language. That will be an ad-hoc choice determined by the native
language of the pupils as much as by the target language itself. Thus,
that would not be an objective (subject-independent) notion, i.e. not a
scientific one.
These two suggestions are not real definitions but only pointers.

Another brand of linguistic research favours subjects such as "The
meaning of soma in Aristoteles" or "The use of virtus by Vergil"
(fictitious examples). Perhaps the search of a definition of
"allophones" would be more like "The use and phraseology of "allophone"
by Martinet and his disciples" than like "What is an allophone?"...

Remy Viredaz, Geneva

Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue