LINGUIST List 10.1785

Tue Nov 23 1999

Disc: What exactly are allophones?

Editor for this issue: Karen Milligan <>


  1. Richard D. Janda, Phonemes, Allophon(e)-emes, Allo-allophones; Minimal Pairs

Message 1: Phonemes, Allophon(e)-emes, Allo-allophones; Minimal Pairs

Date: Tue, 23 Nov 1999 20:23:24 -0500 (EST)
From: Richard D. Janda <>
Subject: Phonemes, Allophon(e)-emes, Allo-allophones; Minimal Pairs

 [Continuing in the same format as the earlier posting....] 

Dear Martin Salzmann (re Linguist List, Vol. 10-1778 (Nov. 23, 1999), 

 Given the already large number of responses to your query, any further pro-
posed answers are likely to get buried in the continuing avalanche, but your
extremely cogent question shows the degree to which fundamental questions of 
phonology remain, frankly, rather muddled (I'll add another such issue at the
end of this message). 

 Joining the chorus of previous similar voices, I would add that I myself have
always emphasized to students that there are at least three kinds of entities:
phonemes, allophones as TYPES of variant phoneme realizations, and the infi- 
nitely many TOKENS instantiating these types (and thus also the relevant pho-
neme). For the allophone-type concept, I have sometimes suggested the term 
"allophoneme" -- i.e., allophon(e)-eme -- although this leaves the allophone-
token concept with the implied stutter-like, haplology-rife label "allo-allo-
phone". ( One of my students this year suggested "smallophone" for this mi-
crolevel notion.) The problem with the term suggested by some responders -- 
phone -- for the infinitely variable allophones is that, at least in American
Structuralist parlance, a phone is a physical speech sound that has not yet 
been classified. (This allows the effective comparison that phones are to al- 
lophones as geology is to geography, since the former member of each pair in- 
volves entities that are considered apart from their assignment to human cul- 
tural entities). For reasons mentioned below, it seems misleading to consid- 
er all allophones-as-types solely as elements of parole; certainly the usual 
sort of definition given for linguistic competence would include a speaker's 
knowledge of what the allophone types for his/her idiolect are. 

 Chomsky's 1964 "Current Issues in Linguistic Theory" (a revision of his 
1962 "The Logical Basis of Linguistic Theory") implies the three-way dis- 
tinction that you (and I and many others) make, since, besides a (systemat- 
ic) [ i.e., morpho-] phonemic level, it recognizes both systematic phonetics 
and physical phonetics. For an earlier, Post-Bloomfieldian American Struc- 
turalist view of this complex issues, see especially Bloch's 1948 "A Set of 
Postulates for Phonemic Analysis" (in _Lg._; reprinted, with a later work 
adding revisions, in V. Makkai , ed., 1972 = _Phonological Theory: Evolu- 
tion and Current Practice_). A work which introduces (I believe) a very use- 
ful term is Gleason's 1965 _Linguistics and English Grammar_, which em- 
ploys _co-allophones_ for 'allophones of the same phoneme' (for 'allophones 
of different phonemes', I have sometimes used "anti-allophones", for want 
of a better term). Since every phone of a language is an allophone (of some- 
thing, of some phoneme), use of "co-allophone(s)" allows students to avoid 
the practice of saying "These two sounds are allophones" -- which really 
should get the response: "Well of course they are, since every phone is an 
allophone of something".... 

 As for the issue of coarticulation vs. complementary distribution, the add- 
ed notion of physiological inevitability strikes me as a red herring, if not a 
red blue-whale. There are plenty of traditional allophones (allophones as tra- 
ditionally analyzed) which are in complementary distribution but do not re- 
sult from coarticulation: e.g., the aspirated realization of English /p/ which 
occurs initially in stressed syllables, as opposed to the unreleased word-final 
variant which tends to close, say, _yep_ and _nope_. Thus, without arguing 
that this is an ironclad characterization, I would suggest that complementary 
distribution, assimilation, & coarticulation are a set of increasingly specif- 
ic terms which seem to involve proper inclusion, although they may turn out 
not to be quite so neatly related. English /h/ and angma are notorious PHO- 
NEMES occurring in complementary distribution but obviously not syn- 
chronically conditioned by assmilation. And distant assimilations (aside 
from their non-linear reduction to adjacent assimilations) do not involve co- 
articulation, at least not of a traditional sort. Furthermore, the tremendous 
amount of cross-linguistic variation that exists across realizations of "the 
same sequence of sounds" shows that physiological inevitability is far from 
being involved in all such cases. E.g., Russian phonetic palatalization can 
surely be said to involve coarticulation, but myriad other languages have 
nothing like the same degree of palatalization -- or coarticulation. Here, 
crucial reference must be made to the work of Ohala on sound patterns as 
EXAGGERATIONS ("hypo- and hyper-corrections") of what originally 
were purely phonetically-conditioned phenomena. 

 Combining Ohala's emphasis on exaggeration with Labov's work on 
"generational change" as a mechanism by which sound change can take on a 
directionality over decades and so come to involve extreme divergence from 
its phonetic origins, I believe we are forced to the conclusion that complemen- 
tary distribution has been overvalued as a sufficient condition for co-allophony
of sounds. I.e., no one has ever established that the phonetic dissimilarity 
generally assumed to exist to a high degree between English /h/ and angma 
does not also exist between, say, Old High German /u/ and u-umlaut. [I'd 
be happy to send you a copy of a recent paper of mine on phonologization 
as having for its central, crucial step a reanalysis whereby exaggeration of an 
allophonic effect beyond the original phonetic conditioning of some environ- 
ment leads to the phonologization of that effect, regardless of whether the 
(former) allophone in question is still in complementary distribution with 
other (former) co-allophones. The paper, given at the International Congress 
of Phonetic Sciences in San Francisco last summer (1999), is called "Ac- 
counts of Phonemic Split Have Been Greatly Exaggerated -- But Not 

 Finally, to "minimal pair(s) ": one often encounters definitions of this no-
tion something like 'two words with different meanings which show only 
one difference in sound(s) -- whereby this difference must involve a corre- 
sponding locus in the two words, which must therefore have the same num- 
ber of sounds [with diphthongs allowed to count, for this purpose, as one 
sound]) '. But even this rather lengthy typical definition won't work, since 
the majority of minimal pairs that are usually cited (e.g., English _beat_ vs. 
_bead_, _rub_ vs. _run_, _goose_ vs. _geese_, etc.) involve more than one 
difference in sound, due to the fact that the relevant phonemic difference it- 
self correlates with different allophonic effects (above, vowel length vs. non-
length, vowel nasality vs. non-nasality, consonant frontness vs. backness). 
One can of course try to build into the definition of "minimal pair" some ref- 
erence to 'words which show only one difference in sound(s) which is not 
itself conditioned by some other difference in sound(s)', but this verges on 
circularity. So even the notion of "minimal pair", which phonology students 
tend to hold onto as to a lifeline, is not an unproblematic concept. 

 You deserve everyone's thanks, then, Martin Salzmann, for providing an 
opportunity for phonologists to return to these fundamental but as yet unfin- 
ished building blocks of our field! 

 Richard Janda (Department of Linguistics, The Ohio State University) 
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue