THE PHONEME CONCEPT

TWO LEVELS OF REPRESENTATIONS

1
What motivates the postulation of the
phoneme concept
• Could the phonology be organized in levels of abstraction,
with classes and realizations of its own?
• What motivates the postulation of the phoneme concept?
• We will be concerned here with what is often called the
classical phonemic answer to this question an answer that in
one form or another dominated phonological thinking in the
first half of this century and is still with us.

2
Data from a dialect of British English
Data from a dialect of British English, transcribed
in considerable detail.
• [k+ g+] are ‘advanced’ velar stops
• [?p| ?t| ?k|] are pre-glottalized and
unreleased
• [v)] is a nasalized vowel
• [ɫ] is velarized or ‘dark’
3
dialect of British English (cont’d)
(a) (b)
1. k+iɫ ‘keel’ 1. kɫ ‘cool’
2. k+eiɫ ‘kale’ 2. kɫ ‘cull’
3. k+I?k| ‘kick’ 3. kUd ‘could’
4. k+E)n?t| ‘Kent’ 4. k+?t| ‘caught’
(c) (d)
1. k+i ‘ski’ 1. k?p| ‘scoop’
2. k+eI?t| ‘skate’ 2. k ‘skull’
3. k+If ‘skiff 3. k)n ‘scorn’
4. k+Q)m?p| ‘scamp’ 4. k?t| ‘Scot’ 4
dialect of British English (cont’d)
(e) (f)
1. g+i ‘geese’ 1. gɫ ‘ghoul’
2. g+ei?t| ‘gate’ 2. gɫ ‘gull’
3. g+E)n?t| ‘Ghent’ 3. g?k| ‘gawk’
4. g+Q?p| ‘gap’ 4. gUd ‘good’

(g)
1. i?k| ‘leek’
2. Q?k| ‘lack’
3. U?k| ‘look’ 5
Distribution of Velars
• So English has four voiceless velar stops [k+ k+ k]. And if we look at
the voiced velars in (e-f) we find two kinds again: so we add [g g+] to our
inventory. And if we add velars in final position, we find another one:
glottalized unreleased [?k|]. But do we really want to say that English has
seven velar stops?
• As a matter of simple observational fact we do ‘have’ all these velars.
But leaving it at this misses an interesting generalization. Given the two
sets [k+ k+ k k ?k| ] and [g g+], the members of each are distributed so that the
occurrence of one or another is totally PREDICTABLE. Thus all velars are
advanced before front vowels and retracted before back; voiceless velars
are aspirated before stressed vowels unless [s] precedes and glottalized
6
and unreleased only finally.
• But this doesn’t hold for one group AS A WHOLE VS. Substituting the [k+] of keel for
the [k] of cool may give us a rather odd version of cool: but one that’s still cool
and no other word. But if we substitute [g] for [k] we get ghoul.
• So: the difference between front and back velars in English is not lexically
information-bearing (though it may give indexical information, i.e. not about
word-meaning but about speaker-characteristics, like sex, effeminacy, foreignnes ,
etc.).
• Whereas that between voiceless and voiced velars is: the presence or absence of
voice tells us that one word rather than another has been said but frontness and
aspiration are redundant. They are conditioned automatically by properties of
the speech chain. Such variants are not normally under the speaker’s control: he
‘choose’ either voice or voicelessness, but not frontness or aspiration: one might
7
say that he system does the choosing for him.
Phonemes & Allophones
• So we have two kinds of entities:
• 1. belonging to the language as a signalling system or code,
• 2. predictable exponents of the former units,

• We could say that in English there is some ‘abstract structural

element’ corresponding to the range of [k]-types, and another
to the [g]-types.
• The same is true of non-velarized [l] and velarized [ɫ]: the
former only occurs syllable-initially, the latter only syllable-
finally. Or of nasalized vs. non-nasalized vowels, etc. 8
Phonemes & Allophones
• In other words, there are two types of ‘sound’ inventory: the very
large set of actually occurring noises, and the smaller set of funda-
mental units they realize. These units are phonemes and the
realizations of a phoneme are its allophones. (An actual sound, with
no reference to its class membership, is a phone.)
• If we now represent (allo)-phones in square brackets, phonemes in
solidi, we can illustrate the levels of representation as follows:

Phoneme allophones
/k/ [k+] [k+] [k] [k] [?k|] 9
Criteria for phonemic status
• How do we determine which sounds in a language represent
which of its phonemes. How do we go about setting up
phoneme systems, and organizing the phonetic material we
encounter?
• To begin with, any pair of phonemes, like /k/ and /g/, are
contrastive: their function is to separate entities, Two words
that are phonemically different are different words.
• We must now examine the relation between distribution and
contrastiveness, to see how the distribution of phones can, in
a fairly precise way, be used to establish contrast, and thus
determine the phoneme inventory of a language.
10
complementary distribution
• The point of complementary distribution is that two sounds with such a
distribution could never form a minimal pair. For example, the [k] and [k|] of
can and scan cannot form part of minimal pairs because unaspirated [k|] must be
preceded by [s] while aspirated [k] can never be preceded by [s] .Thus, [k] and
[k|] will, by definition, never have the opportunity to be used contrastively.
• Complementary distribution is important in trying to figure out which sounds
correspond to independent phonemes and which are just allophones of phonemes.
If two sounds are in complementary distribution and therefore cannot form
minimal pairs they are unlikely to be independent phonemes. It is more likely
they are conditioned variants of one and the same phoneme. This is the
conclusion we would draw from our example with aspirated and plain plosives.

11
complementary distribution
• In the dialect exemplified above, voiceless stops (not only /k/) are aspirated
before stressed vowels unless /s/ precedes. Thus [k] does not occur before a
stressed vowel unless /s/ precedes, and contrariwise [k] doesn’t occur if /s/ does
precede.
• The two phones occur in mutually exclusive environment: they are in
complementary distribution. In general, two phones in complementary
distribution are allophones of the same phoneme, i.e., if they can never appear in
the same environment, they obviously can’t contrast; the prime situation for
contrast is parallel distribution, e.g. the initials of cat, tat, pat.
• By the above criterion, [k] and [k] must represent one phoneme: therefore cot
[k?t|] and Scot [sk?t|] can be phonemically represented as respectively /kt/
and /skt/ (the rule for [?t|] is obviously the same as that for [?k|]).
12
Minimal pairs & contrastiveness
• We take contrast as a fundamental notion: two phones are contrastive (represent
different phonemes) if substitution of one for the other produces a ‘different
word/morpheme’ in the language, or, to stretch it a bit, if substitution produces a
word that isn’t in the language for either accidental or structural reasons.
• Substitution (or commutation) gives us a useful tool for establishing
contrastiveness: two phones are in contrast if they are commutable in the same
‘frame’. Thus we can establish the phonemes /p b t d k g/ for English by
commutation in the frame ‘before [t]’: pot # bot # tot # dot # cot # got, etc.
• This is minimal contrast (the forms contrast in one segment only, and a pair
like got : pot is a minimal pair.
• In classical phonemics, the isolation of minimal pairs is the basic method of
establishing a phoneme inventory and in practice especially in field-work with an
13
unknown language, this is a prerequisite for almost anything else).
• Of course these techniques aren’t always available: not all
potential phonemes contrast in every context. There are
extreme cases of defective distribution (as with [h] and [N] in
English), which are not commutable in any frame.
• Nonetheless they can be shown to be phonemic in their own
macro-distribution. By contrast with other (independently
established) phonemes: one could use cat:pat:hat,
ham:had:hang, etc.
• Further, ONE minimal contrast with another independent
phoneme may be enough. Thus [Z] in English has an extremely
limited distribution, and minimal pairs are difficult to find. But
if we establish /S/ as phonemic vis-à-vis /s/ (ship : sip), then
azure: Asher establishes /Z/, And even if there are no minimal
pairs, ‘near-minimal’ ones can make the case for contrast 14
 • So a phoneme may be defined — roughly — as ‘a class of phonetically
similar phones in complementary distribution’. This is a procedural or
operational definition: phonemes ‘arise as the product of operations on
data.
• The English data above can be contrasted with that found in a language
such as Hindi, illustrated below (from Ladefoged 1993: 145):

a. pal ‘take care of’

pal ‘edge of knife’
b. tan ‘mode of singing’
tan ‘roll of cloth’
c. tal ‘postpone’
tal ‘place for buying wood’
d. kan ‘ear’
kan ‘mime’ 15
Two different levels of analysis: phonetic and
phonemic
• So far I have spoken as though some sounds are phonemes and
others sounds are conditioned variants, or allophones. So far this has
been a harmless oversimplification. However, it is a very misleading
way of thinking of the notion of the phoneme (though it is
unfortunately the way that many beginners picture this notion).
• The problem is that it is not the case that some sounds are used
contrastively and others are not. Instead we must distinguish two
different notions of ‘speech sound’, one a concrete, phonetically
based notion (the actual sounds of a language as they might be heard
and transcribed in a careful phonetic record), and another more
abstract notion,
16
2 levels of analysis: phonetic and
phonemic
• This means that we should distinguish two different levels of analysis. At the
first, more concrete, level (which we could call, though with caution, the
‘phonetic’ or better, the ‘allophonic’ level) we have the inventory of speech
sounds used in a language. We can call these sounds the phones of the language.
We will continue to write phones between square brackets. These will include in
English [k+] [k+] [k] [k] [?k|]. The phones are all those sounds which a
language makes use of.
• At the next, more abstract, level we distinguish a set of sound types. These are
the phonemes and they do not necessarily correspond in a direct way to any
particular physical (phonetic) sound, Rather, they are the units which serve to
build up individual words and hence which distinguish one word from another
Unfortunately, we are obliged to use the same letter symbols to write down 17
phones as we use for phonemes.
• By removing allophonic minutiae from the transcription, we produce
simplified representations, useful for any kind of additional analysis (except
of course detailed phonetics, which is what the phoneme concept is designed
to get us away from).

• On the simplest level a phonetic transcription is a string of segmental

representations, each standing for the occurrence of a structural category; and
it can, given the relevant allophonic rules be mapped onto a phonetic
transcription.

Phonetic [k Q) n ?t|]

18
Phonemic /k Q n t/
•It might be asked at this point why we choose to
represent the phonemes with the symbols /p t k/ rather
than, say, /p t k/, or, indeed, entirely different symbols
with no relation to phonetic transcription, such as /P T K/.
• First, phonologists tend to assume that the basic form of
a phoneme is identical to its most widespread allophone.
In addition, it is generally assumed that the basic form of
a phoneme will be phonetically somehow ‘simpler’ than
19
other forms.
THE MOTIVATIONS FOR TWO LEVELS OF REPRESENTATIONS

• One argument is economy. Why supply allophonic information in the

lexical entries if it can be stated in a set of allophonic rules that are valid for
all morphemes of the lexicon?
• A second argument for the assumption of two levels of representation is that with a
single level it would not be possible to express the phonological relatedness of
morpheme alternants. We take the English plural suffix as an example. Suppose that
instead of saying that there is a morpheme [z] which in different phonological contexts
is adjusted in order to obey the phonological constraints of English, we were to say that
three allomorphs are listed in the lexicon, [z], [s] and [Iz], each of which is used in a
specific phonological environment: [Iz] after sibilants, [s] after (other) voiceless
segments and [Iz] in other cases. This description of the regular plural formation would
be correct, in that for every noun we can predict the plural form. What the description
fails to express, however, is that, somehow, the three alternants [z], [s] and [Iz] are the
same morpheme. 20
 French Spanish
i. iE) ’nothing’ Roxo ’red’
ii. t) ’rat’ Re ’royal’
iii. mi ’to die’ foro ’lining’
iv. mi ’husband’ foRo ’forum’
v. pE) ’prince’ pRDo ’meadow’
vi. tE) ’train’ t5Rt5R ’to treat’
vii. kie ’to shout’ kReTER ’to grow’
viii. fEne ’to brake’ frt5 ’fruits’
ix. i ’brie’ RBo ’brave’
x. do ’funny’ d5Ro ’drug’
xi. g) ’large’ gRit5 ’cries’
xii. ye ’to shout’ pER ’pearl’
xiii.  ’purse’ mwERt5o ’dead’
xiv. fe ’to do’ d5R ’to give’
xv. me ’mother’ TER ’to make’

21
Study the distribution of [   in French. And [R r] in Spanish
Farsi
i. rh ’road’
ii. zgir ’towel’
iii. ziR ’because’
iv. rn ’paint’
v. fri ’Persian’
vi. rz ’day’
vii. omr ’life’
viii.iRini ’pastry’
ix. rg ’leaf’
x. iRn ’pale’ 22

Study the distribution of [r, R r]

French
• One interesting phonetic observation we can make about the French data is that, in addition
to the voiced uvular fricative, there is a voiceless uvular fricative [] and a voiced uvular
approximant []. Our speaker therefore has to master three kinds of uvular sound in
acquiring French. There is something more to learn about the French sounds. Let us consider
where they occur within a word.
• The voiceless [] occurs only after voiceless sounds (as in 5-8)
• The approximant [] occurs only before consonants and at the end of words (as in 12-15)
• The voiced [] occurs at the beginning of words (1-2), between vowels (3-4) and after voiced
consonants (9-11). It is clear where these sounds occur is not an arbitrary matter, but is
systematic.
• In the language of phonology, we express this systematic relationship between these sounds
by saying that they occur in different environments, and we describe the set of
environments in which a sound occurs as its distribution. Notice that the distribution of the
three types of uvular sound is mutually exclusive: where we find one type, we never find
either of the others. This kind of mutually exclusive distribution is called complementary 23
distribution.
Spanish
• Compare this situation with the distribution of the voiced alveolar tap
[R] and the voiced alveolar trill [r] in Spanish. There we find that the
tap may occur at the beginnings of and ends of words, between
vowels and before and after consonants. The trill may also occur
between vowels, and the distribution of the two sounds is therefore
overlapping: there is at least one environment in which they may both
occur. We refer to this pattern of distribution as parallel distribution.
(We mean by this, not that the two sounds necessarily have identical
distribution, but that their distribution overlap, partially or wholly).
The distinction between them is phonemic (contrastive) 24
•
Farsi
• Examine the data from Farsi. It too contains voiced alveolar
trills (transribed as [r]) and voiced alveolar taps [R], as in
Spanish. We saw that the distinction between the two is
phonemic, rather than allophonic. Is this the case for Farsi?
• Once you have the answer, consider the status of [r] in Farsi.
What is its distribution with respect to the other two sounds:
parallel or complementary? Once you have the answer, you
will be in a position to say whether these three sounds are
realizations of only one, or of more than one phoneme in farsi.
25
Farsi
Farsi

[R] V _____V (between two vowels)

• /r/  [r] ______ # (word-finally)

[r] elsewhere

26
 Daga
asi « prendre »
urase « travail »
nesip « danser »
sino « j’ouvre »
simura « sel »
use « jeune »
anet « regarde »
wagat « trou »
utu « effacer »
tapem « loup »
tabe « je prends »
tuiam « tuer »
watap « pas » 27
Study the distribution of [t] and [s]
Daga
Contexts where [s] Contexts where [t]
occurs occurs

asi a ----- i
urase a ----- e
nesip e ----- i
sino # ----- i
simura # ----- i
use u ----- e
anet e ----- #
wagat a ----- #
utu u ----- u
tapem # ----- a
tabe # ----- a
28
tuiam # ----- u
watap a ----- a
Daga

[s] before front vowels /i, e/

• /t/  [t] elsewhere

29