English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 1

CHAPTER 2

ABSTRACT ELEMENTS
IN PHONOLOGY
Ádám Nádasdy

Contents of this Chapter:

2.1. Abstraction in analysing data
2.2. The phoneme and its allophones
2.2.1. The standard solution
2.2.2. Alternative solutions
2.3. Natural Classes
2.4. Neutralization
2.4.1. The taxonomic handling of neutralization
2.4.2. The generative handling of neutralization
2.5. Lexical representation: the problem of Vowel Shift
2.5.1. The taxonomic handling of Vowel Shift
2.5.2. The generative handling of Vowel Shift
2.6. Representations and rules
——————————————————————

2.1. Abstraction in analysing data

Any scientific description involves an amount of abstractness. When the chemist classifies
water, snow, ice and steam as “water”, he ignores their temperature and state, and considers
only one feature to be relevant: the chemical composition H2O. This is the distinctive feature
serving as the label of the category, while their being liquid or solid or falling in flakes are
considered redundant features, being of secondary importance.
Abstractness in phonology means that we suppose the existence of segments that are
more or less different from the surface (i.e. from phonetic facts).1

2.2. The phoneme and its allophones

Let us take the example of English /l/ to discuss phonological abstraction. The two kinds of L-
sound, Clear-L and Dark-L are both articulated as lateral sonorant consonants, but

1
At all levels of language we find this duality of concrete vs. abstract: physically realizable, con-
crete data (allophones, morphs, word forms) versus their abstract equivalents (phonemes, mor-
phemes, lexemes). Compare:
Concrete ↔ Abstract
sound, allophone ↔ phoneme
morph, allomorph ↔ morpheme
word form, syntactic word ↔ lexeme
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 2

 Clear-L, phonetically [l] (leg), is pronounced with the tongue body in neutral position;
 Dark-L, phonetically [ɫ] (belt), is velarized: the back of the tongue is raised towards
the velum (the [o] position).2
These are the phonetic facts. But what does the phonologist say? How do we analyse this state
of affairs?

2.2.1. The standard solution

The distribution of Clear-L and Dark-L is complementary. They are allophones, the realiza-
tions of an abstract category, a phoneme, which we represent as /l/. Any phoneme is an ab-
straction compared to the actual sounds. You cannot pronounce a phoneme:3 you can only
pronounce an allophone, either Clear-L or Dark-L. The phoneme is, perhaps, more than a
sound; it is the sum of its allophones: /l/ = [l] + [ɫ]. The actual sounds (= allophones) are de-
rived from the phoneme by means of rules. The standard solution is an “L-Darkening Rule”:

(1) L-Darkening Rule

/l/–→ [ɫ]  __ (#)C  = the phoneme /l/ is pronounced “dark” (= velarized)
║  when followed by a consonant in the same or in the next
word, or by a pause.
[l] / elsewhere = the phoneme /l/ is “clear” (= not velarized) elsewhere.

2.2.2. Alternative solutions

(a) Let us point out that we could have chosen Dark-L to be the basic alternant of this pho-
neme, saying that underlyingly all English L’s are dark, and they become clear by an “L-
Clearing Rule” when followed by a vowel.

(2) L-Clearing Rule

/ɫ/–→ [l]  __(#)V = the phoneme /ɫ/ is pronounced “clear” (= unvelarized)
when followed by a vowel in the same or the next word.
[ɫ] / elsewhere = the phoneme /ɫ/ remains “dark” (=velarized) elsewhere.

Rules (1) and (2) are equally logical and scientifically correct. It is arbitrary which allophone
we choose to represent the whole phoneme, i.e. in which direction we abstract away from the
phonetic data. It is for practical reasons (simpler typography!) that we normally choose Clear-
L to be the basic alternant, and speak of L-Darkening rather than L-Clearing.

2
In some positions English /l/ had become so “dark” (= so much like an [o]) that it was historically
absorbed in the previous a or o, and disappeared from pronunciation, remaining only in spelling as
a silent letter: talk, walk; folk, yolk and half, calves, balm, Holmes, Stockholm. Note that the conso-
nant following this “absorbed” silent L is always noncoronal.
3
We often get round this problem by pronouncing the name of the alphabetic letter which the pho-
neme is normally spelt with, so we say “the L phoneme”, or “the Double-U phoneme” for /w/, the
“voiced TH phoneme” for /ð/; or we use traditional names like “yod” for /j/ or “schwa” for /ə/.
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 3

(b) It would be equally logical to use a third symbol for the phoneme, one which is neither
clear nor dark, for example capital /L/, and say that the abstract (and therefore unpronounce-
able!) phoneme /L/ is realized in two ways, either as Clear-L or as Dark-L, according to its
position. This is probably closer to psychological reality: when learning English (either as
mother tongue or later), people do not lexicalize L as clear or dark, but store a general L in
their mental lexicon,4 and pronounce it as clear or dark as appropriate. Of course certain L’s
will always turn up as clear (before a vowel inside the word, leg), others always as dark (be-
fore a consonant, belt), while final L’s will alternate (tell me – tell it), but this does not
weaken the argument. This “general L” is less specified than its allophones, being unspecified
for clearness or darkness, having only the characteristics “lateral sonorant consonant”.
This leads us to turn round our previous statment (2.2.1) that a phoneme is more than a
sound; actually, a phoneme is less than a sound, because it has fewer features – just like, say,
a mammal is less than a horse, or H2O is less than snow, ice, etc. The higher the category, the
less specified and more abstract it is.

(c) A different solution is offered by Generative Phonology. This approach says that any
variant can be taken as the underlying one if, by applying the right rules in the right order, we
can derive the required surface pronunciation. The rules will then “map” the underlying form
of words onto their surface form, that is, they convert phonemes into speech sounds.
The derivation in (3) shows L-darkening (as well as other rules: Aspiration and Un-
stressed Vowel Reduction, to make the picture more complete).
(If a sound is affected, the changed sound is written under its starting form. The rules
appear between the two horizontal lines. n.a. = not applicable.)

(3) Generative derivation: an example

Spelling: leg belt tell us tell me
Underlying Represent. (=UR) leg belt ˈtelʌs ˈtelmi
L-Darkening n.a. ɫ n.a. ɫ
Aspiration n.a. n.a. th th
Unstr’d Vowel Reduction n.a. n.a. ə n.a.
Surface Representation (=SR) [leg] [beɫt] [ˈtheləs] [ˈtheɫmi]

As we see, the L-Darkening Rule applies to belt and tell me (but not to leg or tell us, where
the /l/ is before a vowel). The Aspiration Rule applies to the /t/ of tell (but not belt, where it is
final). The Unstressed Vowel Reduction Rule applies to us (but not to me, because final /i/ is
never reduced to /ə/).

4
Interestingly, this is exactly what English spelling does: it uses the same letter L for both variants.
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 4

2.3. Natural classes

Consider the distribution of English [h] (the glottal fricative) and [ŋ] (the velar nasal).5

(4) Before vowels Before consonants Word-finally

__V __C __#
[h] ham, historical, behave NEVER NEVER
6
[ŋ] NEVER bank, anger, Hungarian sing, belong, tongue

If we follow the principle of complementary distribution, we’ll have to say that [h] and [ŋ] are
allophones of one phoneme! This is counter-intuitive and ought to be avoided. We may point
out that [h] and [ŋ] are not similar enough to be allophones of one phoneme – but this might
sound vague and impressionistic, for what is “similar”? To put it more precisely, we say that
they do not form a natural class: they do not share any feature which would not be shared by
other sounds. Their only common feature is that they are both consonants; but that is not
unique to them, as there are many other consonants. The allophones of a phoneme must ex-
clusively share at least one distinctive feature (or feature-combination) not shared by other
sounds. In the case of [l] and [ɫ] this feature was [+lateral], since no other sounds share this
manner of articulation. The sounds [h] and [ŋ] do not satisfy this “exclusive similarity” re-
quirement, so they must be analysed as two distinct phonemes even though they are in com-
plementary distribution.

2.4. Neutralization
There are cases where two phonemes behave like allophones. Consider English /s/ and /ʃ/.
These are phonemes, producing minimal pairs: so – show, mass – mash, parcel – partial,
(uni)versal – (contro)versial, etc. But there is one position in which their appearance is pre-
dictable: at the beginning of a word when followed by a consonant (the “initial-preconsonan-
tal” position):

(5) Distribution of /s/ and /ʃ/

Position [s] [ʃ]

initial-preconsonantal (not r) #__Cnot r stub NEVER

initial-preconsontl. before r #__r NEVER shrub

The symbolization “#__Cnot r” means “before any consonant except /r/”: in this position only
/s/ can occur (stub), because there are no English words beginning with /#ʃt-/, /#ʃm-/, etc. On

5
Remember that orthographic h is silent before a consonant or word-finally, so words like John,
Sarah, shah, Noah do not have [h] (BEP 4.41).
6
There are words in which [ŋ] is before a vowel, but these are all made up of free stem plus suffix, so
they do not belong here, e.g. sing#er [ˈsɪŋə], slang#y [ˈslæŋi]. A real irregularity is hangar /ˈhæŋə/.
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 5

the other hand, when the second consonant is /r/, only /ʃ/ can occur (shrub), because there are
no English words beginning with /#sr-/. This is complementary distribution. In this position
/s/ and /ʃ/ do not contrast: their opposition is neutralized.

Here are some further examples of neutralization.

 /n/ and /ŋ/ are phonemes (sin ↔ sing), but are neutralized before velars (/k, g/), where
only /ŋ/ can appear (bank, anger).
 /s/ and /z/ are phonemes (seal ↔ zeal, rice ↔ rise), but are neutralized word-finally
after an obstruent consonant, with which they have to agree in voicing (backs /s/ but
bags /z/).
 /ʊ/ and /u:/ are phonemes (look ↔ Luke, pull ↔ pool), but are neutralized word-
finally, where only /u:/ can appear (bamboo, menu, continue). 7

Neutralization means that two phonemes suspend their contrast in a particular position. In this
neutralizing environment the two sounds behave like allophones.
The /s/ – /ʃ/ problem is different from the /h/ – /ŋ/ problem because [s] and [ʃ] are really
similar sounds, forming a natural class: they (and only they) are the voiceless alveolar frica-
tives of English. How can we answer the problem of their behaviour?

2.4.1. The taxonomic handling of neutralization

In introductory courses and practical dictionaries the taxonomic (or “phoneme-inventory”)
approach is followed (BEP 2.18). This is based on minimal pairs: if two sounds contrast in at
least one minimal pair in the language, they are declared to be phonemes of that language, and
are represented as phonemes in all other words. This is the “once a phoneme, always a pho-
neme” principle. Because [s] and [ʃ] do contrast elsewhere (so–show), they are different pho-
nemes (separate members of the phoneme inventory of English, the “taxonomy”); and they
are analysed as such even in those neutralizing environments where they do not contrast. A [ʃ]
sound always represents an underlying /ʃ/, so shrub is transcribed as /ʃrʌb/ (as if a contrasting
*/srʌb/ was possible).
This representation is logical, but it pretends that the /s/ ↔ /ʃ/ contrast is valid every-
where, even though in certain places it remains “unexploited” by the language. Such a solu-
tion is surface-oriented: it says that shrub is pronounced with /ʃ/ simply because it is repre-
sented in the lexicon with /ʃ/. This gives the wrong impression that it is an accidental gap in
the lexicon of English that no words begin with /sr-/ and no words begin with /ʃ/+other con-
sonant. The taxonomic approach then, can be criticized because it misses a generalization.8

7
Similar neutralizations from Hungarian:
 n and ny are phonemes (kén ‘sulphur’ ↔ kény ‘whim’), but are neutralized before ty, gy, where only
ny can be pronounced: hangya [hanygya] ‘ant’, pinty [pinyty] ‘finch’.
 o and ó are phonemes (kor ‘age’ ↔ kór ‘illness’), but are neutralized word-finally, where only ó can
appear: olló ‘scissors’.
8
Hung. ’szem elől téveszt egy általánosítást’.
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 6

2.4.2. The generative handling of neutralization

We can handle the stub/shrub problem in a generative framework by assuming /s/ to be the
underlying phoneme in all word-initial consonant clusters, so shrub would be UR //srʌb//.
(We shall include in double slants //...// those UR forms that are different from the usual tran-
scription.) Now we have to introduce a rule turning this UR /s/ into [ʃ] before /r/. This is the
S-Cluster Adjustment Rule. It is not different from allophonic rules (e.g. L-Darkening) be-
cause it has no exceptions.

(6) S-Cluster Adjustment Rule

/s/ → [ʃ] / #__r shrub //srʌb// → /ʃrʌb/

A sample derivation involving this rule appears below:

(7) Generative derivation of words beginning with /s/ or /ʃ/ + Consonant
Spelling: so show stub shrub
U. R. soʊ ʃoʊ stʌb srʌb
S-Cluster Adjustment n.a. n.a. n.a. ʃ
S. R. [soʊ] [ʃoʊ] [stʌb] [ʃrʌb]

This mapping correctly expresses that in stub/shrub the /s/ ~ /ʃ/ choice is predictable (= rule-
governed), as it is in allophonic rules. The generative approach recognizes that contrast can be
locally determined, i.e. it is valid in some places but not in others.

2.5. Lexical representation: the problem of Vowel Shift

Lexical representation means the string of phonemes with which a word (or morpheme) ap-
pear in the lexicon, the mental “dictionary” of speakers. For example, coach is represented
lexically as 3 phonemes, /k–oʊ–ʧ/. Allophonic detail is not included, namely that this /k/ is
aspirated, this /oʊ/ is clipped, and this /ʧ/ is preglottalized, because all this is predictable.

2.5.1. The taxonomic handling of Vowel Shift

Many stems are pronounced in two forms, with a tense vowel and with a lax one: grave ~
grav-(ity). This is called Vowel Shift9. What could be their lexical representation? The /eɪ/ ~
/æ/ alternation cannot be regarded as allophonic alternation, since – as shown in (8b) – the
same vowels are elsewhere opposed, producing minimal pairs.

9
See BEP 8.17-19 for details.
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 7

(8) Vowel pairs in Vowel Shift and in minimal pairs

Vowel pairs (a) in Vowel Shift (b) in minimal pairs

/eɪ/ ~ /æ/ grave – grav-ity cape  cap
/i:/ ~ /e/ metre – metr-ic feel  fell
/aɪ/ ~ /ɪ/ type – typ-ic-al mile  mill
/oʊ/ ~ /ɒ/ sole – sol-itude road  rod

Once a phoneme, always a phoneme, says the taxonomic school: in that approach Vowel-
Shift must be regarded as alternation between phonemes, that is, allomorphic alternation,
which assumes that these stems simply have two different lexical representations (just like
foot–feet or teach–taught), from which the speaker chooses the appropriate form. But while
foot and teach are really irregular, the vowel-shifting stems like grave behave quite predicta-
bly. So once again we see that the taxonomic approach is logical and disciplined, but it misses
a generalization by treating grave–gravity the same way as foot–feet or teach–taught.

2.5.2. The generative handling of Vowel Shift

The way Vowel Shift is presented at a practical level (so in BEP) is actually a generative so-
lution, suggesting that in grave~grav- the vowel is some abstract element which is realized as
/æ/ in a laxing environment (grav+ity, grav+itate), but as /eɪ/ elsewhere (grave, grav#est,
grave#ly, grave#ness). Let us symbolize this abstract vowel as //A//.

(9) Abstract Vowel Realization Rule in Vowel-Shift

//A// –→ /æ/ / in a laxing environment (Trisyllabic Laxing, Laxing Suffix, etc.)
/eɪ/ / elsewhere

There are four such abstract vowels, symbolized with letters of the alphabet (very much as it
happens in English spelling!), in capitalized form. The absract vowels each have two “daugh-
ter” phonemes. They are listed in (10), with their properties. You will note that the properties
are very vague if they are to embrace both “daughters” of the abstract phoneme.

(10) lexically phonemically properties of the abstract vowel

=UR = SR
//grAv// /greɪv/ ~ /græv-/ //A// = front, nonhigh/high?
//mEtr// /mi:tər/ ~ /metr-/ //E// = front, nonlow?
//tIp// /taɪp/ ~ /tɪp-/ //I// = nonback, nonmid?
//sOl// /soʊl/ ~ /sɒl-/ //O// = back, nonhigh?

In this approach cape, favourite, basic have lexical /eɪ/, and cap, parody, matter have lexical
/æ/, because they do not undergo Vowel Shift; their vowel is stable (= non-alternating). Stems
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 8

with alternating vowels, like grave, nation, vain have lexical //A//, so //grAv//, //nAʃən//,
//vAn//.10

(11) Nonabstract and abstract vowels – an example

spelling UR SR
cape, favourite, basic /eɪ/ /eɪ/ – nonalternating stems
cap, parody, matter /æ/ /æ/ – nonalternating stems
grave, nation, vain } //A// /eɪ/ ~ /æ/ – alternating stems
gravity, national, vanity }

This solution has a weakness: the abstract vowels //A, E, I, O// listed above do not have
well-definable features that would distinguish their “daughter phonemes” on the basis of ex-
clusive similarity (see 2.3 above). For example, the abstract element //A// in the lexical repre-
sentation of grave, nation, vain, etc., has the daughter phonemes /eɪ/ and /æ/, of which [æ] is a
low vowel, while the diphthong [eɪ] starts with mid [e] and ends with high [ɪ]. Unfortunately
these properties cross over with the daughters of abstract //E//, namely /e/ and i:/ (the lexical
representation of metre–metric, severe–severity, etc.). The diagram below shows that the
vowel-height of the abstract vowels //A// and //E// cannot be specified exclusively, since their
realizations cross over:

(12) /i:/ metre — high vowel

//E// /eɪ/ grave — mid-high vowel
/e/ metric — mid vowel
//A// /æ/ gravity — low vowel

We must conclude that the abstract “vowels” in (10) are too vague and too cross-positioned to
be proper phonological segments. They are useful as morpho-phonological abbreviations ex-
pressing the working of Vowel Shift in that stem.
Another solution would be to arbitrarily pick one “daughter” to be the underlying seg-
ment (say, /eɪ/ for grave) and turn it into the other when necessary (so gravity would be UR
//greɪvəti// → SR /ˈgrævəti/); but the problem with this would be that there are cases like ba-
sic /ˈbeɪsɪk/ (and not */ˈbæsɪk/), obesity /oʊˈbi:səti/ (and not */oʊˈbesəti/), where Vowel Shift
fails to take place without apparent reason. These are called “lexical exceptions” because
their sound shape does not explain their not undergoing a rule. They show that the rule is not
a real phonological one, because a pronological rule (like L-darkening) never has any excep-
tions.
All in all, Vowel Shift has to be regarded as an 80 per cent reliable morpho-phono-
logical tendency. True, the spelling usually has the same vowel-letter for the alternants, so

10
An analogous Hungarian example is provided by stable vowels such as gyár ~ gyárak ‘factory/ies’
(lexical /a:/), nyak ~ nyakak ‘neck/s’ (lexical /ɔ/), as opposed to stem-internal shortening vowels
such as nyár ~ nyarak ‘summer/s’ (lexically an abstract vowel, perhaps /A/).
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 9

the letters a, e, i, o act practically like abstract underlying segments – but the existence of
lexical exceptions undermines the phonological status of Vowel Shift.
Study this table carefully:

(13) Various types of alternation

Example Type of Do they Do they Is the choice UR

alternation ever have predictable by
contrast? exceptions? rule?
[l]~[ɫ] allophonic NO NO yes, everywhere /l/
leg~felt
[s]~[ʃ] phonological yes NO yes, in a particu- /s/~/ʃ/
stub~shrub lar position
[eɪ]~[æ] morpho- yes yes 80 % certainty, //A// ?
grave~gravity phonological but has lexical /eɪ/~/æ/
exceptions
[ʊ]~[i:] morphological yes all these are no /ʊ/~/i:/
foot~feet exceptions

2.6. Representations and rules

The phonology of a language is made up of representations and rules. By representations we
mean the phoneme strings of the words and morphemes in the lexicon; by rules we mean the
transformations that turn these underlying representations into surface pronunciation. This
chapter was about making representations more abstract in order to deprive them of unneces-
sary (because predictable) detail. The predictable detail is furnished (“fleshed out”) by the
rules, as shown in (17).
 Column (a) gives the actual pronunciation in a fairly “narrow” phonetic script.
 Column (b) gives the traditional taxonomic analysis (= Gimsonian transcription). Here
only allophonic (= subphonemic, non-neutralizing) rules are needed.
 Column (c) gives a more abstract lexical representation, which has to be mapped onto the
surface by various phonological rules, including the allophonic rules under (b). The unusual-
looking transcriptions enclosed in double slants //...// in column (c) are not incorrect, they are
just more abstract than the taxonomic-Gimsonian transcriptions in column (b). The derivation
goes from (c) through (b) to (a).
If we use surface-close representations (as in the Gimson system), we need few rules; if
we use more abstract (“impoverished”) representations, we need many rules. The advantage
of the more abstract representation is having a single underlying form for alternants of a given
morpheme. Observe for example, that atom and atom-(ic) have a single underlier in Column
(c) even though they are pronounced differently.
English Phonological Analysis – Chapter 2 – ABSTRACTNESS (Autumn 2013) 10

(14) Phonetic, taxonomical, and abstract-lexical representations

(a) Phonetic (b) Taxonomical (c) Abstract

Surface-close represent. Lexical representation
Actual pro- (explicit, rich, phonetic) (implicit, poor, phonological)
nunciation only alloph. rules needed phonological rules needed
bank [bæŋk] /bæŋk/ //bænk//
– vowel nasalized bef. nasal – Nasal Place Assimilation /n/ → [ŋ]
bang [bæŋ] /bæŋ/ //bæng//
– vowel nasalized bef. nasal – Nasal Place Assimilation /n/ → [ŋ]
– Postnasal-Final G-Drop /g/ → 
kissed [khɪst] /kɪst/ //kɪs+d//
– aspiration – Voice Assimilation /d/ → [t]
each [iʔʧ] /i:ʧ/ //i:ʧ//
– pre-voiceless clipping
– preglottalization
beer [bɪ:] /bɪə/ //bi:r//
–smoothing of broken diph. – Pre-R Breaking /i:/ → [ɪə]
– R-Dropping /r/ → 
atom [ˈætəm] /ætəm/ //ætɒm//
– stress assignment
– Unstressed Vowel Reduction /ɒ/ → [ə]
atomic /əˈtɒmɪk/ //ætɒmɪk//
[əˈthɒmɪk] – aspiration – Stress fixed by suffix -ic (atómic)
– Unstressed Vowel Reduction /æ/ → [ə]
gravity /ˈgrævəti/ //greɪvɪtɪ//
[ˈgrævəti] – stress assignment
– Trisyllabic Laxing /eɪ/ → [æ]
– High Vowel Tensing /-ɪ/ → [-i]
– Unstressed Vowel Reduction /ɪ/ → [ə]

We have shown that an abstract representation, which needs complex rules to map it onto the
surface, is more economical and often more insightful than the “rich” (i.e. surface-close) rep-
resentations used in introductory textbooks and practical dictionaries.

END OF CHAPTER 2