McMahon, April (2000) Lexical Phonology and the history of English.
Cambridge Studies in Linguistics 91. Cambridge: Cambridge University
Press. 309pp. ISBN: 0-521-47280-6
Reviewed by David Deterding
Note: because of problems in representing phonetic symbols, I will use the
[I] for the vowel in 'pin' [E] for the vowel in 'pen'
[e] for the vowel in 'face' [ae] for the vowel in 'pat'
This book considers the insights that historical changes in English can
offer for the formulation of Lexical Phonology (henceforth LP),
concentrating particularly on three areas: historical vowel shifts; Scottish
vowel length; and rhoticity. The goal is to develop a strictly-constrained
model of LP that can represent the phonology of all dialects of English
without resorting to the excessively abstract nature of some previous
In Chapter 1, McMahon introduces her approach to LP, in particular
justifying the role that external evidence such as historical change and
dialect variation can play in helping to determine the phonological
structure of English. She argues that the highly abstract, psychologically
unreal, underlying representations of previous formulations of LP arose as a
direct consequence of their over-reliance on internal evidence and rejection
of external evidence. However, while she criticises these versions of LP,
particularly because they have been unable to shake off the influence of the
Sound Pattern of English (henceforth SPE), her goal is not to abandon LP
entirely but instead to develop a new, less abstract model of rule-based
derivational phonology, as she is even more critical of constraint-based
models such as Government Phonology and Optimality Theory.
In Chapter 2, after reviewing the origins of LP, McMahon considers in some
detail its best-known formulations, particularly that of Halle and Mohanan
(1985). She is critical of this model for proposing Strict Cyclicity and
Structure Preservation as constraints on lexical phonological rules but then
finding ways to allow these constraints to be overridden (p.53), and she
further concludes that there is no good reason for the lexical level of the
phonology to have four separate strata, with a loop allowing derivations to
cycle back from Stratum 3 to Stratum 2, and with a non-cyclic stratum
(Stratum 2) being interspersed with the other three cyclic strata. Instead
she proposes (p.84) a simpler phonological model with (in addition to the
postlexical level) just two levels of lexical rules, Level 1 being cyclic
and strictly subject to Structure Preservation and the Derived Environment
Condition, and Level 2 being non-cyclic and not always strictly subject to
Structure Preservation or the Derived Environment Condition. However, not
everything in her model is necessarily simpler, as her rules need to make
reference to bracketing configurations to distinguish affixed words from
compounds, something which Halle and Mohanan's four-stratum model had aimed
to make redundant.
Chapter 3 investigates how alternations that arose out of historical vowel
shifts, such as that between the vowels in the first syllables of
'sane'/'sanity' and the second syllables of 'courage'/'courageous', should
be handled in a synchronic phonology of English. In considering the modern
reflexes of these vowel shifts, particular attention is given to the
consequences of strict adherence to the Derived Environment Condition.
One example that illustrates McMahon's approach concerns the alternation
between the vowels in the second syllable of 'profound'/'profundity'.
Although this alternation arose out of the Great Vowel Shift in the same way
as 'sane'/'sanity', McMahon argues that it has no psychological reality for
speakers today (p.98), so it should not be present in the synchronic
phonology of English. This demonstrates how historical evidence can inform
us, but does not necessarily have the final say in determining how modern
English should be represented.
McMahon is particularly critical of the abstract underlying vowels that tend
to occur in SPE-influenced versions of LP. In contrast, in her model of LP,
as Level 1 lexical rules such as the Vowel Shift Rule (VSR) are strictly
subject to the Derived Environment Condition, only words that have undergone
a process such as affixation can be subject to VSR, and this means that the
underlying representation of morphologically simple (underived) words must
be surface-true (apart, maybe, from the effects of vowel reduction, to be
discussed below). However, there is a cost to this: as both 'courage' and
'sane' are underived forms, neither can undergo VSR, which means that the
[e] in the second syllable of 'courageous' and the [ae] in the first
syllable of 'sanity' must be derived by two separate versions of VSR, one to
deal with the underlying lax vowel /ae/ in 'courage' and create [e] in
'courageous', and the other to deal with the tense vowel /e/ in 'sane' to
create [ae] in 'sanity'. This is unfortunate, as these two alternations
both originate from the same historical shift and are basically the same, so
it is quite inelegant to represent them as two separate rules. McMahon
acknowledges this problem (p.91), but asserts that the benefit of strict
adherence to the Derived Environment Condition for Level 1 rules outweighs
the disadvantage of this duplication of VSR.
McMahon considers the sequence /ju/ in 'cute', in some detail, and she
proposes that this is a rising diphthong (onglide) in modern English (in
contrast to falling diphthongs such as /ai/ and /au/). One issue here is
that, in her analysis, the [j] in 'tabular' is inserted by rule, as this
word is derived from 'table', but the /j/ in 'cute' is part of an underlying
/ju/ diphthong, as this is an underived word. This means that, if we
compare RP with General American (GenAm), we would have to say that the two
dialects have different underlying vowels in the underived 'duke' but a
different application of rules in the morphologically complex 'studious'
where RP has an inserted [j] but GenAm does not, even though the contrast
between RP and GenAm 'duke' and 'studious' is basically the same: GenAm does
not have a [j] after coronal consonants.
One rule that deals with consonants but is discussed in Chapter 3 because it
interacts with VSR is Velar Softening, the rule that derives the medial [s]
in 'criticise' from the final /k/ in 'critic'. It is argued (p.128) that
Velar Softening must occur before VSR on Level 1 of the lexical rules,
because the conditioning environment for the rule must make reference to
vowels before they have undergone VSR. However, if it is a Level 1 rule, it
cannot apply to morphologically simple words, which means that it cannot
apply to 'reduce', even though historically the final /s/ in 'reduce' is
related to the [k] in 'reduction' in exactly the same way that the [s] in
'criticise' is related to the final /k/ in 'critic'.
Finally in Chapter 3, McMahon considers irregular verb alternations such as
that between 'fight'/'fought', and she proposes that all they are simply
held in the lexicon and not derived by rule, with the exception that the
alternation between /i/ and [E] found in words such as 'keep'/'kept',
'sleep'/'slept' and 'feel'/'felt' and also that between /ai/ and [I] in
'bite'/'bit' and 'hide'/'hid' are derived by rule. An issue that arises out
of this analysis (p.134) is that regular inflection belongs on Level 2 of
the lexical rules, but the affixation of the /t/ suffix necessary to feed
VSR for the derivation of the above forms must occur before VSR so it must
be on Level 1. This means that essentially the same rule (past tense
suffixation) applies on two separate levels.
In Chapter 4, the nature of Scottish vowel length is analysed in
considerable detail. The Scottish Vowel Length Rule (SVLR) states that some
vowels become lengthened before a voiced fricative, a final /r/, or a
morpheme boundary. The existence of a morpheme boundary as one of the
conditioning environments results in pairs such as 'need' [nid] and 'kneed'
[ni:d], and also 'tide' [taid] and 'tied' [ta:id], which differ in length
because, in the second word of each pair, the final [d] is a past tense
McMahon considers the connection between SVLR and Low Level Lengthening
(LLL), the rule that lengthens vowels before voiced consonants or at the end
of a word in all varieties of English, and she concludes that the two are
separate for three reasons: first, not all vowels are subject to SVLR but
they are all subject to LLL; second, not all voiced consonants condition
SVLR, as voiced plosives, nasals, and /l/ are not in the conditioning
environment; and finally, for SVLR there is the 'tide'/'tied' distinction,
which makes reference to morpheme boundaries, while LLL never makes
reference to word-internal structure. However, she also concludes that the
two processes were historically linked, but that the extra lengthening in
the SVLR environments resulted in this process becoming sufficiently
perceptually salient for SVLR to become promoted into a lexical rule, while
the lesser lengthening in the LLL environments remained inaudible to most
speakers with the consequence that LLL remains a postlexical rule (p.201).
Initially, McMahon assumes that SVLR takes place on Level 2, because it
interacts with regular past-tense inflection which is a Level 2 lexical rule
(p.193). However, after considering further evidence, she concludes that
SVLR may also apply on Level 1 (p.195).
Two further issues remain with Scottish vowel length. First, SVLR applies
(for some speakers) in words like 'spider' and 'pylon' (p.193) which do not
appear to meet its conditioning environment, unless it is assumed that there
is a false morpheme boundary after the /ai/ in each word. Second, as SVLR
introduces a non-phonemic distinction between vowels such as [i] and [i:],
it is not Structure Preserving, which is one of the defining characteristics
of lexical rules, and even if we assume that Structure Preservation only
strictly applies to Level 1 of the lexical rules, this does not resolve the
issue, as SVLR also applies on Level 1. One solution to these two issues is
that there may be an incipient phonemic contrast between long and short /ai/
that is slowly undergoing diffusion through the lexicon, for if this is
true, then the introduction of length distinctions by SVLR no longer
violates Structure Preservation (p.203).
In Chapter 5, McMahon contends that different dialects of the same language
can differ quite substantially, not just in the sounds in individual words
and the inventory of phonemes, but also in the underlying phonetic features,
and she suggests that while tenseness is an underlying distinctive feature
for GenAm vowels, length may be the corresponding feature for RP vowels.
McMahon also considers underspecification theories, which she claims are
built on the flawed premise that computation is always favoured over storage
so that every last bit of regularity must be teased out and represented in
terms of rules (p.215). She argues that this kind of theory leads to
unreasonably abstract underlying pandialectal representations, and she
further claims that, as underspecification can allow fundamental constraints
such as Structure Preservation to be easily circumvented, it is an unwelcome
component of any properly constrained model of LP.
Finally Chapter 6 investigates linking [r] and intrusive [r] in many
varieties of English such as RP, and it is concluded that these two are
basically the same, being instances of the rule of [r]?Insertion, where an
[r] is inserted after a schwa or [+long][-high] vowel and before any other
vowel. This rule must take place on all three levels (p.280): Level 1 of
the lexical rules, as it occurs before Level 1 suffixes in 'doctoral' and
'dangerous'; Level 2, as it occurs before regular Level 2 inflectional
suffixes in 'colouring' and 'soaring' (and also the intrusive 'saw[r]ing');
and the postlexical level, as it operates across word boundaries as in 'fear
of flying' (and the intrusive 'data[r]analysis').
McMahon proposes that this represents an instance of rule inversion, whereby
the historical rule of /r/?Deletion has now become a modern rule of
[r]-Insertion. Once again, we can see how the historical record can inform
us, but does not always determine the final shape of the synchronic
Anyone who does not have a substantial background in phonology and is not
reasonably familiar with LP would probably find much of this material hard
to understand. For example, applications of the Alternation Condition, the
Elsewhere Condition, and the Strict Cyclicity Condition are discussed in
quick succession on pages 15 to 16, but none of these conditions is
explained till later: the Elsewhere Condition on page 42, the Strict
Cyclicity Condition on page 45, and the Alternation Condition on page 85.
In other instances where prior knowledge is assumed: 'children' is given as
an exception to the blocking effect which prevents the accumulation of
semantically equivalent affixes on a single word (p.41), but this example
might be opaque to anybody who did not already know that 'childer' was once
the plural of 'child'; the CiV Tensing rule is listed in the derivation of
words like 'comedian' (p.90) with no explanation; and there is frequent
criticism of Duke of York derivations, where certain rules undo the effects
of previous rules, but the colourful and rather apt imagery behind this name
would be lost for anyone not already familiar with the nursery rhyme about
the duke marching to the top of a hill and then marching down again. It is,
perhaps, a pity that this allusion is not explained.
However, this book is not intended as an introductory text on phonology.
Instead it aims to be a serious contribution to the debate on phonological
structure, and it achieves this admirably. Moreover, despite the frequent
assumptions about background knowledge, the text is actually very clearly
written with obvious close attention to detail and only very occasional
flaws in the text. (On page 115, 'sulphuric' is shown as undergoing CiV
Lengthening, which is clearly an error, but such oversights are rare.)
Anybody who is familiar with the background in phonology will find this book
an exceptionally coherent and valuable contribution to debates on various
controversies in the field.
Many of the claims made by McMahon will be controversial, not least the
central claim that diachronic evidence can contribute invaluable data for
determining the structure of phonology. In the tradition of SPE, of course,
it was assumed that the goal of phonology is to model the knowledge held by
an individual speaker, and as most speakers are not aware of the historical
changes that took place in their language, such data must be irrelevant.
McMahon, however, makes a compelling case that historical change originated
from somewhere, and if the phonological representation cannot model such
change, then there is something wrong with that representation. It is
somewhat ironic that, in explicitly allowing evidence from historical
processes to help determine the structure of phonology, McMahon's model of
English phonology is actually far less influenced by the past than SPE-based
models which often seem to incorporate every twist and turn in the history
of the language into their synchronic rules for the derivation of words.
A parallel claim is that evidence on dialectal variation can contribute to
the structure of phonology, though this is perhaps a bit less controversial.
Nowadays the work of Labov and others has shown that the concept of an ideal
speaker with a single monolithic pronunciation is not just improbable but
actually misguided, as we all have a variety of accents which we use at
different times. Consequently, it is nowadays widely accepted that, even if
the goal of phonology is to model the knowledge of an individual, this
should include the variation exhibited by the individual.
With respect to dialect variation, McMahon considers the claim that all
dialects of the same language have the same underlying phonological
representation, and she argues convincingly that this viewpoint is simply
not tenable, for this kind of common underlying representation would have to
be highly abstract, which would compel us to adopt an implausibly complex
derivation for the surface form. Moreover, she argues that the boundary
between different languages and dialects of the same language is too fuzzy
to support the contention that the former have different underlying
representations while the latter have a common underlying representation.
Different dialects often evolve over time into different languages, so at
what point does a common underlying representation suddenly become a
Although McMahon claims to be developing the theory of LP, or more
specifically allowing it to develop away from the insidious legacy
bequeathed by SPE, in reality her criticisms of LP run rather deep. One of
the key insights of LP is that there are two different kinds of phonological
rules: those in the lexical level which are idiosyncratic and subject to
exceptions; and those in the postlexical level which are more phonetically
motivated and exceptionless (Carr, 1993: 179). McMahon does maintain this
basic distinction, and she further retains another fundamental aspect of LP,
the separate levels of the lexical rules, particularly to accommodate
affixes such as ?ity and ?ic in Level 1, and -ness, ?hood and regular
inflections in Level 2. However, her re-introduction of bracketing
conventions to distinguish affixed forms from compounds makes her model
rather different from that of Halle and Mohanan, with their four lexical
strata and no such bracketing conventions. She further suggests (p.71) that
their use of four separate strata of lexical rules could, in theory, be
extended to an infinite number; but, in the absence of firm evidence that
all human languages can be modeled with two levels of lexical rules, it is
not entirely clear that her two levels are actually much further away from
infinity than their four strata.
In fact, as the rules in Level 2 of McMahon's model are not always strictly
subject to Structure Preservation and the Derived Environment Condition, two
of the basic characteristics of lexical rules, in reality one might conclude
that the fundamental division into lexical and postlexical rules is not that
clear, and indeed, she does suggest (p.55) that the lexical-postlexical
distinction may be more gradient than absolute. (Note, also, that the clear
division of rules into separate, autonomous levels is not really maintained
when a single rule, [r]-Insertion, can apply on three different levels.)
One might, therefore, see McMahon's model as consisting of three levels: a
fully lexical level, which is strictly subject to Structure Preservation and
the Derived Environment Condition; an intermediate level, that is not
strictly subject to those constraints, but which is phonetically somewhat
idiosyncratic, does have exceptions, and results in binary distinctions
which speakers are often aware of; and a postlexical level, which has strong
phonetic motivation, is entirely exceptionless, and introduces small,
perceptually inaudible features of pronunciation. If, then, there are three
levels on this gradient analysis, is there any evidence that there might not
be a fourth level, or, in theory, infinite levels?
Let us consider the possibility of a fourth level a bit further. Level 1
lexical rules, like Velar Softening and VSR, are Structure Preserving and
subject to the Derived Environment Constraint. Level 2 rules, such as SVLR,
may occasionally break these constraints but they cannot be fully
postlexical because they can refer to morphological information. In
contrast, postlexical rules can make no reference to morphological
information so that, for example, Regressive Assimilation of a nasal
consonant takes place whether it is inside a word such as 'income' or across
a word boundary such as 'tin cup'. However, note that Low Level Lengthening
takes place at the end of words, and this is one kind of morphological
information (p.192), so LLL does not entirely fit in with the framework for
postlexical rules. It is just possible, therefore, that we may need another
level between Level 2 of the lexical rules and the postlexical level.
The number of levels is perhaps a minor difference from previous
formulations of LP. A rather more fundamental deviation from other models
is McMahon's rejection of abstract underlying phonological forms. While
many phonologists may welcome the abandonment of /fixt/ and /bix/ as the
underlying representations of 'fight' and 'buy' (Halle and Mohanan, 1985:
110), there may be less agreement with the complete elimination of abstract
underlying forms, which seems closer to Natural Phonology (Hooper, 1976)
Although McMahon aims to make the underlying representations of underived
forms maximally surface-true, in some cases it seems that they cannot be
completely surface-true, for the underlying form of a word like 'courage'
must have a full vowel, not a schwa or [I], in the second syllable if
'courageous' is to be derived by rule. Presumably, the schwa or [I] in
'courage' is derived by vowel reduction, though this is never is actually
fully explained. Though we are told (p.92) that the issue of vowel
reduction is being left aside "for the moment", the only subsequent
consideration of this issue is when we are told (p.101) that the schwa in
the second syllable of 'harmony' may either be derived by vowel reduction or
it may be a surface-true schwa with the full vowel necessary for the
derivation of 'harmonic' obtained by means of a Spelling Pronunciation rule.
Unfortunately, McMahon leaves this unresolved, stating "I will not pursue
this analysis further".
Given that, in McMahon's model, most irregular verb alternations are assumed
to be held in the lexicon and not derived by rule, and given that
alternations for 'keep'/'kept' and 'bite'/'bit' cannot be derived entirely
by existing Level 2 rules but need their own Level 1 rule of past tense
affixation, it is perhaps strange that she does not go the whole hog and
assume that all irregular verb alternations are purely lexical, with any
apparent regularity explained by means of cognitive pattern associations.
Pinker (1999) makes a strong case for this dual model, with rules handling
all regular forms and lexical retrieval combined with pattern associations
for all irregular forms, and in some ways this seems neater than assuming
that the past forms of just two small classes of irregular verbs are formed
by rule. This would appear to be a valuable area for future research, to
determine how psychologically real the rule-based past tense formation of
the 'keep'/'kept' and 'bite'/'bit' verbs is.
One underlying theme of the whole book is: what happens if we insist that
the Derived Environment Condition applies strictly to all Level 1 lexical
rules. We have seen some of the consequences of this: we must assume two
separate Vowel Shift Rules, even though they describe essentially the same
process; the difference between RP and GenAm in the occurrence of /j/ in
'duke' and [j] in 'studious' cannot be described by a single rule; and Velar
Softening cannot deal with the alternation between /s/ and [k] in words like
'reduce' and 'reduction', even though this is essentially the same
alternation as that between the /k/ and [s] in 'critic' and 'criticise'.
One conclusion might be that strict adherence to the Derived Environment
Condition causes more problems than it solves, though this is certainly not
the conclusion that McMahon draws, as she insists that properly constraining
the model of phonology by means of the Derived Environment Condition and
Structure Preservation is absolutely essential.
Although many will certainly disagree with some of the positions McMahon
adopts, all phonologists are likely to find this book very interesting and
exceptionally valuable. Overall, it provides many fascinating, well-argued
contributions to an understanding of the nature and structure of phonology,
and it is certain to become a classic text for anybody interested in this
Carr, P (1993). Phonology. London: Macmillan.
Halle, M & Mohanan, K P (1985). Segmental Phonology of Modern English.
Linguistic Inquiry, 16:1, 57-116.
Hooper, J B (1976). An Introduction to Natural Generative Phonology. New
York: Academic Press.
Pinker, S (1999). Word and Rules: The Ingredients of Language. London :
Weidenfeld & Nicolson.
About the reviewer: David Deterding is an Associate Professor at the
National Institute of Education, Nanyang Technological University,
Singapore. His webpage is: