English phonology and morphology Ricardo Bermúdez-Otero & April ...

1
English phonology and morphology
Ricardo Bermúdez-Otero & April McMahon

In:
Aarts, Bas, & McMahon, April (2006). The handbook of English linguistics.
Oxford: Blackwell. 382-410.



1 Introduction

The title of this chapter poses a daunting challenge, since the
morphophonology of present-day English is one of the most intensively
studied areas in the whole of morphology and phonology. Indeed, as key
innovations in phonological and morphological theory have been introduced,
they have frequently been illustrated by means of case-studies from English:
this is true not only for classical rule-based generative phonology (Chomsky &
Halle, 1968; henceforth SPE), but more recently for connectionist and dual-
route approaches to inflection (Rumelhart & McClelland, 1986; Pinker &
Prince, 1988) and for output-output correspondence within Optimality Theory
(OT) (Benua, 1995, 1997). It follows that we must define our aims somewhat
narrowly.

First, then, this chapter focuses on interactions between phonology and
morphology in present-day English and their implications for the shape of the
morphology-phonology interface in natural language. Perforce, we disregard
phonology-syntax interactions, although clearly some key facts and concepts
in morphophonology have close phonosyntactic analogues. Our data are drawn
from both British and American dialects, standard and vernacular, though
obviously no variety is exhaustively described. We focus on facts that have
figured prominently in the wider theoretical debate, but also pay some
attention to phenomena that seem peculiar to English. Even the latter,
however, underscore points of general relevance: as we shall see in section
3.5, for example, some of the idiosyncrasies of present-day English
morphophonology are the product of historical contingencies; this illustrates
how, when contending with the effects of diachrony, morphophonological
theory routinely encounters historically conditioned facts that it can note but
not explain.
From a theoretical viewpoint, we concentrate on major conceptions of
the morphology-phonology interface, abstracting away from other dimensions
of variation between theories. Wherever possible, therefore, our presentation is
neutral between rule-based and constraint-based systems, with ‘rule’ simply
meaning ‘symbolic generalization’ unless otherwise stated or required by
context. We accordingly ignore the differences between rule-based Lexical

---

2
Phonology and Morphology (LPM: e.g. Kaisse & Shaw, 1985; Kiparsky,
1982b, 1985) and Stratal OT (Bermúdez-Otero, 1999, forthcoming; Kiparsky,
1998, 2000; Orgun, 1996), except where the choice of model has affected the
demarcation of phonology, morphology, and the lexicon (section 2) or the
application of concepts such as cyclicity and level segregation (section 3). The
general aim of the chapter is to sift through the intricate debate (often highly
esoteric and theory-internal) that surrounds English morphophonology and to
identify key concepts and issues that deserve our continued attention,
regardless of major shifts in the theoretical landscape.


2.
The division of labour between phonology, morphology, and the
lexicon

2.1 The
problem

We have thus far identified our main concern as being the interaction of
morphology and phonology in present-day English, but the problem can only
be formulated if we can first distinguish between (i) computations performed
in the phonology, (ii) computations performed in the morphology, and (iii)
lexical storage.

Here, however, the spectrum of opinion is extraordinarily wide. SPE did
not countenance an independent morphological module and envisaged lexical
storage as maximally economical, with all alternations derived via
phonological rules. On the other hand, in connectionist and so-called cognitive
approaches (Rumelhart & McClelland, 1986; Bybee, 1995, 2001) the lexicon
is highly concrete and massively redundant: all grammatical knowledge,
whether phonological or morphological, is taken to inhere in the network of
associations between items stored in long-term memory, so that, in effect, the
lexicon is the grammar.

2.2 Testing
the
boundaries

Most practitioners would assume intermediate positions between these two
extremes; but, again, this raises the difficulty of formulating explicit criteria
for drawing boundaries between the phonology, morphology, and lexicon.
The typical approach here has been to propose tests to identify genuine
phonological rules. Below we review a number of these tests, although our list
is not exhaustive.
• SPE allowed unlimited phonological opacity: such restrictions as it
imposed emerged during acquistion from (relatively ill-defined) provisions
in the evaluation measure. In contrast, [Bybee-]Hooper’s (1976) True
Generalization Condition requires genuine phonological rules to be
transparent, and therefore not to be contradicted by surface evidence.

---

3
Although this work has been influential, the proposal seems too strong:
more recent research usually acknowledges that phonological rules may be
opaque, but proposes grammatical architectures that impose severe formal
restrictions upon the complexity of phonological opacity effects, over and
above learnability considerations (see e.g. Bermúdez-Otero, 2003, §2).
• Phonological naturalness has often been seen as a hallmark of genuine
phonological rules, although ‘naturalness’ has variously been defined
formally (e.g. genuine phonological rules operate over features, which
define natural classes of segments, rather than random segment lists), or
functionally (e.g. genuine phonological rules are phonetically grounded), or
typologically. In OT, whether mono- or poly-stratal, naturalness is a key
criterion, as every genuine phonological process must be the best solution
to the problem posed by a given ranking of phonological markedness and
faithfulness constraints. Definitions overlap here, since the notion of
markedness in OT is intrinsically typological, but can be given both formal
and functional readings, as in the recent controversy over the grounding of
constraints (Bermúdez-Otero & Börjars, forthcoming; Hale & Reiss, 2000;
Hayes, 1999a; Hayes et al., 2004).
• In Kiparsky’s (1994, p.16) reading, Ford & Singh (1983) and Spencer
(1991, §4.4) claim that all rules subject to morphological conditioning are
morphological. A more nuanced version of this approach is advanced by
Anderson (1992), who asserts that genuine phonological rules (as opposed
to ‘word-formation’, i.e. morphological, rules) can be circumscribed to a
morphologically defined domain, but cannot refer to specific morphemes or
morphological/syntactic features. This claim is explicitly endorsed in
Stratal OT by Orgun (1996) and, modulo alignment constraints, by
Bermúdez-Otero (forthcoming, ch. 2). The cost of this strategy may be a
proliferation of cophonologies, but see section 4 for some interesting
applications of the concept of cophonology. Monostratal OT, in contrast,
tacitly reverts to the SPE position that all morphological information is
available to the phonology (see Bermúdez-Otero, forthcoming, ch. 2;
Orgun & Inkelas, 2002, p. 116).
• Kiparsky (1994) asserts that morphological rules can be distinguished from
phonological rules (both lexical and postlexical) by the cluster of formal
properties in (1):

---

4
(1)
a. Phonological rules
b. Morphological rules
general
item-specific
manipulate single phonological units
manipulate phonologically arbitrary strings
observe phonological locality conditions
observe morphological locality conditions
follow all morphological rules in the same
precede all phonological rules in the same
cycle
cycle

The properties in (1a) are clearly related to the criteria of transparency and
naturalness: any transparent phonological rule will ipso facto be general
and follow all morphological operations in the same cycle, while any
natural phonological rule will ipso facto manipulate nonarbitrary
phonological constituents and observe phonological locality conditions.
However, it should be clear that (1a) falls far short of requiring absolute
transparency or naturalness. In consequence, Kiparsky’s (1994) proposal
can easily be adopted in post-SPE rule-based frameworks, where opacity is
formally unlimited and naturalness criteria are defined formally rather than
functionally; but it will not work in theories with strong transparency and
naturalness requirements —including, interestingly, Kiparsky’s own (1998,
2000) stratal version of OT.
• More recent work in Stratal OT seeks to derive the typical life-cycle of
phonological rules (Harris, 1989; McMahon, 2000) from properties of the
phonological learning algorithm. From this viewpoint, Bermúdez-Otero
(2003, forthcoming) suggests that phonological alternations triggered by an
independent phonotactic requirement are easier to acquire, and therefore
more resistant to morphologization and lexicalization, than phonological
alternations lacking in phonotactic motivation. The evidence of Berko’s
(1958) classic wug test supports this claim: Berko found that, by age five,
children acquiring English know that the plural noun suffix is an alveolar
fricative, i.e. /-S/; however, when selecting among its surface allomorphs,
i.e. [-z ~ -s ~ - z], children perform best when the choice is phonotactically
determined (e.g. [w
-z], [b k-s]), slightly worse when the choice requires
knowledge of the underlying voice specification of the suffix (e.g. [l n-z],
though *[l n-s] is phonotactically fine), and worst of all when there is
competition between several potential repair strategies (e.g. [tæs- z] with
epenthesis vs *[tæs] with coalescence).
• Finally, in their dual-route approach to morphology Pinker & Prince
(1988) have produced detailed and fairly explicit criteria for distinguishing
between lexical storage and morphological computation, at least for
inflection. These criteria turn out to be relevant to the distinction between
lexicon and phonology, although their applicability is limited. First, if a
morphological item is (or can be) constructed online, the logic of the theory
requires that all phonological alternations associated with that construction

---

5
should also be computable online. Thus, since the past tense and past
participle suffix /-d/ is added to verb stems by a genuine morphological
rule, it follows that the [-d ~ -t ~ - d] alternation must also be generated by
a (phonological) rule. As it happens, this rule is independently required to
capture robust word-level phonotactic constraints, which provide further
evidence for it. However, this argument does not work in the opposite
direction: a phonological pattern may be enforced by a discrete symbolic
generalization represented in the grammar even if it does not trigger
alternations associated with regular morphological processes. An extreme
case would be that of productive phonotactic patterns in isolating
languages, which do not cause alternations but are shown to be
grammatically active in, for instance, the nativization of loans (Yip, 1993,
1996).

2.3 Do the criteria converge?

If the theory of grammar is to have nontrivial empirical content, one should
aim to draw the boundaries between phonology, morphology, and the lexicon
by means of a set of logically independent but empirically convergent criteria.
As we have seen, however, some of the criteria reviewed in the previous
section are mutually incompatible: for example, if phonological rules must be
typologically or phonetically natural, then the scope of phonological
computation will be considerably narrower than if the status of an alternation
depends only on its form and locality properties, as suggested by Kiparsky
(1994). Finding a set of convergent criteria has in fact proved to be rather
hard. In this section we shall illustrate these difficulties by considering the
possible involvement of a phonological process of vowel shift in the
alternations found in strong verbs (e.g. eat~ate) and in irregular weak verbs
(e.g. keep~kept).

As is well-known, present-day English has a number of vowel
alternations triggered by morphologically sensitive processes of shortening
and lengthening (see e.g. SPE, pp. 178ff.; Myers, 1987). Their morphological
conditioning is discussed in section 3 below.
• In stressed antepenultimate syllables followed by a stressless penult, long
vowels are subject to so-called ‘trisyllabic shortening’: e.g. sāne~sănity,
serēne~serĕnity. This should be regarded as the result of trochaic
shortening under final syllable extrametricality: i.e. (săni)<ty>,
se(rĕni)<ty> (Hayes, 1995, §6.1.5). Trochaic shortening also applies in
penultimate syllables before the suffix -ic: e.g. cyclōne~cyclŏnic,
Hellēne~Hellĕnic (see section 4 below).
• Long vowels undergo shortening in closed syllables, assuming word-final
consonants to be extrasyllabic: e.g. dēe<p>~dĕp<th>, fī<ve>~fĭfty.

---

6
• Finally, short vowels undergo lengthening when immediately followed by
CiV sequences: e.g. comĕdy~comēdian, harmŏny~harmōnious.
In SPE, the qualitative aspect of these alternations is handled by means of a
rule of long vowel shift, which largely recapitulates traditional accounts of the
diachronic evolution of long vowels in Early Modern English:

(2)
Long vowel shift in SPE


a.




sane
sanity



UR



/æ /
/æ /
Trisyllabic
shortening
—

æ
Long
vowel
shift

e

—


b.




cyclone
cyclonic



UR



/ /
/ /
Trochaic
shortening




—




Long vowel shift

o
—


c.




deep
depth



UR



/e /

/e /
Closed
syllable
shortening

—

e
Long
vowel
shift


i

—


d.




comedy
comedian
UR





/e/
/e/



CiV lengthening

—
e



Long vowel shift

—
i

Consider now the vowel alternations found in strong verbs such as
eat~ate, dig~dug, and fly~few, extensively discussed in Halle & Mohanan
(1985). Halle & Mohanan’s analysis is ostensibly within LPM, but wears the
restrictions inherent in the architecture of that model very lightly; in fact, it
approximates in abstractness the SPE description on which it is based (see
McMahon, 2000). Following the programmatic assumptions of SPE, Halle &
Mohanan seek to derive these vowel alternations by rule, whilst positing the
smallest possible number of rules and maximizing the application of each rule
(i.e. its ‘functional yield’). To achieve this end, Halle & Mohanan formulate a
number of (essentially morphological) processes of ablaut, and allow their
output to take a free ride on long vowel shift. The alternations are thus
factored out into a morphological and a phonological component.

---

7
(3)
Strong verb alternations in Halle & Mohanan (1985)






eat
ate


UR



/e / /e /

Lowering
ablaut
—
æ


Long
vowel
shift

i
e




Assume, however, that strong past tense and past participle forms are
irregular and therefore stored in long-term memory, as convincingly argued in
Pinker & Prince (1988) and related work. If so, it becomes unnecessary to
divide vowel alternations such as eat~ate into a morphological and a
phonological component. In consequence, even if one countenances a
synchronic phonological rule of vowel shift (and this is a big ‘if’, on which see
below), vowel shift will not need to be involved in strong verb morphology.
Taking advantage of this result, McMahon (1990, 2000) replaces Halle &
Mohanan’s single word-level rule of long vowel shift by two stem-level rules
of long vowel shift and short vowel shift; these two rules apply only in derived
environments created by the previous application of a shortening or
lengthening rule (on the blocking of stem-level rules in nonderived
environments, see section 3.3 below).

(4)
Vowel alternations in McMahon (1990)


a.




eat ate



UR



/i /
/e /
Long
vowel
shift
blocked blocked


b.




sane
sanity



UR



/e /
/e /
Trisyllabic
shortening
—

e
Short
vowel
shift
—

æ
Long
vowel
shift
blocked —


c.




comedy
comedian
UR






/e/
/e/



CiV lengthening

—
e



Short vowel shift

blocked
—



Long vowel shift

—
i


By doing away with problematic free rides, McMahon’s analysis
represents a clear improvement on Halle & Mohanan’s in terms of
concreteness and learnability. Admittedly, Pinker & Prince’s approach to
strong verb morphology does not necessarily prevent one from factoring out
alternations such as eat~ate into a lexically listed part and a part derived by a
free ride through an SPE-style word-level rule of long vowel shift; but it is
hard to see why this should be a desirable option unless one is wedded to the

---

8
notions of maximal lexical economy and maximal rule utilization —in which
case one would not accept the premises of Pinker & Prince’s model in the first
place. In this example, therefore, a measure of convergence is achieved:
applying Pinker & Prince’s dual-route model of morphology results in
considerable gains in terms of the generality, transparency, and learnability of
phonological rules.

Let us now turn to irregular weak verbs such as keep~kept, sleep~slept,
bite~bit, or light~lit. As we saw above, the vowel alternations found in these
verbs are not unique, but are replicated in many other constructions; in this
sense, they fulfil Kiparsky’s generality criterion for genuine phonological
processes (see (1)). By Pinker & Prince’s criteria, however, they are always
associated with irregular (nondefault) morphology: e.g. -t against default -d, -
th against default -ness. Therefore, if Pinker & Prince are right, then both keep
and kept will have to be stored in long-term memory; the question is whether
kept will be listed as /ki p-t/ or as /kep-t/.
In the light of section 2.2, there are good prima facie arguments for
handling the length component of the keep~kept alternation by means of a
phonological rule of closed syllable shortening: this process is natural (and
indeed phonetically grounded), largely transparent, and blind to morphology
within its domain (on the notion of domain, see section 3.2 below). Closed
syllable shortening is also required independently to handle robust phonotactic
constraints on morphologically underived items. In turn, the qualitative
dimension of the alternation could be analysed using McMahon’s (1990,
2000) stem-level rule of short vowel shift. In contrast with closed syllable
shortening, however, short vowel shift is still somewhat problematic: e.g. it
has no independent phonotactic motivation, involves Greek-letter variables
(or, in OT terms, contrived versions of faithfulness), and has a messy
penumbra of (un)gliding and (un)rounding rules.
In the case of the irregular weak verbs, therefore, we are confronted
with an instance of nonconvergence between demarcation criteria. If the
naturalness of closed syllable shortening persuades us to derive the alternating
vowels from a single underlier, then we are also committed to computing the
far less natural qualitative component of the alternation. But, paradoxically,
this would imply that naturalness and transparency (which led us to consider
closed syllable shortening as a plausible phonological rule in the first place)
are not reliable criteria for distinguishing between lexicalized patterns and
genuine phonological generalizations after all.

2.4
Cutting the Gordian knot

How, then, can one solve this impasse? There seem to be two possibilities.

First, we might propose that, at least at the highest grammatical level,
phonological generalizations are not constrained by naturalness: they may be
pure inductive generalizations, and therefore less markedness-driven than
history-driven (in the sense that they simply encapsulate the synchronic

---

9
outcome of processes that were once natural and transparent). If so, the
criterion for the psychological reality of a phonological rule at the stem level
will just be whether the rule can be acquired by induction: this may to a large
extent be determined by the rule’s transparency, but naturalness clearly has
nothing to do with it. The question then arises as to whether this type of purely
inductive rule is essentially different from a morphological rule in the style of
the word-formation processes of Anderson’s (1992) a-morphous morphology.
If they are broadly the same kind of entity, there may be no reason beyond
familiarity of convention to write vowel shift in a feature-based format, with
Greek letter variables and the like, instead of employing notation roughly like
that in (5).

(5)

/i / alternates with /e/



/e / alternates with /æ/ etc.

Of course, this option brings us back to the earlier problem of distinguishing
between morphological and phonological processes. Those not wishing to take
this direction might retreat to the middle-way position defined by Kiparsky
(1994), where genuine phonological rules need not be natural in a typological
or phonetic sense, but only in the purely formal sense of referring to
phonological categories and obeying phonological locality conditions. This,
however, will not be a possibility in frameworks where all phonological levels
are optimality-theoretic.

Alternatively, we may choose to list kept as /kep-t/, thereby
dispensing with vowel shift as a phonological rule. Here, the perceived
difference between closed syllable shortening and vowel shift in terms of
typological and phonetic naturalness, transparency, and independent
phonotactic motivation is directly reflected in their grammatical status:
shortening becomes a static phonological generalization over stem-level
domains, while vowel shift is reduced to a pattern of relationships among
stored lexical entries. Interestingly, this implies that the output of every stem-
level computation is stored in long-term memory (for related arguments, see
Kiparsky, 1982b; Giegerich, 1988, 1999). In turn, this result has significant
implications: in section 3.4 we show that, given certain plausible assumptions
about blocking, storing every stem-level output as a lexical entry produces
results which resemble stratum-internal cyclicity. Cyclicity and the related
concepts of domain and level, however, are the topic of the next section.


3 Misapplication

3.1 The
problem

Once phonology, morphology, and the lexicon have been appropriately
demarcated, the theory of grammar must account for their interactions. In

---

10
particular, the setup of the morphology-phonology interface must explain how
morphological structures can cause phonological generalizations to misapply.
Present-day English abounds in instances of such misapplication.
Underapplication is said to occur when a phonological process p fails
to apply even though a morphological (or syntactic) construction m, or a
phonological process triggered by m, creates the conditioning environment of
p. In certain varieties of Northern Irish English, for example, the coronal
noncontinuants /t, d, n, l/, usually realized as alveolar, become dental when
followed by [r] or [ r] (Harris, 1989, p. 40). This process of dentalization
applies normally when its structural description is met within a single
morpheme (6a) or within a form derived by class-I suffixation (6b), but it fails
when its conditioning environment is created by class-II suffixation (6c),
compounding (6d), or syntactic concatenation (6e). For the terms ‘class I’ and
‘class II’, see Siegel (1974) and much subsequent work.

(6)
Dentalization (Northern Irish dialects)
a.
[t]rain, [d]rain, ma[t]er, la[d]er, spa[n]er, pi[l]ar
b.
sani[t]ary, eleme[nt]ary
c.
shou[t]er, ru[n]er (agentive
-er)


la[t]er, fi[n]er
(comparative -er)
d.
foo[t]rest, su[n]roof
e.
goo[d] riddance, ca[l] Rose

The absence of dentalization in (6d) and (6e) can conceivably be explained in
purely phonological terms; the process may simply be blocked by prosodic
word boundaries: e.g. [ω'[ωfoot][ωrest]], [φ[ωgood][ωriddance]]. In (6c),
however, the cause of underapplication is clearly morphological.

Other phonological processes that underapply in the presence of class-
II suffixes include trochaic shortening and closed syllable shortening,
discussed in section 2.3: e.g. prov[
]ke, prov[ ]c-ative, but prov[
]k-ing-
ness; d[i ]p, d[e]p-th, but d[i ]p-ness. However, Northern Irish dentalization is
special in that it is a purely allophonic process, as the alveolar and dental
realizations of the coronal noncontinuants are in complementary distribution;
see section 3.3 for the theoretical implications of this fact.

In cases of overapplication, a phonological process p applies even
though its conditioning environment is destroyed by a morphological (or
syntactic) construction m, or by another phonological process triggered by m.
In Canadian English, for example, the diphthongs /a / and /
/ undergo raising
to [ i] and [
] when immediately followed within the same prosodic word by
a voiceless obstruent that does not belong to a syllable with stronger stress
(Chambers, 1973). Like Northern Irish dentalization, this process is
allophonic, in that the surface distribution of the raised and unraised
diphthongs is entirely predictable. As observed in Bermúdez-Otero (2003,

---