LINGUIST List 12.595
Sat Mar 3 2001
Review: Giegerich, Lexical strata in English
Editor for this issue: Terence Langendoen <terrylinguistlist.org>
What follows is another discussion note contributed to our Book Discussion
Forum. We expect these discussions to be informal and interactive; and
the author of the book discussed is cordially invited to join in.
If you are interested in leading a book discussion, look for books
announced on LINGUIST as "available for discussion." (This means that
the publisher has sent us a review copy.) Then contact Simin Karimi at
simin
linguistlist.org or Terry Langendoen at terry
linguistlist.org.
Directory
Mike Maxwell, Review of Giegerich, Lexical Strata in English
Message 1: Review of Giegerich, Lexical Strata in English
Date: Fri, 2 Mar 2001 21:08:53 -0500
From: Mike Maxwell <Mike_Maxwellsil.org>
Subject: Review of Giegerich, Lexical Strata in English
Review of Giegerich, Heinz J. 1999. Lexical Strata in
English: Morphological causes, phonological effects.
Cambridge Studies in Linguistics 89. Cambridge: Cambridge
University Press. $69.95.
Reviewed by Mike Maxwell, Summer Institute of Linguistics
Depending on your bias, a book describing a rule-based (as
opposed to constraint-based) phonological analysis of
English may seem either antiquated or refreshing.
Regardless of your viewpoint (the reviewer confesses to
belonging to the latter camp), a rule-based analysis which,
in this day and age, does not even make reference to
Optimality Theory--if only to argue against it--is likely
to strike you as out of touch with reality. For better or
worse, the book being reviewed here is just that: an
analysis of English phonology in a rule-based LP framework,
with Optimality Theory not so much as mentioned.
I will briefly summarize the chapters, then turn to some
aspects of the book's analysis which strike me as less
defensible in the context of a rule-based approach.
Giegerich (henceforth G) begins with a brief summary
(chapter 1) of the demise of the theory of Lexical
Phonology (LP), suggesting that it was not the theory as a
whole that was flawed, but rather certain key assumptions--
assumptions which could be modified while leaving much of
the rest of the theory intact. Chapter 2 explores these
assumptions in more detail, arguing that the fatal flaw was
that of assigning affixes, rather than the bases to which
they attach, to strata. It may be worth pointing out here
that the original use of the term 'strata' (due to Saciuk
1969), the membership of a morpheme in a particular stratum
was a matter of loan word vs. native phonology. The
distinction in English between Romance and Germanic
(native) vocabulary is a typical example. LP usurped the
term, using it to encode the SPE (Chomsky and Halle 1968)
distinction between '+' and '#' boundary markers. I found
this chapter especially enlightening, as G teases apart the
Romance - Germanic distinction from the boundary marker
distinction.
In Chapter 3, G proposes his revision of LP, in which
stratal behavior is determined not by the affix, but the
base to which an affix attaches. This is a pivotal chapter,
and I will return to some of the claims below. Chapter 4
derives the effect of the Strict Cycle Condition
(structure-changing cyclic rules do not apply in underived
environments) from other aspects of G's theory (again, I
will say more about this below). Chapter 4 also argues that
there is no pre-morphology cycle, that is, cyclic
phonological rules do not apply before affixation. This has
several salutary effects, including the elimination of
"free ride" analyses, in which e.g. the mid vowel in a word
like name would be derived from an underlying form with a
low vowel through the application of the English rule of
Vowel Shift, despite the absence of any alternations
affecting this morpheme (it always surfaces with a mid
vowel).
Chapter 5 explores a topic which in my opinion deserves
greater discussion: the effect of literacy, and especially
orthographic conventions, on the lexical representations in
speakers' minds. G suggests that the effect is strong, and
in particular that it drives vowel alternations like those
in Anderson--Andersonian (schwa - /o/) and Mendel--
Mendelian (schwa - /i/). This claim has far reaching
implications for issues of representation and the
application of phonological rules.
Chapters 6 and 7 turn to issues of 'linking-r' and
'intrusive-r' in non-rhotic varieties of English, nicely
tying together the appearance of [r] with that of the
semivowels [j] and [w].
Finally, chapter 8 discusses syllabification in English,
arguing that there is no process of resyllabification in
the lexical phonology (i.e. syllabification is entirely
structure-building, not structure-changing), thereby
explaining the dependence of syllabification on the
morphology (in particular on differences among prefixed
words, suffixed words, and compound words), without
recourse to morpheme boundaries.
Turning now to some of G's specific claims, he notes that
under the traditional analysis of LP, an affix belonged to
a single stratum--with few exceptions, and where those
exceptions might be treated as homophonous affixes (see
e.g. Aronoff's (1976) analysis of the two -ables, predating
the theory of LP). One of G's major claims is that the
exceptions are numerous (at least in derivational
morphology)--in fact so numerous as to invalidate the
traditional analysis. Instead, the stratum on which a given
affix attaches is determined by the base to which it
attaches. G proposes that there are two strata in English
(some linguists assumed more than two), and calls the bases
of Stratum 1 (the deepest stratum) 'roots', and the bases
of Stratum 2 'words'. For languages which have three
strata, the bases of the intermediate stratum are called
'stems'. (The use of the term 'root' for morphologically
complex bases of Stratum 1 is surely an unfortunate choice
of terminology, but does not of course affect the validity
of the theory.)
Some roots in Stratum 1 are bound, such as 'matern' (as in
'maternal'), while others are free (such as 'go'). G argues
that all "roots" in Stratum 1--bound or free--lack
categories. That is, a root like go is not listed as a verb
in the lexicon. This raises a host of problems. It does
seem plausible that bound roots lack categories (assuming
for the sake of argument that bound roots are lexemes at
all), since it is often impossible to determine what
category they would belong to: 'matern', for example, takes
both the suffix -ity (which normally attaches to
adjectives) and the suffix -al (which normally attaches to
nouns on Stratum 1). But the claim that all lexical items
of Stratum 1 lack categories results in a very odd analysis
of unbound "roots". For example, 'moment', 'courage', and
'harmony' are analyzed not as nouns, but as roots-which-
will-become-nouns if they do not receive further Stratum 1
affixes.
Even stranger is the analysis of forms with Stratum 1
affixes, such as 'maternity' or 'courageous': these
are also analyzed as roots-which-will-become-X (X = noun or
adjective, in this case). The problem here, as I see it, is
that G must postulate two different properties: root-which-
will-become-X for Stratum 1 words, and ordinary categories
for Stratum 2 words. But the only difference between these
two properties (apart from the stratum on which they are
relevant) is that the former is supposed to be invisible to
any process in its stratum except the process that converts
a root-which-will-become-X into an X at the end of the
stratum! Specifically, what G claims is that each nonbound
root in the Stratum 1 lexicon, as well as each suffix, is
diacritically marked for one of the subrules of the
following Root-to-Word Conversion rule (G's (10) page 76):
[ ]r --> [[ ]r]L (L = N,V,A)
(where 'r' stands for 'root')
Worse, if we leave aside bound roots for the moment, the
root-which-will-become-X property is demonstrably not
invisible to the Stratum 1 morphology. On the contrary,
when a Stratum 1 affix attaches to something other than a
bound root, the affix nearly always attaches to a
particular class of bases--a class which can be
characterized by the category which (on G's analysis) it
would become in the absence of Stratum 1 affixation. That
is, the unbound roots to which -ity attaches are a roots-
which-would-become-adjectives (if they did not take -ity),
while the unbound roots to which Stratum 1 -al attaches are
roots-which-would-become-verbs. Moreover, the category
resulting from the attachment of a Stratum 1 suffix is
usually predictable, both for purposes of attachment of
further Stratum 1 affixes, and for forms which do not
receive further affixes. G attributes this to the fact that
Stratum 1 suffixes are also diacritically marked for the
subrules of the Conversion rule; but the result is again
precisely the same as if the suffixes were marked for the
categories themselves, as presumably happens with Stratum 2
suffixation. And finally, if as G argues, many suffixes can
attach on both Stratum 1 and Stratum 2, we are faced with
the fact that for many of these, the subrule of the
Conversion Rule which the suffix selects when it attaches
on Stratum 1, assigns the same category as the suffix
itself assigns on Stratum 2. (Another issue is how English
stress rules can exhibit their well-known sensitivity to
the category of the form to which they apply. G takes the
position that these rules belong to Stratum 1 and are
cyclic, but apply after the Conversion rule. It is not
obvious how these properties are reconciled in his model.)
Why is G led to this odd analysis? His argument is based
on a perceived need for consistency: since -ous, say, can
attach in Stratum 1 both to bound roots (ambitious,
raucous) and to unbound "roots" (courageous, advantageous),
and since the former lack categories, the latter must also.
In G's words:
...in suffixation processes all members
of that category--complex or simple--are
freely interchangeable with a subclass
of that category, namely that of bound
roots, and that, therefore, the
essential properties of that subclass
must be shared by all members of the
category... Rather than arbitrarily
calling moll- [of mollify] an adjective
root and gorm- [of gormless] a noun
root, we adopt the opposite analysis
and treat sensation, sensational and,
indeed, sensationality as roots that
crucially bear no lexical categories.
[page 75-6]
(Two other arguments are given on page 77, both are
arguments for the Conversion rule above, but neither
establishes that this rule must have subrules assigning a
category to the output.)
There is, however, a way between the horns of this dilemma,
which is compatible with much of G's overall theory, while
at the same time eliminating the need for the root-which-
will-become-X categories. Suppose that rather than
specifying the category to which the base must belong
(thereby preventing attachment to bound roots which lack
categories), an affix's input category must be able to
unify with the category of the base. This would allow -al,
for example, to specify the input category of noun,
allowing attachment to unbound roots bearing the category
of noun, such as tone and hormone, as well as to bound
roots whose category is uninstantiated, such as 'matern'
and 'fin'.
Another questionable claim of G concerns Kiparsky's (1982)
proposal concerning the Elsewhere Condition
(oversimplifying: a set of rules can apply disjunctively,
such that the first rule to apply in such a set prevents
later rules from applying). Kiparsky intended to derive the
Strict Cyclicity Effect from this principle (as G intends
to derive it from his own principles). Crucial to
Kiparsky's derivation is the idea that lexical entries are
a sort of rule, which block cyclic structure-changing rules
from (inappropriately) applying to those entries. G argues
that this is wrong: "[lexical] items and processes have
different ontologies." It seems to me that this is reading
too much into the word "rule." If one looks at the ontology
of linguistic concepts as an inheritance hierarchy, it is
quite plausible that there is an abstract node in that
hierarchy from which both lexical entries (or at least the
phonological parts of such entries) and phonological rules
inherit. A trivial re-wording of the Elsewhere Condition
would make it refer to that abstract node, getting around
G's objection (and telling us something interesting about
the human language capacity in the process).
The detail of G's coverage of English derivational
morphology and phonology is impressive. (German data is
also brought to bear on occasion.) But as with many such
works, I found myself at times wondering whether a given
word was really morphologically complex in the synchronic
grammar (or if a word even exists in the synchronic grammar
of most native speakers, such as gormless--a word to which
G refers often, but which not even my spell checker knows
about). Other doubtful data on which arguments are based
includes a putative distinction between the suffixes -ant
and -ent; while these may be spelled differently, they are
certainly homophonous in my dialect (both in stressed and
unstressed forms, cf. coincidental - consonantal). (The
contrast is argued on page 31 to be real, but there are
exceptions both ways to the generalizations used to
motivate the distinction.)
I wish that G had taken more time to compare his approach
with other theories. For example, his description of
Stratum 1 affix and root co-occurrences invites comparison
with Koenig's (1999, particularly section 5.2) theory of
stem selection, while his treatment of allomorphy bears
interesting resemblances with Natural Generative
Phonology's treatment of 'morphophonemic' alternations
(Hooper 1976), and with Declarative Phonology's use of
lexically listed allomorphs together with nonviolable
constraints (Bird 1995, Coleman 1998). The discussion of
syllabification in chapter 8 would also have benefited from
contrasting the analysis there with other analyses in the
now voluminous literature on this topic.
For those who have already decided that declarative
phonology is correct, this will be the wrong book at the
wrong time. But for those who want to see how Lexical
Phonology might have turned out if declarative theories had
not taken over, this book should prove of interest. The
treatment of certain English-specific phenomena,
particularly that of linking and intrusive 'r', will also
appeal to many. And finally, as I have said above, G
devotes a chapter to the influence of orthography on the
spoken language, a question which deserves greater
discussion.
Aronoff, Mark. 1976. Word Formation in Generative Grammar.
Linguistic Inquiry Monographs 1. Cambridge, MA: MIT
Press.
Bird, Steven.1995. Computational Phonology: A constraint-
based approach. Studies in Natural Language
Processing. Cambridge: Cambridge University Press.
Chomsky, Noam; and Morris Halle 1968. The Sound Pattern of
English. New York: Harper and Row. Reprinted 1991,
Boston: MIT Press.
Coleman, John. 1998. Phonological Representations.
Cambridge Studies in Linguistics 85. Cambridge:
Cambridge University Press.
Hooper, Joan. 1976. An Introduction to Natural Generative
Phonology. New York: Academic Press.
Kiparsky, Paul. 1982. "Lexical phonology and morphology."
Pages 3-91 in I.S. Yang Linguistics in the morning
calm, volume 2. Seoul: Hanshin.
Koenig, Jean-Pierre. 1999. Lexical Relations. Stanford
Monographs in Linguistics. Stanford: CSLI.
Saciuk, Bohdan. 1969. "The Stratal Division of the
Lexicon." Papers in Linguistics 1: 464-532.
- ---------------------------
Mike Maxwell works in the development of computational
environments for syntactic, morphological and phonological
analysis for the Summer Institute of Linguistics, where he
developed a parsing engine which implements a form of
lexical morphology and phonology.