Review of Asymmetry in Grammar
|
|
|
|
|
Review:
|
Date: Mon, 08 Sep 2003 14:45:03 -0400
From: Andrew Ira Nevins <anevins@MIT.EDU>
Subject: Asymmetry in Grammar, vol. 2: Morphology, Phonology,
Acquisition
Di Sciullo, Anna Maria, ed. (2003). Asymmetry in Grammar, Volume 2:
Morphology, Phonology, Acquisition, John Benjamins Publishing Company,
Linguistik Aktuell 58.
Reviewed by Andrew Ira Nevins, Massachusetts Institute of Technology,
Cambridge, Mass.
This is the second volume of papers from the conference on Asymmetry in
Grammar, held at UQAM in May 2001, organized under the auspices of the
Asymmetry project <http://www.asymmetryproject.uqam.ca>.
The unifying theme of the papers is exploring "whether asymmetry is
given by Universal Grammar", that "asymmetry is basic in grammar and
thus is a property of grammatical relations across the board" (p.3).
One use of the term "asymmetry" may arise in linguistic descriptions of
"argument/adjunct asymmetries", "subject/object asymmetries" and
"production/perception asymmetries". It is important to realize that
"asymmetry" just means "different behavior" in these contexts, and has
no theoretical status.
The second use of the term "asymmetry" is much narrower, and refers to
a formal property of n-ary relations: a relation R is asymmetric if aRb
implicates NOT bRa. Defining R in terms of sets of ordered pairs, an
asymmetric relation holds between ordered pairs if <x,y> is in R but
<y,x> is not.
It would an unexpected and interesting result if the fundamental role
of "asymmetry" in grammar was restricted to the second definition. Of
course, it would also be interesting to find out that transitivity
characterizes every grammatical relation, or that "associativity is a
property of grammatical relations across the board". But looking at
some of the most familiar grammatical relations reveal that none of
these formal properties of relations hold across the board. (Take
sisterhood, which describes two constituents with the same mother; it's
not associative: spec,X is-sister-of [X is-sister-of complement]. Take
government or theta-assignment, neither of which are transitive.)
As for falsifying the hypothesis that asymmetry characterizes
grammatical relations across the board, we need only consider the most
basic relation one can think of: identity. Identity is a symmetric
relation, and there is no way to describe "homorganic", "the OCP",
"binding theory", "coordination" or any of the dozens of components of
any grammatical theory without it.
Given that "asymmetry as fundamental" cannot refer to the formal
property of relations where the order of arguments matters (unless we
are to ignore identical-to, sister-of, tautosyllabic-with, clausemate-
of, etc.), it must be that the unifying theme of these papers is the
broader, phenomenological use of "asymmetry": different behavior of
different objects. And, to that (extremely general) end, the papers do
make for an interesting collection. (Of course, there are papers in the
volume that *do* concentrate on asymmetry as a formal property of
relations; notably DiSciullo, Marquis, Raimy, and Reiss).
CONTENTS and CRITICAL EVALUATION
Introduction (DiSciullo): It raises the question of the place of
asymmetry in grammar, and provides a brief overview of the papers in
the collection. There is no bibliography provided for the numerous
references that appear in the discussion.
Chapter 1 (DiSciullo) "Morphological relations in asymmetry theory"
This chapter examines relations in Morphological trees, which are
argued to be built by different principles of construction than
syntactic phrase markers. The two operations are Shift and Link, which
bear resemblance to Merge and Move. Shift creates a sisterhood relation
between two items and projects one of them; one property of Shift is an
asymmetric c-command relation between elements. Derivational affixes
are posited to project specifier-head and head-complement structure.
The second operation of morphological tree composition is Link, which
relates two positions: the spec(ifier) of a derivational affix to
either the spec or complement of the verbal base. For example, the
specifier of -er is Linked to the specifier of the verb, the specifier
of -ee is Linked to the complement of the verb, the specifier of -ive
is linked to the specifier of the lower verb, and the specifier of -
able is linked to the complement of the lower verb. These structural
anchors of the Link operation are claimed to derive argument-structural
restrictions on affixation, i.e. the putative impossibility of -er
affixation to unaccusatives (but cf. examples like "a late arriver"
(attested all over Google), "that ship is a sinker"), the impossibility
of -ee affixation to unergatives (but cf. "retiree", and moreover,
examples such as "amputee" which have no direct object (Barker 1998),
the impossibility of -ive suffixation with unaccusatives, and the
impossibility of -able suffixation with unergatives (but cf.
"danceable", attested all over Google).
Though clearly distinct from overt Movement in the syntax, it is not
clear to me why the relation of these two argument positions needs to
be performed by M-link; perhaps there is some reason why binding
relations or operator movement in the syntax could not relate these
positions, but if so, it should be argued for. DiSciullo claims that
"the derivation of spec-head relations in syntactic derivations is
obtained by Move", but this is certainly not an agreed-upon assumption;
it is a large consensus that expletives are directly merged in Spec,
TP, and that "why" is directly merged in Spec, CP (see e.g., Ko 2002).
The second half of the article is an application of the operator-
variable structure of M-shells to the decomposition of functional
constructs: "what" is created by a Shift operation that substitutes the
spec-head-comp structure headed by "-at" into the complement position
of the spec-head-comp structure headed by "wh-". Decomposition of
functional words allows spellout dependent on features. For example, if
the variable is +human, it is spelled out as "-o", if -human, as "-at".
If the operator is [-wh], it spelled out as "th-". (I suppose something
extra must be said to rule out "tho", the [+human,-wh] combination).
This decomposition is extended with interesting consequences to
Romance, e.g. Italian "per-che" (why) vs. "prima-che" (before).
Chapter 2 (Angela Ralli) "Prefixation vs. Compounding"
This chapter is an examination of the three Greek preverbs ksana-, kse-
, para- (very roughly, over-, re- and de-), and their categorization
into the categories prefix or first-member-of-compound, due to
idiosyncrasies in their semantic productivity and compositionality, and
phonological behavior. Ksana can occur as an independent word, while
the other two cannot. The separability question and its implementation
(e.g. incorporation vs. phrasal combination) are of significant
interest to researchers interested in the independence of and level of
syntactic attachment of affixes (for example, English -able can occur
as an independent word, often in periphrastically equivalent
sentences). Kse- and para- show selectional restrictions on the
Aktionsart of the verb they combine with, and can alter its argument
structure. Ralli is up to an ambitious task in uniformly characterizing
the semantic contribution of these prefixes. One formulation of the
contribution of kse- is that it "assumes the reversative meaning when
its ability of inducing the verbal properties to a high degree of
realization crosses the limits of the notional domain of the verb"; it
would help the reader to know the role of "notional domain" within the
theory: is it the essential, encyclopedic meaning of the verb (If so,
how can its limits be crossed)? In the end, Ralli proposes that there
is an internal and external para-, the former of which, along with kse-
is a Stem-attaching prefix, the latter of which is a Word-attaching
prefix, while ksana- is a Word-attaching Word.
Chapter 3 (Réjean Canac Marquis) "Asymmetry, syntactic objects, and the
Mirror Generalization"
This chapter begins with a fascinating examination of the types of
attested and unattested syntactic objects generated by the principles
of Chomsky's (1994) Bare Phrase Structure. Marquis shows that the Chain
Uniformity Condition does not exclude (i) adjunction of an XMax to a
head (e.g. a specifier undergoing "head movement" and not projecting at
the target; this is licit since the chain contains XMax at both
positions), nor does it exclude (ii) movement of a head to a specifer
position, where it would then project (this is licit since the chain
contains XMin at both positions). Next, Marquis reminds us that even
head-movement does not obey Chain Uniformity, and must be exempted by
Chomsky's (1995:322) word-independent invisibility; but then, why are
X-zero and XP movement even subsumed under the same operation Move?
Finally, Marquis shows that (iii) movement of an X' projection to a
specifier where it subsequently projects is also unattested (and notes
that any formulation of a principle of X' invisibility is directly
contravened by the step of moving a complex head in multiple head
movement). One generalization over unattested (though predicted by
Chomsky) syntactic objects (ii) and (iii) is that a moved object should
not be able to project at its target position. Marquis' solution is as
follows: dominance is a primitive asymmetric relation. When Y dominates
X, if (a) the features of Y = X, then X is non-maximal (XNMax),
otherwise (b) X is maximal (XMax). (Notice of course that these
definitions rely on identity, a fundamentally symmetric relation).
Having established these definitions, he proposes a Non-Equivocal
Requirement on syntactic objects: they must be XMax or XNMax in all
occurrences. (ii) and (iii) are immediately ruled out, as follows: if a
head or X' moves and projects at the target site, then the XP
dominating its base position is simultaneously an XMax (by virtue of
dominating the featurally identical base position) and an XNMax (by
virtue of being dominated by the featurally identical new projection at
the target site). Similarly, (i) and head-movement as adjunction are
ruled out, since in head-to-spec movement, the base position is an
XNMax, but the target position is an XMax. Réjean thus proposes that
complex-head-formation is achieved through head-substitution, whereby a
head H1 adjoins to another head H2, but a feature-union H1+H2 projects.
Under this implementation, H1 will retain its XNMax status in both
positions, and the H1P dominating the base position will remain an
XMax, since it its not featurally identical to the projection H1+H2P.
Having established a typology of attested syntactic objects through
these elegant means, Rejean proceeds to examine Mirror Principle
effects resulting from complex head formation in a few schematic
crosslinguistic patterns.
Chapter 4 (Abdelkader Fassi Fehri) "Synthetic/analytic asymmetries in
voice and temporal patterns"
Fassi-Fehri examines variations in phrase structure arising from
whether the Mood-Tense-Aspect-Voice-Verb chain is expressed as a single
inflected verb or distributed across a series of auxiliaries. He
notices that the choice between analytic and synthetic means of
expression is determined by properties of the complexity of the tense
and agreement system, is organized hierarchically from lower to higher
grammatical functions, and by the nominal character of the Agr
auxiliary. Contrasts between English, French, and Arabic perfects and
passives are analyzed in detail. While my own expertise in this area is
insufficient to provide a further detailed evaluation, interested
readers are encouraged to consult Fassi-Fehri's article for its fresh
look at connections between seemingly disparate variation in nuances of
morphological exprssion.
Chapter 5 (Eric Raimy) "Asymmetry and linearization in phonology"
Raimy's central hypothesis, developed in his thesis and book, is that
phonology allows multiple precedence: a segment can be in the
asymmetric relation immediately-precede with more than one segment.
Multiprecedence structures can lead to "loops", whereby the last
segment of a word immediately precedes the first segment of a word.
Linearization of a multiprecedence structure of this sort yields total
reduplication. The empirical consequences of Raimy's model yield
explanations for many phenomena, including over- and underapplication
(Raimy 2000), fixed-segment reduplication (Nevins & Vaux 2003),
multiple-reduplication, retriplication, and inherent reduplication
(Fitzpatrick & Nevins 2003). This chapter, however, focuses on details
of the linearization algorithm for simple cases. I will make a few
narrow remarks on the content of the chapter. First, Raimy claims that
precedence is "non-transitive", since vowel harmony across consonants
is a long distance effect, while strictly adjacent processes are
"atransitive" (p.130). This is an unfortunate formulation of the
problem; it seems to me, rather, that precedence is always transitive;
it is the particular phonological processes under discussion that may
be sensitive to immediate precedence, while others are sensitive simply
to transitive precedence. Second, for the benefit of the reader-to-be,
there is a potentially confusing typo on p.132 that says that
linearization occurs to eliminate "asymmetrical" precedence relations;
surely "symmetrical" is intended here.
For reasons of space, I cannot summarize all of the interesting aspects
of Raimy's proposal here (unfortunately, the majority of the field has
ignored Raimy's response to McCarthy and Prince's (1995) challenge that
overapplication cannot be handled in a derivational model). Raimy asks
the deep and interesting question why syntax and phonology are
different: in other words, why, when as the result of movement, syntax
creates two occurrences of an item, is only one pronounced, while in
phonology, when reduplicative multiprecedence creates two occurrences
of an item, both are pronounced? Raimy seeks an answer in the
definition of chains, which "require identity to be formed". But given
that the deepest parallels between syntax and multiprecedence
phonology, in my opinion, can be drawn with a syntactic model of
multidominance (Gartner 2002, Abels 2002), why make a comparison with a
theory of chains? Rather, it seems that the difference between syntax
and phonology is as follows: even with movement, there will never be
symmetric dominance: it is never the case that a moved item will
dominate itself. The reason is due to one simple fact about phrase
structure: projection. Moving XP to the root of the tree, it will not
project, hence never dominate its base position. There are thus no
dominance "loops" in syntax. Suppose that phonology must realize all
transitive precedence relations; then if a segment transitively
precedes itself, it must be pronounced twice. In syntax on the other
hand (assuming that covert movement does not participate in
linearization), if all pairwise transitive dominance relations must be
realized, only one copy will need to be pronounced.
Chapter 6 (Harry van der Hulst and Nancy A. Ritter) "Levels,
constraints, and heads"
This paper contains a broad discussion of architectural issues in
phonology theory: intrinsic ordering of constraints, levels of
derivation, and parametric variation. van der Hulst and Ritter make
very many provocative claims e.g., 1) allomorphy is not in the
phonology, it is "the result of phonetic interpretation" (p.151),
without any commitment to a theory of the latter; 2) the distribution
of internal superheavy syllables (e.g., in "mountain") is expected by
"the distinctions between L-words and L-clitic groups" (p.155). 3)
There are no repairs in phonology; the argument coming from the
rhetorical question "Are there also L-morphological or L-semantic
repairs?" (I suppose the answer is supposed to be an obvious "no", but
it's far from obvious that periphrasis, haplology, vehicle change, or
existential closure aren't "repairs".) 4) There is a postlexical
phonological level, where many of the inelegant things in phonology
(e.g., improper bracketing) occur: "In short, the PL system [a level of
representation that is by no means necessary, but vdH&R posit -- AIN]
is largely a terra incognita" (p.159). 5) "Phonological representations
do not express linear order"; perhaps linear order within the
constituents is predictable: onsets universally precede nuclei, but
there is absolutely no consistent way to predict that a given syllable
S1 should precede syllable S2 from higher-level relations. I should
mention that van der Hulst and Ritter present a number of interesting
ways of looking at Optimality Theory, e.g., that a number of
faithfulness constraints only serve the purpose of anti-repair. The
central asymmetry-related proposal is that phonological constituents
are asymmetrically headed; i.e., that complex onsets are internally
asymmetric.
This hypothesis has many exciting consequences (see, e.g. Shaw 1987 on
headedness in explaining many puzzling aspects of reduplication, or
Harris 1994 on headedness in explaining distribution within onset
clusters). van der Hulst and Ritter illustrate how the licensing
principle s of Government Phonology (Kaye, Lowenstamm, and Vergnaud
1990) insighfully explain closed-syllable shortening and vowel-zero
alternations in Yawelmani. The article contains 67 footnotes (actually,
66; the last one somehow escaped print), some of which I wish had been
elaborated in order to fully consider, such as fn10: "Later we suggest
that the set of phonological expressions may not be infinite, because
there is no recursion."(!),
Chapter 7 (G.L. Piggott) "Obstruent neutrality in nasal harmony"
In contrast to Piggott's (1992) feature-geometric account of nasal
harmony, he presents here a constraint-based account of the behavior of
voiceless obstruents in nasal harmony domains. One fascinating aspect
of the proposal is the suggestion that constraints can be ordered
according to the elsewhere condition. For example, the faithfulness
constraint MaxStop (which I think bears more resemblance to an Ident
constraint than a Max constraint), prohibiting nasalization of stops,
is ranked higher than NasHarm, enforcing nasal harmony, and this
ranking is due to the fact that the structural description for the
former constraint (stops) is more specific than the structural
description for the latter constraint (all segments). Piggott's 1992
paper was, in my own opinion, quite insightful; he claims its weakness
was the "failure to explain why voiced obstruents are never transparent
to nasal harmony" (p. 190). It's hard to see how the constraint-based
implementation, with MaxStop, gets any closer to this goal; on p.198
Piggott claims that the constraint does independent work in
characterizing static inventories: "since stops are present in every
language, MaxStop must be part of every grammar". But it is doubtful
that every instance of nonparticipation in spreading (e.g. the
transparency of sonorants to voice-spreading in Russian, summarized in
Steriade 1995) will have independent justification based on the
universal structure of inventories.
Chapter 8 (Charles Reiss) "Towards a theory of fundamental phonological
relations"
Reiss examines conditions on vowel deletion rules discovered by Odden
1988. One class of rules deletes a vowel only if flanking consonants
are identical. Reiss points out that there is no way to compute
identity of all feature values between two consonants given the current
technology of feature geometry, which does not make use of variables,
and introduces universal quantification as a means for computing
identity: for all features f(i), segment 1 and segment 2 have the same
value for f(i). Another class of rules deletes a vowel only if flanking
consonants are non-identical. Reiss employs existential quantification:
two segments are non-identical if there exists some feature f(i) for
which segment 1 and segment 2 have different values. Reiss's
introduction of existential and universal quantification are extremely
useful tools in formulation of structural description, and it is
important to realize that they actually limit they class of possible
conditions on rule application: if only existential and universal
quantification are allowed, we will not predict systems that delete a
vowel just in case any three arbitrary features (could be voice, place,
and continuancy OR stridency, voice and palatalization, OR ...) are
identical. In my own opinion, Reiss's decision to discard feature
geometry altogether, because it contains more expressive power than
feature-geometry, hence rendering the latter unnecessary, is unfounded.
To describe Yapese vowel deletion, which requires homorganicity of
flanking consonants, Reiss's structural description requires identity
of values for all features in the subset of features
{[coronal],[labial],[dorsal]}. Having to enumerate all of these
features as a conjunction misses the generalization that these are all
Place features, and not some other haphazard set. Reiss' formulation of
the conditions on Koya vowel deletion, requiring total identity except
for retroflexion, catalogues an even longer set of features, and misses
the generalization that the only feature missing is a secondary place
of articulation, a subdistinction among classes of features. The
hierarchical organization of features captures many generalizations
about the types of features that behave as a natural class, for all of
the reasons discussed by Clements (1985). The addition of
quantification is a useful limitation on structural descriptions, but
does not warrant rejection of the representational vocabulary used to
organize the features quantified over.
Chapter 9 (B. Elan Dresher) "Contrast and asymmetries in inventories"
Dresher begins with a useful illustration of the fact that logical
redundancy of feature specification is not enough to make that feature
redundant (and hence, subject to underspecification) in a particular
contrastive hierarchy. Consider two elements which are different for
all three relevant features: a small shaded square and a large clear
circle. Clearly any one feature is predictable given the other two. But
omitting every logically redundant feature results in no features at
all! More generally, the set of contrastive specifications depends on
the overall symmetry of the inventory. Dresher argues for a contrastive
hierarchy, in which each feature has relative scope over the rest, as
determined by an algorithm that performs successive division of the
inventory. He concludes with an illustration of how successive
contrastive specification in the 7-vowel inventory of Written Manchu
has the effect of underspecifying the back high vowels as labial. These
specifications precisely accord with the fact that these vowels do not
trigger labial harmony. Xibe, however, a descendent of Manchu which
lost the -ATR high back vowel, now needs Labial specification, and,
concomitantly, /u/ triggers labial harmony in Xibe. Dresher's approach
to contrast (part of a larger project:
<http://www.chass.utoronto.ca/~contrast>) within inventories appears
to have broad consequences for the relationship between syntagmatic
processes with the structure of the inventory, and possibly for the
acquisition sequence as well.
[Below, the sequence ",e" stands for "e-ogonek". --Eds.]
Chapter 10 (María Luisa Rivero and Magdalena Gol,edzinowska) "The
acquisition of constructions with reflexive clitics in Polish"
This article argues for a version of the Maturation Hypothesis (Borer &
Wexler 1992) to syntactic development: young humans have a proto-UG
with the Unique External Argument Proto Principle: every predicate
associates with a unique external argument. The evidence comes from the
omission of reflexive clitic "si,e" and the impersonal modal "wolno",
and the late development of extrinsic reflexives, reciprocals, and
impersonals, all of which involve more than one external argument.
Developmental data comes from five children in the CHILDES database.
Chapter 11 (David Lebeaux) "A subgrammar approach to language
acquisition"
Lebeaux models acquisition as the addition of Licensing relations to
the grammar, rather than a process of parameter-setting. His example is
conjunction, the acquisition of which is, in his model, the addition of
a licensing relation (a generalized transformation) among the
conjuncts, and "against Merge as a monolithic structure building
operation" (p. 289). Lebeaux argues that licensing relations are added
monotonically, with theta-subtrees coming in before X'-theory, which
comes before Project-alpha, a transformation that takes trees composed
of open class categories and projects them into a closed class frame, a
pure representation of Case. The phenomenon of telegraphic speech in
young children is thus before they acquire Project-alpha. Additional
evidence for Project-alpha (or, more generally, Case structure as a
distinct level of representation from theta structure) is mustered from
studies of speech errors (Garrett 1975) such as "my frozers are
shoulden" in which open class items are permuted while closed class
items remain in place (apparently, the -ers in shoulders is a
functional morpheme). One interesting benefit of the licensing analysis
is that it makes predictions about the types of two-word utterances
encountered in child language: phrases such as "Mommy sock" [meaning
either "Mommy's sock" or "Mommy puts on the sock"] are more common (no
actual counts are provided) than "Mommy Kathryn" [meaning "Mommy puts
on Kathryn's sock"], because two theta-subtrees are involved.
REFERENCES
Abels, Klaus. 2002. Move? Generals Paper, UConn.
Barker, Chris. 1998. Episodic -ee in English: A thematic role
constraint on new word formation. Language 74.4.
Borer, Hagit & Ken Wexler 1992. Bi-unique relations and tha maturation
of grammatical principles. Natural Language and Linguistic Theory
10:147-189.
Chomsky, Noam. 1994. Bare Phrase Structure. MIT Occasional Papers in
Linguistics.
Clements, G.N. 1985. 'The geometry of phonological features'. Phonology
Yearbook 2.
Gaertner H.-M. 2002. Generalized Transformations and Beyond:
Reflections on Minimalist Syntax.
Garrett, M. 1975. The analysis of sentence production. In G. Bower
(ed). Psychology of Learning and Motivation Vol. 9
Harris, John. 1994. English Sound Structure. Blackwell.
Fitzpatrick, Justin & A. Nevins. 2003. Linearizing Nested and
Overlapping Precedence in Multiple Reduplication. Penn Linguistics
Colloquium 27 Proceedings.
Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. 1990.
Constitutent structure and government in phonology. Phonology 7.2.
Ko, Heejeong. 2002. On the Origin of Why-In-Situ. Paper presented at
the Worlshop "On Wh- Movement", Leiden.
McCarthy, John and Alan Prince 1995. Faithfulness and Reduplicative
Identity. UMass Occasional Papers 18.
Nevins & Vaux 2003. Metalinguistic, Shmetalinguistic: The Phonology of
Shm- Reduplication. CLS 39 Proceedings.
Odden, David. 1988. Antigemination and the OCP. Linguistic Inquiry 19.
Piggott, G.L. 1992. Variability and feature-dependency: the case of
nasality. Natural Language and Linguistic Theory 10.
Raimy, Eric. 2000. Remarks on Backcopying. Linguistic Inquiry 31.
Shaw, Patricia. 1987. Non-conservation of Melodic Structure in
Reduplication. In CLS 23 Volume 2.
Steriade, Donca. 1995. Underspecification and markedness. In Handbook
of Phonological Theory, ed. Goldsmith,1995:114.174.
|
| |
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Andrew Nevins is a graduate student at MIT interested, among other
things, in the details of representations in linguistic theory.
|
|
|
|
|
|
