[Moderator's note: Although this review is written by a moderator of the
LINGUIST LIST, it reflects only the personal opinion of the author. It does
not represent the opinion of the other LINGUIST moderators or official
LINGUIST policy. The official editorial position of the LINGUIST LIST on
issues of theoretical controversy is one of total neutrality.]
Newmeyer, Frederick J. (1998) Language Form and Language Function.
MIT Press. Cambridge MA. 428pg.
Reviewed by Andrew Carnie, University of Arizona
In late September of this year, I had occasion to attend our annual
LINGUIST list meeting in Ann Arbor. After many hours of meetings about
how we are going to organize next year's tasks, I -- along with the other
moderators and some other colleagues -- went out for dinner at a fashionably
expensive restaurant. As happens when a group of linguists get together, the
topic of conversation quickly turned to matters of theory. After the main
course was served and we had drunk far too much wine, one of my
colleagues (who I will leave anonymous) took to his favorite game of
provoking me alternately on politics and the nature of linguistic theory. We'd
been talking about a project I've been working on the UG-driven properties
of verb initial languages. My colleague, let's call him B, who is an avowed
(historical-) functionalist, started with the jab that "you so-called
syntacticians don't know what an empirical fact is. All you ever do is probe
obscure and irrelevant 'acceptability' judgments that have nothing to do with
how language is actually used." I growled my deepest growl and the
subsequent debate went something like the following:
AC: "B, you're insane. That isn't at all what we do. Chomksyans are devoted
to empirical arguments. The ideal set of data for linguistic theory includes
both sentences heard in every day use AND information about our basic
competence in language."
B: "Exactly, Andrew, your problem is you all are obsessed with this fictional
'linguistic competence'. Functionalist theory is about real language."
AC: "So is generative grammar, in fact we don't limit ourselves the way you
all do. And by the way, 'functionalist theory 'isn't a theory at all. A theory
needs to make predictions. As far as I can tell, the 'predictions' made by
functionalist aren't anything more than simple descriptive generalizations."
B: "What is so simple about descriptive generalizations anyway? [Waving a
forkful of curried chicken provocatively close to my face] Your problem is
that you people attempt to psychologize grammar, but then ignore all the
results from psychology."
AC: " You are out of your mind [I banged the table, spilling some of his
wine], are you completely ignoring all the work in language acquisition done
by people like Crain and Pinker."
At this point, another colleague, the only diplomat among us, noticed the
reaction of the waiters hovering nearby and the alarmed looks on the faces of
our neighboring restaurant patrons. We were getting a bit loud. She piped in:
C: "Um, boys, perhaps we should talk about something a little less
controversial, like Kosovo, or abortion?"
To which the two of us, shame-faced and cowed, busied ourselves with our
dinners.
The above conversation really happened. Fritz Newmeyer starts out
his book on the nature of the functionalist/formalist debate with a fictional,
but very similar discussion between two graduate students at the LSA.
Perhaps what both the fictional discussion and the real one highlight for us is
the basic failure among linguists of different stripes to communicate with
each other. I came away from my discussion with my colleague with the
impression that we really hadn't understood what the other was saying. We
were using the same terminology, but they didn't mean the same things to us,
precisely because of our differing theoretical orientations. Newmeyer's book
(henceforth N), has gone a long way towards helping me understand what
the debate is about and why functionalists view things the way they do. In
particular, it has helped me understand the importance of a lot of the results
of functionalist approaches to linguistics, and better appreciate the work they
do. In the first chapter, N claims that formalists need to take heed of the role
that meaning(/function) plays in determining form. However, N has also
strongly confirmed my belief that Chomskyan generative linguistics is on
the right track, and that the functionalist criticisms of formalist approaches
are either based on misunderstandings of what formalists are saying, or on
simple factual errors. Through out the book, N points out generative (and
other formalist approaches) are more often than not quite consistent with the
assumptions made by functionalist linguists. He argues quite convincingly
that the basic tenets of formalist grammar are not only conceptionally well-
founded, but are empirically justified as well.
Although this book is a firm and resounding endorsement of formalist
(or more accurately, generative) approaches, I think it bears the more
important moral that we really need to not only talk at one another, but to
*listen* to one another as well.
Synopsis.
The first chapter of N is a lucid scorecards of who's who in linguistics.
The first problem to be dealt with is the puzzle of what sets of assumptions
identify one as either belonging to the functionalist or formalist camp:
"One orientation sees as a central task for linguists characterizing the
formal relationships among grammatical elements independently of any
characterization of the semantic and pragmatic properties of those elements.
The other orientation rejects that task on the grounds that the function of
conveying meaning (in its broadest sense) has so affected grammatical form
that it is senseless to compartmentalize. It is the former orientation, of
course, that I have been referring to as 'formalist' and the latter as
'functionalist'." (N:7)
Within these broad categories there is a wide range of variation in the
precise sets of assumptions being made. N explores the range of theories that
seem to fit the formalist camp (GB, Minimalism, HPSG/GPSG, LFG, and to
some degree RG), and the relatively minor differences between them with
respect to the formalist/functionalist debate. He also points out the bizarre
attitude among some functionalists to view certain varieties of formalist
grammar (such as HPSG or Categorial Grammar) as being more congenial to
functionalism than GB/minimalist approaches (presumably because such
views share a non-transformational/unificational approach to language.) I
wasn't surprised to hear this about functionalists, but it continues to strike
me as odd, because among the "formalist" approaches to grammar,
proponents of HPSG/GPSG/CatG seem to me to be the most "formal" of all
(see for example Pullum's (19XX) NLLT topic/comment on how fuzzy
Chomskyan syntacticians have become.) If anything, despite its
transformational nature, I think minimalism, especially in light of recent
work on motivating transformations, is in fact the most functionalist of the
formalist theories (more on this below in the Critical Evaluation section).
The spectrum of functionalist approaches seems to be much broader
than that of formalist grammar. N adopts Croft's (1995) typology of
functionalist theories. At one extreme, perhaps the view closest to
formalism, is External Functionalism. This includes such theories as Role
and Reference Grammar, Dik's Functional Grammar, Systemic Grammar,
the Competition Model, Construction Grammar, and Cognitive Grammar.
These approach, while rejecting the autonomy of syntax, maintain that there
is a systematic (and often formalizable) grammatical system where semantic
and pragmatic elements are linked to syntactic ones.
The next class of functionalist theories includes the Emergent
Grammar approach of Hopper (1987), and various unnamed "functional-
typological" approaches. Croft labels these as Integrative functionalist
approaches. N observes that such approaches "deny the Saussurian dictum
that it is meaningful to separate langue from parole and synchrony from
diachrony" (N:16). In essence, the Integrative approaches deny that the
cognitive system is self-contained, and that social and historical factors
interact with it to an extremely high degree. I have to admit that I still don't
understand this approach. Perhaps it is my MIT training showing through
and blinding me to the obvious, but I simply fail to see how it is at all
possible that a two-year old child has direct access to diachronic influences
like OE word order or the great vowel shift. As far as I can tell, without
time-machines or university degrees, infants only have access to what they
hear spoken around them, which makes this approach psychologically
incoherent. While I don't understand their philosophical approach to
grammar, I have a healthy respect for the descriptive empirical work that so-
called functional-typologists do. I have often made use of their grammars
and their putative universals in my own work. As we will discuss below, N
takes a slightly more cynical view of typological research.
At the far end of the functionalist spectrum lies theories likes those
developed by the 'Columbia School', and by work by Kalmar. Croft calls
these Extreme Functionalists. As N observes, these theories are so extreme
in their functionalism, that it is hard to imagine, even if you have
functionalist bent, how they could possibly be correct. N states "Advocates
of this approach believe that all of grammar can be derived from semantic
and discourse factors � the only 'arbitrariness' in language exists in the
lexicon." (N:17). There is so much empirical evidence that such a claim is
false, N chooses not to bother arguing directly against them. For example,
Hudson (1996) observes such examples as the arbitrary difference between
'likely' and 'probable' with respect to raising constructions:
1) a. He is likely to be late
b. *He is probably to be late (N:28)
In the second chapter: The Boundaries of Grammar, N addresses the
question of Autonomy in grammar. Chomsky and his followers have often
been criticized for their adherence to three autonomy hypotheses (I have
paraphrased N here):
1) The autonomy of syntax (AUTOSYN): there is a system of
primitive non-semantic/discourse defined terms whose combination makes
no reference to system external factors.
2) The autonomy of knowledge of language (AUTOKNOW):
Knowledge of language (competence) is different from language use
(performance).
3) The autonomy of grammar as a cognitive system (AUTOGRAM):
there is a cognitive system of language that is distinct from other cognitive
systems.
N argues that all three are correct, and that to some extent, the criticisms laid
at the door of such hypotheses are built on misunderstandings of their
content.
N again draws on the insights of Croft 1995, who observes there are
three more primitive assumptions (or claims � depending upon who you are
talking to) that underlie AUTOSYN. These are (i) Arbitrariness, (ii)
systematicity, (iii) self-containedness. Of these really only self-
containedness and systematicity are controversial.
Functionalists have long criticized Chomskyan grammar of ignoring
the clear interactions between meaning and form. Ironically, these criticisms
do not appear to be based in fact. As N observes, Chomsky has always
advocated as the main task of linguists to be fleshing out the points of
contact that exist between form and meaning. For example the theories of
thematic roles, Jackendoff's conceptual structures, or Levin-style argument
structures, are precisely the kinds of interactions where the meaning drives
the form of a sentence. Most recently, in the minimalist program, differences
between languages in terms of syntactic structure is thought to derive from
purely lexical-semantic properties of those languages. It is also worth noting
that, despite functionalist claims to the contrary, there is a long standing
tradition in generative linguistics of the study between syntactic structure
and semantic/pragmatic principles. For example, Diesing, Jelinek, Kroch,
Prince, Heim and many others focus their research on precisely the
interaction between syntactic and semantic/pragmatic form. Jelinek for
example, has long argued that word order in Lummi Salish is a result
between a mapping between the information structure of an utterance and
the formal syntactic structure that underlies it.
At least some formalist approaches then, are not purely self-contained.
N claims that what *really* underlies the differences between functionalists
and formalists on the question of AUTOSYN lies in the domain of
systematicity. One believes in AUTOSYN if one believes that "the
relationship between purely formally defined elements is so systematic that a
grammar should accord a central place to formalizing the relationship among
these elements without reference to their meanings or functions."
N goes on to defend this view with a variety of conceptual and
empirical arguments. In particular, he discusses the role of intuitions, the
fallacy of assuming that frequency of utterance type determines some kind
of 'basic nature', and the confusion Givon rain upon the question by mixing
up explananda and explanans. N points out the longstanding Chomskyan
claim that "S speakers have some internal principles that lead to S
judgments. Surely the study of such principles is logically prior to the study
of how the structures resulting from the principles are actually put to work in
the discourse" (N:43). In the interest of keeping this review to a reasonable
length, I can't go through all the fascinating discussion in this section, I will
however, summarize one of the empirical arguments that N raises in favor of
AUTOSYN. He observes first that there is not necessarily a tight link
between a grammatical construction and discourse function. For example,
Subject/Aux inversion can be found with questions, requests, offers, and
exclamations:
2) Can you take KLM from Seattle to Amsterdam
Could you please pass the salt
Can I help you
Is syntax ever easy (based on N:49)
Within the realm of 'canonical' uses, Subject/Aux inversion is found in main
questions (but not embedded ones), in wh-constructions, after preposed
negative adverbs, with bare subjunctives, and after preposed so clauses:
3) Have you been working late
What have you been eating
Under no circumstances will I take a day off
Had I known the dangers, I would have kept my distance.
So tall is Mary, she can see into second story windows
(modified from N:47-48)
Note further that these are all tokens of the same phenomenon, as shown by
the fact they are subject to the same idiosyncratic formal constraints, such as
the fact that none may occur in embedded contexts
4) *I asked had you been working late
*I wondered what had you been eating
*?I think that under no circumstances will I take a day off.
*I think that had I known I the dangers S
*I'm sure that so competent is Mary, she will get the promotion.
A semantic or pragmatic account of this uniform behavior is greatly
challenged by the fact that there appears to be no uniform semantic
properties that these constructions have in common. A formal approach, by
contrast, while arbitrary, runs into no such difficulties. N raises similar
arguments from Wh-constructions, and lexical government which also show
a one-to-many link between form and function, thus providing
straightforward support for AUTOSYN. Since we can't characterize the form
in terms of the function, the form must be to some extent autonomous of the
meaning.
Turning now to AUTOKNOW, the principle challenge to this
hypothesis comes from the Emergent Grammar literature. Hopper (1987)
describes the basic philosophy of this approach as "Language is, in other
words, to be viewed as a kind of pastiche, pasted together in an improvised
way out of ready-made elements." (Hopper 87:144, cited in N:59). N rejects
Hopper's claim that parts-of-speech categories reflect prototypical discourse
functions, rather than formal categories (see also the discussion of chapter 4
below). He adapts an analogy from Jerrold Sadock and says "an emergent
grammarian is like an anatomist who, realizing that birds can fly, loses all
interest in the structure of their wings" (N:63-64). N discusses at length
theories of discourse-grammar interactions based in formalist systems that
are wholly consistent with the AUTOKNOW hypothesis. Following Prince,
he observes that it is consistent to claim that there is a discourse component
of the grammar, which itself is subject to the competence/performance
distinction. N reviews the arguments from Historical change, creolization
and L2 acquisition, and comes to the conclusion that none are inconsistent
with AUTOKNOW.
On the surface, the claim that that the cognitive principles that govern
other aspects of human behavior also govern language is extremely
appealing. It allows linguists true status as participants in that
interdisciplinary beast we call "Cognitive Science." Many cognitive
grammarians (in particular Lakoff) make the claim that language is simply
part of larger cognitive system. So called grammatical principles reflect
larger cognitive abilities like memory or learning algorithms. This
constitutes a putative denial of AUTOGRAM. N observes that many
functionalists (including for example Givon), explicitly adopt some form of
AUTOGRAM. N soundly trounces Lakoff (1991) for simply
misrepresenting the "cognitive commitment" of Chomskyan linguists.
Lakoff claims that generativists reject a view of linguistic theory situated
within cognitive neuropsychology. N shows that what Lakoff says is simply
wrong. N cites Chomsky (1975) here: "A grammar is a cognitive structure
interacting with other systems of knowledge and belief".
N also deals with the question of arguments for the innateness of
AUTOGRAM knowledge. In particular, he focuses on the questions of
poverty of the stimulus. While this discussion is interesting, I think it is his
basic conclusion that is the most telling. He observes that innateness is *not*
an assumption underlying AUTOGRAM, but rather is a conclusion of
generative grammar. AUTOGRAM can stand independently of one's beliefs
about innateness. Finally, N reviews the literature on Specific Language
Impairment (SLI), and shows how the existence of such a phenomenon is
straightforward evidence in favor of AUTOGRAM.
In chapter 3, N turns to the question of what constitutes an
explanation in functionalist and formalist grammar. N distinguishes internal
explanations (such as the case filter or other formal grammar-internal
devices) to external explanations (such as iconicity, parsing, economy,
innateness, discourse flow, prototypes, metaphor, "playfulness", and text
frequency). At first blush, it may appear as if external explanations are the
domain of functionalists, and internal explanations that of formalists. N
concludes that this is too easy a classification. He shows that many
functionalists assume internal explanations. More importantly, despite the
blatant misrepresentations (or misunderstandings) of some leading
functionalists, generative grammar is open to external explanations as well.
In particular, it is open to such external properties in explaining the historical
(or evolutionary) origins of certain constraints and processes in the grammar.
For example, he notes that Chomsky (1973) suggested that the origins of the
bounding and binding theory may lie in a perceptual strategy that links
predicates to their arguments. Although N only mentions it briefly, more
recent Minimalist approaches to grammar are almost entirely motivated by
external factors such as economy, and the pressures put on the grammar by
the phonological and semantic components. N's book was published before
another important book brought out by MIT press, Juan Uriagereka's (1999)
_Rhyme and Reason_ . Uriagereka's work on the reason why such external
pressures occurs bears greatly on these questions and ties in nicely with N's
work.
Chapter 3 also deals with two other important questions on the nature
of explanation. First, N evaluates external explanations with respect to (i)
whether they can be precisely formulated, (2) can identify a linkage between
cause and effect, and (3) show measurable typological consequences. Of the
various putative external explanations, he shows that only two (parsing � in
particular the Early Immediate Constituents (EIC) analysis of Hawkins
(1994)<and somewhat to my surprise, iconicity) meet these criteria.
Second, he evaluates the claims that it is the interaction of external pressures
that gives rise to typological phenomena. He concludes that due to the open
ended nature of such explanations, there is no way to include them in a
synchronic grammar. Instead, he shows, that while remaining entirely
consistent with generative grammar, such factors can explain the
evolutionary (or historical) origins of certain linguistic phenomena. An
important moral, to my mind, lies in the generalizations developed by
functionalists about the link between iconicity and syntactic form. While
such generalizations are more often than not simply ignored by generative
grammarians, N speculatively shows how they might be incorporated into a
view with autonomous formal syntax. Having never been exposed to this
literature (the fault of which lies at the feet of both my teachers and my own
ignorance), I never would have been made aware of these very real
generalizations if N hadn't brought them to my attention.
The next three chapters of _Language Form and Language Function_
deal with arguments that functionalists have pitted against a formalist (and
primarily generativist) approaches to grammar.
Chapter 4 takes on the question of whether syntactic categories
constitute formal discrete algebraic entities (as is assumed by any formal
approach), or rather stand on continua, based on some notion of semantically
driven prototypes (as claimed by many functionalists). N examines the
claims of 'syntagmatic simplicity' (the claim that prototypical words of a
particular category are usually the most unmarked morphologically) and
'paradigmatic complexity' (the claim that prototypical words of a particular
category will show the greatest range of inflectional possibilities.) He claims
that syntagmatic simplicity is not inconsistent with a discrete category
approach -- as seen in the default Canonical Structural Realizations of Pinker
1984 -- and that the data does not support 'paradigmatic complexity' in any
but its weakest form. N shows that if you have a theory of markedness, then
there is no need for a theory of prototypicality. He shows that the so called
"fuzzy" cases of syntactic categories, such as the non-prototypical behavior
of expletive 'there', may well following from other semantic or pragmatic
properties rather than representing evidence that syntactic categories are not
discrete.
The phenomenon labeled "grammaticalization" (the idea that certain
unidirectional shifts occur towards more functional items over time) is often
touted as evidence in favor of functionalist approaches, as it is something
that is not easily accounted for in a purely synchronic generative approach.
N claims that grammaticalization is not really a theory, instead he claims
that "none of the mechanisms they [Bybee et al] propose to explain
[grammaticalization phenomena] S are specific to grammaticalization."
Instead he claims that grammaticalization is a truly an epiphenomenon due
to factors that any theory of diachronic changes (functionalist or formalist)
must posit anyway. He shows that all aspects of grammaticalization are
attested independently of one another and that unidirectionality is not-
empirically motivated (in fact it is contraindicated by the data).
Finally in chapter 6, Newmeyer casts his critical eye on the field of
crosslinguistic typology. He observes first that despite functionalist claims to
the contrary, linguistic typology is of interest to both functionalists and
formalists, particular given recent advances in parametric variation in
principles and parameters syntax. N is highly critical of the whole
typological enterprise. He has doubts about the shape of the typological
database, both in terms of the sample of languages as being in any way
representative of the set of *possible* human languages, and in the fact that
such material suffers from being overly reliant on secondary sources which
themselves are subject to the biases of the initial researcher. He observes for
example a great inconsistency in the criteria used to determine "basic word
order", even among languages inside a single sample. He is suspicious of all
but the most robust generalizations. This trouble, he claims plagues both
functionalist and formalist approaches to typology.
Critical Evaluation.
N closes his book with the following conclusions, which he claims are
both correct and consistent with one another:
"1) The grammatical properties of human language are best
characterized in terms of autonomous formal systems.
2) The grammatical properties of human language have been shaped
by external pressures. " (N:365)
I think N succeeds in proving both of these points in this book. As such, this
is a book that linguists of all theoretical stripes should read. The moral is
well-taken, the inward-looking attitude of many generativists and the
dismissive attitude of many functionalists are equally inappropriate. The two
approaches are not as far apart as we would like to believe and we should
pay attention to each other's results, in both directions. Having said this, I am
sure this is a book likely to anger many functionalists. I am not in a position
to evaluate the accuracy of N's representation of functionalist theories, but
he is soundly critical of all but the most robust of their claims and
generalizations. It definitely supports a formalist/generativist point of view.
There was one area in which I was greatly disappointed, which lies in
N's presentation of principles and parameters in almost exclusively GB
terms. With a few minor exceptions here and there, N ignores the Minimalist
approach to grammar (MP). I think this was a grave mistake on N's part,
because if anything, Minimalism is a theory that is more amenable to
functionalist thinking than GB and greatly supports his contention that
language is both an autonomous formal system and subject to external
pressures. In fact, it is a tenant of the MP that syntax is a "perfect" system,
and that differences in word order, grammatical marking, etc are *all* due to
external pressures (such as semantic and pragmatic criteria.) Some recent
theorizing is even consistent with the notion that there are no discrete
syntactic categories (only formal (phi) and semantic/lexical (lambda)
features). In Bare Phrase Structure, Chomsky claims that there is no such
category as Noun or Noun Phrase, rather, the relevant features of a word are
projected to the next higher constituent level, thus we are able to contrast the
properties of phrases headed by non-prototypical nouns like "there" and
more prototypical ones like "dog". These structures may well have different
structural properties.
I started this review with the story of a dinner, at which I and one of
my colleagues battled over these discipline dividing issues. I suspected that
while we were debating, neither of us really understood the position of the
other. Having now read N's book, I think I have a better understanding of my
colleague's opinions. This doesn't mean that our dinnertime battles will be
any less vocal or contentious, but at least I'll be able argue with him without
spilling any more of his wine.
|
