syntax/semantics, form/ meaning

[David_Tuggy at SIL.ORG]

I played hookey for a few days from my post in lurkitude (I liked that word!),
and then came back to find that the Funknet had come alive! Seems like a fair
bit of heat and maybe even a decent amount of light has been generated. Not sure
I've absorbed all the light (or even all the heat) that might do me good, but
thought I'd add my two bits' worth anyway. The first part of this is trying to
understand what others have said in this discussion: I'd appreciate being
straightened out if I've gotten it wrong. Also, I'm sorry it runs on so long,
but I think slowly ...

Parsers were interesting, but the questions that most seem to have vexed
people's souls are two closely related ones (if they're not in some sense the
same one):  Is syntax autonomous, particularly from semantics?  Can/should we
study linguistic forms (structures) without reference to their function or
meaning?

As usual, the answers depend on the definitions you give to the technical terms
and the assumptions or presuppositions on which those definitions are based.

After some initial sparks, Tom Giv¢n and George Lakoff seem to have settled it
between them that neither of them really denies that there is such a thing as
syntax, and they are in agreement that it is not autonomous. I don't know
anybody else who denies that syntax exists, but Fritz Newmeyer and others are
ready to argue that it is autonomous.  The form/function thing came up in that
context, and just about everybody seems happy to say "Yes, there is form and
there is function, and they are tied together, but not so closely as to make
them indistinguishable."  And everybody seems to agree that both are worth
investigating, but they're not all in agreement on whether there's any point in
(or possibility of) investigating form without investigating function at the
same time. In other words, are syntax and form separate or autonomous enough
that we can profitably treat them on their own terms without bringing in
functional or semantic information?

The crucial definitions and underlying assumptions include "syntax",
"semantics", "form", "function", and "autonomous". We assume we know what each
other means, and build arguments, or appeal to analogies with other sciences
such as biology, all of which makes sense given our views of these key words but
those analogies and arguments will not be as forceful to someone who means
something different by them.

A standard position (I think) is that "semantics" is "truth-conditional
semantics", and syntax includes word-order or phrase-order information,
constituency information, "grammatical class" information, "grammatical
relations", gender and agreement phenomena, and such-like. All of these things
are considered non-semantic in nature.  In doing syntax you pay attention to
semantics only enough to let you know if two structures (usually sentences) are
synonymous (i.e. have the same truth conditions) or not, but basically you
ignore it. As Dick Hudson put it, syntactic generalizations "refer to words,
word-classes and syntactic relations, without mentioning meaning (or
phonology)."

So what does it mean for syntax to be "autonomous" from semantics? For Fritz it
seems it is enough to prove that some generalizations about syntactic patterns
follow from other syntactic patterns or primitives: "A[utonomous]S[yntax] holds
that a central component of language is a *formal system*, that is, a system
whose principles refer only to formal elements and that this system is
responsible for capturing many (I stress 'many', not 'all') profound
generalizations about grammatical patterning." Similarly for Dick Hudson "The
generalisations that distinguish auxiliary and full verbs are `autonymous'[sic],
in the sense that they refer to words, word-classes and syntactic relations,
without mentioning meaning (or phonology)."

For Tom Giv¢n and some others it seems to be enough to prove that some
generalizations about syntactic patterns follow from something other than
syntactic patterns or primitives to show that syntax is *not* autonomous. Thus
Tom says "Grammar is heavily motivated by semantic/communicative functions. But
-- ...it never is o100% so. It acquires a CERTAIN DEGREE of its own independent
life. This, however, does not mean 100% autonomy."  Yet Tom agrees with George
that syntax is non-autonomous.  Wally Chafe's position is similar, I think (I
loved this posting, Wally): "My own view is that the stuff of which syntax is
made--its elements and the constructions into which they enter--is either
functional stuff (and a great deal of it most certainly is), or consists of the
fossilized (or partially fossilized) remains of things that were functional at
an earlier time."

The positions seem pretty close to me: they are separated by the definition of
autonomy. For the AS people anything that's not explainable by semantics
/pragmatics/ function/non-linguistic cognition/etc. proves autonomy: for the
Giv¢n-functionalists anything that is proves non-autonomy. Both agree that some
important generalizations are "syntactic", not to be accounted for by
semantics/pragmatics/ function/etc. Of course they differ as to how much is to
be accounted for in which way, and in the importance they assign to what is
accounted for in each way, but they seem to be playing basically the same game,
or at least on the same field.

I guess I'm agreeing with Nick Kibre: "Ultimately, it seems that the autonomous
syntax and and functionalist/cognitive position are more edges of a continuum
than strictly opposing viewpoints. Nearly veryone agrees that language is shaped
both by innate cognitive mechanisms, at least partially specialized for
linguistic function, and by the demands of usage; our only point of disagreement
is how tightly the former constraints the range of possible systems, and how
much regularity is due to the pressures of the latter."

And form? I think both these camps basically agree that what is not predictable
from semantics/pragmatics/function/cognition/etc. is formal.  Whatever is
ossified, fossilized, so just because that's the way speakers of this language
do it and not because of some semantic/pragmatic/etc. necessity, is formal.  And
syntax is the major domain of such formality, I think, for both. (The lexicon
too, I suppose, though I'm not as sure.)

Or am I oversimplifying?  Matthew Dryer wrote that "autonomous syntax" means for
some (1) something about innateness (which I don't want to talk about), and "(2)
one can explain syntactic facts in terms of syntactic notions (3) syntax/grammar
exists (although that too can mean different things)"  "Arguments for the
autonomy of syntax," Dryer continues, "(such as some offered in print by Fritz
Newmeyer) often involve no more than arguments for (3). For me (and I assume
that this was what both George and Tom meant), rejecting autonomy of syntax
involves rejecting (1) and (2)."  Matthew, are you saying that no syntactic
facts can be explained in terms of syntactic notions, or that not all can, or
what? And although I hate to bring any division between them when they've made
up so nicely, I'm not sure at all that George and Tom both mean the same thing.

It seems to me that the position that George alluded to and somewhat described,
the one he and Ron Langacker share (and which it suits me to work from), is
different from what's been said so far. Under this view, syntax is
non-autonomous in a much more radical sense. But a rethinking, i.e. different
definitions, of the basic concepts, is necessary.  Semantics is not limited to
what truth-conditions show us, but extends to virtually any kind of cognitive
activity: it includes all kinds of "imagery" and "construal" factors, degrees of
prominence of parts of a concept vis-a-vis each other or of one concept
vis-a-vis another, attitudinal (emotional) information, etc. Significantly, it
includes relations of one concept to another (e.g. the relation of an actor to
the act he performs can be semantic).  Semantics and pragmatics, to the extent
that it is useful to distinguish them, differ only in degree, not in kind, and
for most purposes what others call "pragmatics" is included as part of
semantics. Categories are not expected to be classical, airtight, all-or-nothing
compartments, but rather are organized around prototypes, with fuzzy borders and
surprising extensions.  Any concept of these sorts becomes *linguistic* (i.e.
part of an established *language*) by being cognitively "entrenched", i.e.
routinized, ossified, fossilized, in some degree, and "conventionalized", i.e.
shared and known to be shared by a community of speakers. Although I don't
remember George or Ron using the word in this way, "formalization" is what the
other folk seem to be calling this process, and I think we can usefully make the
connection.  "Formalization", then, is a function, and a very important one: we
couldn't talk coherently to ourselves or to others without it, and all
linguistic structures undergo it to some degree.

"Formalization" in this system does not necessarily mean arbitrariness, however.
What is entrenched typically is entrenched because it makes sense, it functions
well.  In general there is claimed to be a gradation between the fully arbitrary
and the totally predictable, with most language phenomena in between, in the
"motivated" or "reasonable" part of the cline.  Some linguistic patterns may in
fact hold *only* because "that's the way we do it"; most also have some element
of "we do it because ...". In either case, they are established, "formal"
patterns. As Ron puts it, it is important to distinguish between what kinds of
structures there are (only semantic, phonological, and symbolic, says he), and
the predictability of their behavior.  Few if any are fully predictable apart
from "that's just the way we do it"; even if that is the only motivation left,
the structures do not cease to be semantic, phonological, or symbolic.

Phonology, under this framework, is also a subpart of conventionalized,
entrenched cognition, namely the part that deals with the motor and auditory
routines that constitute our pronunciations and perceptions or linguistic
sounds. Included in it are the timing relations that constitute the order of
production of phonological structures.

(Note that much non-linguistic cognition can also become entrenched and even
conventionalized: e.g. the motor routines of driving a stick-shift car
constitute a quite complex and flexible system that is, in this sense,
"formal".)

Lexical items, reasonably uncontroversially, are claimed to have a semantic
structure (i.e. for L&L a non-classical category of entrenched conventionalized
cognitive routines) which is (again by an entrenched, conventionalized cognitive
routine) linked in a symbolization relationship to a phonological structure.
These structures, and their symbolic relation, are "formal" from the start.  So
proving that something is "formal" doesn't make it non-semantic or
non-phonological, requiring us to make a third category of the "syntactic".

And in fact, the claim is that syntactic constructions, of whatever degree of
complexity or abstraction, are basically of the same ilk as simple lexical
pairings. They differ from them only in degree, not in kind; consisting, as
lexical items do, in the symbolic pairing of a semantic with a phonological
structure. Syntax thus is non-autonomous not in that some of it can be accounted
for by non-formalized stuff: it is non-autonomous in that all of it (so the
claim is) can be accounted for by the same sort of formalized cognition that
constitutes semantics and phonology.  Wally Chafe's words come back to mind:
"the stuff of which syntax is made--its elements and the constructions into
which they enter--is either functional stuff (and a great deal of it most
certainly is), or consists of the fossilized (or partially fossilized) remains
of things that were functional at an earlier time." George and Ron would say "it
is semantic, phonological or symbolic stuff. In some of it some functional
motivation besides pure convention can be seen: some of it may indeed be
fossilized to the point where that is about all that is left, but it is still
the same kind of thing."

The arguments I remember being given for the non-semantic nature of the
syntactic "stuff" were based on two ideas: the notion that semantics is only
truth-value stuff (and in fact pretty nearly only real-world referential stuff),
and the notion of classical, all-or-nothing categories.  Thus most of us were
told in our first-year grammar or syntax courses, "Yes, you might think there
was a semantic basis for, say the 'noun' category, and indeed a word designating
a person, place or thing usually is a noun. But some nouns name actions, and so
the definitions don't work." Since semantics (i.e. truth-conditional semantics)
can't predict 100% of the cases, the category is non-semantic: what else could
it be but syntactic?  I.e. the only thing that will do to define it is the way
it interacts with its linguistic neighbors." But if categories needn't be
all-or-nothing, the argument fails.  For surely, if you allow prototype-based
categorization,  the category 'noun' prototypically does in fact mean something
pretty near to the traditional "person, place or thing", and other cases, even
those denoting actions, though correctly seen as non-prototypical, are pretty
clearly relatable to this prototype. The same works for 'verb' and other
categories (See e.g. Ron's 1987 "Nouns and Verbs" in Lg. 63.53-94; he actually
goes beyond this and proposes schematic concepts that do pretty well at covering
all the cases, claiming e.g. that we conceptualize an action differently when we
refer to it via a noun or a verb). The relations between nouns and verbs fit in
beautifully, since the way something interacts with its linguistic neighbors is
in fact part of cognition, and to the extent that that those interactions are
entrenched and conventionalized, they are linguistic, i.e. semantic,
phonological, or symbolically linking the two. Yes, a category such as 'subject'
doesn't always mean 'agent', but if you include volitional agents as
prototypical subjects, and organize the category around them, it works out quite
satisfactorily. Other supposedly "syntactic" concepts fit in the framework just
as well.

The phonological nature of these syntactic structures bears comment. I don't
remember ever being given an argument for the non-phonological nature of syntax.
And yet, why is word order (or phrase or clause order, or morpheme order) any
less a *phonological* fact than phoneme order? Under this perspective it is a
phonological fact.  And in any particular construction that phonological order
relationship symbolizes a semantic relationship (e.g. that of a verb to its
subject or object.) Both semantic and phonological structures can be so
schematic (underspecified) that they are not useful alone, but are useful for
specifying patterns, yet those patterns differ only in degree, not in kind, from
fully-specified structures.

Anyway, it seems to me that Langacker and Lakoff's position on this is different
from the others I've been reading. It does not deny (in fact it insists) that
linguistic structures are formal(ized), but it says that is true of everything,
not just syntax. Certainly it allows for many "profound generalizations" to be
accounted for by the syntactic relations of one structure to another, but it
insists that those syntactic relations themselves are non-autonomous in the
sense that they are the same kind of thing as the rest of language, i.e. they
consist of conventionalized, entrenched, "formalized" cognitive patterns, either
phonological (related to speech sounds) or semantic.

Dan Everett says that we should compare models by their "empirical production".
"Arguments must revolve around the empirical, case-by-case."  I guess he means
by their success in dealing with real language data. If he does, I think I
mostly agree. I've liked Ron's and George's model because it helps me deal with
my real language data where the others I've tried didn't. But it helps me to try
to sort out (as I've tried to here), who's claiming what, before I can evaluate
what particular data really prove with regard to each model.

David Tuggy