form and function

[Tom Givon]

Dear friends,

This exchange has been skirting hilarity, or the Theater of the Absurd.
I get a verigo feeling, something like irrealis, like people not only
forget to listen to the other guy (common enough), but have stopped
listening to what they themselves are saying. Otherwise I can't imagine
why there seems to be such an abundant, recurrent, reflexive need to
jump on reasonable people (Edith Moravcsik, Dick Hudson) and try to push
them into extremist positions which realy don't fit. So how about a
simple exercise in logic, to begin with:

1. If "form correlates with function", or
      "form is iconic with function", or
      "form it motivated by function"
   are all true, them "form exists" is also true.

2. If "form correlates with function" is true,
   then, unless you wish to be tautological, you have no choice but to
   define "form" and "function" independent of each other. And unless
you
   defined structure in purely structural terms, you haven't escaped
this
   tautology. (That's *all* Edith meant).

3. Escaping tautology in (2) above does not mean that you could not or
   should not investigate the functional (semantic, pragmatic)
motivation
   of form/structure. Non sequitus.

4. In most biologically-based system, functions are not simply performed
   by themselves. They are performed by some structures. Why should
   this overwhelming fact, which has bothered no biologist since
Aristotle,
   should be such a shocking revelation to linguists?

5. The notion of "structure" is, by definition, formal and more abstract
than
   unstructured, unconstrained, unorganized reality. The only remaining
   question is -- "How formal? How abstract?" And that question must be
   resolved empirically. Different domains, in language and elsewhere,
   are structured with different degrees of formality ('generativity',
   'rule governedness').

6. When functionalists cite Sapir's famous dictum: "...all grammars
leak...",
   they often forget that Sapir did not say "all grammars leak 100% all
   the time". What he *meant*, I think, is something like this:
"Grammatical
   rules/regularities/structures often, maybe always, retain a ceretain
   measure of flex, rule-ungovernedness" Fact of life.

7. Chomsky's apriori assertion that grammars are 100% algorithmic (see
'On
   the notion 'rule of grammar'", 1961) clearly over-shoots the
empirical
   facts, and was not motivated by them. It was motivated, I suspect, by
   his background in Machine Theory. Of course, the notion of
*competence*
   allowed him to do it, by ruling out facts of natural communication as
   *performance tainted*. But the fact that Chomsky was wrong, and
grammar is
   not 100% generative, does not mean that grammar is 0% generative.
Such an
   assertion is just as much at variance with the facts as Chomsky's
assertion
   of 100% generativity (and damn the rest of the facts).

8. Grammar is just another instance of *automaticity* of processing.
   In all known cases (vision, memory, motor control, music) the
acquisition
   of complex, rhythmic-hierarchic skills entails autoimaticity. And
   automated processing is *highly* structure-dependent and category-
   dependent. That's what 'chunking' is all about in memory
organization,
   kinesiology & elsewhere.

9. Why should both evolution and history conspire to -- repeatedly --
   grammaticalize so many communicative functions if grammar was just
   a mushy affair with near-zero generativity/rule governedness or pre-
   dictability? Why this extravagant machinery that we desvcribe both
   synchronically and diachronically? All spandrells? Come on. Get real.

10. Grammar arises diachronically from pre-grammar, and has a life
trajectory
   during which the degree of 'generativity' changes drastically. At the
   very early stages of grammaticalization of paratactic constructions
with
   only 'pragmatic'('discourse') regularities, one tends to find low
   generativity. A nascent construction reveals variability of behavior,
   it is not 'well governed'. Somnewhere in mid-life of constructions,
rule-
   governedness increases, i.e. efficiency and predictability of
form-fuction
   correlations ('iconicity'). But sooner or later, what John Haiman
calls
   'ritualization' begins to creep in: Contructions and morphology
become
   nearly-100% rule-governed, indeed highly inflexible, but also slowly
   loose their iconicity. This is the *more* Chomskian, *more* arbitrary
   phase of grammatical structure.

11. It is unfortunate that, for whatever reasons, different people
choose
    to look at *only* the earliest stage of grammaticalization
('emergence',
    high motivation, low generativity), or the latest (arbitrariness &
high
    generativity). Sort of reminds you of the three blind men reporting
    on the elephant. Taking one aspect & claiming it represents the
whole.
    It would be nice if we started considering the whole.

12. In the process of early grammaticalization, constraints creep in
    rather gradually, often in a subtle way. The argument about the
    reflexive is of course a case in point. Older reflexives are much
more
    relationally governed, having severed their umbilical cord, their
    connections to old *emphatic* pronouns that were *not* relationally
    governed. Newly-emergent reflexives are a mix, the old 'discourse'
    constraints on contrast/emphasis co-exist with 'goverened true
reflexives'
    that are not emphatic anymore. But 'severing of the umbilical cord'
can
    be both gradual and subtle. And it allows coexistence of older and
    newer, emerging 'structured' or 'governed' constraints.

13. EXAMPLE: It is tempting to say that REL-clauses in Japanese are
*not*
    relationally governed, since they only have zero anaphora,
presumably
    just like in main clauses governed by 'discourse'. But haven't we
    forgotrten someting? Zero anaphora in Japanese discourse is
*overwhel-
    moingly* anaphoric. But zero anaphora in Japanes REL-clauses is 100%
    cataphoric, because REL-clauses in Japanese precede their head noun.
    So something new and construction-specific has crept in here,
whatever
    its diachronic source may have been.

    EXAMPLE: It is tempting to say that in Malagasy (VOS) the zeroes in
    EQUI (verb complements) are just the same as zero anaphora in
    'discourse'. But again, discourse zeroes are overwhelmingly
anaphoric,
    while the EQUI zero in complementation are 100% cataphoric, because
the
    subject of the  main clause *follows* the zero in the complement
verb.
    Again, something new is creeping in during the process of
grammatica-
    lization. Not only functions are re-analyzed and re-organized, but
    structures too.

I suppose I could go on forever, everybody who seriously studies the
process of grammaticalization probably could. So let me just suggest
that maybe it is time we bade farewell to reductionism, and to the
bizarre idea that complex systems can be described and explained by
single principles. Sure, that is the hallmark of much of what Chomsky
has been trying to do. But all of us, functionalists and formalists
alike, know language is much too complex for such reduction. And we
ought to, by now, know enough about conflicting motivation and adaptive
compromise ('OC'?) to know better. So let's get off the dime.

Cheers,  TG
=============


Michael Barlow wrote:
>
> Dick Hudson states:
>
> > >He sees a formal rule linking those forms, but others don't.
> > ## But since the formal rule is on the table (labelled "Rule A") it's now
> > over to you to show how you can describe the covariance of noun and
> > modifying adjective without using formal categories such as Noun, Adjective
> > and Modifying.  It would also be good to see reasons why this gives a
> > better analysis than Rule A.
> >
>
> I am not against the use of formal categories of Noun and Adjective. I need
> them too.
>
> Rule A works well in those cases where (i) the source of agreement, the noun,
> is both present and is fully specified for agreement features and where (ii)
> the agreement features of the adjective don't differ in their value from the
> features of the noun. Looking at more data, however, leads to the discovery of
> examples in which the source is either absent or exhibits fewer "features"
> than occur on the agreement target, a situation, which I noted in my more
> formalist days (1988), could be handled better by a unification account than
> by feature copying or coindexing. More interesting are those cases in which
> there is a feature mismatch (such as those noted by, for example, Edith
> Moravcsik many years ago; by Grev Corbett in various publications; and, for
> French, by Blinkenberg 1950).
>
> Dick Hudson would agree that formal features such as FEM or PLUR have
> interpretations or meanings and, in fact, are often polysemous such that
> within a particular language PLUR may indicate, for instance, something like
> "multiple entities" or "a single entity politely referred to". A FEM feature
> might indicate "grammatical gender" or "natural gender". These relations
> between forms and meanings are conventional; they are a part of a language and
> are to some extent separate from information about actual referents. For Dick
> (and many others) these interpretations have nothing to do with agreement.
>
> If we consider the agreement features of an adjective, we can ask whether the
> agreement relation associated with those features (i) is morphosyntactic and
> depends on the features of the noun sources; or (ii) is based on a
> "consistency" of interpretations of agreement features, or (iii) depends on
> the properties of the referent associated with the noun.
>
> My Rule B is based on a consistency relation between interpretations of
> nominal/agreement morphology and relates to the identification and tracking of
> discourse referents. I believe that this non-syntactic account is a "better
> analysis" because it covers a wider range of data and because it can be shown
> that what at the morphosyntactic level are unmotivated feature mismatches
> typically turn out to "make sense" at the level of interpretation.  A mismatch
> in formal features is nearly always associated with "extra" information about
> the associated discourse referent. (I am far from home and don't have any
> examples at hand.)  Also, when an agreement morpheme shows up in a discourse
> fragment with no accompanying noun, and hence nothing to be modified by, then
> somehow the agreement morpheme is still always there; it is not omitted.
>
> I don't want to state Rule B here---I usually rely on diagrams---but
> conceptually it is quite straightforward and involves (i) a listing of the
> conventional relations between agreement/nominal morphemes and their
> interpretations and (ii) a description of what counts as a coherent chain of
> discourse referents, which is essentially that the associated interpretations
> be consistent. (My Rule B can be found in a recent "agreement" issue of Folia
> Linguistica XXXIII/2 guest-edited by Grev Corbett, which I am happy to send to
> Dick and anyone else interested.)
>
> I guess that the differences between our accounts of agreement come down to
> the range of data to be considered as "ageement" and the level of commitment
> or priority given to a morphosyntactic account, which in turn is associated
> with differing degrees of tolerance of multi-domain accounts along the lines
> of the discussion of reflexives in John Moore's recent posting.
>
> Michael