existence and reality

Rob Freeman rjfreeman at EMAIL.COM
Mon Apr 7 03:09:18 UTC 2003


I don't find the emergentist position, as I understand it, incompatible with
generativism, it just reverses the power relationship: category vs. example.
Generativist insights about categories are still there, in the main. They are
for me, anyway.

In particular I believe that while emergence (or generalization, actually, as
opposed to generalization_s_) is the fundamental driving force of language I
also believe that driving force must be channeled and parameterized in ways
only glimpsed through some of the insights of generativism.

And Chomsky was only ever going for principles and parameters, wasn't he,
competence rather than performance?

But I seem to have stepped in on an argument which has been going on for some
time. Excuse my ignorance of the established positions, because while I
identify strongly with what I hear of emergence I really don't know yet in
depth how the perspective has developed (over the last 15 years!)

Perhaps I can ask some questions to get to the center of it. Take your
analysis of the empirical nub, for instance. Can you tell me why you want us
to undertake `text-distributional studies of variability ("emergence") vs.
stability ("generativity")'?

Isn't that a bit like saying let's test experimentally how much of language
is performance and how much competence?

Isn't emergence (or really I feel it's better to get down to the driving
force and say generalization) a performance model?



On Monday 07 April 2003 9:03 am, you wrote:
> Dear y'all,
> I guess I had better say something less flippant (sorry, Paul), tho
> temptation is still strong.Chomsky with his intellectual Stalinism had
> pushed us to adopt either extreme rationalism ("everything is innate, input
> doesn't matter") or extreme empiricism ("nothing is innate, all knowledge
> derives from input"). In the very same vein, we have been repeatedly pushed
> in the past 15 years or so by our own extreme emergentists to subscribe to
> an equally reductionist position: "If X--be it grammar or phonology--is not
> 100% generative, therefore it is 100% emergent". I find this passion for
> reductive solutions suspect on both theoretical and empirical grounds. I
> have always thought that the worse gift we could give Chomsky is to counter
> his extreme reductionism with an equally extreme reductionism in the
> opposite direction. That way he wins either way, because we concede his
> main philosophical premise--either or, but God forbid a pragmatic middle.
> Theoretical considerations:
> (i)  complex multi-factored systems: We are dealing with
> biologically-based, adaptively- shaped complex systems. Most often, such
> systems display various kinds of adaptive compromises between
> conflicting--but equally valid--functional imperatives (cf.
> Bates-MacWhinney's "competition model"). This is true all over biology,
> cognition & language. For running & flying animals, muscle-weight conflicts
> with speed. For all populations, genetic variability (adaptive flexibility)
> conflicts with genetic coherence (adaptive inheritance). In phonology,
> articulatory speed (sound assimilation) conflicts with auditory
> distinctness (sound dissimilation). In learning, innateness (relying on the
> evolutionary experience of ancestors) conflicts with emergence (sensitivity
> to changing context). Etc. etc.
> (ii) Automaticity & performance speed: Both phonology and grammar are
> highly automated processing systems. Not 100%, but way above 50%,
> probably--depending how one counts in real-time communicative behavior--ca.
> 90% at any given moment. It is all very nice to observe that emergence
> is--in principle--ever present. But at any given time during fluent
> (non-Pidgin) communication, a spoken-language user produces language at the
> speed of, roughly, 1-2 seconds per event/state clause and 0.250 msecs per
> lexical word. This is an extremely demanding processing rate, which can
> only be made possible by high level of automaticity. And high automaticity
> depends heavily on high categoriality; that is, a high rate of
> rule-governedness & either/or decidability. When Sapir said "all grammars
> leak", I doubt it what he meant was "all grammar leaks 100% all the time".
> What's the point of having a grammar then? Sure, grammar has it's
> counter-adaptive "spandrels", but neither 100% or 50% of the total. More
> like 10% percent or even less at any given time.
> (iii) The S-shaped learning curve: This is a well known phenomenon in both
> psychology and the linguistic change-cum-variation. The variation ratios of
> 90-to-10 (at the beginning of the process of  change) and 10-to-90 (at the
> end) both last for a long time. The in-between phase, the transitions
> between 80-to-20 and 20-to-80 ratios, is extremely rapid. This is most
> likely because the more even ratios of variation are not viable as
> processing systems. Most highly-automated systems can cope with 10% residue
> ("garbage") by either ignoring it (if it is adaptively irrelevant), or by
> investing high-energy resources to attend to the fine details of context
> (if it is adaptively urgent). And I suspect that S.J. Gould's "punctuated
> equilibria" phenomenon has a similar fundamental explanation--the adaptive
> instability of intermediate stages.
> Empirical issues:
> One can easily test all this empirically, as I have tried to suggest in
> chs. 2-3 of Bio-Linguistics (Amsterdam:  J. Benjamins, 2002). One can, for
> example, take a text of natural communication (unedited, strictly oral!)
> and compute how many of the actual uses of all grammatical constructions
> there are highly
> conventionalized & rule-governed ("generative")  vs. how many are in the
> midst of change ("emergent"). The computation I did in ch. 2 of B-L yielded
> the ratio of 98%-to-2%, respectively. So this was only one conversational
> text of 5 pp. In ch. 3 of the book I did a similar assessment of
> cliticization (grammaticalization), yielding a ratio of 90-to-10. And
> studies of the frequency distribution of marked vs. unmarked constructions
> in text yield similar ratios (Givón 1991).
> I think those of you who are interested in this as an empirical issue
> should undertake similar text-distributional studies of variability
> ("emergence") vs. stability ("generativity") in both our major linguistic
> coding ('structural') systems-- phonology & grammar. To continue to just
> take for granted either emergence or generativity without the benefit such
> performance tests has become, to my mind, a bit stale. So if one sounds a
> bit impatient, Paul, it is because Chomsky has been ignoring performance
> and taking 100% generativity for granted since 1957.
> Cheers,  TG
> ======================

More information about the Funknet mailing list