This is not a cry for help, Lady;
this is a stick-up.
(Caption
of a New Yorker cartoon)
PROLOGUE: How
on earth did this paper get so long?
I started with no
goal more ambitious than a critical discussion of Fiona Cowie’s
new book about innateness; 1 it seemed to me
that her arguments, unless refuted in detail, were likely to affront
some or other abstract entity whose cause I favor: The Good, The
True, The Beautiful; whatever. But there were so many things that
the book struck me as being wrong about that the proposed critique
became, in effect, an explication of the kind of nativism I think
a rationalist in cognitive psychology should endorse. And the more
of that I came to explicate, the more digressions and elaborations
suggested themselves. And elaborations of the digressions. And digressions
from the elaborations. Things commenced
to be out of hand.
A
quandary. But one of which an appropriately Gilbertian solution
(see `Iolanthe’, Act 2) occurred to me: Construe the project to
be mainly an exposition of the kind of nativism that I think
a rationalist in cognitive psychology should endorse; and construe
the critique of Cowie’s book to be mostly digressions and elaborations.
Voila!
The result
is before you.
PART 1:
Introduction.
Do
you want to know how to tell when you have gotten old? It’s when
a cyclical theory of history starts to strike you as plausible.
It begins to seem that the same stuff keeps coming around
again, just like Hegel said. Except that it’s not “transcended and
preserved”; it’s just back. So, associationism is back (see Elman et al 1996; for an unsympathetic
review, see Fodor 1998b), and likewise
the ancient argument about innate ideas. Cowie’s resurrection of
the nativism controversy, just when I’d begun to hope that its recent
demise might prove permanent, will be the topic in what follows.
I’d be glad to report that something new has happened; but, as it
turns out, the polemics are almost all familiar. As far as I can
tell, it’s just the Eternal Recurrence recurring. I think I must
have gotten old.
Cowie
claims to rebut arguments for nativism that Noam Chomsky and I have
from time to time endorsed. I don’t, in fact, think that she has
done so; but then I wouldn’t, would I? Since my sense of what’s
the bottom line on all of this is, pretty clearly, preconceived,
---and since I’d guess that yours may be too--- I won’t attempt
to change your mind about innateness. But I do want to claim, at
length, that if the problems that Chomsky and I have to worry about
are only the ones that Cowie’s book raises, then at worst we’re
as well off now as we were before she wrote it. Nothing has changed
because, quite uniformly, the arguments Cowie has on offer either
misconceive the issues or are, in crucial respects, unsound. Or
both. So, anyhow, I hope now to convince you.
Cowie’s
book has three main sections. The first is her exegesis of considerations
that prompt the nativist position (specifically on first language
acquisition, but implicitly on cognition at large.) These
Cowie takes to be: `Poverty of Stimulus Arguments’ (often
hereinafter POSAs), and `Impossibility Arguments’ (hereinafter sometimes
IAs.) The second and third sections are devoted to criticisms of
these arguments, set out in reverse of the order I’ve just mentioned.
Except for this reordering, my plan is basically to track the book.
1.1 The polemical situation
according to Cowie: Let’s start with
a way of viewing the rationalism/empiricism debate that Cowie
flirts with but doesn’t in the end endorse; namely “that nativism
---or empiricism, for that matter, is nothing at all… [and] the
great controversy over innate ideas is not worth the paper it’s
written on… (p. 25)” Eventually Cowie rejects this view since, of
course, it can’t both be that the argument about innateness was
empty and that the empiricists won it. But Cowie is prone to phrase
this `no contest’ reading in ways that suggest invidious asymmetries.
For example “The difficulty, in other words, is that the assertion
of nativism often seems to be merely the denial of empiricism. And
if that is so, then nativism is not a theory of the mind at all;
it signifies merely our lack of such a theory.(25)”. Take home exercise:
try rewriting this passage replacing `nativism’ with `empiricism’
and `empiricism’ with `nativism’ throughout. Notice that it works
equally well (or badly) either way. That’s because, prior to examining
particulars, the polemical situation between rationalists and empiricists
is really entirely symmetrical: Nativism is merely the denial of
empiricism insofar as we lack a way of saying what `innate’ comes
to other than not learned. Likewise, empiricism is merely the denial of nativism
insofar as we lack a way of saying what `learned’ comes to other
than not innate.
But it doesn’t follow,
as Cowie sometimes seems to suppose it must, that empiricism and
nativism were tacitly interdefined in the traditional debate; so
that, for example, “[the nature/nurture argument] is a battle that
is largely fought over, and with, metaphors… [which only] mask the
absence of substantive disagreement (17).” It’s worth getting straight,
before we plunge into deeper waters, on how the argument could have
been fruitful, and the issue substantive (as both clearly were and
continue to be) if nobody had any very definite idea what either
innateness or learning amounts to.
The
metaphors, parade examples, agreed cases and such, in terms of which
the issues were largely framed, didn’t “mask” anything; indeed,
they were just what made it possible for illuminating discussions
to proceed. What happened, unsurprisingly, was that each side elaborated
its claims largely by reference to plausible paradigm examples;
for the nativists, these included (eg.) bird song, skin color, and
the Classical reflex. Their claim was that, when the dust settled,
cognition (including learning, perception, memory and thought) would
be seen to resemble phenomena like those a lot more than it does
such empiricist paradigms as rote learning, verbal association,
and the Instrumental reflex. I say this is unsurprising because
science often starts in media res, finding out what it’s `really` about as it goes
along, thereby discovering the essences of issues.
However,
that way of proceeding implies a kind of inductive risk: the danger
that the paradigm cases, reference to which defines the common ground
of argument, may turn out not to be paradigms of anything. In particular, they may not all exemplify
the same natural kind. If so, then the issues have to be framed
some other way, or dropped. On both sides of the traditional debate,
questions about innateness were widely run together with, for example,
questions about a prioricity, necessity, the existence of God and
the warrant of moral principles. But despite such conflations, it
appears in retrospect that the argument really was about something
---some one
thing--- after all: It was about whether there is a characteristic
human psychological phenotype (`human nature’ in earlier editions)
that can be attributed to a characteristic human genetic endowment.
The constellation
of notions that cluster around `genetic determination’, `genome’
`genotype’ and the rest are, to be sure, themselves adequately contentious.
But I suppose nobody sensible denies that they are now deeply scientifically
entrenched, or that biology is in the process of constructing a
concept of genetic specification that is likely to save many of
the rationalists’ paradigms. Skin color really is largely innate
(/heritable/genetically determined), much as everyone had hazily
supposed. Likewise birdsong in a lot of cases; likewise the Babinsky
Reflex. And it seems unlikely that the notion of innateness according
to which such claims are true will prove dispensable for the larger
purposes of biology. Mendel was, presumably, right about something; presumably
what he was right about was the genetic transmission of the phenotypic
traits he studied. We have, in short, good reason to take for granted
that there’s a substantive notion of innateness because biology
needs one however
the rationalism/empiricism issue turns out.
I’m
going on about this since it’s not at all the view of the polemical
situation that Cowie’s exegesis suggests. As she appears to see
it, the burden is on nativists to say exactly what doctrine
they’re endorsing, thereby avoiding the trivialization of their
side of the classical debate. This burden Cowie, in all kindness,
offers to take up on the nativist’s behalf; she proposes, as she
puts it, to “…find some substance for the nativism debate… to be
about. I argue that there are in fact, two substantive issues over
which nativists and empiricists clash.” The one with which Cowie
takes POSAs to be most involved “concerns the natural architecture
of the mind: Has nature equipped us with general-purpose, or domain-specifc,
learning devices?” The other, which Cowie takes to be what’s at
issue in IAs, concerns “the scope [and limits] of natural science:
what are our prospects for domesticating the mind and locating it
within our overall scientific world view. (26).”
But
even this early in the exposition, it seems something has gone badly
wrong with Cowie’s geography. For, it’s hard to believe that a serious
reconstruction of the argument about whether there are innate ideas
could miss the point that it was an argument about
whether there are innate ideas; hence,
presumably not
(or, anyhow, not in the first instance) about whether there are
special purpose learning mechanisms, or whether there’s a place
for the mind in the scientific world view. These latter issues belong,
respectively, to the psychology of learning, and to metaphysics;
neither sounds much like asking what ideas are innate. Likewise,
as we’ll see presently, neither is what IAs or POSAs are about.
There
is also a deeper objection to Cowie’s initial framing of the issues;
it’s my excuse for taking this long way `round getting started.
Suppose it’s agreed that, as things have turned out, the argument
between rationalists and empiricists was `really’ about whether,
or to what extent, a species-characteristic human psychological
phenotype is genetically specified. That would, as I remarked, vindicate
the rationalists’ claim to have all along been holding a substantive
view; one that the advance of microbiology now promises to explicate.
But no such appeal would vindicate the empiricist side of the debate.
So an empiricist still needs what neither Chomsky nor I believe
him to have: an independent characterization of “learned”; one that
doesn’t amount to just the denial of “innate”.
It
is, I think, a remarkable feature of Cowie’s exegesis that
she never considers the question what, if anything, learning is.
To the contrary, remarks like the following are characteristic:
“I do not regard it as in any way destructive of my position or
arguments… that I do not have on hand any worked-out alternative
to the Chomskyan picture of language acquisition (272)” “Humans
learn an awful lot, about a bewildering variety of topics… that
they can do so… is miraculous and mysterious (216)”. 2 Well, Cowie is right
that you don’t need a `worked out [empiricism] …on hand’ to deny
that nativism is true. But what you do need if you are proposing
empiricism as an alternative to nativism (learning as an alternative to innateness)
is some reason to suppose that your paradigm cases of learning are
indeed mostly paradigms of the same thing. The thoroughly modern
rationalist finds in genetics a science where notions like innateness
are entrenched. What offers empiricists the corresponding encouragement?
There is, after all, no program of research except
empiricist psychology that makes play
with the notions that cluster around learning. So
why (other than a prior commitment
to the empiricist program) should one
believe that there is any such thing?
Empiricists really do have what Cowie takes to be the nativist’s
proprietary problem: How to say what they’re endorsing except that
it’s not what they’re rejecting. So the question really does
arise whether there is a substantive empiricist position for nativists
to argue against.
However,
what I just said isn’t true. There is, in fact, a sketch theory
that purports to provide some idea of what being
learned might amount to beside being not innate;
which is all one could reasonably demand of an empiricism that is
itself in media res. Learning might be association; correspondingly, being
acquired by association formation (i.e. by
processes that satisfy the laws of association) might be the property
that makes most or all of the empiricist’s paradigms instances of
learning. It’s thus not an historical accident that empiricists
have been, pretty much without exception, associationists as well.
Nor is it an accident that, empiricism now being back, associationism is back too.
But,
of course, associationism isn’t true; it is, and always has been,
an intellectual disaster. Perhaps you don’t agree? Even so, for
present purposes, please do suspend your disbelief. It’s fair for
me to ask you to do so, because (to her credit) Cowie isn’t an associationist.
(She makes occasional references to connectionism as possibly an
alternative to Chomsky’s rationalism; but they are guarded and far
from an endorsement). I won’t, therefore, digress to rehearse the
standard anti-associationist arguments. Suffice it that there is
a cost to Cowie for thus exempting herself from the traditional
empiricist-associationist alliance. It’s not just that she is left
with no `worked out’ psychology of learning (etc.) An empiricist
who’s not also an associationist has no
cognitive psychology on offer at all;
only the hope that his favorite paradigm cases of not-innateness
will prove to be all of a (natural) kind. That does not count as a theory
of mind; or even as a properly mongered mystery. At most it’s
a propositional attitude in search of an intentional object.
Among Cowie’s recurrent
themes is that, whereas impossibility argument nativists (like me)
have no positive learning theory on offer, it’s characteristic of
empiricists to propose real, testable models of how cognition is
achieved. That, however, is true only
of empiricists who are also associationists,
and it’s true of them in virtue of their associationism, not of
their empiricism. If it’s read just as the thesis that very
little that’s intentional is unlearned, empiricism offers no positive account
of how the mind works. Nor, likewise, does rationalism if it’s read
just as the thesis that there’s lots intentional that’s unlearned.
What you do to get an honest to God psychology out of empiricism
is add the thesis that mental processes are associative; what you
do to get an honest to God psychology out of rationalism is add
the thesis that mental processes are computational. In principle,
the situation between rationalism and empiricism with respect to
whether they offer positive psychological theories is thus exactly symmetrical
(just like the situation between them with respect to whether they
offer positive accounts of the distinction learned/innate, and for
the same reasons; see above.) De facto, however, the current situation
favors the rationalists since, whereas associationism is certainly
false, computationalism might actually be (partly) true. (For which
part of it might be, see Fodor (2000.))
But,
having thus objected to the way Cowie sets the pieces out, I propose
now to waive all further such complaints. As it turns out, most
of Cowie’s book floats free of her general analysis of the rationalism/empiricism
dispute; mostly it’s about the status of POSAs and IAs, her
main thesis being that neither are convincing. So let’s turn to
that. I’ll start by considering what Cowie takes it that IAs and
POSAs are supposed by their proponents to show. Then I’ll discuss
Cowie’s reasons for holding that neither kind of argument is sound.
I claim, under the first head, that Cowie misconstrues the conclusions
of IAs and POSAs. I claim, under the second head, that although
Cowie misreads both POSA and IAs, her doing so doesn’t really matter
much. That’s because the objections she raises against POSAs and
IAs would be ill-founded even if the intended conclusions of these
arguments were as Cowie believes.
What
with one thing and another, this will amount to a
lot of work in what I take to be the
public interest. I do hope somebody is going to thank me for it
when it’s over. Profusely, by preference.
PART 2: The
Arguments.
2.1 What the arguments
claim to show: Cowie observes that
versions of POSAs and IAs have both been floating around for centuries,
neither displacing the other as the flagship argument for nativism.
She speculates that this is because their presumptive conclusions,
though both incompatible with empiricism, are mutually independent.
By contrast, though I do think Cowie is right that IAs and POSAs
serve different polemical intentions, I think she’s got it utterly
wrong what their conclusions are supposed to be. When that’s straightened
out, they are seen not to be independent after all: Roughly, what
follows from POSAs can’t be true unless what follows from the IAs
is; but not vice versa.
In
a nutshell, here’s how Cowie sees the situation. Insofar as he endorses
POSAs, “the nativist’s claim that such and such mental item is innate…
means that that item is acquired by means of a task-specific learning
device.…” Cowie identifies this version of rationalism as having
historical roots in Plato and Descartes; Chomsky, however, is its
primary current proponent, and he’s the main target in Cowie’s discussion
of POSAs. By contrast, according to Cowie, the conclusion of IA
is not a thesis about (for example) language acquisition, but rather
a kind of “methodological gloom” about naturalism. The nativism
that emerges from IAs is just the claim that “…empiricist
boasts to the contrary notwithstanding, we have no idea whatsoever
how [an] item was acquired (67)”. 3 The
historical affinities of this kind of nativism are, according to
Cowie, largely with Leibniz and Descartes. However, it’s primary
current proponent turns out to be ---of all people--- me.
But
though it’s strikingly imaginative, Cowie’s account of what POSAs
and IAs are supposed to show can’t be right. On the one hand, for
reasons I’m about to try to make clear, it’s very implausible to
read Chomsky as holding a thesis about acquisition devices (my emphasis); or, indeed, as holding much of a view
about any
of the mechanisms that mediate language behavior. On Chomsky’s way
of seeing things, such matters fall in the domain of `performance
theories,’ a term Chomsky generally uses with invidious intent.
I’ve never actually asked him, but I’m prepared to bet a dime that
Chomsky really thinks there can’t be serious performance theories,
and that people who try to construct them are wasting time that
they could much more profitably use studying syntax. If I’m right
to read him that way, then that the intended conclusion of the POSAs
isn’t about acquisition mechanisms, domain specific or otherwise. To the contrary, what Chomsky proposes is a nativism of domain specific
propositional attitudes (= PAs), not a nativism of domain specific
“devices.” More on this presently.
As to my view about
IAs, I have introspected carefully and speak with first-person authority.
I do not
think they show ---or even suggest--- that naturalism is impossible.
I am, to be sure, gloomy enough, metaphysically and otherwise; but
not about the kinds of things, or for the kinds of reasons, that
Cowie supposes. To the contrary, I am, perhaps more than anybody
else I can think of who isn’t actually Australian, a crude, crass,
vulgar, old fashioned, simple minded, positivistic Village Reductionist
about (token, intentional) mental states. Indeed, I think that token
reductionism is a substantive constraint that the scientific world
view (or something) imposes on the ontology of all the special sciences; hence on psychology inter alia. I have
suffered for thinking this: I have been repeatedly beaten around
the head and shoulders by experts, including Tyler Burge, Steven
Stich and, come to think of it, Noam Chomsky. But I have kept my
ground, and I have not cried for help. That after such stoicism I should be accused of
arguing that there can’t be a science of the mind… Well, really!
I am seldom moved to exclamation points, but really!!! 4
So, then, what do the rationalists
who propose them take to be the conclusions of POSAs and IAs
respectively?
The
bottom line of Poverty Of Stimulus Arguments, as Chomsky uses them,
is that innate, domain specific information is normally recruited
in first language acquisition. A nativism of domain specific information
needn’t, of course, be incompatible with a nativism of domain specific acquisition mechanisms;
in fact, people who are into `modular’ views of cognitive architecture
generally (though by no means always; see, eg. Karmiloff Smith (1992)) hold both. But
I want to emphasize that, given his understanding of POSAs, Chomsky
can with perfect coherence claim that innate, domain specific PAs
mediate language acquisition, while remaining entirely agnostic
about the domain specificity of language acquisition mechanisms. Indeed,
as far as I can tell, circa Aspects (1965) Chomsky pretty
explicitly held to the soundness of POSAs; and to a nativism of propositional attitudes (he supposed
Universal Grammar (=UG) to be innate); and to the view that language acquisition is implemented
by some hypothesis formation/testing mechanism which could perfectly
well be domain neutral for all anybody knows. According to my understanding
of Chomsky’s understanding of POSAs, they raise
the question whether the innate knowledge
that language acquisition exploits is at the disposal of domain
specific mechanisms. But they are not in themselves committed on
how that question should be answered. Nor is the last word on this
currently audible. 5
However, as previously
remarked, the difference between the conclusions that Cowie thinks
that Chomsky thinks that POSAs invite and the conclusions that Chomsky
thinks that POSAs invite, doesn’t actually matter much in evaluating
Cowies objections to POSAs. For, these are mostly arguments that
the empirical premises of POSAs aren’t true; or, at a minimum,
that there’s reason enough to doubt their truth that one can’t
reasonably rely on POSAs whatever exactly their conclusion are supposed to be.
But, though distinguishing between a nativism of mechanisms and
a nativism of PAs isn’t essential to Cowie’s enterprise, it matters
a lot to Chomsky’s. The point that’s involved here is really central
to understanding how cognitivist explanations are supposed to work,
and so merits one of those digressions.
Here’s
how I think the geography goes: Chomsky wants it very much that
coextensive, `descriptively adequate’ grammars can differ in truth
value. 6, 7 For example, given
the way Chomsky has things set up, it could turn out that G1 and
G2 are both descriptively adequate, but that G2 is unlearnable because
it acknowledges rules that violate universals imposed by UG. Chomsky
thus requires there to be a distinction between descriptive
adequacy and truth
as they apply to theories of language.
He gets the distinction by assuming, on the one hand, that grammars
are the intentional objects of certain of the speaker/hearer’s PAs
(in particular, attitudes of `cognizing’; see below) and, on the
other hand, that the intentional objects of PAs are ipso facto `internally
represented’ as a matter of nomological or (maybe metaphysical)
necessity. This is all he needs to explain why `G is the grammar
of L’ is opaque to the substitution of descriptively adequate Gs.
Because representations can differ
even if their intentional contents do not,
the assumed equivalence of G1 and G2 in respect of descriptive adequacy
does not guarantee that if either is `psychologically real,’ then
both are. And, by assumption, psychological reality is required
for the truth of a linguistic theory. QED.
But
you can’t, of course, run the parallel argument on psychological
devices, mechanisms and the like. For, the distinction between truth
and adequacy I just drew depends on assuming that, qua intentional,
the objects of PAs are internally represented. But internal mechanisms
aren’t (normally)
internally represented; they’re just internal
tout court. A fortiori, you can’t
choose between equivalent theories of an internal mechanism by reference
to how
it is internally represented. So presumably there’s nothing to choose
between equivalent theories of an internal mechanism; nothing, anyhow,
that could distinguish between them in respect of truth. So, since
it’s important to Chomsky that there can be an empirically motivated
choice among equivalent grammars, it’s likewise important that his
nativism is about propositional attitudes rather than mechanisms.
Not to attend to
this aspect of the mechanism/attitude distinction is to miss exactly
the point at which the notion of intentionality gets its grip on
psychological explanation in Chomsky’s kind of theory. That would
be a great shame whatever you think about rationalism and empiricism, since
the ways it plays the notion of content off against the notions
of representation and mechanism is, perhaps, the characteristic feature of contemporary cognitivist
theorizing. So, then, to repeat: The intended conclusion of POSAs
is that innate, domain specific PAs mediate language acquisition,
not (pace Cowie) that innate domain specific devices
do. It’s because Chomsky holds that
the innate information available in the initial state of language
acquisition is ipso facto among the intentional object of the learner’s
propositional attitudes that Chomsky’s theory of mind is indeed
continuous with the traditional rationalist postulation of innate
ideas.
I’ll,
for now, be very quick about what’s the intended conclusion of Impossibility
Arguments; we’ll presently get to a story that’s more fine grained.
First,
if they are sound, IAs imply that lots of concepts are innate. No doubt, among the lots
of concepts that are innate if IAs are sound are probably
lots of linguistic concepts (ones that express such grammatical
properties of linguistic expressions as, for example, being a noun.) But
so, according to impossibility arguments, are very many other
concepts: TRIANGLE, for one example, and CARBURATOR for another.
There’s thus nothing particularly linguistic about IAs; and,
unlike Chomsky’s POSAs, they require no empirical premises about
the informational environments in which languages are acquired.
Also, since IAs imply that many concepts are innate that one would
otherwise have thought pretty certainly aren’t (including DOORKNOB
forsooth), the conclusions IAs lead to are substantive in a way
that cries for help, grindings of teeth and the like are not. The
philosophically interesting issue is not whether IAs are arguments
of substance; it’s whether they aren’t plain crazy. 8
A final exegetical
remark; according to Cowie, the conclusions of POSAs and the conclusion
of IAs, though both incompatible with empiricism, are mutually independent.
Perhaps it’s now clear why I think that’s wrong. What POSAs are
supposed to show entails what IAs are supposed to show because there
can’t be innate PAs unless there are innate concepts. 9
On the other hand, what IAs are supposed to show is independent
of what POSAs are supposed to show since there could be innate concepts
even if there were no innate PAs. 10
The upshot, then,
is that there might be two kinds of reasons for thinking that there
are innate concepts: roughly empirical ones, of the kind that POSAs
allege, and roughly a priori ones of the kind that IAs do. As for
the logical relations between POSAs and IAs on the one hand, and
empiricism on the other, they go like this (according to me): The
conclusions both of POSAs and of IAs are incompatible
with empiricism if you read POSAs as entailing that there are innate
PAs, IAs as entailing that there are innate concepts, and empiricism
as denying that there is anything (much) that’s both innate and
intentional. If, however, you read POSAs the way that Cowie does
(viz. as arguing that learning is mediated by domain specific devices),
what they preclude is not empiricism but associationism. So construed, POSAs are compatible with empiricism
because empiricists can tolerate the domain specificity of learning
so long as it isn’t itself innate (see Cowie’s own “Enlightened
Empiricism,” to be discussed below.) But POSAs are incompatible
with associationism because, if pretty much all of cognition is
associative, then it’s pretty much all domain neutral: Association
is supposed to act on concepts `mechanically,’ without
respect to their contents. 11 Cowie misses all
this because she both misconstrues POSAs, and runs empiricism
and association together.
So
much, then, for what I take to be wrong with Cowie’s account of
what rationalists think that POSAs and IAs are supposed to show.
We now start on the main stuff, which is her criticisms of these
arguments.
2.2. The empirical arguments: POSAs.
In effect, Cowie has three
points to make in Chapters 8-11 of her book:
2.2.1 The inference from empirical linguistic data
to the innateness of UG requires as a premise that grammars
are mentally represented; and the argument that grammars are
mentally represented depends on such dubious ontological and
methodological assumptions as that languages are mental objects
and that linguistics is `part of psychology’
2.2.2 The empirical data that are supposed to demonstrate
the paucity of information in the child’s linguistic corpus
are, in fact, inconclusive.
2.2.3 There is no reason to prefer the thesis that
UG is innate to the `enlightened empiricist’ thesis which says:
`Yes, domain specific information is recruited in language learning;
but, no, this domain specific information isn’t innate.’
I’ll consider Cowie’s arguments under these
three heads.
2.2.1 What POSAs assume about languages
and grammars:
Cowie endorses a criticism of Chomsky’s argument
for nativism that I take it goes like this.
- The thesis that
UG is innate depends on the thesis that only grammars compatible
with UG are `psychologically real’.
- Grammars are
psychologically real only if they are mentally represented.
- So the empirical
case for the innateness of UG depends on assuming that the kinds
of evidence linguists offer for the grammars they write is evidence
that the grammars are mentally represented.
- Whether the kinds
of evidence linguists offer for the grammars they write is evidence
that the grammars are mentally represented depends on whether
linguistics is “a part of psychology;” in particular, on whether
the `truth makers’ for grammars are facts about the psychology
of speaker/hearers.
- The thesis that
linguistics is part of psychology depends on arguments that
are fraught with methodological and ontological premises, many
of which a reasonable person might reasonably refuse to grant.
- Chomsky should
therefore conditionalize his conclusions about the innateness
of UG not only upon the empirical evidence for grammars, but
also upon the dubious methodological/ontological premises above
mentioned.
- So conditionalized,
the argument for UG’s being innate is weak-to-nil even assuming that the empirical data linguists
offer for the grammars they write are often convincing.
In short, according to this
line of reasoning, deciding whether the available linguistic evidence
argues for UG’s innateness requires first answering such questions
as: `What sort of thing is a language?’, `What is the warrant of
inferences from a creature’s behavioral capacities to its cognitive
states?’, `What is the evidential status of the linguistic intuitions
of native informants?’ `How, if at all, should the performance/competence
distinction be drawn?’ and so forth. Given that many such matters
remain (ahem!)
unresolved, the empirical evidence that linguists offer for the
predictive/explanatory successes of grammars that satisfy UG has
no direct bearing on the issue between rationalists and empiricists.
Chomsky’s inclination to suppose ---a priori, apparently--- that
the psychological reality of a grammar and its truth are the same thing
is at the bottom of this confusion. Likewise, all that’s required
to dispel it is to recognize that “a grammar could be true of language…
but false of speakers’ psychologies.” (244) In any case, ”it’s an
empirical psychological question whether grammars provide true theories
of linguistic competence. (246).”
But,
surely, this diagnosis can’t be right? Surely linguists don’t have
to do all that philosophy (or, worse yet, have to wait for us to
do all that philosophy) before they get to do their science? Surely
that would be unprecedented?
No
doubt, somebody really should sort out the methodological and ontological
(not to say the historical) issues involved in understanding the
relations between psychological and linguistic theories. And, quite
right, if an empirical assessment of nativism presupposes such a
sorting out, then we are in no current position to make one. But,
in fact, that is not to the point. For, even if the question whether
UG is innate turns on (inter alia) the question whether grammars
are mentally represented, the central argument that grammars are
mentally represented does not (pace Cowie) invoke methodological
premises about the relations between linguistics and psychology;
or ontological premises about languages being mental objects. Rather,
it turns on the predictive/explanatory success of grammars with
respect to behaviors and behavioral capacities of speaker/hearers.
Here’s the
argument from the explanatory/ predictive success of grammars to
their being mentally represented:
- It would explain
the explanatory/predictive success of grammars if the information
they express is available to speaker/hearers. No other explanation
of the predictive/explanatory success of grammar is on offer.
So, all else equal, we should suppose that the information that
grammars express is available to speaker/hearers.
- Cognitivism is
common ground; a speaker/hearer’s behavior should be explained
by reference to his propositional attitudes.
- Taken together,
(i) and (ii) license the (nondemonstrative) inference that the
information grammars express is part of the what speaker/hearers
know/believe/cognize. 12
- Nobody has the
slightest idea how a creature’s PAs could predict/explain its
behavior unless the intentional objects of its PAs are mentally
represented by the creature whose behavior they predict/explain.
- So, all else
equal, we should infer that well-evidenced grammars are mentally
represented by speaker/hearers.
Please note the brevity of
this argument; also its absolute and endearing freedom from any
assumptions particular to the relation of linguistics to psychology,
or to the ontology of languages or grammars. It could be run, just
as well, on how the information that it’s polite not to dine with
your hat on explains your taking your hat off at table. Likewise,
it could be run by the most ardent Platonist, according to whom
the truth makers for theories of languages are eternal facts about
relations among nonnatural objects. Even
Platonism is neutral on whether a speaker-hearer mentally represents
the grammar of his language;it’s committed
only on whether his doing so is what makes the grammar true. That’s
just as well, since a Platonist might reasonably wish to explain
the empirical success of a grammar in the same way that cognitivists
do; viz. by assuming that the information it expresses is known
to speaker/hearers of the corresponding language. 13 And (have I mentioned
this?) nobody has the slightest idea
how what a creature knows could determine its behavior unless the
propositional content of its knowledge is mentally represented.
Cowie’s
way of proceeding belongs to a tradition of trying to settle issues
about the `psychological reality’ of grammars, and/or of UG, by
taking sides on issues about the ontology, methodology and epistemology
of linguistics (see papers in Block, (1980); including my
own). These issues are of considerable independent interest, to
be sure. But the argument that UG/grammar is mentally represented
simply does not address them. Indeed, though some of the assumptions of that argument
are tendentious, not to say inflammatory, none
of them are ones that Cowie disputes.
Notably, she concedes all the following:
-the predictive/explanatory successes
of grammars that conform to UG;
-a cognitivist construal
of the `know’ in `S behaves so and so because he knows that
such and such;’
-the `representational theory of mind, ’
according to which the causal consequences of a creature’s propositional
attitudes are mediated by mental representations of their intentional
content.
The inference from what Cowie
concedes to the psychological reality of UG/grammar consists largely
of `what else’ arguments: (What else but grammars being mentally
represented could explain their empirical successes? What else but
UG’s being innate could explain the child’s ability to assimilate
the grammars whose predictive/explanatory success the story about
grammars being mentally represented is supposed to account for?)
Well, on what else if not `what else’ arguments would you expect
to ground an empirical inference from data to theory? Empirical
inferences are ipso facto not demonstrative
The
possibility of justifying psychological reality claims by using
arguments to the best explanation suggests reversing the order of
demonstration that Cowie takes for granted: Instead of such claims
depending on the prior vindication of the ontological and methodological
`dubious assumptions,’ the vindication of the dubious assumptions
should rest on the de facto empirical success of theories which
require that grammars and UGs are psychologically real. That’s entirely
as it should be.
One vindicates the ontology and methodology of a science
by appeal to the work they do, NOT THE OTHER WAY AROUND.
So
the psychological reality of grammars explains their success; and
the innateness of UG explains why successful grammars are structurally
similar. I’m almost certain that Cowie’s book doesn’t contain a
refutation of this line of thought; in fact, as far as I can tell,
she says nothing at all about what might be wrong with it . Here
is the passage in which she declines to do so: “[According to Chomsky,]
since the hypothesis that the [language] learning mechanism respects
the [UG] principle of structure dependence enables us to explain
and predict many … linguistic phenomena… we should accept that it
is our innate knowledge of [for example] UG’s principle of structure
dependence that is at work in language-learning… I do not propose
to criticize this inference to the best explanation… [since] it
is hardly fair to expect the Chomskyan to show that his theory is
better than rivals that do not yet exist. 14 Accordingly, I will
accept that Chomskyan nativism is the best available theory of language
acquisition --- and argue that it provides no real explanation of
language acquisition at all. (249)”
-The
innateness of UG can’t provide the best explanation of language acquisition because it can
provide “no real explanation” of language acquisition at all. Why
is that? You might expect, at this point, that Cowie would revert
to the thesis that UG couldn’t explain language acquisition sans an argument that
grammars are mentally represented, and the arguments that grammars
are mentally represented turn on methodological and ontological
principles both suspect and obscure… etc, see above. But, disconcertingly,
she doesn’t; the next long stretch of her polemic isn’t methodological
or ontological, but straightforwardly psycholinguistic. It’s about
the status of hypothesis-testing and parameter-setting models of
first language learning; in particular, whether either could explain
how the language learner uses the information in UG to induce a
grammar from his corpus. This survey leads, finally, to the conclusion
that “parameter setting models are too underdeveloped to be appealed
to in support for such a claim… [and] the hypothesis testing model
has been amply developed, but in the wrong sorts of ways. As a consequence,
Chomsky’s identification of the principles of UG with the information
specified [in the `initial state’ of the language learning device]
remains unwarranted. (270)"
What
on earth is going on? As far as I can make out, Cowie has two different
arguments running in this part of her discussion. One turns on the
methodological and ontological stuff about dubious assumptions.
The other one is this: `UG doesn’t explain language acquisition
unless there’s a theory about how the information it expresses is
employed to get from a corpus to a grammar. But we haven’t got such
a theory. Ergo…’ The present exegetical question is how these two
arguments are supposed to fit together.
God
only knows, and Cowie doesn’t say; but it does seem clear that the
first doesn’t work and the second is unpersuasive if it’s offered
as an alternative
to the first. No doubt, something is badly wrong with Chomsky’s
picture unless there is finally a story about how UG is used
to project a grammar from a PLD (= from a corpus of Primary Linguistic
Data). But a lot of hard empirical work has been done on this problem
over the last several decades; and some pretty good stuff has turned
up.15 Surely, in any case,
the plausibility of Chomsky’s story doesn’t require that one crack
this nut first.
What’s wrong with trying to crack one’s nuts in parallel? I would
have thought that was the usual strategy of scientific research.
Why shouldn’t
Chomsky say (what, in fact, he is forever saying): UGs are about
what information the language acquisition process has access to.
They thus invite (but don’t provide) a theory of how that information
is exploited when a child infers a grammar from a PLD. It does follow
that UG isn’t, all by itself, a “real explanation” of language acquisition.
Cowie’s problem, however, is that nothing interesting follows from
that;
certainly not that postulating a mentally represented UG is other
than essential for providing the `real explanation’ that’s required.
The long and short is that Cowie needs a principled reason for doubting that the problem about how UGs
function in language acquisition can be solved; but all she’s got
is that, to date, nobody has solved it.
It
often seems that Cowie is tempted by a kind of dialectic that goes
like this: Somebody endorses a theory on the ground that it’s the
best (available) explanation of some or other evidence. `T because
it explains E,’ this guy says. `But,’ Cowie replies, `not T unless
D; and maybe not D.’ ( So, for example, maybe UG explains why grammars
have such a lot in common; but they can’t be what’s within unless
there’s a story about how you get from UG and a PLD to a grammar;
and we haven’t got such a story.) `So,’ Cowie seems tempted
to conclude, `not `T because it explains E’ after all.’
But
that way of arguing is no good. `T à
D & maybe not D’ simply does not rebut,
or even get a leg up on rebutting, `T because it explains E’. What
you need, if you’re to do that, is some
reason to believe `not D’ and `maybe not D’ doesn’t,
of course, amount to one of those. To the contrary (and this is
much of their charm), all else equal,
a best explanation argument vindicates those of its own premises
that are otherwise moot. If T à D, then if
T is the best explanation of E, that
is itself a `best explanation’ argument for D. It’s,
no doubt, desperately sneaky of best explanation arguments thus
to underwrite their own premises; they only get away it because
they are so shamelessly nondemonstrative. Be that as it may, it
makes them much harder to kill than Cowie seems to have an inkling
of. Perhaps, on balance, it’s just as well that they’re so hardy
since we’ve very little else to do our science with.
I
think Cowie’s failure to understand how best explanation arguments
work undermines quite a lot of her book. Still, her claim that there
is, de facto, no good empirical evidence for UGs could be true;
and, of course, you can’t run a `T because it explains E’ argument
if you don’t have any E. So I turn now to Cowies’ second objection
to POSAs, which is not that the bearing of their premises upon their
conclusions is dubious (as per 2.2.1), but that the empirical data
that the premises rely on are unpersuasive. In dispute here is primarily
whether, as POSAs suppose, the child’s PLD is so empoverished that
it radically underdetermines the grammar he acquires.
2.2.2. The status of the POSA data
POSA’s strategy is to claim
that there is less information in the PLDs from which children acquire
language 16
than would be needed if language acquisition were a species of learning.
To be sure, such claims are often impressionistic; for who knows
what a language learning process would demand of its input if it
lacked specific, prior information about the kind of language it
is to learn? Who knows, for that matter, anything about empiricist
learning processes, unless they are associationistic (a thesis to
which, as previously remarked, Cowie clearly does not wish to be
committed.)
There
are, however, some respects in which the issues can be focused.
For example, Chomsky often argues that the corpora children have
access to are unlikely to contain evidence that syntactic transformations
are `structure dependent.’ (According to Chomsky, `Is the man who
is wearing the hat bald?’ is the sort of sentence that shows that
question-formation is sensitive to phrase structure rather than
ordinal relations; for discussion see Chomsky (1972) 17.)
Likewise, a lot of recent theorizing about the `learnability’ of
various sorts of grammars proceeds from the assumption that the
child has access to little or no `negative evidence’ about what
expressions are not well-formed in his language. Much of Cowie’s long
discussion of POSAs is about whether, in point of fact, the PLD
really is empoverished in these respects. If it isn’t, then the
putative "poverty of the stimulus… does nothing to brace the
nativist position on language acquisition" (276).
On
this reading of her text, Cowie has nothing against POSAs as a form of argument;
she just doubts that, in the case of language acquisition, its empirical
assumptions are true; even if parents don’t correct a child’s ungrammatical
utterance overtly, their behavior may provide him with "subtle
cues (228)" to its ill-formedness.18
And, Jeff Pullam once found, in a corpus drawn from the Wall Street
Journal [sic!], "several" sentences that illustrate the
structure sensitivity of question formation (including: "How
fundamental are the changes these events portend?" and "Is
what I’m doing in the shareholders’ best interest?") Pullam
also found one in `The Importance of Being Ernest’ where Lady Bracknell
wants to know "Who is that young person whose hand my nephew
Algernon is now holding in what seems to me a peculiarly unnecessary
manner?"
One
might reasonably greet such observations with hilarity. It is, after
all, Oscar’s little joke that only Bracknellish sorts of people
talk in this Bracknellish sort of way. Indeed, in other moods, Cowie
is herself very impressed by how much about a language a child might
learn by attending to statistical properties of his corpus. "There is dramatic
experimental evidence that the statistical properties of the inputs
are used by children in order to abstract higher-level concepts
for apparently `unobservable’ syntactic properties. (191) 19 " Well, what
would you guess is the relative frequency of Bracknell-sentences
in speech that is addressed to (or overheard by) children? And if,
as one might suspect, it must be vanishingly low, why don’t children
who do happen to encounter such sentences prefer the hypothesis
that they are ungrammatical to the hypothesis that the regularities
in the PLD are structure sensitive?
Nevertheless,
Cowie is absolutely right about the state of the data; it is, as
she says "surely premature" to endorse a nativist account
of language acquisition solely ---or even mostly--- on observations
of what is or isn’t in the child’s corpus. Indeed, it always will
be surely premature; in linguistics, as elsewhere in serious science,
the confirmation of theories rests on an interplay between their
explanatory/predictive successes and all sorts of other considerations
about simplicity, economy, plausibility, the availability of alternatives,
and so on familiarly. At most, one is entitled to wonder aloud why,
if negative evidence and instances of the structure dependence of
transformations really are essential to language acquisition, does
the linguistic community make such data so hard for the child to
find? Why make the poor creature search for it in `subtle cues’
or in the back pages of the WSJ? Is there some conspiracy among
adults to keep the structure of their language hidden from their
children? Perhaps the facts of grammar are like the facts of life:
only to be revealed to those who have reached the age of discretion.
Pas devant les enfants? 20
Well, enough of
that; I don’t propose to enter into a detailed review of the empirical
literature on the typical contents of PLDs. Beyond doubt, every relevant observation
is susceptible to rational challenge. It’s an understatement to
claim that current assumptions "may be much too strong"
and that our current picture of the PLD may be "badly skewed"
(263). The trouble is: So what? At the risk of sounding merely pompous,
I offer a methodological observation: Linguistics isn’t philosophy.
(Neither, I suspect, is philosophy).
According
to the standard metatheory, philosophical arguments are supposed
to be knock-down; or better, lethal (for some good jokes about this,
see Nozick, 1981). This means, in
particular, that if you have a dozen arguments that P, all but one
of which prove to be unsound, the one that remains should still
be sufficient to make the case that P. In this respect, Philosophy
is required to be like logic; perhaps, in their most secret fantasies,
philosophers dream that it is logic. Probably that’s why so little philosophy works.
Linguistics, in
any case, is different. Like any other empirical discipline, it
appeals to a balance of plausibility. If, in particular, you consider
the whole range of empirical data currently available, it seems
pretty plausible that the PLD isn’t as rich as one might reasonably
expect it to be if a rich corpus is essential for acquiring a grammar.
My point is that attacking this claim the way Cowie does ---by attempting
to undermine the experiments one by one--- is simply not appropriate
to the polemical situation. What she needs, but clearly doesn’t
have, is an argument that the available data suggests, even remotely, a PLD so
rich that the child can is, as it were, squeeze through with lots
of room to spare. (Notice how, as usual, it’s the counterfactuals
that count; see fn. 15). There is, I venture to say, nothing in the psycholinguistic
literature that suggests this; and, to my knowledge, empiricist
arguments about language learning (Cowie’s definitely included)
never so much as claim it; they claim just that the data aren’t
apodictic. For the rest, one gets a priorisms: Empiricism should
be preferred not because the PLD is independently seen to be saturated
with information germane to acquiring a language, but rather on
grounds of the simplicity, or generality, or neurological plausibility,
or political correctness 21
of the learning theory that an empiricist approach would (/might,/might
some day,/might in principle some day) allow us to construct.
If, in short, you
wish seriously to evaluate the available data about the poverty
of the child’s stimulus, the pertinent question is not `which of
them can I perhaps impugn’; rather it’s whether, if they aren’t
entirely misleading, a move in the direction of empiricism seems
plausibly the way to account for them. Or put it like this: We know
what facts about the PLD are alleged to argue for the face plausibility
of the nativist picture; well, suppose all of those were to disappear.
The question remains: What are the facts about the PLD that are
supposed to argue for the face plausibility of the empiricist picture?
Answer: As far as I know (and, certainly, as far as Cowie tells
us) there are none.
2.2.3. Enlightened empiricism.
Suppose
it turns out (as I’d expect it to on the balance of the evidence
so far) that the PLD isn’t so rich as to make nativist speculations about the language
acquisition mechanism patently otiose. Suppose, even, that it turns
out that language acquisition requires a lot of domain specific
information of the kind that would be expressed by a motivated formulation
of UG. Still, it doesn’t follow that UG is innate. Maybe, rather, children start with
principles that are innate but not
domain specific (or, anyhow, not specific
to the language domain). Couldn’t the integration of such information
with the child’s’ nonlinguistic experience get him into a mind set
that will, when he finally gets around to learning his language,
require his hypotheses about the PLD to conform to UG? "It’s
impossible to think that the learner was told that grammatical rules
are structure-dependent. But it is certainly possible that she may
have had other experiences that would lead her to seek deep rather
than surface regularities.(182)." "Enlightened empiricism"
(=EE) allows that language acquisition may crucially require prior
knowledge of the domain specific sort that UG provides. That’s what
makes EE "enlightened". But it insists that this prior
knowledge is itself acquired rather than genotypically specified,
and that the procedures by which it is acquired are (eventually)
domain neutral. That’s what makes EE empiricism.
I
will not dwell at great length on enlightened empiricism; for, though
its plausibility is a main thesis of Cowie’s book, just think what
is being proposed: Of the three or four years that it apparently
takes a child to work out the grammatical structure of his language
22, some unknown
fraction is first devoted to constructing, on the basis of nonlinguistic experience
(together with general principles of nondemonstrative inference
(assuming there are such things)) a learned UG; viz a theory of
what the sentence structures in all the possible natural languages
that he doesn’t have
to learn have in common with the sentence structures in the one
that he does. What on earth would be the point (to say nothing of
the feasibility) of instituting such an indirection? Has God joined
the adult linguistic community in its plot to keep the grammar of
their language hidden from their children? EE grants to the nativist
that, whatever language a child may eventually learn to speak, he
must be in prior possession of the very same UG as every other child.
That being assumed, why doesn’t God just wire the damn thing in
species-wide and let each child spend his time learning to talk
the language of his speech community? It explains a lot to suppose
that God is sort of stupid (see Hume’s `Dialogues Concerning Natural
Religion’); but could he possibly be that stupid? 23
You will, in any
case, not be surprised by now to hear that Cowie offers no account,
and no examples, of how a domain neutral learning mechanism could
be used to construct a language-specific learning mechanism which
could then be relied on to deliver an adequate grammar of whatever
language the child happens to encounter. Instead, when this problem
starts to loom, Cowie is wont to speak of bootstraps.
`Bootstrapping,’
however, isn’t a theory of language acquisition (or, indeed, of
anything acquisition).
It’s just a name for whatever the process turns out to be that gets
a child first from nonlinguistic experience to knowledge of the
domains specific constraints UG imposes; and then from less good
theories of the PLD that observe these constraints to better theories
of the PLD that likewise observe these constraints; and, eventually,
to the right theory of the PLD (which observes these constraints
by assumption.) To say that the child solves the language acquisition
problem by bootstrapping is to say that he solves it somehow; which
is true, but not news. Since, to repeat, `bootstrapping’ is the
name of
this problem about acquisition, it is a fortiori, not the solution of this
problem. It’s extremely depressing to find cognitive science back in a condition
in which it is once again necessary to say such things.
Look,
is it Cowie’s assumption that the regularities in the child’s nonlinguistic environment
are structure dependent? If so, how
does the child learn that they are? Since
this seems to be another case of the same kind of question that
we started with (viz. how does a child recognize that a regularity
it encounters is structure dependent?) what has enlightened empiricism
bought for us that the old, unilluminated kind did not?
The
preceding paragraph gestures in the direction of what Cowie calls
an `iteration’ argument: If it’s common ground that a child can’t
learn a language unless he knows that P, and if, by assumption,
knowledge that P is learned rather than innate, then it just follows that
the child can’t learn the language unless he (somehow) learns that
P. `Enlightened empiricism’ adds nothing to this truism except the
assumption that the child learns P by (first) learning some (unspecified)
Q that entails P 24.
There’s a dumb joke about an enlightened empiricist who could count
sheep very fast. `How do you do it?’ everyone asked. `I count their
legs and divide by four’ he replied.
This, apparently, is the situation
Cowie has in mind when she admits that enlightened empiricists haven’t
a "detailed" alternative to nativism "on hand"
"yet".25
Considered as a
positive theory of learning, the version of EE that Cowie describes
is empty. But I wouldn’t want it to seem merely that Cowie has got
hold of slightly the wrong kind of EE; so I’ll briefly consider
an alternative formulation. I’d like to get it across that Cowie’s
research program is, as one might say, robustly
empty: tweaking the details doesn’t
make it any fuller.
One
might try holding some species of nonmodular
rationalism, (in effect, what Cowie
calls `weak rationalism’), according to which the child’s innate
endowment includes a domain neutral constraint enjoining him (ceteris paribus) always
to prefer theories that represent experiential regularities as structure
dependent. That would be perfectly all right with empiricists as
far as it goes; they take the principles of inductive inference
to be innate, and maybe a bias towards hypothesizing structure dependence
is one of these.
But
notice that this compromise view won’t work unless the experiential
regularities in nonlinguistic domains are typically structure dependent
in the same way that linguistic regularities
are; eg. they approximate to satisfying
the formal linguistic universals. There is, however, not the slightest
scintilla of evidence that any such thing is true. Indeed, supposing
that the kind of structure dependence UG requires of linguistic
rules will do for the general case would be a nonsense. The linguistic notion of structure
applies only in domains for which a
constituency relation is defined and independently motivated. Who knows which such domains there are? Surely some
domains have ordinal structure of precisely the sort that (if Chomsky
is right) sentences don’t; the months of the year starting with
January, for example. That being so, to insist both that the child’s
pre-wiring determines a domain-neutral preference for structure
dependence of the kind that language
exhibits, is to require the child
to prefer false
theories of such domains as happen not to be language-like. Only
a very
stupid God ---or a plain crazy one--- would endow the child with
a learning rule that’s biased toward a kind of structure that, de
facto, lots of domains don’t have
It
is, I suppose, a truism that domains whose structures can be learned
are ipso facto structured domains. But if you propose to make hay
of this truism, you need to keep in mind that the domains there
are, are structured in many different ways. If you don’t keep this
in mind, it might occur to you that nonmodular rationalism is (not
merely an alternative to Chomskian nativism but) a cognitive architecture
for which transcendental justification can be supplied.26
Cowie often writes as though she is moved by some such thought:
A preference for structure dependence is A
Good Thing As Such because `prefer
dependent regularities’ and `prefer deep, explanatory regularity’
are two ways of saying the same thing. See, eg, p. 189: "a
nonpositivist proponent of domain-neutral learning, taking Chomsky’s
lesson to heart, would surely endow her learner with a bias towards
seeking out the `hidden springs’ (and not the superficial regularities)
in the world, a bias that in the domain of language would manifest
itself as a preference for rules stated in terms of unobservables
over those stated in terms of observables, that is for [the structurally
dependent rule] H1 over [the structure independent rule] H¬¬¬2").
If, however, Cowie
takes this impulse to transcendental argument seriously, she must
be confused about what UG means when it says that linguistic rules
are structure dependent. Linguistic rules are dependent on constituent structure;
as opposed, say, to ordinal or cardinal structure; or the dimensional
structure of visual space; or the Fourier structure of auditory
stimulations; or the vector structure that Connectionists appear
to think that everything depends on. Each of these kinds of structure seems
quite `deep’ enough to be getting on with, so it’s hard to imagine
a kind of argument that would choose among them a priori.
The
kind of structure dependence UG cares about is just one among an
infinity of ways that rules, operations, processes, and the like,
might be sensitive to the organization of their domains. There is,
as far as anybody knows, nothing that prefers any one such domain
structure to any other in general. Nor is it easy to see why a process that is constituent
structure dependent should be especially "unobservable;"
or, indeed, why it should be endowed with any other epistemically
interesting property. That there is nothing especially interesting
about constituent structure is exactly why, if Chomsky is right
about all grammars having rules that are constituent-structure sensitive,
that’s a surprising discovery and it wants an explanation. Nativists
have such
an explanation, though, not one Cowie approves of; namely, that
UG is innate. There isn’t, "yet" an empiricist alternative,
transcendental or otherwise, to the best of my knowledge.
2.3 General learning mechanisms:
Almost everybody thinks that some
things must be learned; and almost nobody thinks that the basic
mechanisms of belief formation could be among them. Well, if it’s
common ground that some things are surely innate and it’s likewise
common ground that other things surely aren’t, what (other than
matters of degree) could there be left for nativists to argue with
empiricists about? One might thus wonder why modern rationalists
take so strong a line on acquisition mechanisms being domain specific.
Even if, pace Cowie, the domain specificity of learning devices
isn’t what they take to be the moral of POSA arguments, it’s clearly
true that most nativists are pretty grumpy about domain neutrality.
Why is that, do you suppose?
Fair
question. I have, however, only a fable with which to answer it.
Fable:
Once upon a time, there was this otherwise
unremarkable guy (history did not record his name, so let us call
him Anon; your local bookstore carries his stuff) who was really
extraordinarily good at answering questions about opera. He could,
for example, tell you every 19th century Italian composer of operas
whose last name ended with `i’ (of which, I assure you, there were
many.) He could likewise tell you who was the first violinist at
the second performance of `Lohengrin’, and who was the second violinist
at the first performance of `Lohengrin;’ and not just at Beyreuth,
but also in Salt Lake City. And he could tell you who manufactured
the swans. Also: Anon could quote the entire libretto of `Die Freishutz’
on request, and he knew where Callas sang on any evening in July
of 1957, and how many elephants Verdi wanted there to be in chamber
performances of `Aida.’ Mirabile dictu, Anon could explain the plot of `Simon Boccanegra’,
a thing that nobody else has ever been able to do. He was, as I
say, quite remarkably good at answering questions about opera, even
by the standards of opera buffs.
So,
of course, sensible people wondered a lot what could account for
his prodigious facility. After some consideration, they converged
upon the following hypothesis: `The reason,’ they said, `that Anon
is so good at answering questions about opera must be that Anon knows
a lot about opera. `That,’ they said,
`would explain it.’ Having arrived at this not implausible view,
they dispersed, each upon his own affairs.
Or
rather, they were just about to when there spoke a philosopher of
the empiricist persuasion. `It is not, after all, his knowing a
lot about opera that explains Anon’s surprising ability to answer
opera questions,’ said , the phthis philosopher. `Instead, it’s
that Anon has in his head what I call a `General Purpose Question
Answerer’ (of which I have discovered that the acronym is GPQA)’.
`Hmm,’ sensible
people replied, `what exactly is a GPQA, and how does one work?’
`I will tell you,’
said the philosopher of Empiricist persuasion. `A GPQA is a black
box that takes as its input any of an inordinately large range of
questions and provides the corresponding answer as its output. Here
is the flow diagram for such a device. It comes from the cutting
edge of cognitive science.’
----------
Q à GPQA à A
----------
Figure 1: Flow diagram for
a GPQA
`Pshaw!’
the sensible people replied; `for how could such a black box work?’
`It works by applying
General Question Answering Principles,’ replied the philosopher
of empiricist persuasion.
`And
what are these
General Question Answering Principles?’ the sensible people demanded
`As to that, inquiry
is proceeding in my laboratory even as we speak.’
Sensible
people thought about this for a while. Certain prima objections
occurred to them. For example: ‘If, as you say, Anon has a GPQA
in his head, what accounts for the domain specificity of his performance?
Why is it that, although he is remarkably proficient at answering
questions about operas, he is not nearly so good at answering questions
about, as it might be, bagels; or about who won The World Series
in 1905’?
`The
available data to that effect are unapodictic’ said the philosopher
of empiricist persuasion with hauteur. `Perhaps Anon is better at answering bagel questions
and baseball questions than has thus far appeared. Perhaps, when
closely examined, his behavior will exhibit subtle cues which show
that he knows the answers after all. Or perhaps there is more information
about opera in the environments where opera-questions are put to
him than cursory investigations have suggested. Let us not, in any
case, close our minds to such possibilities. For,’ said the philosopher
of empiricist persuasion (who had perhaps begun to sound a bit like
Auntie), "if I could urge just
one thing as a `take home lesson’
to be drawn from this discussion… it would be this: …We need to
look everywhere we can for relevant insights, data, and techniques
(Cowie, 308)." `Further research will therefore be required;
as will further funding.’
The
philosopher of empiricist persuasion was about to enlarge upon the
latter themes when sensible people, having, as they considered,
heard enough, commenced to pelt him with small objects. This forced
him to retire.
End fable.
Normal
human children are, as far as we know, quite extraordinarily good
at answering questions of the form: `What grammar underlies the
language of which the following corpus is a sample (insert PLD here)?’
But this competence is, in a number of respects, strikingly narrow.
For one thing, they exhibit no corresponding capacity for answering
questions about bagels. For another, it appears that children can
do their trick only if the PLD is drawn from a language the grammar of which
conforms to UG. It thus seems plausible to many sensible people
that part of the reason children are so good at answering questions
about what grammar underlies a PLD is that they come to the task
already knowing a lot about what kinds of grammars conform to UG; specifically,
they know UG. And since there is no proposal on offer about how
a child could possibly have learned UG before he learned his native language, many sensible people
think that UG must be innate. And even sensible people who don’t
think that it’s exactly UG that’s innate are inclined to think something
must be that is at least equally dedicated and equally complicated.
A fortiori, they think that what’s within isn’t a General Purpose
Learning Machine.
Wherein
does the symmetry fail? If postulating a General Purpose Question
Answerer is not a reasonable alternative to the theory that Anon
knows (innately or otherwise) a lot about opera, why is postulating
a General Purpose Learning Mechanism a reasonable alternative to
the theory that children know (innately) a lot about UG? One wanders
through the empiricist landscape, holding one’s little lantern aloft,
asking this question of the locals one encounters. And never getting
a sensible answer. That’s what makes nativists so grumpy.
2.2.4 Essentialism: I remarked, at the beginning of the discussion of
POSAs, that quite a lot of Cowie’s polemic amounts to reiteration
of points that are familiar from the linguistic and psycholinguistic
literature. She does, however, offer one line of pro-Empiricist
argument that is, as far as I know, quite without precedent: "[According
to Chomsky] Linguistic Theory characterizes the essential properties
of languages; it delimits the set of possible natural languages.
But it is in general false that we need to know about the essential
properties of a thing in order to learn about it. … a child’s grip
on cathood predates her excursions into zoology…. Reflection on
the nature of learning tout court, I’m suggesting, should have alerted
us .. to the possibility that Chomskyan theories of language learning
are on the wrong track. (273)"
I
do think that is confused. For one thing, acquiring the concept
CAT does, of
course, require learning what property is proprietary to cats as
such; namely, being cats. Likewise, in the untechnical sense in which a monolingual Russian speaker
can perfectly well have the concept ENGLISH SENTENCE, his having
it doesn’t
(according to Chomsky or anyone else) require his cognizing the
grammar of English. It requires only that he understand that whatever
ENGLISH SENTENCE applies to is, ipso facto, an English sentence.
Or (a formulation I prefer on balance) it requires only that he
be able to think about English sentences as such.
By
contrast, Chomsky puts in play a quite technical notion of concept
possession according to which mastering ENGLISH SENTENCE requires
becoming a speaker/hearer of English (and hence, by assumption,
cognizing English grammar.) Whether Chomsky’s technical sense of
concept possession actually applies to anything is, of course, a
Very Deep Empirical Issue. If it doesn’t, then he and I have both
wasted quite a lot of our time over the years. In any case, according
to Chomsky’s usage, the typical consequence of having the concept SENTENCE
OF L is being able to recognize and construct arbitrary sentences
that belong to L. Since, to repeat, nothing of the sort is true
of concept possession in the vernacular, learning CAT does not (Pace
Cowie) provide a model for learning ENGLISH SENTENCE. 27
Once that is straightened
out, it’s really quite plausible that when having
a concept of
X requires being able to make and
recognize Xs,
coming to have the concept of X will require mastering a metaphysical theory about
Xs. That’s
why, though people have had the concept WATER for simply ages, it
was only when we learned what the property
of being water is (only, as one says,
when we got the `technical’ concept WATER) that we were able to
make some in the laboratory, and to distinguish arbitrarily close
approximations to water from the real thing. Likewise, though we’ve
had cats and their concept ever since we lived in Egypt, it’s only
quite recently that we’ve begun seriously to contemplate building
a cat from scratch.
So
much for Part 2. Let’s turn to the impossibility arguments.
top | PART
3 |