<< . .

( 19)

. . >>

tweentheinitialstate,thetopicof UniversalGrammar,andtheattained
states, the actual languages. But I think that, at least within the P & P
approach, it is more reasonable to forget about that distinction: the lan-
guage faculty just has states; one state is the initial state; others are the
stable states that people reach somehow, and then there are all kinds
of states in between, which are also real states, just other languages.
If the strong “No Dead End” Condition is met, then the minimalist
thesis would say that all states have to satisfy the condition of infinite
legibility at the interface “ and to do so in an optimal manner, to the ex-
tent that the strong minimalist thesis holds. That is orthogonal to the
dimension of explanatory and descriptive adequacy, because it holds in
both the initial state and the attained states. So it™sboth explanatory and
descriptive, but the distinction is by and large put aside. One nice thing
about the P & P approach, which at least I didn™t realize at that time,
is that it essentially eliminates the distinction: it eliminates the princi-
pled distinction between the initial state and the attained states. That
looked like a principled distinction in the earlier period and it is princi-
pled in the sense that the initial state is an expression of the genes, and
the others are not entirely, but from the point of view of the adequacy of
theories, the distinction doesn™t matter: you want an adequate theory
for all, they all have to be descriptively adequate, meaning true theories
of whatever state you are describing (if it is the initial state, this is what
was called explanatory adequacy). If the minimalist thesis holds, it will
hold for all states, at least on the “No Dead End” assumption. These
questions are really in the process of being formulated, alongside of
efforts “ with some success, I think “ to show that strong minimalist
conditions can be approached in some domains, sometimes attained.

On nature and language

ab & lr: Keeping for a moment this classical distinction, it has
often been said that there are tensions between the goals of descriptive
and explanatory adequacy as the first typically favors the enrichment
of descriptive tools, while the second favors restrictiveness and the im-
poverishment of the descriptive apparatus. It seems to us that partly
analogous tensions could arise between the demands of explanatory
adequacy (in the classical sense of adequacy in addressing the log-
ical problem of language acquisition) and minimalist explanation.
It is conceivable that a less structured, hence more minimal, system
would allow for more alternative analyses of the primary data, thus
making the task harder for the language learner. To give a concrete
example, consider a theory of phrase structure permitting a single
specifier for each head, and one allowing for multiple specifiers. One
could argue, even though the point is not entirely obvious, that the
second is more minimal in that it lacks a specification that the first
has. But consider the problem from the viewpoint of acquisition: the
language learner hears an expression with n phrases and must inte-
grate them into a structural representation. In the first theory, s/he has
no choice: s/he must assume n heads licensing the phrases as spec-
ifiers; in the second theory, s/he has a priori many options ranging
from a single head with n specifiers to n heads, each with a single
specifier. Of course this is crucially related to the question of what can
constitute a possible head, and in practice there are many other com-
plications, but the example is simply aimed at suggesting that some
tensions could arise here. Do you think this tension actually arises?
nc: It could. Minimalist questions are substantive: they ask
whether true theories of states of the faculty of language satisfy the
interface condition in an optimal way. If a proposal yields as options
languages that can™t exist, it is just the wrong theory. The same con-
clusion holds if the proposal does not yield a solution for the logical

An interview on minimalism

problem of language acquisition. So, the first condition that has to be
met is truth for every state of the language faculty. At the initial state
it has been called explanatory adequacy, at a later state, descriptive
adequacy. By now, I think this terminology is basically useless; as I
said, just truth matters. Of course, it is not the case that we are given
the truth and then we ask minimalist questions: life isn™t that simple.
You ask minimalist questions to reconstruct your conception of what
is probably true and so on and so forth. Logically speaking, the con-
dition in the background must be that you have got the true theory.
Take for example the case you mention. There are articles on that in
the current literature. Linguistic Inquiry has a recent article in which the
author says that his way of doing things does not require the special
assumption that there are multiple specifiers. But that puts the matter
backwards: the assumption that there is a single specifier is a special
assumption; to say that there are any number of specifiers is not an
assumption, it™s just to say you may continue to merge indefinitely: it
merely states that language is a recursive system. To say that there must
be a single specifier and no more, is to stipulate that when you merge
twice you have got to start a new category: that™s a special enriching
assumption. So, there is no issue of getting rid of the extra assumption
of multiple specifiers; on the contrary, you would need evidence for
the special assumption that you can only have two things attached to a
head. Selectional properties of roots may “ in fact surely do “ impose
conditions on multiple Merge to a single head. But a strong argument
would be needed to show that the same condition must be restated,
independently, within the theory of phrase structure, complicating
that theory, largely redundantly.
In a bare phrase structure theory the distinction between com-
plement and specifier disappears, there is no difference: it™s just first
Merge, second Merge, third Merge, and so on. So, from this viewpoint,

On nature and language

a lot of analyses which I have given just don™t make any sense. Take
adjectives, for instance; I used to worry about whether the element
selected by an adjective is a complement of its head or a specifier of
its head, rather different things, but in a bare system you can™t ask that
question. It™s attached to the head; we call it complement if it™s first
Merge, but it doesn™t mean anything, there™s no further question to be
asked. And the notations that we use are rather misleading; we put
it in front of a head if we mean it to be a specifier, after the head if
we mean it to be a complement: those are meaningless distinctions in
a bare system. So the whole notion of complement and specifier dis-
appears except as a terminological convenience: you have the things
that you merge first, the things that you merge second, and so on.
Let™s now assume that we have the simplest system, meaning no
extra conditions on how many times you are allowed to merge; so you
can do it once, you can do it twice, in which case we call it a specifier,
three times, in which case we call it multiple specifiers, and so on, but
just merge any number of times you like, plainly the simplest system.
And of course we want to know: is it true? Is language perfect in this
respect? Or does it have this extra requirement that you can only
merge n times, for some fixed head, maybe two? Now let™s go back
to the child acquiring the language. If the child is acquiring the lan-
guage with the principle of Universal Grammar that says you can
merge as many times as you like, the child hears two merges and
OK, that™s fine; then he hears the third thing come along and, you™re
right, the child has two choices. One is to say: “OK, it™s third Merge,”
the other is to postulate a new head. But that™s a hard choice: to postu-
late a new head you have to have evidence, you have to know what head
it is, to find it somewhere and if it is a zero head as it could be in this
case, it is very hard. If it is a head that doesn™t have any semantics, you
are in trouble because that head will have to disappear in the course

An interview on minimalism

of computation, which will leave you with a headless category and
you™ll have to tell some story about that. If there is some universal
set of options, say, Cinque™s hierarchy, then you can pick something
out of that, but then there has to be a semantic consequence and you
have to have evidence for it. So I don™t think it™s a question of harder
or easier choice, it™s just different choices. If the Universal Grammar
has Cinque™s hierarchy and no limitation on merging, then when you
get to that third element, the child will have to ask whether it has the
semantics of something in the hierarchy. If it does, then that™s where
it belongs; if it doesn™t, just merge down below and that™s the answer.
Let™snow take the other approach; suppose that phrase structure
theory is complicated to impose the (largely redundant) requirement
of single or double Merge, not triple Merge. Then the child is forced
to find another head; and if there is nothing around that makes any
sense, it will just have to invent it, and that™s a harder task. So, I don™t
think that the conflict breaks up this way. It seems to me that there are
different factual assumptions about the nature of language. Are there
heads available with the kind of semantics that will compel the child
to merge to them, whether it is third Merge or fourth Merge?
In fact the same question arises for second Merge. Suppose the
child assumes first Merge on a head, and then a second expression
comes along. Let™s assume a Universal Grammar which has no lim-
itation on specifiers and the Cinque hierarchy. After the first Merge,
when the second expression comes along, the child is confronted with
the same question: does this have the semantics of one of the positions
of the hierarchy, because it has some kind of aspect interpretation, or
the like? Well, if so, then the child should postulate a new head; if not,
then the element is a specifier of the first head and the same question
arises on third Merge, fourth Merge, and so on. The situation you are
mentioning could arise and then it would be a question of truth; so

On nature and language

the truth may be that you have more complicated phrase structure,
with conditions on the number of specifiers over and above those
that follow from selectional requirements. For example, take the LCA
(Linear Correspondence Axiom.)15 If that theory is true, then the phrase
structure is just more complicated. Suppose that you find out that gov-
ernment is really an operative property. Then the theory is more com-
plicated. If ECP really works, well, too bad; language is more like the
spine than like a snowflake.16 You can™t change reality, you can only
ask: does reality happen to meet these surprising conditions?

Minimalist questions and other scientific domains

ab & lr: Granting the common background of methodolog-
ical minimalism as a component of scientific inquiry, are substantive
minimalist questions ever asked in other scientific domains?
nc: Not often, I suppose, but they are in some. So, for exam-
ple, there is a standard joke in physics and mathematics that the only
numbers are 1, 2, 3, and infinity; the others are too complicated, so if
anything comes out, say, 7, or something like that, it is wrong. And in
fact that actually shows up in scientific work. It showed up in the de-
velopment of the theory of quarks, apparently: if I remember correctly,
when Murray Gell-Mann and his associates were devising the theory,
it turned out that they had evidence for seven quarks, but nobody was
happy with that, because 7 is too ugly a number; so the assumption
was that the picture must be reconstructed in terms of 2 and 3, which
are nice numbers. And after further experimental work stimulated by
that intuition, the prettier picture turned out to be true. I think that
that kind of reasoning does go on. In a sense the discovery of Pluto
was kind of like that. There were perturbations, so it could be that
the world is ugly and you have to make up some story; but everyone

An interview on minimalism

was very happy when they found a postulated entity out there which
may or may not be a planet, that is debated, but whatever it is, it is
out there and it accounts for the perturbations without complicating
physical theories. You want the systems to look nice. Take the Periodic
Table, for example. The known facts didn™t entirely fit, but it was so
nice that it had to be right, so it didn™t matter if they didn™t fit. There are
famous examples in the history of science that are similar. Chemistry,
which is a rather revealing model for linguistics, provides many exam-
ples. Many chemists were unhappy with the proliferation of elements
and chemical atoms in the theories of Lavoisier and Dalton. Humphry
Davy, for example, refused to believe that God would have designed
such an ugly world. At the same time, in the early nineteenth century,
William Prout observed that the atomic weights of the elements were
pretty close to integral multiples of the atomic weight of hydrogen,
and fudged the data to yield whole numbers exactly. “Prout™s hypoth-
esis,” as it was called, stimulated heavy experimental inquiry trying to
find the exact deviation of the atomic weight of heavier elements from
an integral multiple of hydrogen and to try to find some explanation:
is Prout™s hypothesis right or wrong? Are all elements constructed
from hydrogen, as he speculated? Finally isotopes were discovered in
the 1920s and then it all became clear: it was clear that Prout™s hypoth-
esis was fundamentally correct. Without an understanding of isotopes
and atomic theory generally, the data are a mess. But if you reanalyze
the data in terms of new theoretical understanding, you discover in just
what sense Prout™s hypothesis was correct, because you get a proton,
many protons, its integral multiples, and electrons don™t add much,
and isotopic effects modify the numbers systematically. The research
inquiry was driven by the hope that somehow this pretty law will turn
out right and there will be a reason for it; finally the reason was found,
and incidentally a good deal of the experimental work of a century

On nature and language

went out of the window; nobody cared anymore what the average de-
viations were because you had a fundamental explanation for them.
I suppose the Galilean ideal of perfection of nature is, at some
level, a driving force in all inquiry, but it certainly isn™t very much of a
leading force in most fields, any more than it has been in linguistics.
A good reason is that it is so hard to gain something approaching
descriptive adequacy that you can™t realistically ask further questions.
Take a look for example at Mark Hauser™s recent comprehensive
study Evolution of Communication.17 It really is a comparative study of
communication, comparing communication systems. He reviews a
lot of systems and describes them in very delicate detail. Take the bee
dance. There are extremely detailed descriptions of it, but it™s basically
like descriptive linguistics. Questions that go beyond are apparently
too hard: for example, what is the “generative grammar” of the bee
dance, the internal state that allows for this range of dances and not
some other range? Or questions about neural mechanisms, their role
in action and perception, their evolution. The problem of just giving
a description is hard enough; and then finding some understanding
of the function of the dance. To go beyond that, to get real minimalist
questions is hard, but there were people who were trying to do it also
in biology. A famous example is D™Arcy Thompson.
ab & lr: This leads to the next question. Let us assume that
some form of the minimalist thesis is correct, and human language
is a kind of optimally designed system. You have often stressed that
this would be a very surprising conclusion in the context of biologi-
cal systems, which are characterized by the “bricolage” or tinkering
of evolution, in Fran¸ois Jacob™s terms.18 So, it would be useful to
try to spell out the consequences of this discovery for biology. One
possible line of approach could be to think that language is effec-
tively rather unique among biological systems, possibly in relation to

An interview on minimalism

its combinatorial character; but it could also be that language readily
reveals something that is more common than usually assumed in bio-
logical systems, but only difficult to detect. Could it be that the role of
tinkering has been overstated? And that at different levels of the evo-
lutionary scale “perfect systems” may have come to existence, but are
hard to tease apart from their biological context?
nc: That is, I think, quite reasonable. It is unpopular today, but
the fact is that if you take a look at anything that you don™t understand
it™s going to look like tinkering. That was true of the way people looked
at languages. If you go back to the 1950s a standard assumption “ I am
paraphrasing Martin Joos, one of the major theoreticians “ was that
languages can differ from one another without limit and in arbitrary
ways. Basically, there is nothing much to say about language: almost
anything goes.19 That is certainly what it looks like. If you consider
the range of languages in the world, it looks as though you can find
just about anything. That was a standard point of view in structural-
ist linguistics, which departed from this assumption only in limited
ways: there is some fixed structure of the phonemic system and maybe
a little bit more, maybe some of the morphology, some loose condi-
tions on phrases . . . but essentially anything goes. Sapir said similar
things and in fact it™s pretty common.20 And it™s true: if you look at
anything that you don™t understand it is exactly what it is going to look
like. With regard to evolution, everybody believes Darwin is basically
right, there™s no question about that; but beyond that, not too much
is understood. For evolution of species, there are few cases in which it
can be demonstrated, by the standards of the sciences, that natural se-
lection operated, though everyone assumes that it is true. It is not easy
to measure selective advantages of traits. When you look at what are
called “natural selection explanations,” what you often find is some-
thing different. Hauser™s book is a good source there. He™s trying to

On nature and language

show in detail what everybody believes generally: that natural selec-
tion functions crucially to yield and design an output. But the kind of
argument that he gives doesn™t show that. So he takes bats and shows
that they have an amazing technique of echolocation: they can find an
insect flying somewhere and shoot right at it by some kind of echoes
that manmade systems can™t duplicate. The conclusion is: look at how
beautifully natural selection worked. That is very plausible, but the
argument doesn™t show it; what is shown is that it has these beautiful
characteristics. A recent review of the topic in Science points out that
it is plausible to suppose that piranha teeth evolved for cutting, “but
we have no direct evidence that that was the case.” A creationist might
say, irrationally, that God made it that way. It is just that if you have
a naturalistic approach to the organic world, you assume that it must
have been largely the result of natural selection. A description of the
beautiful adaptation to the organism™s needs is just formulating the
problem to be addressed. The problem is: here™s the object, here are its
strange properties marvelously adapted for survival and reproduction.
That sets the problem, but doesn™t answer it. It is often taken to be
an answer to the problem, on the assumption that the outcome has
to be the result of natural selection. The dogma in this case is pretty
plausible (it™s hard to think of anything else), but that™s not an an-
swer and sometimes, when things have been looked at carefully, the
answer turns out to be something different and unexpected. Things
are what they are, not necessarily what we dreamt of. In fact, at the
moment, little is known about evolutionary processes other than the
main principles, and a huge amount of descriptive work that yields
highly plausible assumptions (like echolocation and piranhas™ teeth),
of course, a lot of special things about what genes do, and so on and
so forth. But it does look mostly like a mess, and it may not be. It
may be that the whole of evolution is shaped by physical processes in

An interview on minimalism

a deep sense, yielding many properties that are casually attributed to
Now, of course, when people say that something is the result
of natural selection they don™t mean it literally. Natural selection can™t
work in a vacuum; it has to work within a range of options, a struc-
tured range of options; and those options are given by physical law
and historical contingency. The ecological environment is in a certain
state and it is going to impose constraints: you could imagine a planet
in which you have different ecological conditions and things would
work in a different way. So, there are contingencies and there™s phys-
ical law and within that range natural selection finds its way, finds
a path through it; but it can never be the case that natural selection
is acting on its own. The logic is rather like that of behaviorism, as
was pointed out by Skinner, incidentally.21 He thought it was an ar-
gument for his radical behaviorism, that it works like unstructured
natural selection: so the pigeon carries out any possible behavior and
you reinforce the one you want, and you get pigeons playing ping
pong, etc. He argued this is the same logic as natural selection, which
is true, but what he missed is the fact that natural selection requires
a structured environment, structured entities, and the conditions im-
posed by natural law, and the same is true of the pigeon. So, it is the
same logic, and the same mistake for both. And it™s common. When
you read these excited pronouncements about “show me good design
and I™ll find natural selection,” “God or natural selection,” taken lit-
erally, it™s worse than Creationism. Creationism at least is coherent;
you can be a rational creationist (Voltaire, Jefferson, etc.), you can even
be a neo-Darwinian. A rational creationist could say OK, all this stuff
happened by natural selection but God was necessary to do X. There
is no point in this vacuous assertion, but it is not incoherent. On the
other hand, a belief in pure natural selection would be totally irrational;

On nature and language

it is assuming that some selectional process can take place in a vac-
uum, which can™t happen. It is always the case that what goes on is
to some extent conditioned by physical law at least. There is a kind of
“channel” set up by physical law and, in addition to that, there are his-
torical contingencies and so on. Within those structured constraints,
natural selection can operate. Well, that raises a question, always: to
what extent is the channel functioning in determining the output? It is
going to be more than zero, it has to be. In some cases, it may approach
100 percent. Take the fact that you find the Fibonacci series showing
up all over the place. Nobody believes that it is either God or natural
selection; everybody assumes that it is the result of physical law and
by now there are non-trivial physical explanations of why you should
find it. So, between 100 percent and something, that™s the effect of the
Now, when you understand very little and it all looks like a mess,

<< . .

( 19)

. . >>

Copyright Design by: Sunlight webdesign