<< . .

( 19)

. . >>

probably some good reason for that, if we could figure it out.
ab & lr: So, the displacement property is an inherent property
of natural languages, one that any theory of language aiming at em-
pirical adequacy must express in some way. As for the question why it
is so, you offer the speculation that displacement may be an optimal

On nature and language

solution to the problem of having to connect two types of seman-
tic properties to expressions, traditionally, deep and surface semantic
Now, we can pursue the speculation and ask why displacement
is the solution chosen by natural language syntax. Clearly there would
be other possibilities.
Consider for instance the model, normally adopted in phonol-
ogy, according to which the sequence of units is on a line at the in-
tersection between distinct planes, such that each plane expresses
certain properties, and a unit can be simultaneously assigned proper-
ties expressed on distinct planes.
A priori, the integration of thematic and informational proper-
ties could work like that, with the same position assigned the property,
say, “patient” on one plane and “topic” on another (with, say, deep se-
mantic properties signaled by one kind of affixes, and surface semantic
properties also signaled in situ by another kind of affixes). Still, natural
language syntax does not seem to work like that in the general case.
Rather, it postulates positions uniquely dedicated to the prop-
erty “patient” (say, under the Hale“Keyser theory of theta roles), and
positions uniquely dedicated to the property “topic,” with the same
element occurring in different positions in the same representation,
and thus picking up both interpretive properties.7 This is the displace-
ment property.
In other words, natural languages seem to prefer to solve the
problem of connecting deep and surface semantics by proliferating
occurrences of elements, rather than by proliferating intersecting
planes, or finding other ways to assign different types of interpretive
properties to the same position.
Could we speculate on why language systematically goes for
this solution? Could this tell us something about the requirements

An interview on minimalism

imposed by the interface systems? Could the requirements of lineari-
zation on the PF side be of relevance here? Or some other constraint
on the format of legible information on the LF side?
nc: It™s a very interesting question, which arises at the outer
limits of current understanding, so anything one suggests has to be
very tentative.
Suppose first that there was only “deep” semantics, so the prob-
lem of displacement does not arise. We now ask: why does language
(apparently) identify semantic roles by configuration instead of by
particular inflectional elements? Actually, it seems to do both. Thus,
Inherent Case (say, Ablative) does identify a semantic role by inflection,
while Structural Case (Nominative-Accusative, or Ergative-Absolutive)
carries no specific semantic role. For elements with Structural Case,
the semantic role is determined configurationally, typically by virtue of
their relation to the element that selects them: subject and object of a
verb, for example. That this is true is by no means obvious; until quite
recently no such distinction was recognized. But it seems to be correct.
Furthermore, configurational relations also seem to enter into deter-
mining the semantic relation of an element that has Inherent Case.
If so, language uses both devices “ both inflection and config-
uration “ to assign semantic relations, quite apart from the matter of
displacement. We therefore want to know why this is so. The natural
place to seek an answer is at the interface between the language faculty
and the systems of thought to which it provides information. Pre-
sumably, these external systems distinguish among various kinds of
semantic relations, and prefer to have them signaled in different ways.
One can proceed to develop further ideas about what these properties
of the thought system might be. We are now in a notoriously difficult
area, because it is so hard to find out anything about these systems
apart from their interaction with the language faculty. We are asking

On nature and language

about thought without language, in traditional terms, a concept often
rejected, though it seems to me reasonably clear that something of the
kind must exist.
Turning to the question of displacement, the question about
configuration vs. inflection once again arises. Why does language pre-
fer to signal the “surface semantics” configurationally rather than by
an inflectional system of the Inherent Case variety? Again, one place
to seek the answer is at the interface. Thus we might ask whether, and
if so why, the external systems require that the surface semantics fall
together with the deep semantics that is not signaled inflectionally
by Inherent Case. But here there are also other possibilities. If surface
semantics were signaled by inflection, the underlying morphologi-
cal system would be complicated. For elements with Inherent Case,
there would be double inflection if they have distinct surface-semantic
properties; for elements lacking Inherent Case, they would have in-
flection only in this case. In contrast, if surface properties are signaled
configurationally, at the edge, the morphological system is uniform
throughout: a single Case inflection always (whether manifested pho-
netically or not). Possibly that is a factor.
Are requirements of linearization on the sound side relevant?
Perhaps so. To pursue the matter further we should introduce into
the discussion languages with more free word order and (typically)
richer manifested inflection “ languages of the kind sometimes called
“non-configurational” (though the term is probably inaccurate).
This is no answer: rather, a suggestion as to where one might
look for answers to questions that definitely do arise, and in interesting
ways, particularly in the context of serious pursuit of minimalist issues.
ab & lr: If it is true that a constitutive characteristic feature
of natural languages is to privilege representations with many dedi-
cated positions, each with simple interpretive properties, it becomes

An interview on minimalism

important to draw a map as precise and fine-grained as possible of this
complex positional system. This is the rationale behind the so-called
cartographic studies, which are pursued intensely in some research
centers in Italy and elsewhere. How can this endeavor relate, in your
view, to the topics and goals pursued by the Minimalist Program?
nc: This work has led to fascinating results in many areas.
To first approximation, the clause seems to be of the general form:
[ . . . C . . . [ . . . T . . . [ . . . V . . . ]]], where V is the verbal head of the con-
figuration in which deep semantic roles are assigned, T is the locus
of tense and event structure, and C (complementizer) is a kind of
force indicator distinguishing declarative, interrogative, etc. But the
cartographic inquiries have made it very clear that this is only a first
approximation: the positions indicated by . . . have a rich structure.
The “left periphery” includes not only force indicators, themselves
differentiated, but also at least fixed positions for topic and focus; and
the Cinque hierarchy yields a very detailed and apparently universal
array of structures in the T-V region.8 Other work in progress has pro-
vided much insight into the positions at and to the left of T, which host
clitics and inflections in various ways; and into apparent parallels be-
tween the T-based configuration and the V-based configuration. There
are no obvious reasons, at least that I see, why the facts of language
should distribute in just this fashion, so once again we are led to the
kinds of questions you raised about configurational vs. inflectional
solutions, here in a much richer and more diverse terrain.
This kind of work leads us to inquire more closely into the na-
ture of interface relations; the traditional two-interface assumption “
sound and meaning “ is presumably only an approximation. And
beyond that, it leads us to investigate the “external” systems them-
selves, and the conditions they impose on a well-designed language
faculty. As is common, these questions have traditional antecedents,

On nature and language

but it seems that they can now be addressed on much firmer grounds,
and in much more promising ways, in large part as a result of such
endeavors as the cartography projects.
ab & lr: What kind of empirical discovery would lead to the
rejection of the strong minimalist thesis?
nc: All the phenomena of language appear to refute it, just as
the phenomena of the world appeared to refute the Copernican thesis.
The question is whether it is a real refutation. At every stage of every
science most phenomena seem to refute it. People talk about Popper™s
concept of falsification as if it were a meaningful proposal to get rid
of a theory: the scientist tries to find refuting evidence and if refuting
evidence is found then the theory is given up. But nothing works like
that. If researchers kept to those conditions, we wouldn™t have any
theories at all, because every theory, down to basic physics, is refuted
by tons of evidence, apparently. So, in this case, what would refute the
strong minimalist thesis is anything you look at. The question is, as
in all these cases, is there some other way of looking at the apparently
refuting phenomena, so as to preserve or preferably enhance explana-
tory power, where parts of the phenomena fall into place and others
turn out to be irrelevant, like most of the phenomena of the world,
because they are just the results of the interactions of too many factors?
That™s one reason why people do experiments. They do experiments
to try to get rid of irrelevant phenomena: the point of the experiment
is to try to throw out most of the phenomena and discover just those
that matter. An experiment is a highly creative act; it™s like creating a
theory. One may not talk about that in methodology courses, but the
working scientist certainly knows it. To try to devise the right experi-
ment is very hard. The first experiment you think of is usually garbage,
so you throw out the experiment and try to get a better experiment and
so on. Finding the right experiment is very much like finding the right

An interview on minimalism

theory and in fact intimately related to it: serious experiment is theory-
guided, sometimes to answer questions that arise in the search for
explanation and understanding, sometimes because you can see that
the phenomena apparently refute your theories and you want to de-
termine whether that is just an artifact. Unanalyzed phenomena don™t
really matter much in themselves. What matters is the results of prop-
erly designed experiments, and “properly designed” means internal to
a theory. That™s true whether the experiment is about the relation be-
tween movement and manifestation of inflectional features, or about
language acquisition, or anything else.
Take a concrete example from linguistics and cognitive psychol-
ogy, one that has been badly misunderstood, the experiment that Bever,
Fodor, and Garrett did on click displacement.9 The idea was to see if
you could find phrase boundaries, perceptually, by looking at the dis-
location of a click. So, you play a piece of tape, put a noise somewhere
and ask people where they hear it, and it turns out that they don™t hear
it where it was, they hear it displaced somewhere; maybe the click was
displaced to the edge of the phrase because of some Gestalt property
that says that you try to maintain closure, you don™t want to be inter-
rupted in a coherent unit, so you perceptually displace it at the edge
of the unit. If that worked, it would be an interesting way of finding
phrase boundaries. What they were interested in were the hard cases,
like Exceptional Case Marking contexts: do you have object raising or
not, etc.? So if you have John expected Bill to leave, where is the phrase
boundary? Is it after Bill or before Bill? This is a real question, and the
way they proceeded was completely reasonable: first let™s design an
experiment that works; if we get an experiment which we have faith
in, because it is working in the cases where we know what the answer
is, then we will apply it in a case where we don™t know the answer,
and that™s what they did. They did a lot of experiments, but what was

On nature and language

published was an attempt to show that the experiment works, not
to provide new results. In other words, you don™t want to have an
experiment that is going to give the wrong result in clear cases, i.e.
one which in John saw Bill would put the break between saw and Bill.
First you have to find an experiment that works. Suppose that it turned
out that the click invariably got displaced to the middle of the phrase,
then it would have been a good experiment, but it would have been
interpreted the other way: the Gestalt property is that you displace the
click to the middle, we™ve shown that, because that™s what happens.
Testing the experiment and deciding how the experiment should be
interpreted, that™s a large part of the work. In fact, in the case of the
click that was essentially all the work. Well, when they got something
that seemed to work (displacement to the edge), then they tried it on
the hard case: unfortunately, it didn™t give very clear results, so it wasn™t
much pursued. But that shows what experiments are like. Now, this
has been seriously misinterpreted. For example, by W. V. Quine, who
has been much interested in methodology of linguistics for a long
time, since the 1940s. At one time, he argued that phrase boundaries
are just an artifact, just as they would be in a formal language, the
model he had in mind apparently, as is pretty common.10 For formal
languages, there is no “right” grammar; it™s arbitrary, you pick any one
you like. So by analogy, in language the linguist can pick any grammar,
depending on one or another concern or interest; the only thing that
is real is the utterances. That™s a false analogy to start with; human
languages are biological objects. What is real “ what is in the brain “
is a particular procedure for characterizing information about sound,
meaning and structural organization of linguistic expressions. The
choice of a theoretical account is no more arbitrary than in the case
of the visual or immune systems. But pursuing the analogy to formal

An interview on minimalism

systems, back around 1970 Quine argued in an article on the method-
ology of linguistics that it is “folly” to assume that there is a real answer
to the question of where the phrase boundary is in something of the
form ABC: it could be between B and C or between A and B. It™s just like
picking an axiom system for arithmetic, any way you like. Later, after
the click experiments came out, Quine changed his mind and said:
“Now it™s real, because the click experiments show that there really is
an answer.” This is a serious misinterpretation. The work on clicks
he refers to was testing the experiment, not the phrase structure. If
the click experiments had given the wrong phrase structure in clear
cases, that would have shown that the experiment is not well designed.
One wouldn™t say: “The phrase boundaries are not where the linguists
thought, they™re in the middle of a word” on that basis. Suppose that
the click was always heard in the middle of the sentence, so usually
in the middle of the word. From Quine™s point of view, you™d say:
“OK, that™s where the phrase boundary is,” but from any scientist™s
point of view, you would rather say: “Well, it™s a terrible experiment.”
And in fact, if the clicks were displaced towards the middle of the
phrase you just would reinterpret the experiment. From within the
framework of the empirical sciences, first you have to test the experi-
ment and that™s hard: most experiments are just irrelevant, and to find
an experimental procedure that really makes sense is very difficult. It™s
a theory-internal task, often undertaken because the phenomena of
the world are apparently refuting everything, and you want to discover
whether, and how, the appearance is misleading.
So, to get back to your question after a long detour. If you want to
know what seems to refute the strong minimalist thesis, the answer is
just about everything you can think of or pick at random from a corpus
of material. That is not particularly interesting because it is the normal

On nature and language

situation in the sciences, even the most advanced. Again, this is one
of the reasons why people do experiments, which are a crucial part
of the “Galilean style”: it is the experiments that matter, and the well-
designed ones, the ones that fit into a sensible theory. They are the ones
that give the data that count, not what you come across. That™s not the
way linguistics was done until pretty recently. When I was a student,
the general idea was to acquire a corpus and try to organize it, to provide
a structural description of it. The corpus could be marginally modified
by field-method procedures “ “elicitation techniques” designed,
basically, to determine the scope of partial regularities in observed pat-
terns. But there are no techniques to try to discover data that might be
relevant to answering theory-determined queries about the nature of
language. That™s a creative act. Now, the point of view is that the corpus
doesn™t matter, it™s like the phenomena that you see out of the window.
If you can find something in the corpus that is interesting, great.
Then you™ll explore that with what amounts to doing experiments. But
in fact, a lot of the most interesting work has been on things that no-
body ever says, like parasitic gaps, for example. You can listen for thou-
sands of years and never hear a parasitic gap, but that™s what seems
to matter. Sometimes there are really striking results like the work of
Dianne Jonas on the dialects of Faroese,11 where she found dialectal
differences that nobody had expected and they showed up mostly in
things people almost never say, like Transitive Expletive Constructions,
and about which speakers are pretty unsure when they say them; but it
turned out that there were systematic differences in a category of con-
structions in areas that people have very little information about, and
moreover they weren™t aware of such dialectal differences. It™s similar
to the parasitic gaps case . . . Which is, incidentally, normal in exper-
imental sciences: the phenomena that turn out to be interesting are
not the normal phenomena of the world, they are usually very exotic.

An interview on minimalism

Explanatory adequacy and explanation in linguistics

ab & lr: In the characterization of the aims of scientific lin-
guistics, one important conceptual distinction introduced in the early
1960s was the distinction between two levels of empirical adequacy:
descriptive adequacy, achieved when a fragment of grammar correctly
describes an aspect of the speaker™s competence, and explanatory ad-
equacy, achieved when a descriptively adequate analysis is completed
by a plausible hypothesis on its acquisition. The Minimalist Program
characterizes a notion of minimalist explanation according to which,
to quote from “Minimalist Inquiries,”12 “a system that satisfies a very
narrow subset of empirical conditions in an optimal way “ those it
must satisfy to be usable at all “ turns out to satisfy all empirical condi-
tions” (p. 9). Clearly, minimalist explanation is a different concept
from explanatory adequacy: explanatory adequacy, in the technical
sense mentioned above, could be met by a system not corresponding
to minimalist desiderata (for instance, the assumption of an innate
list of island constraints could reach explanatory adequacy in certain
domains as well as a unifying, simple locality principle, but only the
latter would probably meet minimalist standards). How do you see
the relations between the two concepts of explanatory adequacy and
minimalist explanation?
nc: The “list of islands” model was, of course, developed in
some of the most important work of the 1960s. When the tension
between descriptive and explanatory adequacy came up, there were
several approaches; one approach, which is in “Current Issues in
Linguistic Theory,”13 was to try to find principles like A over A,
actually also the wh- island was in there, and a couple of other things;
the other approach was to give a taxonomy of properties, that™s basi-
cally Ross™s dissertation,14 a taxonomy of islands, and an interesting

On nature and language

paper by Emmon Bach in which he argued that there should be spe-
cific principles for restrictive relative clauses, maybe in all language,
and other sets of principles for other constructions. These are just two
different intuitions about which way it is going to turn out; and in fact
Ross™s taxonomy of islands is extremely valuable, a core contribution
which everybody goes back to, but that pursues a different intuition, the
one that you are describing. What you suggest seems to me quite right.
If the truth about language turns out to be something like a system of
conditions on rules and constructions, with a unifying locality princi-
ple, then only that principle would satisfy minimalist standards, and
the program would be a false hope: our explanatory sights simply can-
not be set that high “ unless some independent reasons can be found
for the other properties postulated, which does not seem very likely “
and core aspects of language would remain unexplained. There also
seems to be little prospect for improvement. One would still of course
keep to the methodological imperative of seeking the best theory of this
biological organ, however “imperfect” it is. My own view is that we can
hope for a good deal more than that, but that™s a personal judgment.
Assuming so, we might consider a variety of minimalist theses
of varying strength. One, which has come up in seminars in Siena, is
that every possible language meets minimalist standards. Now, that
means that not only the language faculty, but every state it can attain
yields an infinite number of interpretable expressions. That essentially
amounts to saying that there are no dead ends in language acquisition.
You can™t set parameters in such a way that you get a system that will
fail to have an infinite satisfaction of the interface conditions. That is
far from obvious: it is a strong condition on the system. Let™s assume
that condition is met: minimalist conditions hold for all states of the
language faculty, including the initial state. The issue here is not ex-
planatory vs. descriptive adequacy. The standard way to express that

An interview on minimalism

distinction is to take a descriptively adequate theory to be a true theory
of an attained state, whereas an explanatorily adequate theory is a true
theory of the initial state. So, in this view there is a sharp distinction be-

<< . .

( 19)

. . >>

Copyright Design by: Sunlight webdesign