Chapter 7: The Structure of Sentences
Bad News and Good News:
The bad news is that, as I warned, Chapter 7 is the hardest Chapter of
all. The good news is that you don't have to understand the technical
part, you just have to read it to get the "flavour" of how our minds'
language capacity is explained.
POSSIBLE AND IMPOSSIBLE SENTENCES
As you read the chapter, pay attention especially to the sentences with
and without asterisks (*). The ones without asterisks are possible
sentences in a language; the ones with asterisks are impossible.
You might reply: "How can they be impossible? Hasn't someone produced
those sentences -- if no one else, then the linguist who wrote this
Chapter?"
"Besides, how can they be "impossible"? Only a logical contradiction is
impossible: for example: 'All swans are birds and not all swans are
birds'. That really is impossible, because it is self-contradictory.
But even there, it's not that it's impossible to SAY it: it's just
impossible for it to be true."
Well the "impossibility" that grammarians refer to is not logical
contradiction. It is a contradiction of the rules of "Universal
Grammar." This Chapter is about what "universal grammar" is.
Pay special attention to the asterisked sentences with this in mind:
It's true that grammarians who study language produce sentences like
that. So they are not logically impossible to produce; if language had
been otherwise, they might even have been grammatically correct. But
language has not been otherwise: That is why you hear those asterisked
sentences in linguistics departments, but not in the street or in the
home or nursery where children are first learning to speak.
UNIVERSAL GRAMMAR (UG)
The reason you don't hear the kind of error that is asterisked in
Chapter 7 even when a child is first learning to speak is that the
child already "knows" Universal Grammar (UG) from birth. Think of UG as
a very complicated sentence processing machine. The child is born with
that machine already in its head, so language learning is only a matter
of learning to adjust a few of the dials on this machine to "tune it
in" to whichever language -- among all the possible ones that UG does
allow -- the child happens to be learning.
Think about this one: If the ONLY thing a child ever heard from birth
onwards were linguist-invented asterisk-sentences like the ones in this
chapter, then the child would never learn to speak any language at
all!
LEXICON & SYNTAX
We have already discussed some aspects of language: Speech and Reading.
We haven't discussed meaning yet (semantics), and you might expect
that it will be discussed when we discuss language. But it turns
out that some of the important properties of language can be studied
without studying meaning: The SYNTAX of a language is the formal
structure of its utterances. The LEXICON of a language is its
vocabulary. The syntax determines how the words in the lexicon can be
put together into correctly formed phrases and sentences.
(There are also some rules that are similar to syntactic rules but
apply only to parts of words; these are the rules of MORPHOLOGY,
but for present purposes we can treat them as syntactic rules too.
Morphology + Syntax = Grammar. We will use the terms "grammar"
and "syntax" interchangeably.)
WHY RULES?
Why does a language have to have grammatical rules at all? We can
almost always understand what's being said even if the rules of grammar
are violated, so why do we need the rules?
Well, first of all, the only reason you can understand grammatically
incorrect sentences is that they violate some rules and not others.
If they always violated ALL rules then figuring out what such a "sentence"
meant would be even harder than deciphering a secret military code,
because there would BE no code. The words would simply,
Humpty-Dumpty-like, mean what their speaker meant them to mean, but
the hearer would have no clue as to what they meant.
So one of the necessary conditions for a language is similar to the
necessary conditions for a game: There has to be a lexicon that is
shared by all players; these are the "pieces" in the language-game. And
there have to be shared rules about how the pieces can be combined;
rules that all the players know and follow. Those are the syntactic
(grammatical) rules.
SYNTAX, SYMBOLS, RULES
We've come across syntactic rules already, when we were talking about
computation and symbol systems: The rules for combining and recombining
the symbols in a symbol system are syntactic rules: they determine
which strings of symbols are correctly formed and which are not;
they also show how some strings can be "derived" from others, as in:
(2 + 2) = ? When you compute the 4 you are following the rules of
addition and subtraction in arithmetic. Those rules can be taught
to people without explaining to them that they are doing addition.
So the rules can be applied independently of what we take the symbols
to mean. (The Chinese Room Argument is based on this.)
AUTONOMY OF SYNTAX
In the case of language it was also thought that the rules of grammar
were independent of meaning. The idea of the "autonomy of syntax" was
one of the first, and most profound, examples of "modularity": The
syntax module was thought to be independent of the semantic module. So
to know which sentences are grammatically correct or incorrect, one did
not have to consider their meanings, only their FORM. The lexical
"pieces" are, we know, arbitrary. ("Cow" does not look like a cow.) So
syntax would just be the rules for combining those arbitrary pieces
into phrases and sentences.
There was perhaps always reason to doubt that syntax and semantics
could be separated quite as completely as that, however. Consider that
even in arithmetic, there are two kinds of "syntactic" rules. One
would be the kind of syntactic rule that ruled out * "+ 2) 4 2 = (" as
incorrectly formed according to the rules for forming a mathematical
"sentence." The other would be the kind of syntactic rule that ruled
out *"(2 + 2) = 5" as being incorrectly formed according to the rules
of addition for forming a mathematical sentence that can be derived
from the axioms and rules of arithmetic and is hence "true." (That is,
if * "(2 + 2) = 5* were true, it would contradict the axioms of
arithmetic.)
EXAMPLE OF AN IMPOSSIBLE SENTENCE
If language syntax were completely independent of meaning, then
we probably couldn't know why the first three sentences below are
syntactically correct but the last one is not.
"John is eager to please"
"John is eager to please her"
"John is easy to please"
*"John is easy to please her"
In other words, the rules of grammar are not just operating on
squiggles and squoggles, as Searle would be doing in the Chinese Room,
and as you would be doing if you mechanically applied the recipe for
computing the roots of a quadratic equation without understanding what
you were doing. The "lexicon" is already more than squiggles and
squoggles to you when you put together words into grammatically correct
and incorrect sentences.
LOGICAL FORM
But the rules of syntax are independent enough from semantics so
that they can be determined from the "logical form" of a sentence:
"Would you mind getting off my foot?" is actually saying "Get off
my foot!" rather than asking me about what I mind or don't mind.
So there are many aspects of the understanding of this sentence that
go beyond its logical form. The logical form of this sentence
is the same as the one for "Would you mind watching over my food?"
which is perhaps a little closer to a real question rather than
a demand.
Pragmatics and semantics are the fields that study the wider aspects
of meaning, the ones that go beyond logical form and require knowing
the context in which a sentence was spoken, the beliefs and needs of
the speaker and/or the hearer, stylistic conventions, etc. Syntax
applies only to the logical from of the sentence.
CHOMSKY's CRITIQUE OF SKINNER
Noam Chomsky is arguably the greatest cognitive scientist there has yet
been. Many think that it was Chomsky's definitive critique of B.F.
Skinner's "Verbal Behavior" that ended the epoch of Behaviorism and
ushered in Cognitive Science. There were other factors too, but there
was a reason why Chomsky's critique was the crucial one: Language is
uniquely human and is probably what makes us human. Among our
capacities, language is certainly the foremost one. Skinner recognised
this too, and it is the reason he considered his 1957 book
"Verbal Behavior" to be his most important work
According to Skinner's behaviorism, everything we do (all of our
"behaviour") is the result of our history of reinforcement (what we
have been rewarded for doing). The "shaping" of our ability to call an
apple "apple" and our ability to produce all the other sentences I have
typed on this screen that you (I hope) understand -- all this is
analogous to how a pigeon is "shaped" with rewards to peck a key or a
rat to press a lever to get food. In humans it is more complex, but
essentially the same process, Skinnerians would say.
VOCABULARY LEARNING
And this was not an unreasonable thing to believe at the time: After
all, how DO we learn our vocabularies if not by being reinforced for
calling certain things by certain names?
(Chomsky's critique was not actually focused on vocabulary learning,
though even there it could be shown that Skinner's analysis fails to
address any of the real problems: How do we learn to sort and name
things? Not What was our history of reinforcement? but HOW do we do it?
What has to be in our heads to make us ABLE to learn to sort and name
things as we do? Saying we're shaped like pigeons and rats is no
answer, because pigeons and rats can't learn to do what we can and do.
And if someone were trying to design a MODEL that could do it, and
asked Skinner what there was in Verbal Behavior that could help him
model our vocabulary learning capacity, Skinner's message that the
model should be "shape-able by reinforcement" to do what we do would
leave the modeler as utterly clueless about how to do it as he was
before he had asked!)
Chomsky's critique of Skinner focused on grammar: It is quite natural
to suppose that the way we learn grammar is by being corrected after we
make mistakes until we are gradually shaped into speaking
grammatically. This sounds as if it could be done in principle through
Skinnerian reinforcement schedules. And this may indeed be the way we
learn the rules of "style". The phrase "between you and I" is
stylistically wrong. It should be "between you and me," and a child
could learn that through trial and error and reinforcement.
STYLE GRAMMAR VS UNIVERSAL GRAMMAR
There is a good chance, however, that styles will change, and "between
you and I" will become accepted as correct, just as "It is me" has become
correct. (It used to be: "It is I".) A living language is a changing
language, and yesterday's "grammatical" errors become today's accepted
usage. That's why I called these rules "stylistic" rather than
grammatical -- though this kind of thing is what most of us think that
grammar is about.
ANTHROPOLOGIST FROM MARS
But Chomsky had a much deeper form of grammar in mind, and the best way
to appreciate it is from the perspective of an anthropologist from
Mars (to borrow a phrase from Oliver Sacks). Such an anthropologist
would come to earth to describe and explain the habits of earthlings;
and what he would find that earthlings do for an inordinate proportion
of their waking hours is to make sounds at one another.
The anthropologist (with a large Martian research team gathering data
for him) would also discover eventually that these sounds that people
make toward one another are not random. He could, with enough
analysis, discover that a lot of them were just recombinations of
sounds that kept recurring. In this way, he could eventually come up
with the "lexicons" of all the languages on earth: the vocabularies out
of which their vocalisations were composed. And by shrewd observations
of when they said what, he could even decode the meanings of the
lexicon and discover what words mean.
REGULARITIES IN WORD COMBINATIONS
But apart from the correlations between single words and meanings,
the alien anthropologist would also notice that there were regularities
in how the words in the lexicon were combined: Certain combinations
occurred but other combinations never spoken. Now how would word
combinations that were never heard or spoken even occur to the Martian
observer? Well, let's suppose that he is superintelligent and is trying
to decode what the earthling vocalisations mean. He has successfully
decoded the individual words in the lexicon, but when he tries to
decode combinations of words, he runs into trouble.
He assumes there are regularities in the combinations of vocalisations,
just as there were in the vocabulary. So he tries to find out what these
regularities in the combinations might be. For example, he runs into
our three sentences: He notices that people say:
"John is eager to please"
and
"John is eager to please her"
and
"John is easy to please"
but he can't for the life of him say why they never (except if they
are linguists at the Massachusetts Institute of Technology [MIT]) say:
*"John is easy to please her".
This seems like just as reasonable a combination of words as the other
three. So he starts to theorise that the reason some combinations are
heard and others are not is because certain rules are being followed
that determine which combinations are OK and which are not. But what
are those rules, and where do they come from?
LEARNABLE RULES
For a few rules it is easy to discover what they are and where they come
from: When little earthlings say "between you and I," bigger earthlings
say "No, it's 'between you and me'," and eventually the little earthlings
don't say "between you and I" any more. But (except in Cambridge
Massachusetts) no earthling, big or small, ever says *"John is easy to
please her". And there are MANY such word combinations that no one ever
speaks. What are the underlying rules, and how do earthlings learn them?
UNLEARNABLE RULES
The Martian anthropologist has a very long life span and many assistants
to help him. He watches the tinkering going on in Cambridge Mass.
in the vicinity of a remarkable earthling who writes several books a
year (the Martian is by now quite familiar with earthlings' ways) and
who makes tree diagrams with labels that correspond to earthlings'
vocalisations. This singular earthling also writes down rules that
would explain why some combinations of words are spoken and others
(which he tags with a *) are not.
The rule-book is not complete yet, and it has undergone several
revisions, but the Martian grasps that these are the rules that all
earthlings are following. But how did they get them? The Martian has
confirmed over and over that earthlings hardly ever say anything that
breaks these Cambridge Mass. rules. There is only one explanation for
how they get those rules: They must be born with them already in their
heads. Otherwise they couldn't all know which combinations were
right and which were wrong.
INBORN UNIVERSAL GRAMMAR
How could the Martian be sure that the rules were not learned?
Because he sees that the little earthlings, who at first only vocalise
with an irritating Martian-ear splitting noise, are very soon vocalising
like all other earthlings, but almost never do they make mistakes
or get corrected.
The knowledge that these little earthlings are born with must be vast,
the Martian infers, because even for him, the Martian, and for the many
earthlings working around the MIT Wizard who writes all those books, it
takes a long time to figure out what the rules are: How could a
complicated set of rules that has not yet been fully understood in
40 years of work by the MIT Wizard and his elves, and that even
the Martian has trouble learning, be "known" by every little earthling?
That's why the Martian correctly infers that they must be inborn.
POVERTY OF THE STIMULUS
Now let's leave the Martian allegory. As you read Chapter 7,
you will get a flavour of what Chomsky's rules look like. Those
were the rules that Skinner (if he had known about them at all) would
have had to say that children learn by trial and error and corrective
feedback. But there are hardly any errors, and not much feedback.
Most of the errors that could have been made if children were
learning Universal Grammar the hard way, the way the MIT linguists are
learning it, are never made. You never hear those asterisked sentences
except in Linguistics Departments.
So Skinnerian learning is completely ruled out as the source of the
rules; so is every other kind of learning. The information is just not
there, in the vocalisations that children hear and produce. This is
what Chomsky calls the "poverty of the stimulus." There is no basis for
ANY kind of learning system to learn in a year (as the child would be
doing if it got the rules by learning them) what teams of linguists
have still not entirely learned in 40 years of collaborative work!
CHAPTER 7
The MIT Wizard's abstract system of rules and constraints, Universal
Grammar (UG), is what Hans van de Koot is making a heroic effort to
describe to you in the 40 pages of Chapter 7. You owe it to him to at
least pass your eyes over all the lines he wrote, if only to persuade
yourselves how very UNlearnable this all is for a 2-3 year old child!
Yet the child KNOWS all of it -- not explicitly, not to be able to
TELL you the rules, but implicitly, in the sense that his word
combinations are obeying all those rules, and preventing him from ever
uttering those asterisked sentences.
For the examination, you should have 1-2 examples of sentences that
are and are not grammatical, and why, according to UG.
Universal Grammar UG is the complex set of rules that govern which
sentences are grammatical and which are not. We earthlings all
follow the rules of UG, but what we hear and say as children learning to
speak is not a basis for learning UG; no human or machine could learn
UG from that alone. Linguists still haven't figured out all the rules
in 40 years of analysis. So UG must be inborn in the child.
PARAMETER SETTINGS
The child does learn SOMETHING in addition to vocabulary, however. What
he learns is the setting of the PARAMETERS of UG for the particular
language he is learning. Parameters are like the dials on a ham radio:
You must set the dial to tune in to the station you want to hear from
and speak on. There are quite a few dials and settings on the dials
that correspond to all the possible human languages. The
parameter-settings are what the child learns when he learns a language.
But the complex structure (UG) on which the parameters are merely a
a choice among a fixed set of possible choices cannot be learned because
of the poverty of the stimulus.
COMPETENCE VS PERFORMANCE
There are many unsystematic errors that people make in speaking. If you
saw a transcript of an everyday conversation, there would be many
incomplete sentences, stylistic lapses, utterances that could not
be understood if you were not in the room (references to "this"
and "that," where you do not know what the speaker is pointing to) etc.
These are topics for the study of pragmatics and psycholinguistics,
but not for Chomskian grammarians. Chomsky is interested in
what he calls our syntactic "competence," not our actual syntactic
"performance."
What he means is that he wants to explain our syntactic capacity -- our
capacity to produce and understand all and only the grammatically
correct sentences of our language. There's many a slip twixt the cup
and the lip, but it is not those slips that Chomsky is trying
to explain; it is our capability.
An excellent chapter on language acquisition to supplement Chapter 7
of Green et al. is Pinker's Chapter on Language Acquisition:
http://cogsci.soton.ac.uk/~harnad/Papers/Py104/pinker.langacq.html
This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:51 GMT