Jan 21 2009
1. Linguistics as a Theory of Knowledge of Language
The goal of linguistic theory: modeling knowledge of language
Knowledge about Possible/Impossible novel linguistic combinations:
(1) a. blick vs. *bnick phonological (sound)
b. untie vs. *uncheerio vs. *unfly morphological/semantic
(2) Take it out of the fridge vs. *Take out it of the fridge. syntactic
Asterisks prefixed to sentences indicate ungrammaticality
2. Linguistics as a Theory of Language Learning
How did you come to know such things?
No one explicitly told you what things were ungrammatical.
Nor can it be that you only recognize as grammatical those sentences you have heard or already produced. I can come up with plenty of sentences that you’ll accept but you haven’t heard.
(3) a. The customers appreciated my coffee more than my tea? b. *What did the customers appreciate your coffee more than?
Neither of these sentences could have been modeled on anything you’ve heard – this is especially true of (3)b. But nonetheless you can still determine that (b) it is ungrammatical and (a) isn’t.
HOW DID YOU COME TO KNOW THIS?
Incomplete data
The problem of learning the syntax of a language is made harder by the incompleteness of the data we are exposed to in learning the language.
For instance, verbs like say and think and others can be followed by sentences introduced by
that. These sentences can in turn contain another instance of a verb like say or think, which can then be followed by a sentence that contains…you get the point:
(4) They said that Bob wanted Mary to think that Sean was happy that Maria was…
The recursive property of linguistic structures allows for (in principle) infinitely expanded sentences
We can never hear such things and yet by a young age you can recognize (4), and its possible continuations, as good.
HOW DOES A CHILD COME TO KNOW THIS?
[In the reading in the Carnie textbook for this week, you will read about the recursive and “infinite” nature of human language. There’s no denying that linguistic
operations are recursive, but questions about what this tells about language and language acquisition are debated.]
Misleading data
Not only is data missing, some data in the child’s environment is misleading. Take a particularly common type of sentence produced below in (a):
(5) ?This is the pie that I wondered who made it.
There is something weird about (5)a. But you hear these things all the time, so why do we have the intuition they are not fully grammatical?
Adult grammar is also full of utterances that don’t even bear on the kind of grammaticality judgments that native speakers appeal to in judging many sentences:
(6) Umm, so, yeah, I don’t actually know if…ok.
A lot of our utterances aren’t full sentences – much of what we say is full of starts and stops, incomplete sentences.
Underdetermination of the data
(7) a. Who did you say Fred will kiss at the party? b. Who did you say that Fred will kiss at the party?
(8) a. Who did you say will kiss Fred at the party? b. *Who did you say that will kiss Fred at the party?
Given the data in (7), that is optional, but then what about (8)? How did you come to know that (8b) is bad?
no one taught to you that (8b) was not possible
there is nothing in the data that supports the conclusion that (8b) is bad; in fact a reasonable learner would expect it to be good.
HOW DOES A CHILD COME TO KNOW THIS?
Child Language Acquisition
In fact, children fully acquire knowledge of a language quickly and completely even though the language they hear spoken around them is incomplete, misleading, and doesn’t provide the data necessary to provide the child with “models” to make her judgments about sentences.
As they acquire language, children show signs of forming generalizations and reformulating incorrect generalizations. They do this without imitating adults and, in fact, consistently ignore adult correction.
(9) Dialogue 1
C: My teacher holded the baby rabbits and we patted them. A: Did you say your teacher held the baby rabbits?
C: Yes.
A: What did you say she did?
C: She holded the baby rabbits and we patted them. A: Did you say she held them tightly?
C: No, she holded them loosely.
Children are predisposed to making generalizations; i.e. in forming regular past tenses. If they were simply “modeling” or imitating, why would they insist on this?
(10) Dialogue 2
A: Now, listen carefully, say “Nobody likes me.” C: Oh, nobody don’t likes me.
Here the child seems predisposed to treating “not/n’t” negation as non-meaningful. (It turns out there are many languages where this is true, even dialects of English.)
Non-errors in Child Language
Yes/No Question Formation
(11) a. Has Addy eaten a cake? (Addy has eaten a cake.) b. Will Lilah come to the party? (Lilah will come to the party.)
Any number of rules can be devised to derive these interrogative Yes/No questions from their declarative counterparts.
(12) a. Swap the first two words around.
b. Swap the first verb with the first noun phrase c. Swap the subject and the verbal element after it
Try out each of these hypothesized rules on the following test cases:
(13) a. The man has eaten a cake.
b. The woman who is singing is happy.
c. The book that John is reading is on sale.
d. The book that John was telling Sue that Fred was happy about is on sale.
While child language differs in interesting ways from the target adult language, errors of the sort predicted by (12a,b) are never made by children.
(14) Structural Dependency:
Syntactic operations reference structure, not strings of words. (i.e. There are no syntactic operations that “count” words.)
Sentences are not just strings of words; in the sentences above there is intuitively a major division that separates one chunk of the sentence from the other. I notate that division with the symbol “│”.
(15) a. The man │ has eaten a cake.
b. The woman who is singing │ is happy. c. The book that John is reading │ is on sale.
d. The book that John was telling Sue that Fred was happy about │ is on sale.
We can “draw” this split with a tree-like structure:
(16) Sentence
• •
The book is on sale
The formation of a question involves moving a verb like element at the edge of the right-branch in front of all the stuff in the left right-branch.
(17) Sentence
• •
is
The book is on sale
The left-side of the structure can be made longer (with more verbs and nouns), and accordingly the structure will become more complex:
(18) Sentence
• •
the • is on sale
But the rule that forms a question seems to ignore this complexity within the left branch, and the question is formed in the same way.
HOW DOES A CHILD COME TO KNOW THIS?
3. Chomsky’s Answer: Linguistic Theory as a theory of Language Learning1
Noam Chomsky, a famous linguist and philosopher, sought to explain what allows children to acquire a language so effortlessly and easily with such meager data, and why they appear to entertain only certain kinds of linguistic operations and not others. His idea is that the child already knows quite a bit about language before hearing any of it!
Think about our knowledge of some language L as a set of rules that generates the sentences (or, in fact, the structures) of the language L.
One name for this knowledge or set of rules is a grammar (especially when referring to the linguist’s model of this knowledge). Don’t confuse this kind of grammar with prescriptive or school-room grammar.
One of the rules in the grammar of English will be that which forms questions (we’ll investigate quite closely what that rule is in later parts of the semester).
Learning a language, then, involves selecting from all the logically possible grammars for a language the one that generates the strings of the language L to be acquired (so selecting the right rule for question formation among the many possible rules).
Chomsky’s question: how is it that the learning procedure is able to find the grammar of a particular language (GrammarL) when the universe of Grammars (or Gs) is so huge and the evidence so meager.
Chomsky’s answer: The set of Gs available to the child is limited.
Chomsky claims that certain properties of grammars are built-in: these properties – what defines the set of Gs actually available to a little language learner – is sometimes referred to as Universal Grammar. Chomsky believes that UG is genetically-endowed, hence explaining its uniformity and species-specificity.
1 This project is characterized in various ways. Some good first-hand discussions about it are found in N.
Back to our earlier Yes/No question formation examples:
Let’s think of question formation as a rule that maps a declarative (non-question) to a question sentence, by way of relocating parts of the first.
(19) a. the book is one sale declarative
b. is the book on sale? Question/interrogative
The child must be pre-disposed to expect that the rule forming Yes/No questions is stated on structures not strings of words. In particular, the child expects that if a rule moves some element to another part of the sentence, it makes reference to structure not strings. The child can then account for the difference between the order of words in the question and the declarative sentence by postulating a rule that moves the first verb-like element on the right branch of the sentence tree to the front of the sentence.
(20) Sentence
• •
is
The book __ on sale
Actually, we will see that all syntactic operations that move things around in sentences are “structure dependent.”
Some properties of “Universal Grammar”:
Sentences (and smaller units of language) have a structure (hierarchical relationship)
All syntactic operations are structure dependent.
??
Chomsky’s goal for syntactic theory: find out how the set of available Grammars is structured, i.e. to find the properties of “Universal Grammar”
4. Language Variation
The set of available grammars can’t be too narrow.
The rule that forms English questions clearly cannot be operative in Irish.
Instead of switching the location of the subject and the finite (auxiliary) verb, Irish keeps the word order of the declarative sentence the same but adds a separate word at the front of the sentence to indicate a YES/NO question:
(21) a. Art hit Seán? Irish Q fall John
“Did John fall?”
b. Hit Seán fell John “John fell”
But there still appear to be very limited ways in which languages form YES/NO questions:
I know of no language that forms YES/NO questions by removing the verb altogether or by deleting the vowel of the second last word. While a little crazy, these are possible rules of a grammar but appear not to be among the available rules.
One popular idea: English and Irish aren’t so different, but what words you pronounce at the front of the sentence in a question is a little different:
(22) has John fallen? English Question/interrogative
move the verb to front of sentence…
(23) Art hit Seán? Irish
fall John
5. Data – Ungrammatical vs. Hard
Linguists distinguish between knowledge or grammar, often called competence, and a speaker’s use of language, called performance.
Competence: a speaker’s knowledge of language is what allows them to produce and understand an infinite number of utterances, including new utterances.
Performance: a speaker’s use of language, in specific situations at specific times. Performance depends and interacts on other cognitive systems (memory, for instance) and the nature of the outside world. But it is driven by linguistic competence.
To appreciate the difference between the two, read the following sentence:
(24) The girl the boy the dog Jim owns bit kissed left.
The sentence is built by using a rule of English that rule is part of competence: you can add a sentence to a noun describing that noun (this is called a relative clause).
(25) a. The girl … left.
b. The girl that the boy …kissed left.
c. The girl that the boy that the dog … bit kissed left.
d. The girl that the boy that the dog that Jim owns bit kissed left.
Obviously, the human cognitive system has all kinds of limitations. And maybe it’s just too complicated to understand a sentence like the one above – even though it is built by rules from the grammar we know! So here is an important idealization we make in linguistics: we try to give an analysis of English how it would be by a perfect speaker that didn’t have any limitations.