[ Up to Syntax]
This sentence doesn't seem to make sense, but it leaves us with the impression that it is 'well formed' anyway.
This is cited as evidence of a 'level of syntax' which functions apart from other linguistic levels.
In fact, it is possible to coerce metaphorical meaning out of this sentence.
[ Up to Syntax]
The level of morphology provides us with a stream of words, each of which can be assigned a part of speech.
Major parts of speech are:
Nouns (eg:cats, dogs, ideas...)
Pronouns (eg:he, her...)
Posessive Pronouns (his, their,...)
Relative Pronouns (eg:who, which..)
Determiners (eg: a, the, that)
Quantifiers (eg: some, all,...)
Adjectives (eg:red, good)
Prepositions (eg: on,in,...)
Adverbs (eg:quickly, certainly,...)
Verbs (eg:eat, think,...)
Verbs break down into a number of subclasses:
Tense (past, present, future), eg 'was, is, will be'
Aspect (Simple, Progressive: '-ing', Perfect: 'have ...-en')
Mood (Active or passive) eg:'I made a mistake' vs 'Mistakes were made' (very convienient!).
Modality (...can..., ...may...)
Adverbs (eg: absolutely, certainly)
Conjunctions (eg: but, and)
Some of these classes are 'closed', i.e. their members are fixed, and generally it is harder to pin down their 'meanings'.
Other parts of speech refer to 'open classes', which have a very large number of members, and to which we frequently add new members.
[ Up to Syntax]
Somehow we have to negotiate our way from a sequence of words, each of which is assigned a part of speech, to entire statements about the world ('clauses'). We can do this by grouping words into phrases, and phrases into larger phrases.
These phrases form patterns which invite us to formulate rules which describe these patterns.
It is commonly believed among linguists that 'competence' in a language involves internalizing rules about the patterns of the language. This contrasts with speakers' actual 'performance', which may contain a number of errors. It is common practice among many linguists to proceed by proposing a number of contrasting statements that exemplify the grammaticality or ungrammaticality of various patterns, and coming up with rules that express these observations. This contrasts with the practice of corpus linguists, who typically use statistical techniques to make generalizations about language.
[ Up to Syntax]
Made up of 'rewrite rules'.
Rules only have one symbol on the left-hand side (LHS), that's what makes them 'context free'
Rules can have any number of symbols on the right-hand side (RHS), unless they are terminals, i.e. actual word forms.
'Generative': a finite vocabulary, with a finite number of rules can generate an infinite number of sentences.
How? Recursion.
See figure 3.3 in NLCL for an example of a simple CFG.(CFG-eg.jpg)
See figure 3.4 in NLCL for example of lexicon.
Criteria for a good grammar:
Generates all sentences in the language.
Doesn't overgenerate, e.g.:'NP -> Det NP' can generate *'the the dog'
[ Up to Context-free grammar (CFG)]
Mary washed the dishes, John, the carpet.
rule:NP->NP NP TrV
Note: example of recursion)
The dog the man kicked
The man the woman slapped
The woman the cat likes
The dog the man the woman slapped kicked.
The dog the man the woman the cat likes slapped kicked
According to Chomsky this is a competence/performance thing.
Rule: S->NP VP
'She smiles.'
'They smile.'
But not: *'They smiles.'
Revised rules:
S->SingNP SingVP
S->PlNP PlVP
or:
S->NP[number=alpha] VP[number=alpha].
So we need to use features to account for the different configurations of possibilites.
EG: pronouns have:
number (sing/plural)
person (1st, 2nd, 3rd)
case (subject/object)
gender (masc/fem)
possessive (+/-)
reflexive (+/-)
[ Up to Context-free grammar (CFG)]
Each rule is a little finite-state machine.
We can 'augment' them with registers to keep track of features.
We can use a stack to generate subordinate phrases.
[ Up to Syntax]
EG:John ate the turkey vs The turkey was eaten by John.
Transformations take a common 'deep structure' and make changes in the 'surface structure'.
Problem: Too powerful!