Syntactic Structures

Author: Noam Chomsky
Date: 1957

Seminal work in 20th-century linguistics.

Laid the foundations of Chomsky's idea of transformational grammar.

Contains the famous sentence "Colorless green ideas sleep furiously", which Chomsky offered as an example of a sentence that is completely grammatical yet completely nonsensical.

Note, the subsections of this work are simply numbered. I have given them my own titles for easier reference.




1   Introduction

2   The Independence of Grammar

## 2.1

## 2.2

## 2.3

## 2.4

3   An Elementary Linguistic Theory

3.1   x

Assuming the set of grammatical sentences of English to be given, what sort of device can produce this set?

Linguistic descriptions proceed in terms of a system of levels of representations. The joint description of levels is simpler than direct description.

A phonemic description of language is very complex, so for practical purposes it often makes sense to talk about elements as morphemes (and talking about them together is either than any alone)

  • What grammar is necessary to describe the morphemic structure of sentences?
  • Describe a finite state machine
    • Can transition to different states via emitting words
    • The sequences of words produced is a sentence
    • Each machine describes a certain language (the set of sentences that it produces)
    • This language is called a finite state language
    • The machine itself is a finite state grammar
    • Mathematically, these are known as finite state Markov processes
    • Each state represents the grammatical restrictions that limit the choice of the next word at any point in an utterance

English is not a finite state language; it is impossible (not just difficult) to construct a finite state grammar which represents English.

3.2   3.2

A language is defined by giving:

  • Its alphabet, the finite set of symbols out of which its sentences are constructed
  • Its grammatical sentences

There are a number of languages finite automata cannot describe. For example, XY where Y = reverse(X).

No regular grammar can explain English.

3.3   3.3

If we declare that the process of finite state machines is limited to repeat only a certain number of times we can make them finite state languages Of course it will be so complex as to not be worth it (practically just enumeration).

In general we assume languages are infinite.

"If a grammar does not have recursive devices it will be prohibitively complex"

4   Phrase Structure

Linguistic description on the syntactic level is formulated in terms of the constituent analysis (parsing).

4.1   4.2

Theorem:Every regular language is a terminal language, but there are terminal languages which are not regular languages.

We have not shown English is describable by phrase structure, but we have shown it to be more powerful than finite state machines

Terminal strings of phrase structures represents set of strings rather than individual words

4.2   4.3

  • Let A be some grammar which generates all grammatical sequences of morphemes of a language
    • To complete it we must state the phonemic structure of the morphemes
    • This transformation can also be given as rules of "rewrite X as Y"
      • e.g. take + past = /tuk/
    • Order of rules matters
  • We the mapping of morpheme to phoneme, we can connect that to the grammar for a single process of generating phonemes sequences
    • This make it appear as if the distinction is arbitrary, however the rules are essentially different

5   Limitations of Phrase Structure Description

5.1   5.2

  • The problem with PSG is that it forgoes history
    • While it can handle it via large rules, adding history considerably simplifies the grammar

5.4   5.5

> If we examine carefully the implications of these supplementary rules, however, we see that they lead to an entirely new conception of linguistic structure. Let us call each such rule a "grammatical transformation". A grammatical transformation T operates on a given string with a given constituent structure and converts it into a new string with a new derived constituent structure. To show exactly how this operation is performed requires a rather elaborate study which would go far beyond the scope of these remarks, but we can in fact develop a certain complex but reasonably natural algebra of transformations having the properties that we apparently require for grammatical description.

  • Some essential properties of transformational grammar
    • We must define an order of application on transformations
    • Certain transformations are obligatory and others optional
    • With this, we can define the 'kernel' of the language as the set of sentences that are produced when apply obligatory transformations to the terminal strings of the grammar. (A sentence in the kernel is called a kernel sentence.) Thus, every sentence of the language will either belong to the kernel or will be derived from the strings underlying one more more kernel sentences between a sequence of one or more transformations.
  • Grammars seem to have a tripartite arrangement
    • Phrase structure
    • Transformation structure (sequence of rules)
    • Morphophonemics
  • To produce a sentence from such a grammar we construct an extended derivation beginning with _Sentence_
  • When transformation analysis is properly formulated we find that it is essentially more powerful than description in terms of phrase structure (just as the latter is essentially more powerful than finite state Markov processes)
  • Grammar is simply a description of utterances, namely, those which it generates.
    • It is not a way of synthesizing or analyzing utterances


6   On the Goal of Linguistic Theory

7   Some Transformation in English

8   The Explanatory Power of Linguistic Theory

> So far we have considered the linguist's task to be that of producing a device of some sort (called a grammar) for generating all and only the sentences of a language, which we have assumed were somehow given in advance.

9   Syntax and Semantics

10   Summary