MIDDLE EAST TECHNICAL UNIVERSITY

Course Code, Title and Credit:

CENG 729. Syntax, Semantics and Computation: Combinatory Categorial Grammar. (3-0)3. Fall 2008

Catalog Description:

Natural Language syntax and computational transparency; logical form; type-dependent vs. structure-dependent syntax; formal principles for possible human languages; substantive principles; syntactical and semantical categories, and their projection; lexicalising a grammar; glue languages for syntax-semantics correspondence; linguistic constructions as artefacts of computational principles; ergativity, accusativity and the lexicon.

Lecturer

Cem Bozsahin

Background Requirements

Basic understanding of logic and formal languages.

Complementing/overlapping Courses:

CENG563 and COGS532 mention CCG briefly, but only as one of the theories. This course is an in-depth study of CCG with its computational and linguistic implications. There is no overlap with CENG565, which covers a related topic, Automata theory.

Check out our sister course at Univ. of Texas, Austin.

Course in relation to the programs:

The course is intended as a graduate elective in Computer Engineering, Cognitive Science, Linguistics and Philosophy.

It is a core elective in Linguistics track of CogSci, and a depth course in CS.

CCG is a theory that has its roots in combinatory logic, automata theory and formal languages. Its substantive component has its roots in Philosophy of Language (such as that of Husserl, Frege, Russell, Chomsky and Curry).

The course is interdisciplinary in nature; it aims to bring together research on natural language syntax and semantics from the perspective of Computing, Linguistics, Cognition and Logic.

Course Objectives:

CCG's main claim is that linguistic expressions are type-dependent, and that co-dependence of semantic and syntactic types can be captured in a lexicalised grammar so that the universal component simply projects them onto phrasal derivations. Computation (from lexicon to PF and LF) is transparent because the lexical types materialise under the guidance of UG to begin with.

This is a radical departure from Chomskyan thought which claims that if you get your syntax right, you get semantics for free, with a helping hand from extraneous mechanisms such as transformations or checking operations. CCG says you cannot get the semantics right unless your syntactical categories are semantically well-motivated, and then you get phrasal semantics for free with no extra help.

(If you've read this far, thank you for your patience; please read on. None of the above is supposed to make sense at the beginning.)

The aim of the course is to flesh out these ideas from the perspective of language and computation.

At the end of the course, students are expected to understand the basic philosophy of lexicalising a grammar, and the computational reasons for doing so; they are expected to formulate their hypotheses explicitly in Combinatory Categorial Grammar, and test them out on computational tools developed for CCG.

They are expected to understand the kind of mental computations CCG claims to be taking place when people put their linguistic competence to use. They are also expected to understand the differences with other theories in terms of computation.

Course Outline:

(download
Latex style files and bibliography databases and lecture notes in latex to do latexing yourself without colours etc.). I provide everything else in pdf.

week 1: Introduction

PART I: Foundations

week 2-5:
Logical form and Syntactic Types: The notion of Category in a lexicalised grammar (fixed).
Substantive and Formal aspects of CCG: Universal Grammar and possible Lexicons.
CCG's principles and computation: More restrictions on UG and Lexicon?
Computational tools for CCG.

PART II: Constructions

week 6-9:
Scrambling;
Control and Boundedness;
Relativisation and Unboundedness;
Coordination and type-dependent syntax;
Transitivity and Universal Grammar;

PART III: Linguistic Diversity and CCG

week 10-12:
Ergativity and Accusativity;
Transitivity, intransitivity and argument structure; (here is a copy of Hopper & Thompson's transitivity paper.

week 13-14: Example analyses from several languages (this part includes student presentations).

Textbook:

None. The course is supplemented by a course reader, which consists of research papers. (look under `files').

Course Conduct:

-11 to 13 weeks of lectures (1 or 1.5 hour lecture, 1 hr student commentary; 1 hr discussion)

-1 to 3 weeks of student presentations

- Each student will pick a language to study, and provide written and applied CCG analyses of data from that language throughout the course.

Grading:

4 pencil and paper HW

1 computational HW

1 Term research

Several reading assignments

1 final exam (take home)