Computer Laboratory

Course pages 2012–13

Syntax and Semantics of Natural Language

Principal lecturers: Prof Ted Briscoe, Dr Stephen Clark
Taken by: MPhil ACS, Part III
Code: L107
Hours: 16
Prerequisites: L100 Introduction to Natural Language Processing and R07 Introductory Logic for students who have not taken a course in logic before.

Aims

This module provides an introduction to the formal syntax and semantics of natural language, in particular Montague-style compositional semantics using a Categorial Grammar model of syntax. Half of the module will focus on the theory of syntax, followed by an example of how recent advances in parsing technology allow such a theory to be implemented in practice, operating on naturally occurring text. The other half of the module focuses on truth-conditional compositional semantics of sentences, computational implementation of this approach, and probabilistic inference for semantic interpretation.

Syllabus

  • Introduction to (Combinatory) Categorial Grammar (CCG). [1 lecture]
  • English syntax in the CCG framework. [3 lectures]
  • Introduction to statistical parsing. [1 lecture]
  • Constructing a wide-coverage CCG English grammar. [1 lecture]
  • Wide-coverage robust statistical parsing with CCG. [2 lectures]
  • Introduction to natural language semantics. [1 lecture]
  • Montague semantics; compositional semantics, typed lambda calculus, quantification, intensionality. [3 lectures]
  • Robust, wide-coverage implementation and underspecification. [2 lectures]
  • Probabilistic inference and semantic interpretation. [2 lectures]

All lectures will be given by Professor Briscoe or Dr Clark.

Objectives

On completion of this module, students should:

  • understand how the syntax of natural language sentences can be modelled using a type-driven (Combinatory) Categorial Grammar;
  • understand how a wide-coverage grammar of English can be constructed;
  • have studied one approach to statistical parsing in detail;
  • understand how the meaning of natural language sentences can be modelled using a logical, model-theoretic approach;
  • understand how the meaning of natural language sentences can be constructed using Frege's principle of compositionality;
  • understand how this approach to meaning can be combined with probabilistic inference;
  • gain an appreciation of how syntactic and semantic theory can be implemented in practice.

Assessment

  • Four ticked take-home tests or short practicals. Each ticked test is worth 5% of the final assessment for the course. Tests will be due in one week after assignment and ticked (with feedback) by Professor Briscoe and Dr Clark.
  • One final take-home exam covering all the material taken at beginning of Easter Term. Final take-home exam will contribute 80% to the final assessment. Questions set and marked by Professor Briscoe and Dr Clark.

Recommended reading

Steedman, M. (with Baldridge, J.) (forthcoming). Combinatory categorial grammar. To appear in Non-transformational syntax (eds. Borsley, R. & Borjars, K.). Available here.
Clark, S. & Curran, J.R. (2007). Wide-coverage efficient statistical parsing with CCG and log-linear models. In Computational linguistics, 33(4), pp.493-552. Available here.
Cann, R. (1993). Formal semantics. Cambridge University Press. (Google Books, UL, etc.)
Bos, J. & Blackburn, P. (2004). Working with discourse representation theory. Available here.
Bos, J., Clark, S., Curran, J.R. & Hockenmaier, J. (2004). Wide-coverage semantic representations from a CCG parser. In Proceedings of COLING-04, pp.1240-1246, Geneva, Switzerland. Available here.
Bos, J. (2008). Wide-coverage semantic analysis with Boxer. 2nd Conference on semantics in text processing. Available here.
Garrette, D., Erk, K. & Mooney, R. (2011). Integrating logical representations with probabilistic information using Markov logic. International workshop on computational semantics. Available here.