Module 02451 (2001)

Syllabus page 2001/2002

06-02451
Reasoning about Mental States
Level 3/H

jab
10 credits in Semester 2

Links | Outline | Aims | Outcomes | Prerequisites | Teaching | Assessment | Books | Detailed Syllabus

The Module Description is a strict subset of this Syllabus Page. (The University module description has not yet been checked against the School's.)

Relevant Links

EXTRA INFORMATION, including ASSIGNMENTS and PAST EXAMINATIONS.

Outline

The challenging question of how to represent and reason about people's mental states, such as beliefs and intentions, has long been central in Artificial Intelligence (and related fields within Cognitive Science, notably Philosophy, Psychology and Linguistics). AI systems that interact with or reason about people must often have the ability to reason about their mental states. In particular, since text and speech utterances often report, refer to or convey mental states, the handling of mental states is important for natural language discourse understanding systems. The module will cover various methods for representing and reasoning about mental states, with some bias towards natural language applications, and with considerable attention to the handling of uncertainty, a crucial aspect of realistic reasoning about mental states. Much of the module is relevant to central issues in other areas of Cognitive Science, and in particular to recent debates concerning the ``simulation theory'' of mental state handling in Philosophy and Psychology.

Aims

The aims of this module are to:

  • give an appreciation of the importance in artificial intelligence of reasoning about mental states such as beliefs and intentions, and of understanding natural language sentences that report mental states
  • introduce some of the main problems involved in understanding mental state reports and in reasoning about mental states, with a special focus on the role of uncertainty
  • convey representational and reasoning techniques aimed at handling those difficulties, with emphasis on techniques useful in practice, and with emphasis on natural language applications
  • link the material to issues in Psychology, Philosophy and Linguistics
  • provide an appreciation of the complexities of representation and reasoning in AI generally, beyond what is provided by previous modules

Learning Outcomes

On successful completion of this module, the student should be able to: Assessed by:
1assess the role of the above-mentioned problems when they arise in artificial intelligence issues that involve the understanding of mental state reports or reasoning about mental states Exercises & Examination
2analyze de-dicto/de-re and related ambiguities in English sentences about mental states Exercises & Examination
3apply various schemes for representing mental states, such as modal, quotational and episodic logic and semantic network schemes Exercises & Examination
4explain, in detail, how mental states can be reasoned about in specific scenarios, potentially by means of both simulative and non-simulative reasoning techniques, and showing how uncertainty of the various types covered in the module is handled Exercises & Examination

Restrictions, Prerequisites and Corequisites

Restrictions:

None

Prerequisites:

06-08764 (Mathematics & Logic B) (logic), 06-11352 (AI Techniques A) & 06-11353 (AI Techniques B), 06-11349 (AI Programming A) & 06-11351 (AI Programming B).
Some knowledge of natural language processing is helpful but not essential.

Co-requisites:

None

Teaching

Teaching Methods:

2 hrs lectures per week
The two class sessions per week include conventional lectures and class discussions. The conventional lecturing itself includes frequent opportunities for students to offer questions and comments.
The module content is conveyed through lectures, student presentations (in some circumstances), handouts, web material, textbooks and research articles. Some required material from readings is not covered in class sessions (except in response to student questions). Conversely, much of the material in lectures is not readily to be found in any small set of books, articles, etc.
The lectures emphasize key issues and problems, explain representational formalisms and algorithmic techniques in detail, and go through detailed examples and solutions to formative and assessed exercises.
The within-semester Assessed Exercise Set (see below) serves, in part, as direct preparation for the exercises included in the Examination. Feedback from the lecturer includes, at an appropriate point, detailed written answers to, and/or other notes on, the exercises. These answers and notes are generally made accessible via a link from the current webpage.
The exercises in the Formative Exercise Sets (of which two are planned) are, in part, similar to the assessed exercises and serve to prepare students for the latter and for the Examination. At an appropriate point the lecturer provides detailed answers to, and/or other notes on, the exercises, but does not mark the students' papers.
The material accessed in obligatory reading tasks may be used in the Formative and Assessed Exercise Sets and in the Examination, even when that material has not been covered in lecture-room sessions.
The material accessed in non-obligatory reading tasks will not be necessary for first-class performance on Exercise Sets or the Examination, but may nevertheless help the student to boost his/her performance.

Contact Hours:

24

Assessment

  • Supplementary (where allowed): As the sessional assessment
  • The Assessed Exercise Set will be assigned just before the Easter break and will be due at some point within the first two weeks after the break.

Recommended Books

TitleAuthor(s)Publisher, Date
Representations of Commonsense KnowledgeE DavisMorgan Kaufmann, 1990
Reasoning about KnowledgeFagin, Halpern, Moses & VardiMIT Press, 1995

Detailed Syllabus

  1. General Issues (2 class sessions)
    • Importance of mental states in AI generally and in natural language processing in particular.
    • Overview of relevant pragmatic aspects of natural language.
    • Types of reasoning about mental states.
    • Various ways in which uncertainty arises in reasoning about mental states.
    • Context-sensitivity of reasoning about mental states.
    • Necessity of reasoning about the absence as well as the presence of mental states.
  2. Opacity Problems (4 class sessions)
    • De-dicto/de-re distinction in the interpretation of mental states reports in natural language.
    • Styles of interpretation intermediate between de-dicto and de-re.
    • [Brief] Extension to nested mental state reports.
  3. Modal logic for mental states (6 class sessions)
    • Representational forms. Coping with de-dicto/de-re distinction, incl. in nested cases.
    • Representational inadequacies.
    • Possible-worlds semantics.
    • Reasoning about mental states using simulation (a natural-deduction technique).
    • Reasoning about mental states without using simulation.
    • Computational benefits of simulation.
  4. First-order quotational logic for mental states (3 class sessions)
    • Representational forms. Coping with de-dicto/de-re distinction, incl. in nested cases.
    • Representational advantages. Danger of paradox.
    • Semantic issues [brief].
    • Reasoning issues, with focus on simulation.
  5. Introduction to other representation methods for mental states (2 class sessions)
  6. Mixing simulative and non-simulative reasoning about mental states in a rule-based uncertain-reasoning framework, and handling of conflict-resolution between lines of reasoning (3 class sessions)
  7. BDI (belief/desire/intention) frameworks, agent theory, multi-agent systems (2 class sessions)

Last updated: 28 February 2002

Source file: /internal/modules/COMSCI/2001/xml/02451.xml

Links | Outline | Aims | Outcomes | Prerequisites | Teaching | Assessment | Books | Detailed Syllabus