JOHN BARNDEN (Professor of AI)

PROJECTS: Undergraduate (3-year or 4-year), MSc (any type), MPhil and PhD

My suggestions can be tailored to suit your status (undergraduate, etc.).

I have a special interest in INTERDISCIPLINARY Cognitive Science projects,
notably those involving Psychology, Philosophy, Linguistics, Literature or Education

I will be happiest to supervise projects in the general areas listed below.

However, I also welcome your own suggestions. Often, projects suggested to me can be made yet more interesting by giving them a bit of an AI flavour (to an extent appropriate to your previous experience, if any, with AI.)

You may not fully understand what the following areas are about. Don't hesitate to come and ask me for clarification. Some of the things may look very advanced, but there's plenty of scope for carving out student projects at any appropriate level.

You can get a rough idea of the nature of my research interests from my home page:


PHONE: Extension 4-3816

Room: 136


The following details were mainly written with PhD projects and Advanced MSc projects in mind, but the ideas could also form the basis of undergraduate undergraduate projects and other types of MSc/MPHil projects.

For ideas to do with symbolic reasoning, default reasoning, mental-state reasoning, metaphor and metonymy, please read the ``ATT-Meta Background and Description'' below and then look at sections after it. For the other items, please look at later sections.

Metaphor and metonymy also feature strongly in the E-DRAMA area mentioned above. This document contains no further details about that area, but you are most welcome to ask me about it.

Although the ``ATT-Meta'' description focuses on metaphor and mental states, the research in question provides opportunities for projects on NON-metaphorical reasoning about mental states, metaphor-based reasoning about things OTHER than mental states, and uncertain reasoning that is related NEITHER to metaphor nor to mental states.

There is scope for projects to be either THEORETICAL or IMPLEMENTATIONAL, or both. Also, projects could be INTER-DISCIPLINARY. Particular, non-CS disciplines that could be involved include Psychology, Philosophy, Literature, Linguistics and Education.

ATT-Meta Background and Description

A major function of utterances in written or spoken discourse is to report or otherwise convey the mental states, or changes of mental state, of people who are mentioned in or who are taking part in the discourse. Mental states include thoughts, beliefs, hopes, intentions, desires, and so forth. This has led to work by numerous investigators on how to represent and reason about mental states. The present research project is also in this central arena. However, the previous computational work on this had almost entirely ignored a salient fact about mental state description in real-life discourse, namely that it is often metaphorical. The mind is frequently metaphorically portrayed as a physical space, with ideas as physical objects located at various points within the space, or moving around within the space. Alternatively, mental states are often cast as visual states. There are various other widely used metaphors.

The metaphorical nature of much mental state description is not mere literary coloration: it can make a substantial difference to the meaning of perfectly mundane discourse and to the inferences one needs to perform in understanding the discourse. So far, the present research project has concentrated on belief-like mental states, and has developed and implemented an automated reasoning system, called ATT-Meta, that can reason about people's beliefs, and, in particular, beliefs that are metaphorically described. Importantly, the system reasons about people's reasoning from their beliefs. That reasoning can itself be about people's beliefs, and so on to any level of nesting.

The system reasons uncertainly about beliefs, whereas most existing implemented systems for reasoning about mental states do not deal with uncertainty. Yet uncertainty is a fundamental requirement: we can never have certain knowledge about other people's mental states, we cannot be certain what they infer from their beliefs, and their own reasoning (about anything, including yet other people's beliefs and reasoning) is itself full of uncertainty.

ATT-Meta reasons in a rule-based way, but the rules and hypotheses in the system are annotated with qualitative confidence factors. Conflicts between different lines of reasoning are addressed by means of a conflict-resolution mechanism that tries to assess which line is the stronger.

ATT-Meta's main technique for reasoning about beliefs, metaphorically described or not, is ``simulative reasoning,'' a well-known technique in AI. When one is reasoning simulatively about person X's beliefs, one pretends to adopt their already-conjectured beliefs as one's own, and applies one's own reasoning mechanisms to them. One then tentatively deems the results to be beliefs of the person. ATT-Meta can also do non-simulative reasoning about beliefs, and smoothly combines the two forms of reasoning.

In ATT-Meta, the results of belief reasoning can be inhibited or enhanced by metaphor-based reasoning. The function of metaphor of mind in discourse is often to supply such inhibition or enhancement.

But the system's metaphor-based reasoning techniques are not special to metaphors of mind: they can be applied to metaphors for topics other than mental states. ATT-Meta is currently limited to dealing only with metaphors it has built-in knowledge about. Despite this restriction, the system can deal with creative manifestations of its built-in metaphors in discourse. (A metaphor is just a conceptual view; a manifestation is a piece of language that exploits it.)

Metaphor-based reasoning is itself done by a technique similar to simulative reasoning about beliefs. In effect, when reasoning on the basis of a metaphor of one thing X as another thing Y, the system pretends to believe that X really is Y, and reasons simulatively on that basis. This approach has a number of practical and theoretical advantages.

Apart from developing the ATT-Meta system, the research project has also built a large web-accessible databank of real discourse examples of the use of metaphors of mind.

The research has been published in various conference and journal papers, which are available from me via my website (or on request if not available there).

Current and Future Work on ATT-Meta

Current and future developments include:

Natural Language Front End for ATT-Meta

The ATT-Meta system is currently only a reasoning system --- it does not directly take natural language input. The development of a natural language front-end is a major topic for future work.

Projects on Uncertain Reasoning about Mental States

The project ideas in this section need not have anything to do with metaphor, or indeed have any explicit involvement with natural language processing in general. Also, although the project ideas are inspired by the research on the ATT-Meta system, a project need not be closely involved with that system itself.

As explained above, uncertainty is fundamental in the study of mental states. ATT-Meta contains some quite powerful mechanisms for dealing with uncertainty, but they need further testing, refinement and extension. Here are some issues that could be addressed in your projects:

Projects on Explicit Methods for Resolution of Reasoning Conflicts

A possible project is to investigate non-specificity-based, explicit methods for resolving conflicts. An explicit method here means one in which rules contain special, explicit information aimed at resolving conflicts. The ATT-Meta research has developed a new, potentially advantageous explicit method, but it needs further research and development.

Such a project need not have anything to do with mental states or metaphor, but there would be interesting issues to investigate if it did.

Projects on Metaphor-Based Reasoning

The section above on ``Current and Future Work on ATT-Meta'' provides plenty of scope for projects on metaphor. Again, although such projects would be inspired by the research on the ATT-Meta system, a project need not be closely involved with that system itself.

ATT-Meta's mechanisms should be adequate for some types of mixed metaphor, and they should be adequate for reasoning in certain ways about people's metaphorical reasoning. However, these adequacies have not actually been extensively tested, and in any case the mechanisms need extension and refinement.

Projects on Metonymy

Metonymy (in the broad sense of the term adopted in AI) is the use of a phrase to refer indirectly to something X by means of referring directly to something Y that is saliently related to it. An example is for a waiter to say ``That table over there wants ice-cream,'' where of course it is not the table itself that wants anything at all! Metonymy is often mixed in a complex way with metaphor. Another interesting point is that it is not clear whether some locutions involve metonymy or metaphor, and it is possible that there are intermediate cases.

A project in this area could examine, and possibly implement, the types of processing that are raised by metonymy, especially when mixed with metaphor.


Projects on Analogy/Case-Based Reasoning

Analogy-based reasoning and case-based reasoning are, essentially, techniques for reasoning about a current situation on the basis of records of individual, already-known situations. The two techniques are basically the same, but some researchers like to identify some differences. In the present document they can be taken to mean the same thing. The technique is also related to the topic of example-based reasoning in Psychology. An example of analogy/case-based reasoning is when you reason about what to do in one airport on the basis of memories of experiences in other, specific airports. Case-based reasoning appears to have several advantages over traditional rule-based reasoning, including increased flexibility and increased opportunities for effective learning.

Analogy/case-based reasoning has been widely studied in AI, and has been applied to many different real-world problems.

One particular interest of mine is the use of case-based reasoning in reasoning about people's reasoning and mental states. A former PhD student built a prototype system for this, and there is plenty of scope for further work. Case-based reasoning is interesting in the mental state arena because the question of how people reason, or how their mental states develop in general, is dependent in complex ways on various messy contextual factors, making traditional rule-based reasoning difficult to apply in practice.

Maintained by
Last update: 28 April 2010