Research
I have a broad interest in Natural Language Processing spanning computational logics, constraint solving, inductive logic programming, machine learning, morphology learning, grammar learning, lexical learning, question-answering systems, dialogue systems, virtual agents and interfaces for the smart home.
- Enhanced QA systems and Persuasive Dialogue
Current work is focussed on extending existing Question-Answering (QA) systems to make them more effective for different users. One direction has been on incorporating User-Models (UM) into QA systems to tailor both the content and the presentation of answers to better suit the user. The other direction is on understanding human-human argumentation dialogues and mimicking these within QA systems.
- Embodied Conversational Agents
During my work in Lexicle (2001-2003) I worked on question answering systems and building conversational agents. I programmed the core component of Lexicle’s natural language processing engine. Together with my PhD students (Marco De Boni and Jose-Luis Jara Valencia) we competed in a QA (Question-Answering) track of the (TREC) Text Retrieval Conferences.
- Stochastic Constraint Programming
Jointly with Toby Walsh and Armagan Tarim our work on Stochastic Constraint Programming has developed a new constraint programming framework that permits applying constraint programming methods to stochastic problems such as stock market portfolio management and inventory control. This work was funded by an EPSRC grant. We have developed a new programming language, Stochastic OPL that extends existing constraint programming language OPL by allowing the specification of stochastic decision variables.
- Unsupervised learning of natural language
I have a strong interest in unsupervised learning of natural language. We have worked on unsupervised categorical grammar learning (jointly with Stephen Watkinson); unsupervised learning of morphology (jointly with Dimitar Kazakov); unsupervised learning of lexical relations (jointly with Enrique Alfonseca) and disambiguation using the WWW (jointly with Ioannis Klapaftis).
Current work is focussed on unsupervised learning of lexical relations and sense disambiguation using Google as a text repository.
- Inductive Logic Programming for Natural Language Proceesing
Inductive logic programming (ILP) has a strong potential for application to natural language learning. My work focussed on implementing highly efficient first order decision list learning system. This resulted in the implementation of the CLOG first order decision list learner which is available for download.
- Computational logics and constraint solving for natural language processing
My PhD at Edinburgh (1988-1993) was focussed on computational logics for natural language processing. During this time I worked on unification algorithms, constraint logics, and efficient constraint solving methods for HPSG (Head-driven Phrase Structure Grammar). This work extends feature logics with set valued features and word-ordering constraints. It shows the computational complexity and constraint solving methods needed for checking consistency in a number of closely related logics. It also also shows the close connection between feature logics used in natural language processing and frame based knowledge representation languages popular within AI.
I also worked on hybrid categorical grammar logics jointly with Jochen Doerre (IMS, University of Stuttgart). In this work we showed how arbitrary logics can be embedded within the Lambek Calculus while preserving the soundness and completeness properties.
More recently our work (jointly with Alistair Willis) extends the constraint solving framework to tree logics and in particular to quantifier scoping trees.