Natural Language Generation

Our current research addresses these issues mostly in the context of automatic software documentation, the generation of instructions, and summarisation. In doing so, we look at producing not only English, but also various other European languages. We thus study the phenomena involved from a multilingual perspective. We are also concerned with identifying how language varies depending on its context of use, which includes the person for whom it is intended. For example, how is a set of instructions different when aimed at different readers. Such variation is crucial in order to design useful and flexible systems. Our research involves active interaction between AI specialists, linguists, psycholinguists, and experts in technical writing and translating.

Multilingual generation of instructions
In two closely related projects, DRAFTER and GIST, we are concerned with the automatic drafting of multilingual instructional texts from an underlying knowledge base that represents relevant objects, functions, processes and actions. Both projects are committed to delivering prototype systems, in which an author or a domain expert specifies the information to be included in a given section of the text from the underlying knowledge base. The different language versions are then automatically generated in parallel, according to defined requirements which might match a house style or a controlled language. These drafts can be revised as necessary. The AGILE project aims to extend DRAFTER to three Eastern European languages (Bulgarian, Czech and Russian).

Legal explanations
As a development of this approach the CLIME project aims to provide natural language answers and explanations to legal queries. Using a natural alnguage generator interfaced to a legal inference engine, it will generate responses to queries relating to shipping regulations and laws, from ship engineers and surveyors.

Generic research
In addition to these applied generation projects, we also continue more generic, theoretical research in natural language generation. In the RAGS project, we aim to develop (in collaboration with the Unversity of Edinburgh) a standard 'reference architecture' for generation systems, and to provide standard resources to support the development, testing and evaluation of such systems. The GNOME project (also with Edinburgh and Durham) aims to develop specific techniques for one particularly important component of generation technology - the generation of nominal expressions.

Summarisation
In another project, we are looking at the processes involved in producing summaries. More specifically, we are concerned with issues of conciseness of the text and explicitness of the information. Using discourse theory, we are examining the structural and grammatical decisions involved in the compression of a text. This project is being conducted jointly with the Instituto de Fisica, São Carlos - USP, Brazil. Financial support is from the National Council for Scientific and Technology Development (CNPq), and the Fapesp Project, Brazil.

For further information, please contact Donia Scott (+44 1273 642901) - see our contact page for full contact details.