Research of Achim Hoffmann

Research Interests:

My research interest concerns the problem of building intelligent systems. More specifically, the problem of putting the knowledge into the system. I am mainly interested in both, 'manual' incremental knowledge acquisition techniques as well as automatic techniques, i.e. machine learning. The incremental knowledge acquisition approach I am pursuing is based on Ripple Down Rules, which have been proved to be extremely effective, also in commercial settings, such as pathology labs (see Pacific Knowledge Systems for details). The approach allows to build large knowledge bases by integrating the acquired rules in afashion that they do not intefere with each other. Rather, a new rule which repairs the poor performance of the existing knowledge base represents just a local patch. The applicability of a new rule is strictly governed and it is ensured that it only applies in cases where the existing knowledge base was unsatisfactory. The focus of the practical problems being addressed lies in natural language processing systems, in particular, systems that assist the human coping with the enormous wealth of available information in written form. This includes automatic text summarisation, question answering, machine translation, localization of relevant information within a collection of documents or within a given document itself. Recently I have been busy with developing a large piece of infrastructure software for allowing the systematic and efficient acquisition of knowledge for machine translation. While further development is ongoing, this platform is also currently being extended to be applicable to other NLP problems.

Current Research Students:

Graduated Research Students: