In this talk I will describe our participation to the BioNLP shared task and the BioCreative II.5 competitions. Our approach is based on a common core: a pipeline of NLP tools and a dependency parser. The adaptation for the BioNLP shared task consisted of suitable input filters and a transformation-based approach which maps syntactic dependencies to event structures. Despite the very simple approach, results were satisfactory (34.78 F-score). The adaptation for BioCreative requires the detection and disambiguation of domain entities, while candidate interactions are proposed on the basis of a simple learning approach.

If time allows I will then describe our approach to finding the 'focus organisms' i.e. the organisms in which the experiments have been conducted or which are the source of the interacting proteins. This information is of crucial importance for the correct disambiguation of other entities mentioned in the article.

Speaker Biography:
Fabio Rinaldi graduated in computer science from the University of Udine, Italy (1995) with a dissertation focusing on knowledge representation for natural language processing. Since then he has worked at four european universities and research centers (ITC/IRST - Trento, Italy; GMD/IPSI - Darmstadt, Germany; University of Manchester, UK; University of Zurich, Switzerland) in different research projects, with a focus on Information Extraction, Question Answering and Text Mining. At the University of Zurich he also obtained a doctoral degree in computational linguistics. In 2005 he initiated the OntoGene project, which targets biomedical text mining. Currently he is a lecturer and researcher at the University of Zurich, acting as team leader of the OntoGene project.

http://www.cl.uzh.ch/rinaldi/