Grammars for Question Answering Systems Based on Inteligent Text Mining in Biomedicine

-A top-down approach to the determination of grammar engineering methods for the construction of Natural Language Grammars for Question Answering Systems based on Intelligent Text Mining for Biomedical applications is presented in the present paper. Our proposal is to start with the formulation of the ultimate task goal such as specialized question answering and derive from it the specifications of the grammar engineering methods such as hand crafting or machine learning. The task goal we have experience in involves syllogistic inference with scientific texts and models in biomedicine. An important kind of knowledge that we deal with is causal knowledge expressed either linguistically or formally. The linguistic expressions we deal with involve causal verbs that connect noun phrases that denote entityprocess pairs. An illustrative example of a Text Grammar is given and a pilot experiment towards Memory Based Induction using TiMBL is described. The results obtained so far render problematic the automatic induction of multilevel grammars.