Labelled Abduction and Relevance Reasoning

The aim of this paper is to explore modelling a more non standard interaction between a human and a database We try to equip the database with the capability of understand ing queries in natural language and answering them in a more intelligent manner than by mere yes no factual answers This capability of the database requires at least two compo nents First the database must possess a front end logic module which can parse the natural language input I and translate it into a query or update QI in its own logical query lan guage Second and in fact more di cult it must cope with the logical patterns of human answer input query which is logically more non monotonic common sense than ordinary query languages In this paper we rst address the preliminary parsing translation component We present an approach to utterance interpretation which takes as input sequences of natural language words and builds linked labelled databases as output and also involves additional abductive choices which are made on line and captured through the labelling algebra We describe the outlines of the formal model we are developing and provide an experimental implementation of the on line construction of linked labelled databases and interpretation relative to a con text We then turn to discussion of the second component and consider in detail the formal properties of the abduction principles required to equip a database with the common sense faculties required to participate in question answering and dialogues