Automated reasoning about the design of effective visual problem representations is possible when we adopt the view that visual problem representations, along with the pemeptual procedures that humans use to manipulate them, can be described using information-processing models of the sort introduced by Newell and Simon (1972). This approach provides us not only with a means of characterizing visual problem representations in a formal syntax but also with a means of automatically mapping between "logical" and "perceptual" problem representations and procedures. An automated system called BOZ is described that begins with a logical problem representation and solution procedure, and generates an informationally-equivalent visual problem representation and procedure that allows the human user to obtain the solution more efficiently. BOZ's representation mapping technique proceeds by: (1) replacing demanding logical inferences the solution procedure with efficient perceptual inferences; and (2) structuring information the visual representation such that search is minimized. The extent to which the visual representations and procedures produced by BOZ agree with what users actually see and do is discussed.
The major aim is to develop representations that are understandable by a reasoning engine and can be used to answer questions. We use abduction to map natural language sentences into concise and specific underlying theories. Techniques for automatically generating usable data representations are discussed. New techniques are proposed to obtain semantically correct and precise logical representations from natural language, in particular in cases where its syntactic complexity results in fragmented logical forms.
To this purpose, a knowledge representation formalism that fully paxallels the logical form is introduced - allowing for both underspecified semantic representations and encapsulation of contextual knowledge in the form of meaning postulates. In this way, the gap between researchers in computationalinguistics - interested in expressiveness and in computing explicit derivations - and researchers in knowledge representation - interested in devising rigorous logic-based representations with robust and complete computational properties - can be filled. This paper tries to give an answer to most of the problems presented by (Allen 1993). As a matter facts, the approach advocated here is exactly along the lines indicated by Allen. However, Allen's conclusions were in the direction of forgetting about logical forms, and of representing the literal and contextual meanings of utterances in a suitable NL-oriented knowledge representation formalism.