Memory and Inference

A set of general knowledge structure manipulation primitives is proposed. These processing elements largely eliminate the need for a separate theory of inference. The elements are presented in terms of an abstract characterixation of the understanding task. The processing theory is given as a particular realisation of this characterixation. 1 . I n t r o d u c t i o n It has been generally recognized that world knowledge is important for making inferences, and that the structure and organization of knowledge in memory has important consequences in determining what inferences are made. For example, Schank and Abelson (1977), Charniak (1078), Colling ford (1078), Wilensky (1078), Lehnert (1082), and Dyer (1082) are all describe systems that use a particular kind of knowledge to make particular classes of inference. However, the overall design the inference engine for text understanding remains obscure. Most of research has focused on the nature of the knowledge needed. The theory of processing per se has generally been relegated a secondary status, and usually developed only as far as was necessary to demonstrate the efficacy of one's knowledge structures. The view taken here is that adequate theories of memory and the structure of knowledge largely eliminates the need for a theory of inference. In this formulation, what gets inferred is a function of what gets accessed in memory. This, in turn, is determined by the structure of knowledge, and by general memory access routines that are sensitive only to the structure and organization of knowledge, not to its content. This view is parsimonious to the one espoused by Quillian (1068). Quillian's notion of representation was primitive by today's standards, however, and he was not concerned with the notion of inference per se. A more sophisticated view of inference is found in Charniak (1082). His approach is similar to the one pursued here, in that both theories endorse highly declarative representations and fairly genera] schemes for manipulating them. However, I am concerned less wi th the details of the implementation, and more with the overall architecture of the understander. ls this paper, I first present an abstract characterization of the understanding task. Then a particular realization for this characterization is given. In this discussion I refer to the representation produced by the story understander as a construst of the input. I call the process of deciding to include a particular fact in a construal determinstion. In addit ion, I shall use the term assumption to refer to the kinds of inferences used in story understanding, to distinguish these from the more general collection of processes that include logical deduction. 2 . Pr inc ip les o f T e x t Comprehension I first attempt to characterize the text understanding process by characterizing those facts whose determination constitute a satisfactory construal of the input. This characterization is given in terms of a set of principles. These are as follows. The Principle of Consistent Construal View the inputs as instantiating a set of frames that are consistent with one's world knowledge. The Principle of Concretion (Overcommitment) Determine as specific a fact as is possible from an input. The Principle of Least Energy Make only the minimal assumptions necessary to determine a fact. The Principle of Exhaustion Determine enough facts to account for all of the input. The Principle of Poignancy Determine those frames that are pertinent to the intent of the discourse. The Principle of Consistent Construal states that, as elements of a text are viewed as instantiating general frames, one should favor interpreting inputs as instantiating those frames that provide a consistent construal of the input. For example, suppose we were given the sentence "The bottle fell off the table and it broke." One construal of this sentence assumes that (1) the bottle hit the floor, (2) it was the bottle (and not the table) thai broke, (3) the bottle was somewhat britt le, and (4) the bottle's hi t t ing the floor, together with the bottle's brittleness, was the cause of the breakage. The Principle of Consistent Construal approves of this construal because it relates individuals in the story ( i . e., the falling event, the breaking event, and the bottle) to knowledge in long term memory in a consistent manner. However, the Principle of Consistent Construal does not inform us how to decide among alternative construals. For example, another construal of our example sentence assumes someone caught the falling bottle mentioned above and then broke it with a hammer. However, this construal seems more ad hoc. We require a way of distinguishing among such alternatives. To address this problem, we introduce the Principle of Least Energy. Least Energy states that one should make those decisions that