Representing the Semantics of Natural Language as Constraint Expressions

The issue of how to represent the "meaning" of an utterance is central to the problem of computer understanding of natural language. Rather than relying on ad-hoc structures or forcing the complexities of natural language into mathematically elegant but computationally cumbersome representations (such as first-order logic), this paper presents a novel representation which has many desirable computational ahd logical properties. It is proposed to use this representation to structure the "world knowledge" of a natural-language understanding system. Working Papers are informal papers intended for internal use. CONTENTS 1 The general problem-domain 3 2 The proposed solution-domain 2.1 Experts 5 2.2 Some history 6 2.3 Advantages of a common representation 7 3 The conceptual layers 9 4 Using the high-level constructs 4.1 Entities (noun groups) 12 4. 2 Events and the case-frame hierarchy 13 4.3 Events and difference descriptions 15 4.4 Events, Minsky's frames, and defaults 15 4.5 Disambiguation-Domain-Range restrictions 18 5 Relation to other work 22 Appendix 25 Bibliography 1 The general problem-domain This project is primarily concerned with the "representation problem": How complex interrelated symbolic information can be structured so that a computer can use it "intelligently". For this project, the information to be structured is the "knowledge of the world" which a person must bring to bear in order to understand some simple natural language utterances (such as found in children's stories), and the "intelligent" use of such information is exactly this process of understanding. Following the lead of other Artificial Intelligence research into the problem of natural-language understanding (primarily Charniak [1972] and McDermott [1974a]), we can consider the computer to have "understood" an utterance if it can answer relevant questions from it. For example, consider the following story fragment from Charniak's thesis (the line numbers are for later reference): (1) Fred was going to the store. (2) Today was Jack's birthday (3) and Fred was going to get a present. Now consider some questions which might reasonably be asked: "Why did Fred go to the store?", and "Who is the present for?" For a human reader the answers are trivially obvious. However, in answering these questions, a significant amount of "knowledge of the world" must be used. The first question requires knowing that presents can be gotten at stores (so that "going to get a present" is sufficient motivation for "going to the store"), and the second requires knowing that one often gives presents to another on …