Robert F. Simmons

Answering English questions by computer: a survey

Fifteen experimental English language question-answering I systems which are programmed and operating are described ) arid reviewed. The systems range from a conversation machine ~] to programs which make sentences about pictures and systems s~ which translate from English into logical calculi. Systems are ~ classified as list-structured data-based, graphic data-based, ~! text-based and inferential. Principles and methods of opera~4 tions are detailed and discussed. It is concluded that the data-base question-answerer has > passed from initial research into the early developmental ~.4 phase. The most difficult and important research questions for ~i~ the advancement of general-purpose language processors are seen to be concerned with measuring meaning, dealing with ambiguities, translating into formal languages and searching large tree structures.

Natural language question-answering systems: 1969

Recent experiments in programming natural language question-answering systems are reviewed to summarize the methods that have been developed for syntactic, semantic, and logical analysis of English strings. It is concluded that at least minimally effective techniques have been devised for answering questions from natural language subsets in small scale experimental systems and that a useful paradigm has evolved to guide research efforts in the field. Current approaches to semantic analysis and logical inference are seen to be effective beginnings but of questionable generality with respect either to subtle aspects of meaning or to applications over large subsets of English. Generalizing from current small-scale experiments to language-processing systems based on dictionaries with thousands of entries—with correspondingly large grammars and semantic systems—may entail a new order of complexity and require the invention and development of entirely different approaches to semantic analysis and question answering.

A Computational Approach to Grammatical Coding of English Words

As a firs l~ step in many computer language processing systems, each word in a natural language sentence must be coded as to its form-class or part of speech. This paper describes a computational grammar coder which has been completely programmed and is oper~tional on Lhe IBM 7090. It is part of a complete syntactic annlysis system for which it accomplishes word-class coding, using a computational approach rather than the usual method of dictionary lookup. The resulting system is completely contained in less than 1~,000 computer words. It processes running English text on the IBM 7090 at a rate of more than 1250 words per minute. Since the system is not dependent on large dictionaries, it operates on any ordinary English text. In preliminary experiments with scientific text, the system correctly and unambiguously coded over 90 percent of the words in two samples of scientific writing. A fair proportion of the remaining ambiguity can be removed at higher levels of synvactic analysis, but the problem of structural ambiguity in natural languages is seen to be a critical one in the development of practical language processing systems.

A deductive question-answerer for natural language inference

The question-answering aspects of the Protosynthex III prototype language processing system are described and exemplified in detail. The system is written in LISP 1.5 and operates on the Q-32 time-sharing system. The systems data structures and their semantic organization, the deductive question-answering formalism of relational properties and complex-relation-forming operators, and the question-answering procedures which employ these features in their operation are all described and illustrated. Examples of the systems performance and of the limitations of its question-answering capability are presented and discussed. It is shown that the use of semantic information in deductive question answering greatly facilitates the process, and that a top-down procedure which works from question to answer enables effective use to be made of this information. It is concluded that the development of Protosynthex III into a practically useful system to work with large data bases is possible but will require changes in both the data structures and the algorithms used for question answering.

A computational model of verbal understanding

The long-term goal for computational linguistics is to increase our understanding of linguistic and conceptual structures and to formally describe them so that computers can deal effectively with natural languages in such applications as question answering, stylistic and content analysis, essay writing, automated translation, etc. The eventual realization of this goal requires not only a satisfactory model of linguistic structures, but also models for verbal understanding and verbal meaning. In this paper we outline a theory and a model of verbal understanding and describe Protosynthex III, an experimental implementation of the model in the form of a general-purpose language processing system. The effectiveness of the model in representing the process of verbal understanding is demonstrated in terms of Protosynthex IIIs capability to disambiguate English sentences, to answer a range of English questions and to derive and generate meaning-preserving paraphrases.

The clowns microworld

About fifteen years of active research in natural language question-answering systems has provided reasonably concise and elegant formulations of computational semantics for understanding English sentences and questions about various microworlds. These include the Woods Lunar Data Base, the Winograd world of a pictured hand and blocks, the Heidorn world of a fueling station, the Hendrix, Slocum, Thompson world of transactions, John Seely Browns power circuit and Schanks sketches of motivated humans. (See Woods <u>et al</u> 1972, Winograd 1972, Hendrix <u>et al</u> 1973, Heidorn 1972, Schank 1975 and Brown <u>et al</u> 1974.) In each of these worlds, a natural language processor is able to understand an ordinary subset of English and use it conversationally to accept data and to respond to commands and questions.

Relating sentences and semantic networks with procedural logic

A system of symmetric clausal logic axioms is shown to transform a thirteen-sentence narrative about a v-2 rocket flight into semantic case relations. The same axioms translate the case relations into english sentences. An approach to defining schemas in clausal logic is presented and applied in the form of a mini-flight schema to two paragraphs of the text to compute a partitioning of the semantic network into the causal organization of a flight. Properties of rule symmetry and network condensibility are noted to be of importance for natural language processing. Because of the conciseness of the logic interpreter and the clausal representation for grammars and schemes, it is concluded that the procedural logic approach provides an effective programming system that is promising for accomplishing natural language computations on mini- and microcomputers as well as on large mainframes. 29 references.

Storage and retrieval of aspects of meaning in directed graph structures

An experimental system that uses LISP to make a conceptual dictionary is described. The dictionary associates with each English word the syntactic information, definitional material, and references to the contexts in which it has been used to define other words. Such relations as class inclusion, possession, and active or passive actions are used as definitional material. The resulting structure serves as a powerful vehicle for research on the logic of question answering. Examples of methods of inputting information and answering simple English questions are given. An important conclusion is that, although LISP and other list processing languages are ideally suited for producing complex associative structures, they are inadequate vehicles for language processing on any large scale—at least until they can use auxiliary memory as a continuous extension of core memory.

A text knowledge base from the AI handbook

Abstract This research aims at defining a consistent set of text representation conventions for organizing fifty pages of the AI handbook as an inferential knowledge base founded on a procedural logic system of general inference schemes for answering questions from it. As a result of research on the AI handbook project, we have developed a prototype, natural-language, text-knowledge system that includes a data base manager to compile the text knowledge and to make it available to navigational commands. The text is represented as logical propositions that form a set of text axioms to model its content. English questions and commands are translated to corresponding logical formulas and treated as theorems to be proved with respect to the text model. The logical form is that of semantic relations (SRs)—logical predicates with varying numbers and ordering of arguments. To compute effectively with such a free form, a relaxed unification procedure was defined as the basis of the SR theorem prover. The use of procedural logic, augmented with fast compiled LISP functions, has shown that questions can be answered in times ranging from a few tenths of a second to minutes of CPU time on a DEC2060 system. The navigational capabilities of the data base manager make available larger contexts surrounding the text and offer the user complete freedom to explore the text and to extract any desired information from it.

RULE FORMS FOR VERSE, SENTENCES, AND STORY TREES

Rule forms and their interpreters are described for deriving sensible and nonsensical verse, for analyzing sentences into case structures, for generating sentences from case structures, and for generating story trees. A system of inference rules and assertions in the form of Horn clauses and their interpreter are presented as a computational method for generating narrative story trees that have the property that their terminal propositions form the story, while nodes closer to the root provide summaries. The story trees and their generator are proposed as a promising computational model for the macrostructure theorized by Kintsch and van Dijk to account for a human readers memory and understanding of narrative text.