Search

Reports on publications and presentations done during the quarter. Summarizes research plan going forward. Describes new research questions identified during the quarter.

This report describes research done at the Artificial Intelligence Laboratory of the Massachusetts Institute of Technology. Support for the laboratory's artificial intelligence research is provided in part by the Advanced Research Projects Agency of the Department of Defense under Office of Naval Research contract N00014-75-C-0643., This paper is written for an Area Examination on the three papers: "A Framed PAINTING: The Representation of a Common Sense Knowledge Fragment" by Eugene Charniak, "Reporter: An Intelligent Noticer" by Steve Rosenberg, and "Using Plans to Understand Natural Language" by Robert Wilensky. Surprisingly, these papers share a common view of what it means to understand a story. The first part of this paper reviews the previous notions of "understanding", showing the progression to today's consensus. The content of the consensus and how the individual papers fit within it is then described. Finally, unsolved problems not adequately dealt with by any of the approaches are presented briefly., MIT Artificial Intelligence Laboratory

This report describes research done at the Artificial Intelligence Laboratory of the Massachusetts Institute of Technology. Support for the laboratory's artificial intelligence research is provided in part by the Advanced Research Projects Agency of the Department of Defense under Office of Naval Research contract N00014-75-C-0643., This paper discusses what it would mean to have a psychological model of the language production process: what such a model would have to account for, what it would use as evidence. It outlines and motivates one particular model including: presumptions about the input to the process, a characterization of language production as a process of selection under constraint, and the principle stipulations of the model. This paper is an introduction, which is largely nontechnical and uses only simple examples. A detailed presentation of the architecture of the model, its grammar, and its interface to the speaker will be forthcoming in other papers., MIT Artificial Intelligence Laboratory Department of Defense Advanced Research Projects Agency

Work reported herein was conducted at the Artificial Intelligence Laboratory, a Massachusetts Institute of Technology research program supported in part by the Advanced Research Projects Agency of the Department of Defense and monitored by the Office of Naval Research under Contract Number N00014-70-A-0362-0005., This paper discusses the problem of getting a computer to speak, generating natural language that is appropriate to the situation and is what it wants to say. It describes, at a general level, a program which will embody a theory of how the various types of available information are used in the linguistic process as well as the possible packaging for some of that information and the experimental situation in which the program will be developed., MIT Artificial Intelligence Laboratory

Generation is the process of deliberately constructing a natural language text in order to meet specified communicative goals. The goals come from another program, perhaps an expert reasoning system or a ICAI tutor, that is motivated to talk to a human user. The texts that are produced may range from a single phrase given in answer to a question, through multi-sentence remarks and questions within a dialog, to full-page explanations. This article describes AI research on natural language generation with a historical perspective, emphasizing the special character of the problems to be solved. It begins by contrasting generation with language understanding, establishing basic concepts about the breakdown of the process into components and the flow of information and decisions through it. A section of excerpts from the output of illustrative generation systems follows, showing what kinds of performance are possible and where the difficulties are. In the remainder of the article the common approaches to generation are surveyed, including characteristic messages and the nature of a generator's lexicon.

This paper examines the nature of generation systems today, the problems they have been designed to deal with, their strengths and their weakness. Its goals to give the MT community a sense of what has been accomplished, and indirectly to show where MT researchers could consider adopting or adapting some of the AI work. This work on generation need not be done by AI people alone: MT can, for example, contribute to AI research on the planning-level by sharpening our collective understanding of the 'carrying capacity' of the different parts of a language through cross-language comparisons that try to fit the ideas carried by the linguistic devices of a source language into the alternative devices of a target language. At lower levels, MT as a task can provide more linguistically demanding sources for generation than most any of today's expert systems. At the same time it is clear that generation is done for very different reasons in two camps. The AI context is more like that of people dealing with each other in normal life--of which translation is not a customary part. Nevertheless, translation is a normal human capacity, and a considered comparison of the generation process in both contexts should tell us more about the nature of generation as a module within the human mind than could either by itself.

In this paper we present a means of compensating for the semantic deficits of linguistically naive underlying application programs without compromising principled grammatical treatments in natural language generation. We present a method for building an interface from today's underlying application programs to the linguistic realization component Mumble-86. The goal of the paper is not to discuss how Mumble works, but to describe how one exploits its capabilities. We provide examples from current generation projects using Mumble as their linguistic component.

This report consists of two documents describing the state of the art of computer generation of natural language text. Both were prepared by a panel of individuals who are active in research on text generation. The first document assesses the state of the art, identifying four kinds of technical developments which will shape the art in the coming decade: linguistically justified grammars, knowledge representation methods, models of the reader, and models of discourse. The second document is a comprehensive bibliography on text generation, the first of its kind. In addition to the citations of documents, it includes descriptions of ongoing research efforts. (Author), Prepared in cooperation with Pennsylvania Univ., Philadelphia, Rept. no. MS-CIS-81-9 and SRI International, Menlo Park, CA. Contracts N00039-80-C-0645 and N00039-80-C-0575.