Artificial Intelligence 11 (1978), 197-223.

The paper discusses the incorporation of richer semantic structures into the Preference Semantics sys­tem: they are called pseudo-texts and capture some­thing of the information expressed in one type of frame proposed by Minsky (q.v.). However, they are in a format, and subject to rules of inference, consist­ent with earlier accounts of this system of language analysis and understanding. Their use is discussed in connection with the phenomenon of extended use: sentences where the semantic preferences are broken. It is argued that such situations are the norm and not the exception in normal language use, and that a lan­guage understanding system must give some general treatment of them. A notion of sense projection is proposed, leading on to an alteration of semantic for­mulas (word sense representations) in the face of unexpected context by drawing information from the pseudo texts. A possible implementation is described, based on a new semantic parser for the Preference Semantics system, which would cope with extended use by the methods suggested and answer questions about the process of analysis itself. It is argued that this would be a good context in which to place a lan-

* Editor's note: The abstracts in this issue are not as "current" as I would like, but, because we did not publish the Journal in 1979, there is still much material to get caught up on.

Department of Language and Linguistics University of Essex Wivenhoe Park

guage understander (rather than that of question-answering about a limited area of the real world, as is normal) and, moreover, that the sense projection mechanisms suggested would provide a test-bed on which the usefulness of frames for language under­standing could be realistically assessed.

Artificial Intelligence 11 (1978), 225-265.

Notions like "frames," "scripts," etc. are now being used in programs to understand connected discourse. We will describe a program in this vein which under­stands simple stories about painting. (Jack was paint­ing a chair. He dipped a brush into some paint. Q:Why?) In particular, problems of matching, read time inference, and undoing false conclusions will be stressed. The program makes heavy use of real world knowledge, and there is an extensive discussion of various issues in knowledge representation and how they affect frame representations: modularity, the need for problem solving, worldly vs control knowledge, and cleanliness. The paper concludes with an extensive discussion of the program's shortcomings.

An Artificial Intelligence Approach to Language Instruction Ralph M. Weischedel

Department of Computer and Information Sciences University of Delaware Newark, Delaware 19711 Wilfried M. Voge Department of German University of California Irvine, California 92717

School of Social Science University of California Irvine, California 92717

Artificial Intelligence 10 (1978), 225-240.

This paper describes an implemented, prototype system for a sophisticated, intelligent tutor for instruc­tion in a foreign language. The system is an applica­tion of artificial intelligence research in natural lan­guage, but it implements several ideas that depart from standard approaches to natural language understand­ing. For instance, the semantic analyzer diagnoses several kinds of comprehension problems and semantic errors that a student might make. Some fine distinc­tions in meaning are represented to detect misuse of words. Not only is a model of good syntax included in the tutor, but also a model of incorrect forms, rich enough to pinpoint specific syntactic mistakes. Find­ing the intended interpretation is complicated by the likelihood of student errors. Therefore, perfect syn­tactic form is not necessary for semantic analysis of the student's input. The problems discussed and solu­tions presented are closely related to the more general problem of how to respond to a natural language input that surpasses the computer's model of language or of context.

Cognitive Science 3 (1979), 1-28.

Bobrow and Winograd have presented to the AI community two descriptions of KRL (Bobrow & Wino­grad, 1977, Bobrow, Winograd et al., 1977) which explicate both a high level AI programming language and a theory of knowledge representation. In actual practice, the line between these roles is necessarily vague. As is the case with all programming languages, commitments made to specific data formats or control structures profoundly affect design decisions made by the user. In KRL, there are additional commitments to knowledge representation in the programming lan­guage as well. While these commitments are neutrally presented as convenient features of high level lan­guage, their impact on the user would be far less neu­tral. To a user who has not previously investigated problems of knowledge representation first-hand, KRL either suggests a particular approach or imposes that same approach. In either case, the user is liable to be unconscious of the continual trade-off between low level design options and high level programming con­venience.

Terry Winograd Computer Science Department Stanford University Stanford, California 94305

Cognitive Science 3 (1979), 29-42.

Wendy Lehnert and Yorick Wilks (pp. 1-28 of this issue of Cognitive Science) have written a lengthy pa­per raising a number of issues concerning KRL. [See

abstract above.] Much of their paper is an excellent explanation of some of the features and problems of KRL, and will serve to clarify things which we have explained poorly or not at all in previous papers. Oth­er parts of what they say we find more contentious, and much of this response will be an argument against views of theirs which we feel are confused or wrong. The decision to focus on the disputes does not imply a general rejection of the paper. It was clearly intended in the spirit of constructive criticism and makes a number of valid and important points. We feel it is useful to write a response, not as a defense, but as a further step in a dialog through which we will all come to a better understanding of language and cognition.

Cognitive Science 3 (1979), 43-66.

This paper explores some issues which a humanlike story system ought to encompass, but which are usual­ly not in the main focus of narrative models since Propp. We argue that knowledge about actions and interactions can account not only for how stories are constructed, but also for why some stories are more interesting and enduring than others. We analyze a traditional English folktale in these terms, and show how classes of surface expressions make recoverable the underlying structures and dependencies that schema-based understanders comprise.

Cognitive Science 3 (1979), 67-90.

Coherence in conversations and in texts can be partially characterized by a set of coherence relations, motivated ultimately by the speaker's or writer's need to be understood. In this paper, formal definitions are given for several coherence relations, based on the operations of an inference system; that is, the relations between successive portions of a discourse are charac­terized in terms of the inferences that can be drawn from each. In analyzing a discourse, it is frequently the case that we would recognize it as coherent, in that it would satisfy the formal definition of some coherence relation, if only we could assume certain noun phrases to be coreferential. In such cases, we will simply assume the identity of the entities referred to, in what might be called a "petty conversational implicature," thereby solving the coherence and core­ference problems simultaneously. Three examples of different kinds of reference problems are presented. In each, it is shown how the coherence of the dis­course can be recognized, and how the reference prob­lems are solved, almost as a by-product, by means of these petty conversational implicatures.

Artificial Intelligence Center SRI International 333 Ravenswood Avenue Menlo Park, California 94025

Technical Note 170, Oct. 1979.

As part of the continuing development of the LAD­DER system, we have substantially expanded the ca­pabilities of the data ' base access component that serves as the interface between the natural-language front end of LADDER and the data base management systems on which the data is actually stored. SODA, the new data base access component, goes beyond its predecessor IDA, in that it accepts a wider range of queries and accesses multiple DBMSs. This paper is concerned with the first of these areas, and discusses how the expressive power of the query language was increased, how these changes affected query process­ing in a distributed data base, as well as what are some limitations of and planned extensions to the current system.

Computational Models of Beliefs and the Semantics of Belief-Sentences Robert C. Moore and Gary G. Hendrix

Technical Note 187, June 1979.

This paper considers a number of problems in the semantics of belief sentences from the perspective of computational models of the psychology of belief. We present a semantic interpretation for belief sentences and show how this interpretation overcomes some of the difficulties of alternative approaches, especially those based on possible-world semantics. Finally, we argue that these difficulties arise from a mistaken at­tempt to identify the truth conditions of a sentence with what a competent speaker knows about the meaning of a sentence.

Artificial Intelligence Center SRI International 333 Ravenswood Avenue Menlo Park, California 94025

Technical Note 185, April 1979.

When two people talk, they focus their attention on only a small portion of what each of them knows or believes. Both what is said and how it is interpreted depend on a shared understanding of this narrowing of attention to a small highlighted portion of what is known.

Focusing is an active process. As a dialogue prog­resses, the participants continually shift their focus and thus form an evolving context against which utterances are produced and understood. A speaker provides a hearer with clues of what to look at and how to look at it - what to focus on, how to focus on it, and how wide or narrow the focusing should be. As a result, one of the effects of understanding an utterance is that the listener becomes focused on certain entities (both objects and relationships) from a particular per­spective.

Focusing clues may be linguistic or they may come from knowledge about the relationships between enti­ties in the domain. Linguistic clues may be either explicit, deriving directly from certain words, or im­plicit, deriving from sentential structure and from rhet­orical relationships between sentences.

This paper examines the relationships between fo­cusing and definite descriptions in dialogue and its implications for natural language processing systems. It describes focusing mechanisms based on domain-structure clues which have been included in a comput­er system and, from this perspective, indicates future research problems entailed in modeling the focusing process more generally.

Artificial Intelligence Center SRI International 333 Ravenswood Avenue Menlo Park. California 94025

Technical Note 188, June 1979.

Communication in natural language requires a com­bination of language-specific and general common-sense reasoning capabilities, the ability to represent and reason about the beliefs, goals, and plans of multi­ple agents, and the recognition that utterances are multifaceted. This paper evaluates the capabilities of natural language processing systems against these re­quirements and identifies crucial areas for future research in language processing, common-sense reason­ing, and their coordination.

Technical Note 197, Oct. 1979.

A framework is proposed for developing a portable natural language interface to large data bases. A dis­cussion of problems arising from portability leads to the identification of a key concept of the framework: a conceptual schema for representing a user's model of the domain as distinct from the data base schema. The notions of conceptual completeness and linguistic coverage are shown to be natural consequences of this framework. An implementation of the framework, called D-LADDER, is presented, and some preliminary performance results reported.

Theoretical Foundations of Linguistics and Automatic Text Processing Jane J. Robinson

Technical Note 199, Oct. 1979.

Texts are viewed as purposeful transactions whose interpretation requires inferences based on extra-linguistic as well as on linguistic information. Text processors are viewed as systems that model both a theory of text and a theory of information processing. The interdisciplinary research required to design such systems has a common center, conceptually, in the development of new kinds of lexical information, since words are not only linguistic objects, they are also psychological objects that evoke experiences from which meanings can be inferred. Recent developments in linguistic theory seem likely to promote more fruit­ful cooperation and integration of linguistic research with research on text processing.

Technical Note 205, Feb. 1980.

This paper presents an explanatory overview of a large and complex grammar, DIAGRAM, that is used in a computer system for interpreting English dialogue.

Artificial Intelligence Center SRI International 333 Ravenswood Avenue Menlo Park, California 94025

DIAGRAM analyzes all of the basic kinds of phrases and sentences and many quite complex ones as well. It is not tied to a particular domain of application, and it can be extended to analyze additional constructions, using the formalism in which it is currently written. For every expression it analyzes, DIAGRAM provides an annotated description of the structural relations holding among its constituents. The annotations pro­vide important information for other parts of the sys­tem that interpret the expression in the context of a dialogue.

DIAGRAM is an augmented phrase structure gram­mar. Its rule procedures allow phrases to inherit at­tributes from their constituents and to acquire attrib­utes from the larger phrases in which they themselves are constituents. Consequently, when these attributes are used to set context-sensitive constraints on the acceptance of an analysis, the contextual constraints can be imposed by conditions on dominance as well as conditions on constituency. Rule procedures can also assign scores to an analysis, rating some applications of a rule as probable or as unlikely. Less likely ana­lyses can be ignored by the procedures that interpret the utterance.

In assigning categories and writing the rule state­ments and procedures for DIAGRAM, decisions were guided by consideration of the functions that phrases serve in communication as well as by considerations of efficiency in relating syntactic analyses to proposition-al content. The major decisions are explained and illustrated with examples of the rules and the analyses they provide. Some contrasts with transformational grammars are pointed out and problems that motivate a plan to use redundancy rules in the future are dis­cussed. (Redundancy rules are meta-rules that derive new constituent-structure rules from a set of base rules, thereby achieving generality of syntactic state­ment without having to perform transformations on syntactic analyses.) Other extensions of both grammar and formalism are projected in the concluding section. Appendices provide details and samples of the lexicon, the rule statements, and the procedures, as well as analyses for several sentences that differ in type and structure.

Interpreting Natural-Language Utterances in Dialogs About Tasks Ann E. Robinson, Douglas E. Appelt, Barbara J.

Technical Note 210, March 1980.

This paper describes the results of a three-year research effort investigating the knowledge and processes needed for participation in natural-language dia­logs about ongoing mechanical-assembly tasks. Major concerns were the ability to interpret and respond to utterances within the dynamic environment effected by progress in the task, as well as by the concomitant shifting dialog context.

The research strategy followed was to determine the kinds of knowledge needed, to define formalisms for encoding them and procedures for reasoning with them, to implement those formalisms and procedures in a computer system called TDUS, and then to test them by exercising the system.

Principal accomplishments include: development of a framework for encoding knowledge about linguistic processes; encoding of a grammar for recognizing many of the syntactic structures of English; develop­ment of the concept of "focusing," which clarifies a major role of context; development of a formalism for representing knowledge about processes, and proce­dures for reasoning about them; development of an overall framework for describing how different types of knowledge interact in the communication process; development of a computer system that not only dem­onstrates the feasibility of the various formalisms and procedures, but also provides a research tool for test­ing new hypotheses about the communication process.

A Plan-Based Approach to Speech Act Recognition James F. Allen

Technical Report No. 131179, Feb. 1979.

This thesis concerns how people use language to communicate. While the literal meaning and syntactic structure of a sentence tell us what the speaker said, often they do not capture what was actually meant. For instance, when we hear "Can you pass the salt?" at the dinner table, why do we take this to be a re­quest to pass the salt rather than a query about our salt-passing ability? The answer, in part, is that un­derstanding what was meant requires recognizing what the speaker intended, i.e. what he hopes to achieve. Thus, we consider "Can you pass the salt?" to be a request because we recognize that the speaker said it intending to get us to pass the salt.

In addition, people often communicate using only short phrases or even single words. For instance, were you to walk into my office, I could request that you close the door by simply saying "the door".

This thesis examines the language comprehension process using a model of goal-directed behaviour. Ut­terances are considered to be instances of speech acts, which are actions that can be reasoned about in a

planning system. A large part of a hearer's under­standing of an utterance involves recognizing the plan (of the speaker) that produced it. This plan may con­tain obstacles, goals that the speaker cannot achieve without assistance. The responses are formed on the basis of these obstacles.

Speech acts are defined in terms of the plan that the hearer believes the speaker intended him to recog­nize. Speech act identification is then accomplished by the plan recognition process. This method accounts for many sentences that are not intended with their literal meaning (the indirect speech acts). In particu­lar, this thesis concentrates on the indirect forms of requests and informs.

There are other linguistic phenomena that can be explained by this model and that are considered in the thesis. Answers that provide more information than explicitly asked for often arise from the method of basing responses on the obstacles in the speaker's plan. Some sentence fragments can identify the speaker's plan and obstacles. Thus, these can be an­swered appropriately without ever having to construct a syntactically complete sentence from the fragment. This work also extends naturally to explain the genera­tion and participation in subdialogues whose purpose is to clarify or correct previous utterances.

Finally, it examines possible extensions to this model. In particular, it considers the generation (the planning) of indirect forms of speech acts and discuss­es how this work relates to the other classes of speech acts.

Research Report RR-79-77, Nov. 1979.

Natural dialogue does not proceed haphazardly; it has an easily recognized "episodic" structure and co­herence that conform to a well developed set of con­ventions. This report represents these conventions formally in terms related to speech act theory and to a theory of action. The major formal unit, the Dialogue Game, specifies aspects of the communication of both participants in a dialogue. We define the formal no­tion of Dialogue Games and describe some of the im­portant Games of English. Dialogue Games are con­ventions of interactive goal pursuit. Using them, each participant pursues his own goal in a way that some­times serves the goals of the other. The idea of Dia­logue Games can thus be seen as part of a broader theoretical perspective characterizing virtually all com­munication as goal pursuit activity. We also define and exemplify the property of Motivational Coherence of dialogues. Motivational coherence can be used as an interpretive principal in explaining language com­prehension. Actual dialogue games have a kind of casual connectedness that is not a consequence of their formal properties. This is explained in terms of a theory of action, which is also seen to explain a similar attribute of speech acts.

Computer as Author - Prospects and Results William C. Mann and James A. Moore

Research Report RR-79-82, Jan. 1980.

For a computer program to be able to compose text is interesting both intellectually and practically. Artifi­cial Intelligence research has only recently begun to address the task of creating coherent texts containing more than one sentence.

Our recent research has produced a new paradigm for organizing and expressing information in text. This paradigm, called Fragment-and-Compose, has been used in a pilot project to create texts from se­mantic nets. The method involves dividing the given body of information into many small propositional units, and then combining these units into smooth coherent text. So far the largest example written by Fragment-and-Compose has been two paragraphs of instruction about what a computer operator should do in case of indications of a fire.

This report describes the text generation problem and anticipates a specific way to disseminate and use technical developments. It presents the research that led to creation of Fragment-and-Compose, including the largest example of computer-produced text. It also discusses the immediate problems and difficulties of elaborating Fragment-and-Compose into a general and powerful method.

HAM-RPM Report No. 11, March 1979, (In German).

This paper discusses adaptations of the ATN for­malism for semantic and pragmatic processing in natu­ral language understanding and generation. First, the various uses of ATNs are classified and a set of defini­tions of terms like 'semantically guided parsing', 'semantic grammar' and 'pragmatic grammar' is pro­posed. The use of ATNs for semantic and pragmatic

processing is investigated by discussing the generation of definite noun phrases. It is shown that when we try to describe the generation of noun phrases as a cogni­tive process whose results enable the listener to identi­fy an object intended by the speaker - as it is done in the dialogue system HAM-RPM - we are confronted with several limitations of the ATN formalism. Final­ly, ten critical theses are formulated which summarize the strengths and limitations of ATNs within the framework of semantic and pragmatic processing.

HAM-RPM Report No. 12, May 1979.

HAM-RPM is a dialogue system which converses with a human partner in colloquial German about lim­ited, but interchangeable, scenes. The objective of this report is to give a detailed, complete and self-contained description of the system in its present state of implementation. After a discussion of the goals and methodological principles which guide our research and a short introduction to the implementation lan­guage, an overview of the system's architecture, of its knowledge base and of the domains of discourse is given. Then each processing phase from the analysis of natural-language input to the generation of a natural-language utterance is described in detail. The examples used during these descriptions are supple­mented by transcripts of complete dialogue sessions. Finally HAM-RPM's programming environment is described.

Cooperative Dialogue Behaviour in the Natural Language System HAM-RPM. W. Hoeppner, A. Jameson Universität Hamburg Germanisches Seminar Von-Melle-Park 6

HAM-RPM Report No. 13, April 1979, (In German).

This paper describes certain cooperative features of the natural language dialogue system HAM-RPM. Two of the system's components employed in analys­ing the natural partner's input are discussed with par­ticular attention to cooperative dialogue behaviour. The realization of such behaviour in response to un­known forms encountered during lexical analysis pres­ents several sequentialization problems. The second component - NP-resolution - provides for clarification dialogues in natural language and verbalizations of time-consuming processing steps ('Thinking aloud').

HAM-RPM Report No. 14, Nov. 1979.

Techniques for dealing with fuzziness are described. First, the components of a natural language dialogue system in which fuzziness can figure importantly are enumerated and the communicative and cognitive functions of vagueness are surveyed. Second, some techniques employed within the dialogue system HAM-RPM for handling fuzziness during the analysis and generation phases are illustrated. Finally, a new model of fuzzy reasoning is presented which is based on a many-sorted fuzzy logic and includes a corrobo­ration procedure for multiple derivations.

Technical Report 79.04.002, April 1979.

It is argued that research in powerful grammar structures opens one possible way to solve the problem of growing complexity in natural language processing for endusers. In automatic indexing there are first indications that analysis can be reduced to "powerful grammar structures", which are seen as function of statistical frequency, simplicity of algorithmic solution and high information value. In the area of query lan­guages the idea of using a subset of language proves feasible. This was shown by evaluating USL, a natural language based question answering system.

Technical Report 79.04.003, May 1979.

The question is asked whether the concept of using subsets of natural languages as query languages for data bases turns out to be feasible in actual applica­tions using the question answering system 'USER SPE­CIALTY LANGUAGES' (USL). Methods of evaluat­ing a natural language based information system are discussed. The results (error and language structure evaluation) give first indications of how to form the general architecture of application systems which use a subset of German as query language.

4 Proctor Ave.

SISTM Quarterly II, 3 (Spring 1979), 16-20.

Labelling a data base management system as "natural language" is a little like labelling a package "natural food": the label doesn't tell us much about the contents. For the purposes of this paper, I want to adopt an absolutely minimal definition of "natural language," and then go on to what seems to me to be the more important question for the user: How well does it work? How much help will it give me in solv­ing my problems?

I want to suggest several criteria for comparison, evaluation, and selection of natural language query systems for use with structured data bases. The crite­ria should also be useful in comparing natural language systems with artificial languages in similar applications. More importantly, I want to suggest that there are "natural language" systems which are working, and which are capable of serving useful purposes, and that it is perhaps best to look at them from the point of view of the practical user, rather than from that of someone trying to simulate the full range of human cognition.

Report M.S. 10.933, 1979.

PHLIQA1 is an experimental computer program which answers English questions about a limited sub­ject domain. The design of the system is based on the distinction between different, successively "deeper" levels of semantic analysis. At each of these levels, a different formal language is used to express the mean­ing of a question. The languages have the same syn­tactic and semantic structure, but they differ in the constants they contain. The most important represent­ation languages are: (1) the English-oriented Formal Language, with constants that correspond to the "content words" of English, (2) the World Model Language, with constants that represent the primitive concepts that characterize the subject domain of the system, and (3) the Data Base Language, with con­stants that represent the "record types", "attributes" and "sets" that constitute the data base. The system consists of a series of modules which translate a ques­tion from one level to the next lower level, until it finally results in an answer. The paper describes the overall structure of the system, the formal languages used for semantic representation, and the translation steps between the different levels. It also shows how the multilevel semantics set-up makes it possible to treat ambiguous and unanswerable questions in a care­ful way.

An Anatomy of Graceful Interaction in Spoken and Written Man-Machine Communication Phil Hayes and Raj Reddy Department of Computer Science Carnegie-Mellon University Schenley Park

Technical Report CMU-CS-79-144, Sept. 1979.

There have recently been a number of attempts to provide natural and flexible interfaces to computer systems through the medium of natural language. While such interfaces typically perform well in re­sponse to straightforward requests and questions with­in their domain of discourse, they often fail to interact gracefully with their users in less predictable circum­stances. Most current systems cannot, for instance: respond reasonably to input not conforming to a rigid grammar; ask for and understand clarification if their user's input is unclear; offer clarification of their own output if the user asks for it; or interact to resolve any ambiguities that may arise when the user attempts to describe things to the system.

We believe that graceful interaction in these and the many other contingencies that can arise in human conversation is essential if interfaces are ever to ap­pear cooperative and helpful, and hence be suitable for the casual or naive user, and more habitable for the experienced user. In this paper, we attempt to circum­scribe graceful interaction as a field for study, and identify the problems involved in achieving it.

To this end we decompose graceful interaction into a number of relatively independent skills: skills in­volved in parsing elliptical, fragmented, and otherwise ungrammatical input; in ensuring robust communica­tion; in explaining abilities and limitations, actions and the motives behind them; in keeping track of the focus of attention of a dialogue; in identifying things from descriptions, even if ambiguous or unsatisfiable; and in describing things in terms appropriate for the context. We claim these skills are necessary for any type of graceful interaction and sufficient for graceful interac­tion in a certain large class of application domains. None of these components is individually much be­yond the current state of the art, and we outline the architecture of a system that integrates them all. Thus, we propose graceful interaction as an idea of great practical utility whose time has come and which is ripe for implementation. We are currently imple­menting a gracefully interacting system along the lines presented; the system will initially deal with typed input, but is eventually intended to accept speech.

Department of Computer Science Carnegie-Mellon University Schenley Park

Technical Report CMU-CS-79-130, June 1979.

A novel implementation of the basic mechanisms of a semantic network is presented. This constitutes a merging, in terms of the underlying language architec­ture, of a powerful problem-solving mechanism, production-rule systems, with a proven representation formalism. Details are presented on the most basic aspects of the network, namely on representing nodes and on mechanisms for their access. Commands for definition, modification, and search-based displays of network information are discussed. The relations of the network are divided into six groups: taxonomic, structural, functional, descriptive, means-ends, and physical. The "further specification" relation is put forward as an improvement over concepts such as "ISA", superset, instantiation, individuation, and the type-token distinction. The importance of uniformly representing methods and network as rules and the importance of distinguishing temporary from perma­nent states are discussed. Since the system is rule-based, it includes a simple but powerful augmentation capability, embodied in a language for expressing me­thods. Though evidence is not provided here for ad­vanced semantic net capabilities, there is sketched a production system position on a number of relevant issues for those capabilities. The domain of applica­tion is the symbolic description and manipulation of computer structures at the PMS (processor-memory-switch) level. The system will ultimately be used for computer-aided design activities.

Computer Models of Human Personality Traits Jaime G. Carbonell

Technical Report CMU-CS-79-154, Nov. 1979.

A goal-based analysis of human personality traits is presented with the objective of developing a compre­hensive simulation model. It is shown that under­standing trait attributions is an integral part of story understanding and therefore much of natural language processing. The model of personality traits is derived from the goal trees of the POLITICS system, the no­tion of social prototypes, and planning/counterplan-ning strategies.

Computer Science Department General Motors Research Labs. Warren, Michigan 48090

University of Michigan Ph.D. thesis, 1978.

A document retrieval system that has knowledge of the subject matter of its data base is used to explore the relationship between memory structures and infor­mation retrieval processes. Six semantic relations are used to define a semantic network model of legal knowledge in the area of Negotiable Instruments Law. These are: set/member relations, superclass/subclass relations, constituent relations, property relations, role relations, and event-condition relations. A formal language is developed that allows the user to create complex relational expressions describing situations of interest to lawyers. A data base of 400 cases and statutes is defined using this language, called the Situ­ation Description Language (SDL), to describe the situations that are discussed in each document.

The retrieval system accepts queries in the Situa­tion Description Language, and uses its knowledge of legal relationships to determine when a SDL expres­sion that has been entered as part of a document de­scriptor satisfies another SDL expression that has been entered as a query. Some logical and semantic rela­tionships between sets of SDL expressions are ex­plored in the context of different types of queries, and the realization of these relationships is demonstrated. The system is implemented in LISP on the Michigan Terminal System.

Technical Report, 1979.

The Legal Research System is a knowledge-based computer system for legal information retrieval. The system uses a semantic model of legal knowledge to understand and interpret user queries, extending them to include terms that are implied, but not mentioned by the user. This paper analyzes some of the reasons why subject area knowledge is needed in an informa­tion retrieval system, describes how legal knowledge is encoded in the Legal Research System, and gives ex­amples of the kind of inference the system can per­form.

Research Report 159, Jan. 1979.

The problem of how to organize particular experi­ences in a long-term memory has been largely neglect­ed in Natural Language Understanding research. This paper addresses that problem. It shows why good memory organization is necessary for doing intelligent tasks such as story understanding and conversation. A computer memory organization modeled after human memory is proposed, as well as strategies for accessing the computer memory (again based on human informa­tion retrieval). This proposal can also be viewed as a new approach to the organization of intelligent data­bases. The CYRUS system, a computer memory mod­el which implements this theory of memory organiza­tion, is described.

Research Report 168, Oct. 1979.

This paper reports on some recent developments in natural language analysis. We address such issues as the role of syntax in a semantics-oriented analyzer, achieving a flexible balance of top-down and bottom-up processing, and the role of short term memory. Our results have led to improved algorithms capable of analyzing the kinds of multi-clause inputs found in most text.

Graph Grammars and Their Application to Computer Science and Biology, edited by V. Claus, H. Ehrig, & G. Rozenberg, Lecture Notes in Computer Science, No. 73, Springer Verlag, Berlin, 1979, 426-439.

Conceptual graphs formalize the semantic networks used to represent meaning in artificial intelligence and computational linguistics. This paper presents mecha­nisms for defining new types of concepts, conceptual relations, and composite entities having other entities as parts. Type expansion allows graphs containing high-level concepts to be expanded into graphs containing only low-level primitives; alternatively, type contraction allows graphs containing many low-level primitives to be contracted into graphs with a smaller number, but higher level, of concepts and relations. Inferences can then be performed on any level that is appropriate: the high-level concepts in the contracted form or the low-level primitives in the expanded form.

Computer and Communication Sciences 2076 Frieze Building 105 South State Street University of Michigan Ann Arbor, Michigan 48104

Computer Studies in Formal Ling. N-13, May 1979.

Montague (1973) translates English into an inten­sional logic which is an extension of the typed lambda-calculus. In this paper we consider the ways in which Montague's Meaning Postulates can be used to justify the introduction of extensional forms of translations. We define a function XTOU, which replaces quantifi­cation over individual concepts by quantification over entities, and a function FIXEXT, which then intro­duces extensional constants in place of the original intensional forms of lexical items. These obtain the forms of translations displayed in examples in PTQ. We prove that the two functions preserve logical equi­valence in all models satisfying the Meaning Postu­lates.

Using Slots and Modifiers in Logic Grammars for Natural Language Michael C. McCord

Department of Computer Science 915 Patterson Office Tower University of Kentucky Lexington, Kentucky 40506

Technical Report No. 69-80, April 1980.

In this paper, ideas are presented for the expression of natural language grammars in clausal logic, follow­ing the work of Colmerauer, Kowalski, Pereira, and Warren. A uniform format for syntactic structures is proposed, in which every syntactic item consists of a central predication, a cluster of modifiers, a list of fea­tures, and a determiner. The modifiers of a syntactic item are again syntactic items (of the same format), and a modifier's determiner shows its function in the semantic structure. In the rules for syntax, the notions of slots and slot-filling play a central role, following previous work by the author. The Appendix contains an example grammar and samples of parses and se­mantic interpretations into logical form. The system is implemented in Prolog.

Technical Report CS-1979-9, July 1979.

An experiment is described which gives data related to the usefulness and efficiency of English as a pro­gramming language. The experiment was performed with the NLC system described herein and used twenty-three paid volunteers from a first course in programming. The subjects were asked to solve two problems, one on the experimental system and one using the PL/C language studied in their course. The subjects typed a total of 1581 English sentences, 81 percent of which were processed correctly. The re­maining 19 percent were rejected because of question­able user syntax or system inadequacies which are discussed.

None of the standard concerns about natural lan­guage programming related to vagueness, ambiguity, verbosity, or correctness was a significant problem, although minor difficulties did arise occasionally. The time required to solve the problems using English was much less than for PL/C. However, the natural lan­guage dialogues do not presently produce programs as general as their PL/C counterparts.

Using Pragmatic Knowledge for Natural Language Understanding: One Realization on Cooking Recipes M.O. Cordier

Laboratoire de Recherche en Informatique Bâtiment 490 Université de Paris - Sud F-91405 Orsay, FRANCE

Research Report No. 48, 1979.

A program understanding a natural language re­quires a large pragmatic knowledge on the domain (in addition to semantic and syntactic ones). How to transit it? How to use it? That is the central problem of our work. The program which we have written is oriented to robot control. It analyses a text written in a natural language, infers all the necessary informa­tion, and outputs the elementary actions to be execut­ed by the robot. The context used for the evaluation is that of cooking recipes. This system is designed to be highly adaptable to a new context. Therefore the large amount of necessary knowledge is entirely pro­vided as data, via an appropriate easily usable lan­guage.

Research Report No. 54, 1979.

The attempts to model the process of natural lan­guage comprehension with a computer have shed some light on the very nature of natural language itself. We discuss here some issues concerning language process­ing and show that any computational choice entails a commitment about what natural language really is. We conclude that what makes a language to be natural is its ability to be interpreted at variable depths.

Department of Artificial Intelligence University of Edinburgh Forrest Hill

DAI Research Paper No. 123, 1979.

Artificial Intelligence, as well as neurophysiological studies of recent decades, suggests a unitary view of language, in which natural language constitutes only part of the total language of the organism; the latter is a single but complex structure containing also the symbol-systems and their transformations responsible for mental processes. Following Sloman, the dogma that communication is the main function of language is opposed, and implications of the unitary thesis in re­spect to a number of linguistic and psychological ques­tions are briefly discussed.

Research Report 12, May 1979.

Some pragmatic problems of man/computer tactile display dialogues are discussed in the light of experi­ence from interactive computer systems used at the Helsinki University of Technology. Various dialogue support features, techniques and concepts are de­scribed and assessed from human engineering and system design point of view.

Universite de Paris IV Dissertation, 1978.

The aims of this study are first to provide epistemo-logical foundations for a scientific study of descriptive or propositional semantics, independent of particular theories of syntax, and then to develop certain aspects of semantic theory to show that for a semantic theory to be adequate, it must integrate the propositional meanings of successive sentences together into larger units. Such theories are called integrative, and the last major aim is to present one such integrative semantic theory, the C-net theory developed by the author and J-P. Paillet, in a semi-formal fashion, and to develop some of its implications about the lexicon. This work shows that integrative semantics is both necessary and possible, and it adds some detail to the structure of an integrative semantic theory.

Clues to Vocabularic Structure: Comparative Suffixal Productivity in the Five Books of Rabelais Heath Tuttle

North Carolina Educational Computing Service Research Triangle Park, North Carolina 27709

Lingua e Stile, Anno XIV, n. 1, March 1979.

Quite typically in word-formation studies, and more particularly in studies of suffixal derivation, productiv­ity, i.e. the extent to which a suffix is or is not used to form new derivatives, is described in qualitative rather than quantitative terms. By using a computer, one can group words together by suffix facilely, and then date the derivatives using historical-etymological dictionar­ies. So for a given time span all new derivative prod­uctions can be obtained and counted, subject, of course, to their datability in such dictionaries.

The some 12,000 words in the Marty-Laveaux Glossaire were punched onto computer cards. A com­puter program was written to reverse alphabetize the words. The reverse dictionary thus produced con­tained the words spelled in their normal order, but alphabetized from the right instead of from the left. The result was that the words were grouped together by suffix rather than by'prefix (as in a normal diction­ary). This grouping allowed all the words in Rabelais' vocabulary with any particular suffix to be examined. Given a list of the suffixes active in the sixteenth cen­tury, the final, major task was to isolate those words with these suffixes which were actually derived in the sixteenth century, the period investigated; they had to have been produced by the joining of a suffix to a root in the sixteenth century, and not be merely a sixteenth-century spelling of an already existing French, Latin, or Greek word. To determine whether or not a word was of sixteenth century genesis, stand­ard dictionaries were consulted. Graphs were then generated for the some seventy suffixes considered, showing the number of sixteenth-century derivatives in each of the five books.

Department of Computer Science North Carolina State University P.O. Box 5972

SIGLASH Newsletter 13, 1 (March 1980), 2-20.

This paper summarizes a project to develop a com­puterized dictionary of American Sign Language. The topics discussed are: project scope and development, the resulting system of programs, applications in sign language studies, and problems for future develop­ment. Appendices describe details of the notation used for sign descriptions, show an example dictionary, and tell how some of the programs are used.