Showing papers on "Natural language understanding published in 1994"

Toward systems that understand spoken language

Abstract: MIT's Voyager system is an attempt to explore issues related to a fully interactive spoken-language system and natural language understanding. The system helps users get from one location to another within a specific geographical area, and can provide information about certain objects in the area. The current version of Voyager focuses on the city of Cambridge, Massachusetts, between MIT and Harvard University. Voyager's domain knowledge (or backend) is an enhanced version of an existing direction assistance program (J.R. Davis and T.F. Trobaugh, 1987). The map database includes the locations of various classes of objects (streets, buildings, rivers) and their properties (address, phone number, etc.). To retrieve information, the Summit speech recognition system converts the user's speech signal into a set of word hypotheses, the Tina natural language system interacts with Summit to obtain a word string and a linguistic interpretation of the utterance, and an interface between the two subsystems converts Tina's semantic representation into the appropriate function calls to the back-end. Voyager then responds with a map, highlighting the objects of interest, plus an textual and spoken answer. The current implementation has a vocabulary of about 350 words and can deal with various types of queries, such as the location of objects, simple properties of objects, how to get from one place to another, and the distance and travel time between objects. >

GEMINI: A Natural Language System for Spoken-Language Understanding

Abstract: Gemini is a natural language understanding system developed for spoken language applications. The paper describes the architecture of Gemini, paying particular attention to resolving the tension between robustness and overgeneration. Gemini features a broad-coverage unification-based grammar of English, fully interleaved syntactic and semantic processing in an all-paths, bottom-up parser, and an utterance-level parser to find interpretations of sentences that might not be analyzable as complete sentences. Gemini also includes novel components for recognizing and correcting grammatical disfluencies, and for doing parse preferences. This paper presents a component-by-component view of Gemini, providing detailed relevant measurements of size, efficiency, and performance.

Hidden understanding models of natural language

Abstract: We describe and evaluate hidden understanding models, a statistical learning approach to natural language understanding. Given a string of words, hidden understanding models determine the most likely meaning for the string. We discuss 1) the problem of representing meaning in this framework, 2) the structure of the statistical model, 3) the process of training the model, and 4) the process of understanding using the model. Finally, we give experimental results, including results on an ARPA evaluation.

Models of natural language understanding

Abstract: This paper surveys some of the fundamental problems in natural language (NL) understanding (syntax, semantics, pragmatics, and discourse) and the current approaches to solving them. Some recent developments in NL processing include increased emphasis on corpus-based rather than exampleor intuition-based work, attempts to measure the coverage and effectiveness of NL systems, dealing with discourse and dialogue phenomena, and attempts to use both analytic and stochastic knowledge. Critical areas for the future include grammars that are appropriate to processing large amounts of real language; automatic (or at least semiautomatic) methods for deriving models of syntax, semantics, and pragmatics; self-adapting systems; and integration with speech processing. Of particular importance are techniques that can be tuned to such requirements as full versus partial understanding and spoken language versus text. Portability (the ease with which one can configure an NL system for a particular application) is one of the largest barriers to application of this technology. Natural language (NL) understanding by computer began in the 1950s as a discipline closely related to linguistics. It has evolved to incorporate aspects of many other disciplines (such as artificial intelligence, computer science, and lexicography). Yet it continues to be the Holy Grail of those who try to make computers deal intelligently with one of the most complex characteristics of human beings: language. Language is so fundamental to humans, and so ubiquituous, that fluent use of it is often considered almost synonymous with intelligence. Given that, it is not surprising that computers have difficulty with natural language. Nonetheless, many people seem to think it should be easy for computers to deal with human language, just because they themselves do so easily. Research in both speech recognition (i.e., literal transcription of spoken words) and language processing (i.e., understanding the meaning of a sequence of words) has been going on for decades. But quite recently, speech recognition started to make the transition from laboratory to widespread successful use in a large number of different kinds of systems. What is responsible for this technology transition? Two key features that have allowed the development of successful speech recognition systems are (i) a simple general description of the speech recognition problem (which results in a simple general way to measure the performance of recognizers) and (ii) a simple general way to automatically train a recognizer on a new vocabulary or corpus. Together, these features helped to open the floodgates to the successful, widespread application of speech recognition technology. Many of the papers in this volume, particularly those by The publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. ?1734 solely to indicate this fact. Makhoul and Schwartz (1), Jelinek (2), Levinson (3), Oberteuffer (4), Weinstein (5), and Wilpon (6) attest to this fact. But it is important to distinguish "language understanding" from "recognizing speech," so it is natural to ask, why the same path has not been followed in natural language understanding. In natural language processing (NLP), as we shall see, there is no easy way to define the problem being solved (which results in difficulty evaluating the performance of NL systems), and there is currently no general way for NL systems to automatically learn the information they need to deal effectively with new words, new meanings, new grammatical structures, and new domains. Some aspects of language understanding seem tantalizingly similar to problems that have been solved (or at least attacked) in speech recognition, but other aspects seem to emphasize differences that may never allow the same solutions to be used for both problems. This paper briefly touches on some of the history of NLP, the types of NLP and their applications, current problem areas and suggested solutions, and areas for future work. A BRIEF HISTORY OF NLP NLP has a long, diverse history. One way of looking at that history is as a sequence of application areas, each of which has been the primary focus of research efforts in the computational linguistics community, and each of which has produced different techniques for language understanding. A number of excellent references are available that survey the field in various ways (7-11). In the 1950s, machine translation was the first area to receive considerable attention, only to be abandoned when it was discovered that, although it was easy to get computers to map one word string to another, the problem of translating between one natural language and another was much too complex to be expressible as such a mapping. In the 1960s the focus turned to question answering. To "understand" and respond to typed questions, most NL systems used a strongly knowledge-based approach, attempting to encode knowledge for use by a system capable of producing an in-depth analysis of the input question. That arnalysis would then be used to retrieve the answer to the question from a database. In the 1970s interest broadened from database interfaces to other kinds of application systems, but the focus was still on the kinds of natural language that would be produced by a person interacting with a computer system-typed queries or commands issued one at a time by the person, each of which needed to be understood completely in order to produce the correct response. That is, virtually every word in the input had some effect on the meaning that the system produced. This tended to result in systems that, for each sentence they were given, either succeeded perfectly or failed completely. The 1980s saw the first commercialization of research that was done in the previous two decades: natural language

Feature-based inheritance networks for computational lexicons

Abstract: The virtues of viewing the lexicon as an inheritance network are its succinctness and its tendency to highlight significant clusters of linguistic properties. From its succinctness follow two practical advantages, namely its ease of maintenance and modification. In this paper we present a feature-based foundation for lexical inheritance. We argue that the feature-based foundation is both more economical and expressively more powerful than non-feature-based systems. It is more economical because it employs only mechanisms already assumed to be present elsewhere in the grammar (viz., in the feature system), and it is more expressive because feature systems are more expressive than other mechanisms used in expressing lexical inheritance (cf. DATR). The lexicon furthermore allows the use of default unification, based on the ideas of default unification, defined by Bouma. These claims are buttressed in sections sketching the opportunities for lexical description in feature-based lexicons in two central lexical topics, inflection and derivation. Briefly, we argue that the central notion of paradigm may be defined in feature structures, and that it may be more satisfactorily (in fact, immediately) linked to the syntactic information in this fashion. Our discussion of derivation is more programmatic; but here, too, we argue that feature structures of a suitably rich sort provide a foundation for the definition of lexical rules. We illustrate theoretical claims in application to German lexis. This work is currently under implementation in a natural language understanding effort (DISCO) at the German Artiffical Intelligence Center (Deutsches Forschungszentrum fur Kunstliche Intelligenz).

From English to formal specifications

Abstract: Specifications provide the foundation upon which a system can be formally developed. If a specification is wrong, then no matter what method of design is used, or what quality assurance procedures are in place, they will not result in a system that meets the requirements. The specification of a system involves people of different profiles who favour different representations. At the beginning natural language is used because the specification document acts as a contract between the user and the developers. Most of the time, the only representation that users understand and agree on is natural language. At the other end, developers find natural language specifications ambiguous and incomplete and may therefore prefer formal specifications. The transition from informal specifications to formal ones is an error prone and time consuming process. This transition must be supported to ensure that the formal specifications are consistent with the informal ones. In this research we propose an interactive approach for producing formal specifications from English specifications. The approach uses research in the area of natural language understanding to analyse English specifications in order to detect ambiguities. The method used for analysing natural language text is based on McCord’s approach. This method consists of translating natural language sentences into a logical form language representation. This helps to identify ambiguities present in natural language specifications and to identify the entities and relationships. These entities and relationships are used as a basis for producing VDM data types. We also investigate the production of data type invariants for restricted sentences and the production of some common specifications. We test our approach by implementing it in Prolog-2 and apply it to an independent case study.

Concurrent, object-oriented natural language parsing: the ParseTalk model

Abstract: The ParseTalk model of concurrent, object-oriented natural language parsing is introduced. It builds upon the complete lexical distribution of grammatical knowledge and incorporates inheritance mechanisms in order to express generalizations over sets of lexical items. The grammar model integrates declarative well-formedness criteria constraining linguistic relations between heads and modifiers, and procedural specifications of the communication protocol for establishing these relations. The parser's computation model relies upon the actor paradigm, with concurrency entering through asynchronous message passing. We consider various extensions of the basic actor model as required for distributed natural language understanding and elaborate on the semantics of the actor computation model in terms of event type networks (a graph representation for actor grammar specifications) and event networks (graphs which represent the actor parser's behavior). Besides theoretical claims, we present an interactive grammar/parser workbench, a graphical development environment with various types of browsers, tracers, inspectors and debuggers, that has been adapted to the requirements of large-scale grammar engineering in a distributed, object-oriented specification and programming framework.

A natural language understanding system combining syntactic and semantic techniques.

Abstract: A large proportion of the medical record currently available in computerized medical information systems is in the form of free text reports. While the accessibility of this source of data is improved through inclusion in the computerized record, it remains unavailable for automated decision support, medical research, and management of medical delivery systems. Natural language understanding systems (NLUS) designed to encode free text reports represent one approach to making this information available for these uses. Below we describe an experimental NLUS designed to parse the reports of chest radiographs and store the clinical data extracted in a medical data base.

A Functional Theory of Creative Reading

Abstract: : Reading is an area of human cognition which has been studied for decades by psychologists, education researchers, and artificial intelligence researchers. Yet, there still does not exist a theory which accurately describes the complete process. We believe that these past attempts fell short due to an incomplete understanding of the overall task of reading; namely, the complete set of mental tasks a reasoner must perform to read and the mechanisms that carry out these tasks. We present a functional theory of the reading process and argue that it represents a coverage of the task. The theory combines experimental results from psychology, artificial intelligence, education, and linguistics, along with the insights we have gained from our own research. This greater understanding of the mental tasks necessary for reading will enable new natural language understanding systems to be more flexible and more capable than earlier ones. Furthermore, we argue that creativity is a necessary component of the reading process and must be considered in any theory or system attempting to describe it. We present a functional theory of creative reading and a novel knowledge organization scheme that supports the creativity mechanisms. The reading theory is currently being implemented in the ISAAC "Integrated Story Analysis And Creativity" system, a computer system which reads science fiction stories.

A uniform computational model for natural language parsing and generation

Abstract: In the area of natural language processing in recent years, there has been a strong tendency towards reversible natural language grammars, i.e., the use of one and the same grammar for grammatical analysis (parsing) and grammatical synthesis (generation) in a natural language system. The idea of representing grammatical knowledge only once and of using it for performing both tasks seems to be quite plausible, and there are many arguments based on practical and psychological considerations for adopting such a view (in section 2.1 we discuss the most important arguments in more detail). Nevertheless, in almost all large natural language systems in which parsing and generation are considered in similar depth, different algorithms are used - even when the same grammar is used. At present, the first attempts are being made at uniform architectures which are based on the paradigm of natural language processing as deduction (they are described and discussed in section 2.3 in detail). Here, grammatical processing is performed by means of the same underlying deduction mechanism, which can be parameterized for the specific tasks at hand. Natural language processing based on a uniform deduction process has a formal elegance and results in more compact systems. There is one further advantage that is of both theoretical and practical relevance: a uniform architecture offers the possibility of viewing parsing and generation as strongly interleaved tasks. Interleaving parsing and generation is important if we assume that natural language understanding and production are not performed in an isolated way but rather can work together to obtain a flexible use of language. In particular this means a.) the use of one mode of operation for monitoring the other and b.) the use of structures resulting from one direction directly in the other. For example, during generation integrated parsing can be used to monitor the generation process and to cause some kind of revision, e.g., to reduce the risk of misunderstandings. Research on monitoring and revision strategies is a very active area in cognitive science; however, currently there exists no algorithmic model of such a behaviour. A uniform architecture can be an important step in that direction. Unfortunately, the currently proposed uniform architectures are very inefficient and it is yet unclear how an efficiency-oriented uniform model could be achieved. An obvious problem is that in each direction different input structures are involved - a string for parsing and a semantic expression for generation - which causes a different traversal of the search space defined by the grammar. Even if this problem were solved, it is not that obvious how a uniform model could re-use partial results computed in one direction efficiently in the other direction for obtaining a practical interleaved approach to parsing and generation. Liegt nicht vor.

A temporal connectionist approach to natural language

Abstract: Time is an essential dimension of human natural language understanding but most of the symbolic models applied to linguistic data do not account for temporal structure. In contrast, the models from the connectionist paradigm have a natural ability to perform dynamic processing.After a presentation of some networks with a concern for time, we describe the model for Coincidence Detection which can be thought of as encoding spatio-temporal regularities of the input data. The architecture of the model is inspired from neurobiological studies of the cerebral cortex. It performs a dynamic interpretation of nominal composition and is analyzed in terms of micro-symbolic co-occurrences. The relevance of the Coincidence Detection machinery in language processing shows the significance of time in computational language understanding.

Integrated connectionist models: building AI systems on subsymbolic foundations

Abstract: Symbolic artificial intelligence is motivated by the hypothesis that symbol manipulation is both necessary and sufficient for intelligence. In symbolic systems, knowledge is encoded in terms of explicit symbolic structures, and inferences are based on handcrafted rules that sequentially manipulate these structures. Such systems have been quite successful, for example, in modeling in-depth natural language processing, episodic memory, and symbolic problem solving. However, much of the inferencing for everyday natural language understanding appears to take place immediately, without conscious control, apparently based on associations with past experience. This type of reasoning is difficult to model in the symbolic framework. In contrast, subsymbolic (distributed connectionist) networks represent knowledge in terms of correlations, coded in the weights of the network. For a given input, the network computes the most likely answer given its past experience. A number of human-like information processing properties such as learning from examples, context sensitivity, generalization, robustness of behavior, and intuitive reasoning emerge automatically in subsymbolic systems. The major motivation for subsymbolic AI, therefore, is to give a better account for cognitive phenomena that are statistical, or intuitive, in nature. >

Building embedded conceptual parsers

Abstract: This dissertation describes how to build conceptual parsers (that is, natural language understanding systems built on semantic and pragmatic principles) that are embedded into application programs. A new architecture for building such parsers, indexed concept parsing, is described. Indexed concept parsing is a case-based reasoning approach to parsing, in which underlying target concepts (that is, those conceptual representations of the application program identified as important to recognize) are associated with sets of index concepts. Each index concept is associated with sets of phrasal patterns. At run time, the parser looks for phrasal patterns in input text, and the index concepts recognized thereby are used to appraise the best matching target concepts. The architecture defines a range of parsers, in which the complexity of the index concept representations can vary according to the needs of the application program: index concepts can be key words, synonym sets, representations in an abstraction hierarchy, or representations in a partonomic hierarchy. Indexed concept parsing was developed to build parsers for Casper, an interactive learning environment designed to teach customer service representatives how to solve customer problems, and TransAsk, a multimedia system for transportation planners. Indexed concept parsing proved robust (for example, the Casper parser had an accuracy rate ranging from 83-96%), yet required minimal knowledge representation. A methodology for building an indexed concept parser is given, and evaluation metrics are described. Another parser, based on Direct Memory Access Parsing (DMAP), and developed for the Creanimate biology tutor, is also described, as well as a DMAP parser for Casper. Indexed concept parsing and DMAP are contrasted as architectures for building embedded conceptual parsers.

NL understanding with a grammar of constructions

Abstract: We present an approach to natural language understanding based on a computable grammar of constructions. A construction consists of a set of features of form and a description of meaning in a context. A grammar is a set of constructions. This kind of grammar is the key element of MINCAL, an implemented natural language speech-enabled interface to an on-line calendar system. The architecture has two key aspects: (a) the use of constructions, integrating descriptions of form, meaning and context into one whole; and (b) the separation of domain knowledge (about calendars) from application knowledge (about the particular on-line calendar).

Reasoning with background knowledge–a three-level theory

Abstract: We address the problem of reasoning in cases when knowledge bases containing background knowledge are understood not as sets of formulas (rules and facts) but as collections of partially ordered theories. In our system, the usual, two-part logical structures, consisting of a metalevel and an object level, are augmented by a third level–a referential level. The referential level is a partially ordered collection of theories; it encodes background knowledge. As usual, current situations are described on the object level, and the metalevel is a place for rules that can eliminate some of the models permitted by the object level and the referential level. As a logic of reasoning the system generalizes the standard model of a rational agent: deducing actions and deriving new information about the world from a logical theory–its knowledge base. It is a natural logical system in which priorities on the possible readings of predicates, not special rules of inference, are the main source of nonmonotonicity. We introduce a theory forming operator PT(x) to exploit the priorities, and we investigate its basic logical properties. Then we show how such a system can be augmented by metarules. Although this paper concentrates on basic logical properties of the new theory, this formalism has already been applied to model a number of natural language phenomena such as the notion of text coherence, Gricean maxims, vagueness, and a few others. The paper also briefly compares it with the model of background knowledge of CYC, as proposed by Lenat and Guha.

Situating natural language understanding within experience-based design

Abstract: Building useful systems with an ability to understand "real" natural language input has long been an elusive goal for Artificial Intelligence. Well-known problems such as ambiguity, indirectness, and incompleteness of natural language inputs have thwarted efforts to build natural language interfaces to intelligent systems. In this article, we report on our work on a model of understanding natural language design specifications of physical devices such as simple electrical circuits. Our system, called KA, solves the classical problems of ambiguity, incompleteness and indirectness by exploiting the knowledge and problem-solving processes in the situation of designing simple physical devices. In addition, KA acquires its knowledge structures (apart from a basic ontology of devices) from the results of its problem-solving processes. Thus, KA can be bootstrapped to understand design specifications and user feedback about new devices using the knowledge structures it acquired from similar devices designed previously.In this paper, we report on three investigations in the KA project. Our first investigation demonstrates that KA can resolve ambiguities in design specifications as well as infer unarticulated requirements using the ontology, the knowledge structures, and the problem-solving processes provided by its design situation. The second investigation shows that KA's problem-solving capabilities help ascertain the relevance of indirect design specifications, and identify unspecified relations between detailed requirements. The third investigation demonstrates the extensibility of KA's theory of natural language understanding by showing that KA can interpret user feedback as well as design requirements. Our results demonstrate that situating language understanding in problem solving, such as device design in KA, provides effective solutions to unresolved problems in natural language processing.

Integrating Natural Language Understanding with Document Structure Analysis

Abstract: Document understanding, the interpretation of a document from its image form, is a technology area which benefits greatly from the integration of natural language processing with image processing. We have developed a prototype of an Intelligent Document Understanding System (IDUS) which employs several technologies: image processing, optical character recognition, document structure analysis and text understanding in a cooperative fashion. This paper discusses those areas of research during development of IDUS where we have found the most benefit from the integration of natural language processing and image processing: document structure analysis, optical character recognition (OCR) correction, and text analysis. We also discuss two applications which are supported by IDUS: text retrieval and automatic generation of hypertext links

The roles of language processing in a spoken language interface

Abstract: This paper provides an overview of the colloquium's discussion session on natural language understanding, which followed presentations by M. Bates [Bates, M. (1995) Proc. Natl. Acad. Sci. USA 92, 9977-9982] and R. C. Moore [Moore, R. C. (1995) Proc. Natl. Acad. Sci. USA 92, 9983-9988]. The paper reviews the dual role of language processing in providing understanding of the spoken input and an additional source of constraint in the recognition process. To date, language processing has successfully provided understanding but has provided only limited (and computationally expensive) constraint. As a result, most current systems use a loosely coupled, unidirectional interface, such as N-best or a word network, with natural language constraints as a postprocess, to filter or resort the recognizer output. However, the level of discourse context provides significant constraint on what people can talk about and how things can be referred to; when the system becomes an active participant, it can influence this order. But sources of discourse constraint have not been extensively explored, in part because these effects can only be seen by studying systems in the context of their use in interactive problem solving. This paper argues that we need to study interactive systems to understand what kinds of applications are appropriate for the current state of technology and how the technology can move from the laboratory toward real applications. This paper provides an overview of the natural language understanding session at the Colloquium on HumanMachine Communication by Voice held by the National Academy of Sciences (NAS). The aim of the paper is to review the role that language understanding plays in spoken language systems and to summarize the discussion that followed the two presentations by Bates (1) and Moore (2). A number of questions were raised during the discussion, including whether a single system could provide both understanding and constraint, what the future role of discourse should be, how to evaluate performance on interactive systems, and whether we are moving in the right direction toward realizing the goal of interactive human-machine communication. Background: The ARPA Spoken Language Program Much of the research discussed at the natural language understanding session was done in connection with the Advanced Research Projects Agency's (ARPA) Spoken Language Systems program. This program, which started in 1989, brought together speech and language technologies to provide speech interfaces for interactive problem solving. The goal was to permit the user to speak to the system, which would respond appropriately, providing (intelligent) assisThe publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. ?1734 solely to indicate this fact. tance. This kind of interaction requires the system to have both input and output capabilities, that is, for speech, both recognition and synthesis, and for language, both understanding and generation. In addition, the system must be able to understand user input in context and carry on a coherent conversation. We still know relatively little about this complex process of interaction, although we have made significant progress in one aspect, namely spoken language understanding.a In the ARPA Spoken Language Systems program, multiple contractors are encouraged to develop independent approaches to the core problem of spoken language interaction. To focus the research, common evaluation is used to compare alternate technical approaches within a common task domain. To ensure that test results are comparable across sites, the sites choose a task domain, collect a common corpus of training material, and agree on a set of evaluation metrics. The systems are then evaluated periodically on a set of (previously unseen) test data, using the agreed upon evaluation metrics. The evaluation makes it possible not only to compare the effectiveness of various technical approaches but also to track overall progress in the field. For the Spoken Language Systems program, the Air Travel Information System (ATIS) (3) was chosen as the application domain for the common evaluation. This is a database interface application, where the data were drawn from a nine-city subset of the Official Airline Guide, containing airline, schedule, and ground transportation information.b To support this effort, sites cooperated to collect a training corpus of 14,000 spontaneous utterances (4), and, to date, there have been four formal evaluations in this domain (5-9). At the start of the Spoken Language Systems program in 1989, an accepted metric had evolved for speech recognition, namely word accuracy (10); however, no comparable metric was available for measuring understanding. Over the past 4 years, the research community has developed an understanding metric for database interface tasks, using either speech or typed input (4, 11). To date, there is still no agreed upon metric for the rich multidimensional space of interactive systems, which includes the system's ability to communicate effectively with the user, as well as an ability -to understand what the user is trying to accomplish. The remainder of this paper is divided into four sections: "The Dual Role of Language Processing" discusses the role of language processing in providing both understanding and constraint; "The Role of Discourse" outlines several sources of discourse and conversational constraints that are available at the inter-sentential level; "Evaluation" returns to the issue aSpoken language understanding focuses on understanding user input, as opposed to communicating with the user, which is a bidirectional process that requires synthesis and generation technologies. bThere is now an enlarged 46-city version of the ATIS database; it will be the focus of the next round of evaluation.

A parallel computational model for integrated speech and natural language understanding

Abstract: Presents a parallel approach for integrating speech and natural language understanding. The method emphasizes a hierarchically-structured knowledge base and memory-based parsing techniques. Processing is carried out by passing multiple markers in parallel through the knowledge base. Speech specific problems such as insertion, deletion, substitution, and word boundary detection have been analyzed and their parallel solutions are provided. Results on the SNAP-1 multiprocessor show an 80% sentence recognition rate for the Air Traffic Control (ATC) domain. Furthermore, speed-up of up to 15-fold is obtained from the parallel platform which provides response times of a few seconds per sentence for the ATC domain. >

Peirce: a tool for abductive inference

Abstract: Abduction, or inference to the best explanation, has been used in artificial intelligence as a framework for solving problems ranging from diagnosis, to test interpretation, to theory formation, to natural language understanding, to perception. Previous research on computational models of abduction has suggested that a single generic strategy may be used to perform abductions in a variety of domains and across very diverse problems. This paper describes a generic strategy for abduction and a tool, Peirce, for constructing abductive problem solving systems. >

A classification of ellipsis based on a corpus of information seeking dialogues

Abstract: The standard classification of ellipsis has determined the way it is handled in natural language understanding (NLU) systems. This work provides a novel classification of ellipsis based on the analysis of ellipsis usage rather than forms in a corpus of information seeking dialogues. The aim is to demonstrate that pragmatic analysis is necessary for the interpretation of ellipsis. The context, in terms of the dialogue participants' belief states, determines interpretation and in turn the interpretation of subsequent utterances. The dialogues produced in a NLU system using this classification are presented.

Disabled person's control, communication and entertainment aid: an investigation of the feasibility of using speech control and natural language understanding to control a manipulator and a software application and development environment

Abstract: The work reported in this thesis is a feasibility study of the possibilities and practical problems of applying speech control and natural language understanding techniques to the use of a computer by a physically disabled person. Solutions are proposed for the overcoming of some of the difficulties and limitations of the available equipment, and guidance given for the application of such systems to real tasks. The use of voice control with a low cost industrial robot is described. The limitations introduced by the speech control hardware, such as restricted vocabulary size and artificial manner of speaking are partially overcome by software extensions to the operating system and the application of natural language understanding techniques. The application of voice control and audio response to common application packages and a programming environment are explored. Tools are developed to aid the construction of natural language understanding systems. These include an extension to the use of an existing context-free parser generator to enable it to handle context-sensitive grammars, and an efficient parallel parser which is able to find all possible parses of a sentence simultaneously. Machine readable dictionary construction is investigated, incorporating the analysis of complex words in terms of their root forms using affix transformations, and the incorporation of semantic information using a variety of techniques, such as semantic fields, the previously mentioned affix transforms, and object-oriented semantic trees. The software developed for the system is written in Borland Pascal on an IBM compatible P C , and is produced in the form of library modules and a toolkit to facilitate its application to any desired task.

Pragmatic linguistic constraint models for large-vocabulary speech processing

Abstract: Current systems for speech recognition suffer from uncertainty: rather than delivering a uniquely-identified word, each input segment is associated with a set of recognition candidates or word-hypotheses. Thus an input sequence of sounds or images leads to, not an unambiguous sequence of words, but a lattice of word-hypotheses. To choose the best candidate from each word-hypothesis set (i.e. to find the best route through the lattice) , linguistic context needs to be taken into account, at several levels: lexis and morphology, parts-of-speech, phrase structure, semantics and pragmatics. We believe that an intuitively simple, naive model will suffice at each level; the sophistication required for full Natural Language Understanding (NLU) (e.g. Alvey Natural Language Toolkit (ANLT)) is inappropriate for real-time language recognition. We describe here models of each linguistic level which are simple but robust and computationally straightforward (hence `pragmatic' in the everyday sense) and which have clear theoretical shortcomings in the eyes of linguistic purists but which nevertheless do the job.

A Relational Goal-Seeking Agent using Conceptual Graphs

Abstract: A goal-seeking, satisficing, rational agent using conceptual graphs is described A particular theoretical stance on the use of case roles has enabled goals, observations, actions and surface-level language parses to be represented in a common conceptual form, permitting fruitful interactions The agent is demonstrated to be capable of choosing from its repertoire only those actions which are likely to satisfy its goals and which are appropriate with respect to its observations of the world A detailed action specification, instantiated with local detail from world observations, is activated and an appropriately parameterised demon is called The ability to use natural language parses to inform action specification is being added The agent is being developed to serve a natural language understanding navigation system capable of obeying human instructions to traverse maps of the physical world

CHAPTER 10 – Syntactic Semantics: Foundations of Computational Natural-Language Understanding

Abstract: Publisher Summary This chapter discusses the way by which it is possible to understand natural language and whether a computer could do so. It presents the argument that although a certain kind of semantic interpretation is needed for understanding natural language, it is a kind that only involves syntactic symbol manipulation of precisely the sort of which computers are capable, so that it is possible, in principle, for computers to understand natural language. The chapter highlights the recent arguments by John R. Searle and by Fred Dretske to the effect that computers cannot understand natural language. A program is like the script of a play. The computer is like the actors, sets, and a process is like an actual production of the play—the play in the process of being performed. The computer must be able to do things like: to convince someone, to imitate a human, that is, it must not merely be a cognitive agent, but also an acting one. In particular, to imitate a human, the computer needs to be able to reason about what another cognitive agent, such as a human, believes.

Knowledge Representation for Lexical Semantics: Is Standard First Order Logic Enough?

Abstract: Natural language understanding applications such as interactive planning and face-to-face translation require extensive inferencing Many of these inferences are based on the meaning of particular open class words Providing a representation that can support such lexically-based inferences is a primary concern of lexical semantics The representation language of first order logic has well-understood semantics and a multitude of inferencing systems have been implemented for it Thus it is a prime candidate to serve as a lexical semantics representation However, we argue that FOL, although a good starting point, needs to be extended before it can efficiently and concisely support all the lexically-based inferences needed

Encoding Conceptual Graphs by Labeling RAAM

Abstract: The meaning of medical texts is not automatically recognized by computers. A representation of this information is strongly recommanded to allow medical texts databases queries. The conceptual graph formalism developed by Sowa [Sow84] is a knowledge representation language initially designed to capture the meaning of natural language. Conceptual graphs have been used in many natural language understanding works [BRS92, VZB+93, Ber91]. In this paper we discuss the possibility to memorize and retrieve natural language sentences and especially medical language sentences given in this kind of formalism with the use of the LRAAM model [Spe93b, Spe93a]. In Section 2 we explain the idea underlying conceptual graphs. In Section 3 we briefly expose the access by content capabilities of the LRAAM and suggest a generalization of the access by content procedures introducing the concept of Generalized Hopfield Network. A discussion on the impact of this generalization on knowledge extraction from a database of conceptual graphs is given in the conclusion.

Pragmatic Semantics by Conceptual Graphs

Abstract: The traditional approach to natural language understanding is to list all polysemous meanings of a word or idiom in the lexicon. For each word a choice is made between one of its lexical entries, and this choice is used to construct the meaning of the sentence. In this paper we will propose an alternative that derives the appropriate meaning by starting with a single more general lexical entry that is expanded to the appropriate polysemous meaning. The semantic details can be provided by the textual context, the background context, and pragmatic knowledge.