Showing papers on "Natural language understanding published in 1993"

Gemini: a natural language system for spoken-language understanding

Abstract: Gemini is a natural language understanding system developed for spoken language applications. This paper describes the details of the system, and includes relevant measurements of size, efficiency, and performance of each of its sub-components in detail.

Gemini: a natural language system for spoken-language understanding

Abstract: The demands on a natural language understanding system used for spoken language differ somewhat from the demands of text processing. For processing spoken language, there is a tension between the system being as robust as necessary, and as constrained as possible. The robust system will a t tempt to find as sensible an interpretation as possible, even in the presence of performance errors by the speaker, or recognition errors by the speech recognizer. In contrast, in order to provide language constraints to a speech recognizer, a system should be able to detect that a recognized string is not a sentence of English, and disprefer that recognition hypothesis from the speech recognizer. If the coupling is to be tight, with parsing and recognition interleaved, then the parser should be able to enforce as many constraints as possible for partial utterances. The approach taken in Gemini is to tightly constrain language recognition to limit overgeneration, but to extend the language analysis to recognize certain characteristic patterns of spoken utterances (but not generally thought of as part of grammar) and to recognize specific types of performance errors by the speaker.

The BBN/HARC spoken language understanding system

Abstract: The design and performance of a complete spoken language understanding system under development at BBN are described. The system, dubbed HARC (Hear And Respond to Continuous speech), successfully integrates state-of-the-art speech recognition and natural language understanding subsystems. The system has been tested extensively on a restricted airline travel information (ATIS) domain with a vocabulary of about 2000 words. HARC is implemented in portable, high-level software that runs in real time on today's workstations to support interactive online human-machine dialogs. No special-purpose hardware is required other than an A/D (analog-to-digital) converter to digitize the speech. The system works well for any native speaker of American English and does not require any enrollment data from the users. Results of formal DARPA tests in Feb. and Nov. 1992 are presented. >

Natural language, knowledge representation, and logical form

Abstract: : Current natural language understanding systems generally maintain a strict division between the parsing processes and the representation that supports general reasoning about the world. This paper examines why these two forms of processing are separated, determines the current advantages and limitations of this approach, and identifies the inherent limitations of the approach. I will point out some fundamental problems with the models as they are defined today and suggest some important directions of research in natural language and knowledge representation. In particular, I will argue that one of the crucial issues facing future natural language systems is the development of knowledge representation formalisms that can effectively handle ambiguity.

Language understanding and subsequential transducer learning

Abstract: The application of the Onward Subsequential Transducer Inference Algorithm (OSTIA) recently introduced by J. Oncina et al. (1993) to (pseudo-) natural language understanding is considered. For this purpose, a task proposed by J.A. Feldman et al. (1990), as a touchstone for comparing the capabilities of language learning systems has been adopted and three increasingly difficult semantic coding schemes have been defined for this task. In all cases the OSTIA was consistently proved able to learn very compact and accurate transducers from relatively small training sets of input-output examples of the task.

Modelling for natural language understanding.

Abstract: Natural Language Understanding (NLU) is a rapidly growing field in medical informatics. Its potential for tomorrow's applications is important. However, it is limited by its ability to ground its components on a solid model of the domain. This opens the way for the emergence of the discipline of medical domain modelling, as part of the vast field of Knowledge Base (KB) engineering. This article aims at describing the current development of a multilingual natural language system, strongly oriented towards the semantics of the domain. Special emphasis is presently given to the task of building a domain model, and to establish direct links with the language platform. The result is a model-driven NLU system. Numerous benefits are expected in the long term.

A bilingual Voyager system.

Abstract: This paper describes our initial efforts at porting the VOYAGER spoken language system to Japanese. In the process we have reorganized the structure of the system so that language dependent information is separated from the core engine as much as possible. For example, this information is encoded in tabular or rule-based form for the natural language understanding and generation components. The internal system manager, discourse and dialogue component, and database are all maintained in language transparent form. Once the generation component was ported, data were collected from 40 native speakers of Japanese using a wizard collection paradigm. A portion of these data was used to train the natural language and segment-based speech recognition components. The system obtained an overall understanding accuracy of 52% on the test data, which is similar to our earlier reported results for English [1].

CARAMEL: A step towards reflection in natural language understanding systems

Abstract: The authors wish to extend to natural language understanding (NLU) systems a paradigm now seen as essential for AI: the use of meta-knowledge and the capability a system may have to observe its own functioning. First, they recall why multi-expert systems seem the best architecture for dealing efficiently with most NL constraints. Then they give general information about reflective reasoning models and propose some extensions to currently admitted ideas in the domain of distributed AI. Lastly, an illustration of these ideas with CARAMEL (in French: Comprehension Automatique de Recits, Apprentissage et Modelisation des Echanges Langagiers), the system developed in the group and able to perform various tasks using NLU.

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. >

Plan recognition strategies for language understanding

Abstract: In recent years the emphasis in natural language understanding research has shifted from studying mechanisms for understanding isolated utterances to developing strategies for interpreting sentences within the context of a discourse or an extended dialogue. A very fruitful approach to this problem has derived from a view of human behavior as goal-directed and understanding as explanation-based. According to this view, people perform actions and communicate to advance their goals, and language understanding therefore involves recognizing and reasoning about the goals and plans of others. This paper explores plan inference in natural language understanding. It presents a core set of ideas on which most models of plan recognition are based and illustrates these by critically analysing three systems in detail. It then discusses issues that have been addressed by various research efforts, explores the major problems that limit the capability of current plan recognition systems, and describes current research directed toward solving some of these problems.

University of Manitoba: description of the NUBA system as used for MUC-5

Abstract: Abduction is the inference to the best explanation. Many tasks in natural language understanding such as word-sense disambiguity [1], local pragmatics [4], metaphor interpretation [3], and plan recognition [5, 8], can be viewed as abduction.

A Semantic Expert Using an Online Standard Dictionary.

Abstract: A system has been developed to find the most likely attachment for prepositional phrases in English sentences in a fairly unrestricted way. The system receives as input a syntactic sentence parse provided by a general-purpose computational grammar called PEG (PLNLP English Grammar) The semantic decision that is necessary to make the right attachments is made (a) by parsing (also with PEG) the natural language definitions of an online standard dictionary, in this case Webster's Seventh New Collegiate Dictionary; (b) by relating words to other words in the dictionary; and (c) by reasoning heuristically about the comparative likelihood of different possible attachments. The basic assumption of this research is that natural language itself is a knowledge representation language that can be conveniently accessed and richly exploited. Techniques such as those presented here offer hope for eliminating the time-consuming and often incomplete hand coding of semantic information that has been conventional in natural language understanding systems.

Massively-Parallel Inferencing for Natural Language Understanding and Memory Retrieval in Structured Spreading-Activation Networks

Abstract: One of the most difficult parts of the natural language understanding process is forming a semantic interpretation of the text. A reader must often make multiple inferences to understand the motives of actors and to causally connect actions that are unrelated on the basis of surface semantics alone. The inference process is complicated by the fact that text is often ambiguous both lexically and pragmatically, and that new context often forces a reinterpretation of the input’s meaning. This language understanding process itself does not exist in a vacuum -as people read text or hold conversations, they are often reminded of analogous stories or episodes. The types of memories that are triggered are influenced by context from the inferences and disambiguations of the understanding process. A full model of the language understanding and memory retrieval processes must take into account the interaction of the two and how they effect each other.

Module-level testing for natural language understanding

Abstract: We have examined a module-level evaluation methodology for natural language understanding, illustrating insights gained during implementation. It is possible to produce useful quantitative and qualitative results using such a methodology, provided certain pitfalls are avoided.

Natural language understanding based on background knowledge

Abstract: Natural language understanding can be divided into five hierarchical levels, each level related to other levels. Background knowledge consists of all level's units and their usage in the real language environment. Co-occurence information of any two units of the same level can be provided by background knowledge. This paper discusses the usage of background knowledge in automatic Chinese word segmentation, and text understanding based on semantic relationships. >

Semantic ambiguity resolution by parallel distributed, associative inference and contradiction detection

Abstract: Proposes a PDAI&CD architecture aimed at constructing natural inference systems. The kernel consists of a mutually associative neural network which processes numerical patterns and of a logical system processing symbols. The associative part calls on context-dependent free-association of concepts based on the relations of concepts acquired from a dynamically changing outer world. In the logical part of the architecture, the results obtained by the neural network are checked, and emerging contradictions create feedback to the associative network and thus find a final optimum solution. WAVE, an implemented system, is introduced, and the authors show its application to the problem of ambiguity resolution in natural language understanding.

NALIG: A CAD system for interior design with high level interaction capabilities

Abstract: In the last few years, CAD systems have been evolving from simple drafting tools to much more complex solid modeling environments. Nevertheless, experience has shown that effective use of such systems relies on the characteristics of their user interfaces: the user should have the possibility of describing a particular scenario in full detail or giving the system only a raw description of it. The authors describe NALIG, a system able to understand and reason about high-level descriptions of spatial scenes. The user interacts with the system by using a simple natural language fragment that is expressive enough to describe complex configurations of objects. NALIG replies by drawing on the screen an image mirroring its own understanding of the scene that has been described. The comprehension process involves various forms of common-sense reasoning carried out at two different levels of abstraction. This has required the integration of different AI techniques (e.g. natural language understanding, spatial reasoning and default reasoning).

Script and Frame: Mixed Natural Language Understanding System with Default Theory

Abstract: Minsky's frame [7] and Schank's script [10] are two leading representation languages in the natural language understanding system for understanding a story. Very different inference mechanisms are embedded in the two representations: property inheritance along taxonomic structure in the frame system vs. causal-effect connectivity among a sequence of events in the script. This paper attempts to merge them into a mixed understanding system, specially via default logic to formalize its inference process and create a causalized default theory. The idea is to retain the frame system as a basic structure for events, and then to organize a script specifying the normal way of events happening in the specific situation via connecting some concerned lower level frames with causal relationship.

Understanding natural language for virtual reality: An information theoretic approach

Abstract: The approach to the solution of the natural language parsing (NLP) problem has traditionally taken a linguistic route. A more basic approach using the theory of information is exploited here. Parsing is accomplished based on the information content of the sentence. This reduces processing time, allows for the handling of ungrammatical sentences, and produces a structure that allows the computer to gain information on a continuous basis. The approach is entirely integrable with much of the important NLP work which has been done in the past. An application program which has been developed to illustrate the basic technique is also described. >

Tools for constraint satisfaction

Abstract: Constraint satisfaction problems involve finding values for problem variables that satisfy constraints on what combinations of values are permitted. They have applications in many areas of artifical intelligence, from planning to natural language understanding. Constraint satisfaction can be very difficult. A variety of tools are available for promoting successful satisfaction.

FEATS: knowledge representation and processing issues

Abstract: FEATS is an intelligent training system that supervises a structural/mechanical engineer using a finite element package such as COSMIC/Nastran. Within this goal, FEATS involves the investigation of various aspects of artificial intelligence including natural language understanding and processing, goal extraction and planning, user modelling, and knowledge representation and manipulation. The paper describes an approach to understanding the engineer's conversation with FEATS, the consequent extraction of knowledge, and its representation. The knowledge representation and processing is implemented in a FEATS prototype. This prototype was written in Quintus prolog and Quintus ProWindows, and it runs on a Sun SparcStation 1.

Assembly instruction manual understanding by fusing natural language understanding and technical illustration understanding

Abstract: This research aims to understand an assembly instruction manual by fusing the result obtained from understanding technical illustrations (TIU) and that obtained from understanding instructions (IU). In the first version of the system, we assume a TIU and an IU are carried out independently. >

A Conceptual Framework For Semantic Integrity In Large Knowledge-Based Systems - The World According to GARP ver 1.0 -

Abstract: Knowledge representation has become recognized universally as the core issue in AI research. This paper describes the fundamental principles of GARP, a conceptual framework fer the representation of knowledge in a semantically consistent manner. An argument supporting the necessity of such consistency is presented, with a description of the ways in which this consistency can be achieved and exploited in the field of natural language understanding.