Showing papers on "Natural language understanding published in 1999"

Adaptive command predictor and method for a natural language dialog system

Abstract: A command prediction system for natural language understanding systems, in accordance with the present invention, includes a user interface for receiving commands from a user. A command predictor receives the commands from the user interface and predicts at least one next command which is likely to be presented by the user based on a command history. A probability calculator is included in the command predictor for determining a probability for each of the at least one next command based on the command history such that a list of predicted commands and their likelihood of being a next command are provided.

The cu communicator system 1

Abstract: The CU Communicator system is our initial testbed for researchleading to advanced dialog systems enabling robust and gracefulhuman computer interaction. It is a DARPA hub compliantsystem for the DARPA Communicator task, and wasdemonstrated at the DARPA workshop in June 1999.Robustness and portability of spoken dialog systems are two ofthe issues we attempt to address in the project. We use robustparsing and dialog control strategies to be as flexible as possibleto user variance. In order to make the systems easier to develop,we have adopted a largely declarative representation where thebulk of the domain specific information is provided in externalfiles. 1. INTRODUCTION In April 1999, the University of Colorado speech group begandevelopment of the CU Communicator system, a Hub-compliant implementation of the DARPA Communicatortask[1][2].The system combines continuous speech recognition,natural language understanding and flexible dialog control toenable natural conversational interaction by telephone callers toaccess information about airline flights, hotels and rental cars.In June 1999, the system was fully functional, and wasdemonstrated at the June 1999 DARPA Communicator meeting.This system connects live over the web to get real up-to-date airtravel, hotel and rental car information. Users call the system onthe telephone and make travel plans. We are using the system asa testbed for conversational system technology development. Forthis system, we developed a new Dialog Manager using an eventdriven strategy. Our natural language and dialog specificationfor the task is very much declarative.Our approach to robustness in natural language understanding isto use robust parsing strategies and event driven dialogstrategies. These strategies stress semantic content andcoherence over syntactic form. We have extended the Phoenixsystem as the basis for our robust parsing. This system uses ahierarchical frame parse representation. Our “event driven”dialog manager does not use an explicit script. The developercreates a set of hierarchical forms, representing the informationthat the system and user interact about. Prompts are associatedwith fields in the forms. The dialog manager decides what to donext from the current system context, not from a script.We also attempt to address issues of authoring and portability.Building a spoken dialog system is labor intensive and requiresa great deal of expertise. A major focus of our research is thedevelopment of tools to facilitate rapid creation of spokendialog applications by non-expert developers. Our initial stepshave been:• Declarative representation – To the extent possible, we arerepresenting information in external files rather thanhaving the developer write code.• Libraries – We are creating a library of common grammarsthat would be applicable to many tasks (times, dates, etc).

Qualia Structure and the Compositional Interpretation of Compounds

Abstract: The analysis of nominal compound constructions has proven to be a recalcitrant problem for linguistic semantics and poses serious challenges for natural language processing systems. We argue for a compositional treatment of compound constructions which limits the need for listing of compounds in the lexicon. The Generative Lexicon (Pustejovsky, 1995) provides us with a model of the lexicon which couples sufficiently expressive lexical semantic representations with mechanisms which capture the relationship between those representations and their syntactic expression. In our approach, the qualia structures of the nouns in a compound provide relational structure enabling compositional interpretation of the modification of the head noun by the modifying noun. This brings compound interpretation under the same rubric as other forms of composition in natural language, including argument selection, adjectival modification, and type coercion. We examine data from both English and Italian and develop analyses for both languages which use phrase structure schemata to account for the connections between lexical semantic representation and syntactic expression. In addition to applications in natural language understanding, machine translation, and generation, the model of compound interpretation developed here can be applied to multi-lingual information extraction tasks.

Method and system for hierarchical natural language understanding

Abstract: A method for hierarchical translation of input to a formal command in natural language understanding systems includes presenting an input command to be translated to a natural language understanding engine. At least two translator levels are provided in the natural language understanding engine. A first translator level of the at least two translator levels translates the input command into at least one category by associating the input command with the at least one category for the next level of translators. A formal language command is output for the input command from a last of the at least two translator levels based on the input command and the at least one category.

Method and system for ensuring robustness in natural language understanding

Abstract: A method and system, which may be implemented by employing a program storage device readable by machine, and tangibly embodying a program of instructions executable by the machine to perform method steps for ensuring robustness of a natural language understanding (NLU) system, includes tagging recognized words of a command input to the NLU system to associate the command with a context, and translating the command to at least one formal command based on the tagged words. A top ranked formal command is determined based on scoring of the tagged recognized words and scoring translations of the at least one formal command. Whether the top ranked formal command is accepted is determined by comparing a feature vector of the top ranked formal command to representations of feature vectors stored in an accept model. The top ranked formal command is executed if accepted and incorrect commands are prevented from execution to provide a robust NLU system.

Search optimization system and method for continuous speech recognition

Abstract: A system and method for continuous speech recognition (CSR) is optimized to reduce processing time for connected word grammars bounded by semantically null words. The savings, which reduce processing time both during the forward and the backward passes of the search, as well as during rescoring, are achieved by performing only the minimal amount of computation required to produce an exact N-best list of semantically meaningful words (N-best list of salient words). This departs from the standard Spoken Language System modeling which any notion of meaning is handled by the Natural Language Understanding (NLU) component. By expanding the task of the recognizer component from a simple acoustic match to allow semantic information to be fed to the recognizer, significant processing time savings are achieved, and make it possible to run an increased number of speech recognition channels in parallel for improved performance, which may enhance users perception of value and quality of service.

Information extraction as a stepping stone toward story understanding

Abstract: Historically, story understanding systems have depended on a great deal of hand-crafted knowledge. Natural language understanding systems that use conceptual knowledge structures [SA77, Cul78, Wil78, Car79, Leh81, Kol83] typically rely on enormous amounts of manual knowledge engineering. While much of the work on conceptual knowledge structures has been hailed as pioneering research in cognitive modeling and narrative understanding, from a practical perspective it has also been viewed with skepticism because of the underlying knowledge engineering bottleneck. The thought of building a large-scale conceptual natural language processing (NLP) system that can understand open-ended text is daunting even to the most ardent enthusiasts. So must we grit our collective teeth and assume that story understanding will be limited to prototype systems in the foreseeable future? Or will conceptual natural language processing ultimately depend on a massive, broad-scale manual knowledge engineering effort, such as CYC [LPS86]?

The IBM conversational telephony system for financial applications.

Abstract: We describe our development work on a telephonebased conversational system in the domain of mutual fund transactions. This system uses several components including robust large vocabulary continuous speech recognition, natural language understanding, dialog management, and text-to-speech synthesis technologies.

Spoken Natural Language Dialogue Systems: User Interface Issues for the Future

Abstract: Technology advances in automatic speech recognition (ASR) and natural language understanding (NLU) in recent years have brought us closer to achieving the goal of communicating with machines via unconstrained, natural speech. Until recently, most applications of speech recognition technology required that the user know a restricted set of command words. In contrast, the research system that is described in this chapter can understand and act upon fluently spoken language. This markedly changes the nature of the dialogue between the human and the computer. The objective of this research is to evaluate user interface design alternatives for these new natural spoken dialogues between humans and machines. In this chapter the focus is on a particular experimental vehicle, that of automatically routing telephone calls based on a user’s fluently spoken answer to the question “How may I help you? ”. This chapter summarizes results from several studies conducted to determine how best to design the user interface for a spoken natural dialogue system.

Modeling disfluency and background events in ASR for a natural language understanding task

Abstract: This paper investigates techniques for minimizing the impact of non-speech events on the performance of large vocabulary continuous speech recognition (LVCSR) systems. An experimental study is presented that investigates whether the careful manual labeling of disfluency and background events in conversational speech can be used to provide an additional level of supervision in training HMM acoustic models and statistical language models. First, techniques are investigated for incorporating explicitly labeled disfluency and background events directly into the acoustic HMM. Second, phrase-based statistical language models are trained from utterance transcriptions which include labeled instances of these events. Finally, it is shown that significant word accuracy and run-time performance improvements are obtained for both sets of techniques on a telephone-based spoken language understanding task.

On the automated interpretation and indexing of American Football

Abstract: Combines natural language understanding and image processing with incremental learning to develop a system that can automatically interpret and index American Football. We have developed a model for representing spatio-temporal characteristics of multiple objects in dynamic scenes in this domain. Our representation combines expert knowledge, domain knowledge, spatial knowledge and temporal knowledge. We also present an incremental learning algorithm to improve the knowledge base as well as to keep previously developed concepts consistent with new data. The advantages of the incremental learning algorithm are that is that it does not split concepts and it generates a compact conceptual hierarchy which does not store instances.

The Role of Evaluation in the Development of Spoken Language Systems

Abstract: In this article, several criteria and paradigms are described tomeasure the performance of spoken language systems developed in theframework of national and international research projects. Theseevaluations are carried out in the domain of spontaneous human-humaninteraction as supported by machine translation systems. They are alsoapplied in the domain of spontaneous human-machine interactiontypically used in information retrieval applications. Some evaluationparadigms are discussed in more detail. It is also shown that officialperformance tests and site-specific evaluation criteria arecomplementary in use.

Approximate Natural Language Understanding for an Intelligent Tutor

Abstract: Intelligent tutoring systems (ITS’s) have a rich history of helping students in certain scientific domains, like geometry, chemistry, aztd progranmdng. These domains are ideal for ITS’s, because they cazl bc easily represented and because the type of interaction between the student and the tutor can be limited to entering a few simple numbers, symbols, or ke)~’ords. Students need help in other areas, but without the ability to robustly understand a student’s input, ITS’s in these areas are inherently limited. Recently a tec-hnique ("ailed Latent Semantic Analysis has offered a corpus-based approach to understanding textual input which is not sensitive to errors in spelling or grammar -- in fact, it pays no attention to word order at all. We are using this technique as part of an ITS system which promotes learning using natural human-like dialogue between the human and the student. This paper describes the tutoring system and Latent Semantic Analysis, and how they operate together. Then it describes our evaluation of LSA’s performance by comparing its judgments with those of human raters.

Pragmatic Reasoning: Inferring Contexts

Abstract: Pragmatic reasoning is defined as the process of finding the intended meaning(s) of the given, and it is suggested that this amounts to the process of inferring the appropriate context(s) in which to interpret the given. This suggestion is illustrated by examples from natural language understanding and visual object recognition. A formal, model-theoretic, definition of pragmatic reasoning is then presented and discussed.

A pervasive conversational interface for information interaction.

Abstract: In this paper, we describe a new pervasive conversational system that provides access to multiple desktop applications, from multiple client devices, using multiple input modalities. Client devices currently supported include desktop and telephone, and the applications incorporated include email, calendarand address book. When the access is from a desktop, both conversational natural language and graphical inputs are supported. The paper describes the overall architecture to support such a pervasive conversational system, along with innovations in continuous speech recognition, statistical natural language understanding, and dialog management that were developed to build the system. The paper also describes the partially unsupervised Wizard-of-Oz style setup to collect real-use data, and the performance of the statistical models constructed using this data for speech recognition and natural language understanding.

Speech repairs, intonational phrases, and discourse markers

Abstract: Interactive spoken dialogue provides many new challenges for natural language understanding systems. One of the most critical challenges is simply determining the speaker's intended utterances: bot...

On phoneme—to—character conversion systems in Chinese processing

Abstract: We propose a new goal for constructing a Chinese phoneme‐to‐character automatic conversion system. Instead of trying to get the largest number of correct characters converted, this goal strives for extracting the correct events from a given sentence. We believe this objective is more fruitful in that the extracted events can be better utilized in other applications such as information retrieval, extraction and dialogue systems. Furthermore, it would motivate researchers to work on the more challenging problem of natural language understanding. An information map is proposed for knowledge representation. These maps give very concise summary of different aspects of an object. They also provide more convenient ways to implement semantic template matching.

Autonomy control software

Abstract: For a number of years, researchers in AI and robotics have been sharing organisation principles for software development called “architectures” (Arkin, 1998; Hexmoor et al., 1997; Kortenkamp et al., 1998). Most recently, there has been an interest in extending the software architectures originally designed for robotic applications to accommodate autonomous operation beyond robotics (Pell et al., 1998). These architectures attempt to state principles for organising software for intelligent systems that exhibit cognitive functionalities such as natural language understanding, planning and learning, as well as organism-like functionalities to cope in the world such as reflexive and reactive behaviours.

Connectionist Analysis and Creation of Context for Natural Language Understanding and Knowledge Management

Abstract: Context affects many aspects of the behavior. Natural language understanding is one of the prime examples. This paper summarizes how an artificial neural network, the self-organizing map, can be used in modeling contextuality in data analysis and natural language processing. Important aspects are adaptivity gained by using a learning system, autonomous nature of the processing based on unsupervised learning paradigm, and gradedness of the representation. Examples in the application areas of information retrieval and knowledge management are considered. For instance, the visualization of self-organizing maps provides meaningful context for documents.

Multimodal speech systems

Abstract: This chapter describes the various knowledge sources required to handle human-machine multimodal interaction efficiently: they constitute the task, user, dialogue, environment and system models. The first part of the chapter discusses the content of these models, emphasising the problems occurring when speech is combined with other modalities. The second part focuses on spoken language characteristics, describes different parsing methods (rule-based and stochastic) using a task model, and briefly presents the integration of the rule-based method in an end-to-end information retrieval system.

Search Techniques for Learning Probabilistic Models of Word Sense Disambiguation

Abstract: The development of automatic natural language understanding systems remains an elusive goal. Given the highly ambiguous nature of the syntax and semantics of natural language, it is not possible to develop rule-based approaches to understanding even very limited domains of text. The difficulty in specifying a complete set of rules and their exceptions has led to the rise of probabilistic approaches where models of natural language are learned from large corpora of text. However, this has proven a challenge since natural language data is both sparse and skewed and the space of possible models is huge. In this paper we discuss several search techniques used in learning the structure of probabilistic models of word sense disambiguation. We present an experimental comparison of backward and forward sequential searches as well as a model averaging approach to the problem of resolving the meaning of ambiguous words in text.

Method for computer-aided foreign language instruction

Abstract: A foreign spoken language instructional method for use by a student utilizing a multi-media computer in which the computer is programmed to display virtual settings that would be encountered by the foreign student in everyday life in the country for the foreign language being learned is spoken. For example, one of the settings could be a fast food restaurant. The computer is also programmed to create a virtual host that is appropriate for that setting, and to allow, through the use of automatic speech recognition and natural language understanding technologies, a real-life conversation to occur between the student and the host that is appropriate to the setting. The computer will have stored or have access to a vocabulary and library of possible responses and statements by both the host and the student, along with a conversational tree utilizing those vocabularies and libraries that will enable the conversation to follow any of several twists and turns, and not a precise, pre-set, structured command-response protocol.

A Bayesian approach for understanding information-seeking queries

Abstract: Explores a Bayesian approach for handling natural language queries for information access or retrieval Natural language understanding is a key technology in human-computer interfaces, and often constitutes the front-end to information systems In this work, we devised a Bayesian methodology for the identification of the user's informational goals in information enquiries We focus on a restricted domain, the air travel information systems (ATIS) domain, where a user's informational goal often falls within a finite set of within-domain goals We formulated the problem into N binary decisions, to allow for the identification of queries with multiple goals, as well as queries with out-of-domain goals Experiments with the ATIS corpus shows that between 846% to 880% of the user queries are correctly handled via goal classification, rejection or multiple-goal identification

A connectionist inference mechanism for a natural language understanding system

Abstract: We have developed the Connectionist Inference Mechanism (CIM) to serve as a connectionist alternative to the symbolic inference module of P. Buchheit's (1991) Informational Network For A Natural Thinking (INFANT) System. CIM consists of several modules working together, including memory, neural networks, and a binding set. Its main task is to perform inference generation with the capability of variable binding. CIM is essentially a hybrid cognitive model that combines the advantages of a symbolic approach, local representation, and parallel distributed processing. It can be shown that each approach complements and supports the others.

Ambiguity resolution in natural language understanding, active vision, memory retrieval, and robot reasoning and actuation

Abstract: A cognitive robot is being designed to be utilized in four research projects studying the resolution of ambiguity in human and robotic systems: in natural language understanding systems, carried out in the Sentence Processing laboratory; in active vision systems, carried out in the Computer and Robot Vision laboratory; in memory retrieval systems, carried out in the Cognitive Science laboratory; and in robot reasoning and actuation, carried out in the Artificial Intelligence and Robotics Laboratories. The goal is to integrate these projects through the use of a cognitive robot which would allow ambiguity resolution in each of these research areas to be enhanced by the resolution capabilities of the other areas.

Using Natural Language and Video Data to Query and Learn American Football Plays

Abstract: In this paper we describe methods currently under investigation for learning, recognising and representing American Football plays. This is a complex spatio-temporal pattern recognition problem which we propose to solve by integrating image understanding and natural language understanding. Specifically, this paper describes the model representation used at different levels of abstraction incorporating several types of knowledge: expert knowledge, domain knowledge and spatio-temporal knowledge. We also describe a decision tree based incremental learning algorithm which combines context knowledge, attribute selection and the ageing of knowledge to keep the model representation of the plays consistent with the information extracted from the American Football video tapes.

A cognitive robot with reconfigurable mind for studying theories of ambiguity resolution

Abstract: Presents the conceptual design of a cognitive robot for investigating issues in ambiguity resolution. The robot will be intelligent, mobile, and anthropomorphic. It is envisioned that the robot will provide an alternative for studying cognition, providing a useful methodology complementing more traditional experimental methods. The ambiguity resolution will be studied in spoken natural language understanding, context driven active robot vision, memory system, and robot reasoning and actuation. Beyond the immediate goal of this research of creating and testing multi-modal theories of ambiguity resolution this project is concerned with validating the methodological idea of using such robots as basic experimental tools for research in the cognitive sciences and for helping to integrate together the research of different cognitive sciences.

Text, Speech and Dialogue: Second International Workshop TSD '99, Plzeň, Czech Republic, September 13-17, 1999

Abstract: This book constitutes the refereed proceedings of the 2nd International Workshop on Text, Speech, Dialogue, held in Pilsen, Czech Republic, in September 1999. Written for researchers and advanced students in area of Natural Language Processing.