Expectation-Based Speech Recognition

Mudler, Joachim; Paulus, Erwin

doi:10.1007/978-3-642-83476-9_43

Cited by 9 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Fink and Biermann [14] used historical information about a dialog to predict the content of future utterances and perform error correction based on these utterances. [24] interleave the generation and verification of expectations at a lexical level. Carbonelf and Pierrel [10] have developed an architecture for using dialog sources for assisting in voice communication, but dld not integrate this with actual linguistic components nor was there any attempt to generalize their expectations.…”

Section: Related Workmentioning

confidence: 99%

Using expectation to enable spoken variable initiative dialog

Smith¹,

Hipp²

1992

Proceedings of the 1992 ACM/SIGAPP Symposium on Applied Computing: Technological Challenges of the 1990's

View full text Add to dashboard Cite

This paper discusses how dialog expectations may be exploited in solving some important problems of dialog processing. Our model for expectation-driven dialog processing extends the theoretical model of attentional state by Grosz and Sidner as it provides a practical application of the model. In addition to a presentation of our model and a discussion of its application to dialog processing problems, the paper reports on results obtained from the use of our voice-interactive dialog system. The results are baaed on over 140 dialogs and over 2800 spoken user utterances.Among researchers in natural language processing it is common knowledge that contextual information must be used in order to properly understand and respond to utterances in a dialog (see Allen [5] for a general discussion). This is true for all types of dialogs including tack-oriented dialogs, the object of study in our research. Consider the following subdialog (C denotes the computer, and U the user. )(1) c:You need to measure the voltage between cormectora 24 and 83. (2) u:Uhere do I connect this red wire? (3) c:To connector 24.(4) u: Where is it?(5) c: To the right of the switch.(6) U: I have made the connection.In utterance(2) theuser refers to "this red wire" although no wire has been explicitly mentioned. Itcame into context when the discussion of meaeuring the voltage was introduced. This implicitly introduced the voltmeter into context along with its wires, one of which is this red wire. Sinrilarly, the "it" of utterance (4)refers to connector 24, which is mentioned in utterance (3). User utterances (2) and (4) cannot be understood in isolation.They can only be understood by using the contextual information available as the result of previous utterances in the dialog.Permission to copy without fee all or part of this material is grantsd provided that the oopiss ars not made or distributed for direot commercial advantage, ths ACM copyright notice and the titls of the publication and its date appear, and notioe is given that copying ia by parmiasion of the Association for Computing Machinery.To oopy otherwise, or to republish, requirss a fee andtor specific permission. e 1992 AcM 0.88791-502.X/92/0002/01 23...$1.50This paper discusse8 how a particular feature of dialogs, eqectation, may be exploited in providing a solution to problems of dialog interpretation.Specifically, we discuss how expectation may be exploited in handling the following important problems:1. Understanding ill-formed inputs. Using context for proper interpretation of utter-ances. Participating in variable initiative dialogs. A variable initiative dialog is a dialog where either dialog participant can have control at different points in the dialog. A variable initiative dialog system contrasts with other N L dialog systems such as [6], [7], [8], [15], [19], [21], [27], and [29], for which the dialog is either purely user controlled or purely computer controlled.Our research has focused on taak-oriented dialogs, dialogs held in conjunction with the performance of some task ...

show abstract

Section: Related Workmentioning

confidence: 99%

Using expectation to enable spoken variable initiative dialog

Smith¹,

Hipp²

1992

Proceedings of the 1992 ACM/SIGAPP Symposium on Applied Computing: Technological Challenges of the 1990's

View full text Add to dashboard Cite

show abstract

“…The linguistic levels, syntax Manuscript received May 20, 1991;revised Febreuary 19, 1993. Recommended (the structural relations between the words of an utterance), semantics (the interpretation of the meaning of an utterance), and pragmatics (finding truth values for the semantic interpretations in a concrete situation), are represented in most natural language (NL) systems for the analysis of written language and also in some speech understanding systems (see, for instance, [6], [8], [17], [33]) using representation techniques like the predicate calculus (e.g., [ 11, [lo]), frames, or semantic networks (e.g., [7], [2], [31]). At least for the recognition phase in systems for speech understanding statistical methods are used ( [6], [ 151, [3], [ 181).…”

mentioning

confidence: 99%

A speech understanding and dialog system with a homogeneous linguistic knowledge base

Mast

Kummert

Ehrlich

et al. 1994

IEEE Trans. Pattern Anal. Machine Intell.

View full text Add to dashboard Cite

“…Der beschriebene Versuch ist allerdings noch in mancher Hinsicht unvollständig und liefert eher Ansätze für weitere Untersuchungen als abschließende Ergebnisse.2.2 Erwartungsorientierte AnalyseDie zusammenfassende Schilderung des hier nä-her betrachteten Systems folgt auszugsweise der bereits aus[8] bekannten Darstellung. Ausführlichere Beschreibungen sind in[6,7,9] zu finden. Das System verwendet silbenorientierte phonetische Einheiten und bildet Worthypothesen in einem Silbenraster (siehe Bild 1).…”

unclassified

Grundzüge eines sprachkundigen Systems zur automatischen Spracherkennung / Principles of a Language Modelling System with Application to Automatic Speech Recognition

Paulus¹,

Mudler²

1989

it - Information Technology

Self Cite

View full text Add to dashboard Cite

Prof. Dr.-Ing. Erwin Paulus studierte Nachrichtentechnik an der TU Wien. Nach mehrjähriger Tätigkeit am IBM-Laboratorium Wien wurde er wissenschaftlicher Mitarbeiter am Lehrstuhl für Datenverarbeitung der TU München. Nach Promotion und Habilitation an der TU München wurde er 1976 Professor am Institut für Nachrichtentechnik der TU-Braunschweig und übernahm dort die Leitung der "Abteilung für digitale Signalverarbeitung und automatische Mustererkennung". Dr.-Ing. Joachim Mudler, 1977-1982 Studium der Informatik an der TU Braunschweig, anschließend von 1982-1987 wissenschaftlicher Mitarbeiter bei Professor Paulus am Institut für Nachrichtentechnik der TU Braunschweig. Promotion 1987. Seit 1988 selbständig.The automatic recognition of continously spoken prose in terms of a sequence of words generally cannot be based upon short-time signal processing techniques alone. Only if various types of knowledge about the actual language (phonology, prosodies, vocabulary, grammar, semantics) are taken into account, the recognition task may be treated as a dynamically changing system of sub tasks, only a few of which must be solved by means of short-time signal processing techniques, and -if the conditions are not too badare indeed easy enough to make a reliable solution feasible. This paper is treating the structure and the performance of a speech recognition system that conforms to the guidelines mentioned above. In particular, it is examined how difficult the subtasks that are subject to signal processing techniques really are and as to which limits the difficulty can be reduced when a given knowledge base is exploited exhaustively.Die automatische Erkennung des Wortlauts frei formulierter fließend gesprochener Äußerungen kann grundsätzlich nicht mit Methoden der Kurzzeitsignalverarbeitung allein bewältigt werden. Erst durch die Ausnutzung umfangreicher Vorkenntnisse über die verwendete Sprache (Phonologie, Prosodie, Wortschatz, Grammatik, Semantik) läßtsich die Erkennungsaufgabe in einem dynamischen Prozeß in ständig wechselnde Teilaufgaben aufgliedern, von denen nur noch wenige unbedingt mit Methoden der Signalverarbeitung gelöst werden müssen und unter einigermaßen günstigen Bedingungen auch gelöst werden können. Der Beitrag behandelt den Aufbau und die Wirkungsweise eines den oben genannten Leitlinien entsprechenden Systems. Insbesondere wird dabei überlegt, welchen Schwierigkeitsgrad die der Signalverarbeitung zugewiesenen Teilaufgaben haben und bis zu welcher Grenze dieser Schwierigkeitsgrad bei vollständiger Ausschöpfung einer gegebenen Wissensbasis gesenkt werden kann.

show abstract

Expectation-Based Speech Recognition

Cited by 9 publications

References 0 publications

Using expectation to enable spoken variable initiative dialog

Using expectation to enable spoken variable initiative dialog

A speech understanding and dialog system with a homogeneous linguistic knowledge base

Grundzüge eines sprachkundigen Systems zur automatischen Spracherkennung / Principles of a Language Modelling System with Application to Automatic Speech Recognition

Contact Info

Product

Resources

About