* * * * * * * * * * * * * * * * * * * * * * * *
LINGUIST List logo Eastern Michigan University Wayne State University *
* People & Organizations * Jobs * Calls & Conferences * Publications * Language Resources * Text & Computer Tools * Teaching & Learning * Mailing Lists * Search *
* *
LINGUIST List 18.1848

Tue Jun 19 2007

Diss: Gieselmann: Computational Ling: 'Fehlerbehandlung in Mensch-M...'

Editor for this issue: Hunter Lockwood <hunterlinguistlist.org>


To post to LINGUIST, use our convenient web form at http://linguistlist.org/LL/posttolinguist.html.
Directory
        1.    Petra Gieselmann, Fehlerbehandlung in Mensch-Maschine-Dialogen


Message 1: Fehlerbehandlung in Mensch-Maschine-Dialogen
Date: 19-Jun-2007
From: Petra Gieselmann <petra_gieselmannhotmail.com>
Subject: Fehlerbehandlung in Mensch-Maschine-Dialogen


Institution: University of Stuttgart
Program: Institute for Natural Language Processing
Dissertation Status: Completed
Degree Date: 2007

Author: Petra Gieselmann

Dissertation Title: Fehlerbehandlung in Mensch-Maschine-Dialogen

Dissertation URL: http://elib.uni-stuttgart.de/opus/frontdoor.php?source_opus=3065&la=de

Linguistic Field(s): Computational Linguistics

Dissertation Director:
Christian Rohrer
Alex Waibel

Dissertation Abstract:

In recent years, spoken dialog systems became more and more popular. The
awareness of the problems emerging from system errors, especially in
recognizing user input and understanding the user intention, increased.
These errors may lead to a consequent confusion for both users and the
system itself. The need to devise better strategies for detecting problems
in man-machine dialogs and dealing with them gracefully has become
paramount for spoken dialog systems.

As an example for spoken dialog systems, a household robot which helps
users in the kitchen is chosen in the present study. Within this scenario,
the error robustness is of special importance because the users are naive
and want to talk to the robot in the same way as to another human.

The main contribution of this thesis is a detailed analysis of errors
within man-machine dialogs and the development of solutions to all the
errors found there. This results in a more efficient man-machine
interaction as shown in a user study.

The present thesis evaluates errors and methods to deal with them in
man-machine dialogs. The first part is about related work. The second part
deals with user studies which result in an error classification with three
classes of errors: New concepts, elliptical and anaphoric utterances and
complex utterances. Before analysing in detail these error classes, the
focus switches to the user's as well as the robot's communication strategy
to evaluate its influence on errors and the communication in general.
The third part deals with error handling: Since the error classification
reveals that lots of errors are due to missing concepts in the linguistic
resources, their automatic extension is explored: Novel mechanisms for
dynamic extension of the vocabulary and for the integration of the internet
as an additional knowledge source are developed. The tenth section is about
the resolution of ellipsis and anaphora which represent the second error
class. Since the third group of errors in the error classification are
complex utterances, the eleventh section deals with a generic method to
resolve them. The final section of the third part explores new concepts
users utter within clarification dialogs and meta communication and
develops mechanisms to deal with them.

The fourth part consists of an evaluation of all these error handling
strategies in an overall user test which shows that the users are much more
successful and can accomplish more tasks in less time compared to the
baseline version without any error handling. The final section gives a
conclusion and an outlook on future work.





Respond to list|Read more issues|LINGUIST home page|Top of issue




Please report any bad links or misclassified data

LINGUIST Homepage | Read LINGUIST | Contact us

NSF Logo

While the LINGUIST List makes every effort to ensure the linguistic relevance of sites listed
on its pages, it cannot vouch for their contents.