The work presented in this proposal describes an approach to solve an actual
problem in the conversational systems area: to enable non-experts to design
and develop in which the user can establish a spoken dialogue systems. In
most of the systems of this type, the specific information relative to the
domain, or the specific information that delimits the specific capacities
and limitations of the system is embedded within the source code. Due to
this, people who want to modify or change actual systems experiment a great
number of difficulties, because they must search and investigate in which
parts of the source code this information is represented. Nowadays there
are some efforts to solve this problem, some of them on the part of private
sector companies, and some others by research centers mainly in universities
around the world; the objective is to separate the domain specific information
from the application, making the creation of new systems easier for people
with few or no programming skills. / (cont.) Our proposal was born, when some years
ago, in our research laboratory; we tried to part from an existing conversational
system to modify it, first to make it work in Spanish language, then to change
the domain of the dialogue. However these efforts did not bear satisfying
results because the specific information for the tasks that were carried
out was in several different parts of the source code, in different programming
languages. The idea of the creation of a conversational system framework
where the task specific information could be separated from the source code
was born, and this idea was approved in 2003 by a committee conformed by
UDLAP researchers and external participants, Dr. Luis Villaseñor from
INOEP and Dr. Wayne Ward from the Center for Spoken Language Research of
University of Colorado at Boulder.
Our proposal consists of the creation of a new module that deals with all
the aspects relative to the specific information of every task and instance
of a dialogue system. / (cont.) This task specific information is specified in text
files with XML format that are loaded first by this module, that scans and
analyzes them, looking for errors, validating these archives against its
respective document type definition (DTD) file. Next a consistency check
is made to make sure that these configuration files keep the consistency
with others according to the type of information handled. Once the information
and its format is validated, this module instructs the rest of the modules
in charge of the dialogue (DM, TTS, NLG, NLP, ASR, etc.) how to carry out
all interaction in the dialogue system, where they can find the data which
they require. The proposal presents the complete architecture of the system,
together with a Graphical User Interface by means of which, a user with few
or null programming skills will be able to specify a task describing all
the elements a conversational system requires.
The main contributions of this work are:
The idea of a conversational system module that verifies and manages all
the task specific information in a conversational system
A GUI to allow non-expert users to create his/her own conversational system
guided step by step in a simple web based interface. / (cont.)
This work has been published in:
"An approach to separate the Task-Specific-Information from the source code
in Galaxy Based Conversational Systems" D. Perez, , I. Kirschning, WSEAS
Transactions on Communications, Issue 1, Volume 3, January 2004.
"TLATOA COMMUNICATOR:
A framework to create Task-Independent Conversational Systems" D. Perez,
I. Kirschning CISSE 2009 Volume 2: Innovations in Computing Sciences and
Software Engineering SCSS, presented on December 16, 2009 (to be published).
Identifer | oai:union.ndltd.org:UDLA-Thesis/oai:ciria.udlap.mx:u-dl-a/tesis/4033021001091 |
Date | 08 April 2010 |
Creators | Pérez Rojas, Daniel |
Contributors | Dra. Ingrid Kirschning Albers, Dr. Mauricio Javier Osorio Galindo, Dr. Gerardo Ayala San Martín |
Publisher | Universidad de las Américas Puebla |
Source Sets | UDLA-Thesis |
Language | Spanish |
Detected Language | English |
Type | Electronic Thesis or Dissertation, Tesis o Disertación Electrónica |
Format | application/pdf, text/html |
Coverage | Doctorado |
Page generated in 0.0025 seconds