Information Technology Reference
In-Depth Information
simulator, the tests can take many shapes, especially if we add the possibility
of involving a computational simulation of a user rather than a human user.
The choice between the different possibilities mentioned above is based on
technical and methodological criteria. We present these aspects in this chapter,
with section 3.1 illustrating some representative study cases, and section 3.2
detailing the methodologies involved in the main design stages.
3.1. Comparing a few development progresses
The following scenarios are purely indications; even if they are based on
real experiences, they tend to caricature them slightly and their main purpose
is to illustrate methodological concepts.
3.1.1. A scenario matching the 1980s
In the 1980s, an MMD system could be developed by a single person who
had a vast knowledge in computer science. The designer imagined here is
trying to test a theoretical dialogue model, which is, for example, the
innovative proposal of his/her dissertation research.
To design his/her system, he/she defines a chain architecture, with
modules that are executed in a chain fashion, in the following order:
automatic speech recognition, linguistic analyses, solving references, speech
act detection, natural language generation and text to speech synthesis. To
implement this system, he/she works on each module one after the other,
following the machine's processing order. He/she uses an existing speech
recognition module, sets it to have a vocabulary matching the tasks he/she is
aiming for and trains it with his/her own voice for a few hours in order to
optimize the performances. To operate, this module needs a button or a pedal
to be pressed when the utterance is pronounced. Once the utterance is
finished, which is indicated by releasing the pedal, the module starts to
process the recorded signal. It creates a written transcription of the
pronounced utterance, which enters a linguistic analysis module. This
analysis, using unification grammar, for example, includes lexical, syntactic
and semantic aspects: in such a grammar, the lexicon, the grammar rules and
the sentence representation use the same formalism, i.e. feature structures.
Thus, the same feature structure unification mechanisms allow us to carry out
different stages of automatic understanding. The designer writes the resources
Search WWH ::
Custom Search