An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 19

outline that allows us to privilege one hypothesis over another. We thus expect
various indications from the prosodic analysis module: locating the
focalization accents, temporal breakdown of the utterance, word by word, to
match words and gestures in a multimodal dialogue, and a coding of the
intonation's main characteristics. More in-depth analysis, with, for example,
the detection of periods, requires additional indications, but for now falls
more in the domain of subsequent oral corpus analysis than the domain of
real-time analysis for the MMD. This is actually a criticism that can be used
against many of the current systems: they do not use prosody, even though it
is an essential component of oral language. Initiatives such as that of
[EDL 05], who presents Nailon, an automatic prosody analysis system able to
detect in real-time various prosodic characteristics of an utterance in MMD,
are important.
One last aspect in which automatic speech recognition module has a role
to play is speaking turn management. The MMD systems have long remained
limited to an alternating operation of interventions, the system never
interrupting the user and only starting to speak once the user has finished
his/her utterance. More than that, we saw in section 1.1.2 with the
push-to-talk button or pedal that it was on the user to let the machine know
the beginning and the end of his/her intervention. We are now able to expect
that an MMD system will let the user express himself at any point, with no
constraints, and it is up to the machine to detect the beginning and the end of
the interventions. This is actually one of the functions of the Nailon system,
which uses prosodic clues of fundamental frequency and rhythm to
automatically detect the end of a user's intervention.
1.2.2. Analysis and reasoning abilities
Once the signals have been received at the system's input and transcribed
into appropriate representation, many analyses and reasonings will be carried
out so that the system can understand the meaning of the user's utterance,
his/her intent and, thus, the answer to give him/her. The analyses fall in the
domain of automatic understanding of natural language, that is of NLP, and
cover the following aspects: word identification (lexical analysis) so as to find
their meaning (lexical semantics) stored in the system according to a
well-defined formalism; the identification of the sentence's structure and the
grammatical functions of the identified components (syntactic analysis): the
construction of the sentence's semantics by combining the meaning of words