Multimodal Output Management - Man-Machine Dialogue: Design and Challenges - page 157

managed, can lead the user not to draw a link between the different
materializations of the same information. In conclusion, redundancy should
only be used in situations in which it is obvious that they are redundant.
Moreover, we should abstain from using redundancy within a single
communication channel, for example by adding the generation of an oral
utterance to that of a sound, because both of them occupy the audio channel
that ends up overloaded.
Redundancy management and information allocation over communication
channels falls within the scope of a single issue, that of multimodal fission. At
the level of audio or visual signals, it focuses on directing the information to
the appropriate communication channel depending on its nature. Presenting a
video involves transmitting the soundtrack in the audio channel and the image
on the visual channel, which is a multimodal fission directed by constraints.
On the level of semantics, it is to dissociate the information content over
several modalities so as to better manage its complexity by obtaining
simplified monomodal messages. One example falling within the scope of
multimodal fission directed by preferences consists of displaying the part of
the information that requires maintained attention and verbalizing the part
that only requires selective attention. At the level of pragmatics, it is
dissociation of the act of dialogue of the message over several modalities to
obtain simpler dialogue acts, as in the case when we dissociate the
components of a composite act to turn it into simple acts. In this case again, it
is multimodal fission, that allows us to progress to the three levels identified
for multimodal fusion that have equivalents in generation.
9.3.3. Generation of referring expressions
The example of “the flashing one” involves several reference phenomena
mentioned in Chapter 6: the mention of a property, here given as a relative,
which singles out the object and allows the system to identify it, as well as the
use of a pronoun that is deictic and anaphoric, deictic because it refers in a
deictic manner to an object that has been focalized and anaphoric because it
takes up from an antecedent its nominal head, that is “itinerary”. All these
phenomena give us an idea of the processes [MEL 06] that a referring
expression generator must manage. A referring expression must enable the
user to identify the referent in a non-ambiguous manner. To this end, a now
classic algorithm, the incremental algorithm [REI 00], finds the properties