Reference Resolution - Man-Machine Dialogue: Design and Challenges - page 93

The result of the reference resolution is an updating of the results of
linguistic analyses in which the variables that have remained free are now
affected to referents, preferably the right ones, i.e. those which match the
user's intention. In case of ambiguity, several alternative representations are
created. In case it is impossible to affect a referent, an underspecified
representation is generated and transmitted to the pragmatic analyses
modules. Thus, in the example given in the introduction, the U2 utterance has
a referring expression “this itinerary which seems shorter”, which is
accompanied with a pointing gesture and thus refers in a multimodal way to
one of the itineraries displayed on the screen. The temporal markers for the
oral utterance and the gesture allow the system to check if they are temporally
synchronized at the (approximate) moment of the referring act. Prosody can
also provide support for this argument, for example if “this” is slightly
accentuated due to the simultaneous gesture. The gesture contributes the
referent identity, the oral utterance contributes the semantic representation
which was made and in which the referring expression has remained as a
variable. The fusion process of multimodal information allows the system to
solve the reference, i.e. assign the referent identifier to the referring
expression.
In this chapter, we will discuss this process: first with object references as
is the case for train journeys (section 6.1), then with action references (section
6.2), and finally in the specific case of references that call upon a referent in
the dialogue history, with anaphora and coreference phenomena (section 6.3).
6.1. Object reference resolution
Reference has been the focus of many pieces of work, from language
philosophy, to logic and linguistics [ABB 10], work which has highlighted
the notion of reference, the referring expression categories (demonstrative for
“this itinerary”), and a whole set of phenomena characterizing language, as
the presentation mode of a referent or the distinction between the attributive
use of an expression (“the train for Palaiseau, no matter which one it is,
always stops at Villebon”) and referential use (“the train for Palaiseau is eight
minutes late”) that implies a specific referent [CHA 02]. In MMD, the issue
always comes back to building the link between a linguistic form and an
element from a database managed by the application, whether it is a specific
train or a type of train, or even the generic class of all the trains as it is defined