Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 74

Chapter 5
Semantic Analyses and Representations
This second part is focused on the processing of the user's utterances at the
system's input, and it starts with what we can consider to be the core part of a
system's understanding abilities: determining the meaning of the utterances.
The goal is to achieve a computational representation of this meaning from
the input audio signal and the system's internal context representation. The
aim is to achieve a representation that does not involve pragmatic aspects
such as reference resolution or the determination of the speech act (see
Chapters 6 and 7). Calling upon context is, however, crucial because in a
dialogue an utterance can follow up on the previous utterance, for example by
adding an adverbial that can only be interpreted thanks to the previous
utterance, or an utterance can take up a term that had been the focus of a
lexical explanation, or even an utterance can be elliptic, that is omit a term,
such as a verb or a noun, due to the previous use of this constitutive element.
Determining the whole meaning of “I want the shortest one” after talking of a
journey to Paris can thus integrate the concept of journey, in the same way as
“and how long with the other?” after “how long will this journey take?” can
integrate “take” or at least integrate the fact that the question focuses on a
lapse of time and not a distance. In the example given in the introduction, if
we consider that it is an oral dialogue, we can also consider that, due to
context, the utterance “how long with this itinerary which seems shorter?”
does not match the noun “seam” like in “how long with this itinerary which
seams ...” (in French, with the same example “combien de temps par ce
chemin qui semble être le plus court?” there can similarly be a confusion