This section presents our model of temporal reference resolution in scheduling dialogs. Section 4.1 describes the cases of deictic reference covered and Section 4.2 presents the anaphoric relations defined. Section 4.3 gives some background information about focus models, and then describes the focus model used in this work.
Anaphora is treated in this paper as a relationship between a Temporal Unit representing a time specified in the current utterance ( TUcurrent) and one representing a time specified in a previous utterance ( TUprevious). The resolution of the anaphor is a new Temporal Unit representing the interpretation, in context, of the contributing words in the current utterance.
Fields of Temporal Units are partially ordered as in Figure 3, from least to most specific. The month has the lowest specificity value.
In all cases of deictic reference listed in Section 4.1 and all cases of anaphoric reference listed in Section 4.2, after the resolvent has been formed, it is subjected to highly accurate, obvious inference to produce the final interpretation. Examples are filling in the day of the week given the month and the date; filling in pm for modifiers such as ``afternoon''; and filling in the duration of an interval from the starting and end points.
In developing the rules, we found domain knowledge and task-specific linguistic conventions to be most useful. However, we observed some cases in the NMSU data for which syntactic information could be exploited [9,33]. For example, ``until'' in the following suggests that the first utterance specifies an end time.
``... could it be until around twelve?''A preference for parallel syntactic roles might be used to recognize that the second utterance specifies an end time too. We intend to pursue such preferences in future work.