Learning Content Selection Rules for Generating Object Descriptions in Dialogue

A fundamental requirement of any task-oriented dialogue system is the ability to generate obje t des riptions that refer to obje ts in the task domain. The subproblem of ontent sele tion for obje t des riptions in task-oriented dialogue has been the fo us of mu h previous work and a large number of models have been proposed. In this paper, we use the annotated o onut orpus of task-oriented design dialogues to develop feature sets based on Dale and Reiter's (1995) in remental model, Brennan and Clark's (1996) on eptual pa t model, and Jordan's (2000b) intentional in uen es model, and use these feature sets in a ma hine learning experiment to automati ally learn a model of ontent sele tion for obje t des riptions. Sin e Dale and Reiter's model requires a representation of dis ourse stru ture, the orpus annotations are used to derive a representation based on Grosz and Sidner's (1986) theory of the intentional stru ture of dis ourse, as well as two very simple representations of dis ourse stru ture based purely on re en y. We then apply the rule-indu tion program ripper to train and test the ontent sele tion omponent of an obje t des ription generator on a set of 393 obje t des riptions from the orpus. To our knowledge, this is the rst reported experiment of a trainable ontent sele tion omponent for obje t des ription generation in dialogue. Three separate ontent sele tion models that are based on the three theoreti al models, all independently a hieve a ura ies signi antly above the majority lass baseline (17%) on unseen test data, with the intentional inuen es model (42.4%) performing signi antly better than either the in remental model (30.4%) or the on eptual pa t model (28.9%). But the best performing models ombine all the feature sets, a hieving a ura ies near 60%. Surprisingly, a simple re en y-based representation of dis ourse stru ture does as well as one based on intentional stru ture. To our knowledge, this is also the rst empiri al omparison of a representation of Grosz and Sidner's model of dis ourse stru ture with a simpler model for any generation task.