编辑: 阿拉蕾 | 2019-07-16 |
attention, but also can activate the relevant context (e.g., domain context and visual context). This con- text can constrain the type of intention associated to attention and thus provide useful information for intention prediction. Driven by this assumption, this paper presents an empiri- cal investigation on the role of deictic gestures in intention recognition. We apply model-based methods and instance- based methods to incorporate gestural information to rec- ognize users'
intention. We examine the effects of using gestural information for user intention recognition in two stages C speech recognition stage and language understand- ing stage. Our empirical results have shown that utilizing gestural information improves intention recognition and the performance is further improved when gestures are incorpo- rated in both speech recognition and language understanding stages compared to either stage alone. In the following sections, we ?rst give an introduction to multimodal input interpretation, then describe in detail in- tention recognition with gestural information, and ?nally present results from empirical evaluations. RELATED WORK Intention has been addressed in a number of ways to serve different purposes in different research areas. For example, in the ?eld of Robotics, intention has been used to represent an idea or a mental state associated with human actions [32], a sequence of human behaviors or robot actions [16, 20], and a speci?c robot action [30]. In dialog systems, intention ........