Miscommunication handling in spoken dialog systems based on error-aware dialog state detection

• DOI: 10.1186/s13636-017-0107-3
• Publication Date: 2017-05-08T13:21:48Z
• Authors: Chung-Hsien Wu; Ming-Hsiang Su; Wei-Bin Liang
• Keywords: Utterance; Semantic interpretation; Word error rate; Dialog system

With the exponential growth in computing power and progress speech recognition technology, spoken dialog systems (SDSs) with which a user interacts through natural has been widely used human-computer interaction. However, error-prone automatic (ASR) results usually lead to inappropriate semantic interpretation so that miscommunication happens easily. This paper presents an approach error-aware state (DS) detection for robust handling SDS. Non-understanding (Non-U) misunderstanding (Mis-U) are considered this study. First, understanding evidence (UE), derived from confidence, is adopted Non-U followed by recovery. For Mis-U recognized sentence containing uncertain words, partial sentences obtained removing potentially misrecognized words input utterance organized, based on regular expressions, as tree structure tolerate deletion or rejection of keywords resulting misrecognition DS modeling. Latent analysis then employed consider verified their n-grams detection, including predefined Base DSs. Historical information-based find most likely Several experiments were performed corpus restaurant reservation task. The experimental show proposed achieved promising performance recovery repair well satisfactory task success rate dialogs using method.