Summary

Spoken dialog systems (SDS) are becoming increasingly pervasive in our everyday lives for information access. Efforts devoted to the design and development of SDS aim to bring the right information to the right people at the right time for a diversity of application domains, such as finance, air travel, train schedules, and weather. These domains typically involve dynamic information of recurring interests to the user. SDS encompasses a suite of speech and language technologies to offer a conversational interface to dynamic information, including speech recognition, natural language understanding, dialog modeling, and speech synthesis. Hence, the user can present queries to the system by speaking naturally, and the SDS can respond in real time in synthetic speech. Numerous commercial SDS have been deployed for multiple languages. For example, the Speech Works¹ voice-activated stock-trading system enables callers to get real-time stock and market information, place orders, and trade stocks at any time. The entire human-computer interaction is conducted over a fixed-line/mobile telephone channel—that is, in a screenless setting. An example is provided in Table 11.1 to illustrate the directed nature of the dialog interactions, in which the system elicits a series of information attributes from the user in a scripted order. Interactions of this style are often known as directed dialogs or system-initiative dialogs. The system is always in control to guide the dialog at each step and constrain possible user input to a small set of options. In this way, the system is able to attain a decent level of robustness in recognizing and interpreting user input in order to achieve a high task-completion rate for informational transactions. However, the ease of use of these human-computer spoken-language interfaces is partially sacrificed due to constraints in interactivity. Recent research efforts in the field strive to relax the constraints in system-initiative dialogs.