RavenClaw介绍_012_结论

4. Conclusions - 结论

We described RavenClaw, a new dialog management framework for spoken dialog systems operating in complex, goal-oriented domains.

我们描述了RavenClaw，一个针对在复杂、目标导向领域中运行的口语对话系统的新对话管理框架。

The framework separates the domain dependent and domain-independent components of the dialog manager and focuses system development effort on defining a hierarchical decomposition of the underlying task.

该框架将对话管理器的域依赖和域无关组件分开，并将系统开发工作重点放在定义底层任务的层次分解上。

A Dialog Engine uses the task representation to drive the dialog forward towards its goals, and uses separate, generic, conversational strategies to maintain dialog coherence and continuity.

对话引擎使用任务表示来推动对话朝着其目标前进，并使用单独的、通用的对话策略来维持对话的一致性和连续性。

---

RavenClaw-based dialog managers were constructed for five dialog systems spanning qualitatively and quantitatively different domains.

为五个在定性和定量上均不同的域构建了基于RavenClaw的对话管理器。

Work in one domain (information exploration) resulted in the addition of new functionality but no major changes in the overall structure or core mechanisms were required.

在一个域（信息探索）中的工作增加了新功能，但整体结构或核心机制没有发生重大变化。

Moreover, the framework easily adapted to all these domains, indicating a high degree of versatility and scalability.

此外，该框架很容易适应所有这些域，表明具有很高的通用性和可扩展性。

---

Currently our efforts are focused on managing grounding in a separate layer based on the continuous updating of the system’s beliefs about the validity of information in the task tree, and decision making under uncertainty.

目前，我们的工作重点是基于任务树中系统对信息有效性的信念的连续更新，在单独一层中管理接地，以及在不确定性下的决策制定。