Kristiina Jokinen
CDM architecture

CDM: Constructive Dialogue Model

In the Constructive Dialogue Model, interaction starts with Contact, being close enough for communication or being connected by some media (phone, skype, etc.) to a potential dialogue partner. Many neurocognitive studies show how brain activation is triggered by the mere appearance of a human in the vicinity, while the attention is directed to human face. Perception concerns the recognition of the speaker’s face belonging to a particular person, sounds belonging to a particular language, or gesturing having a communicative meaning. Together the enablements of Contact and Perception deal with being aware of the interaction, and account for the intention to maintain the contact (move closer) and continue perception of the signals (keep head turned towards the partner).

The part of the interaction commonly regarded as the main body of the participants’ engagement is enabled by understanding the partner’s intentions and by providing an appropriate reaction. The enablement of Understanding deals with the ability to interpret communicative signals as having semantic meaning in the given context, and the enablement of Reaction refers to carrying out an action that addresses the changed situation in which the agents find themselves.

In spoken dialogue systems, Contact and Perception are implemented via signal detection and signal analysis modules (Speech Recognition, Gesture Detection, Gaze Tracking), while Understanding and Reaction are complex and commonly partitioned into modules such as Natural Language Understanding, Dialogue Management, and Language Generation. For instance, NLU usually includes components for parsing, ambiguity resolution, topic and intention recognition as well as decision making of how to react to the situation, i.e. the dialogue manager proper. Reaction refers to language generation of an utterance, motoric actions, and internal updates, and also contains rendering of the planned reaction into behavioural correlates (speech, gaze, gesture, motion).

The Constructive Dialogue Model was applied in the WikiTalk robot dialogue system. It supports the multimodal interaction between the robot and the human partner.