Project A1

When persons communicate, they constantly evolve models of their communication partners, covering aspects on different levels and being partial in that only relevant aspects are stored. The project A1 is concerned with establishing and maintaining individual partner models that make use of implicit memory recall in order to allow for alignment processes. Partner models comprise mechanisms and representations which are crucial for adaptation to the situational context and they play an important role for effective conversation. Traditionally, explicit partner models have been investigated, with logical inference used to reason about the interlocutor's intentions and beliefs. But as these processes tend to be effortful, we investigate how better and faster alignment between communication partners can take place, e.g., by implicit, automatic activation processes of an autobiographical memory. In this project, we address high-level, cognitive dimensions of alignment phenomena arising in the context of a task-oriented conversation. We want to investigate face-to-face situations with up to three interlocutors. In this context two aspects are relevant: On the one hand, partner models on general aspects that can be applied to any communication partner and in which conventions such as turn-taking and focus of attention are important; on the other hand, partner models on individual features which help to align with interlocutors previously known. A typical example for the latter is the activation of the interlocutor's anterior intentions from a similar situation. This cognitive process aids in anticipating what the interlocutor is now referring to, thinking about, or trying to achieve. To this end, the project aims at investigating and implementing partner models which take into account general as well as partner specific aspects both by simulative and empirical methods.

Project team:

Prof. Dr. Ipke Wachsmuth

Dipl. Inform. Nadine Pfeiffer-Leßmann

Dr. Stefan Kopp

Cooperations:

Collaboration is planned with project "A2 - Construction of implicit common ground" to investigate how the partner model can contribute to solutions to problems concerning the interface of implicit common ground and full common ground. A2 findings on the general terms of processing of implicit common ground will guide the realisation of partner models.
The multi-party scenario will be subject of experiments in cooperation with "A4 - Alignment of situation models" investigating the effect of conversational roles and kind of communication partners with regard to spatial alignment processes. In contrast to the project A4 which concentrates on relational and perceptual components of situation models, A1 draws its attention towards the higher and more abstract levels of situation models such as the causal and motivational dimensions of the situation the interactants are currently engaged in.
A1 will cooperate with "A5 - Alignment of attention" on the investigation of models for gaze in turn-taking behavior taking place in face-to-face interactions in 3D environments. While A5 will provide quantitative models developed by use of innovative eye-tracking techniques, A1 will utilize eyetrackers for real-time detection of gaze in natural interaction scenarios to be accounted for by artificial agents.
A1 will cooperate with "B1 - Speech-gesture alignment" on representations of dialogue states, common grounds, and communicative intent to build up a simulator system. A1 will profit from the multi-modal expressiveness of the artificial agent investigated in B1 to be applied during the interactions. In exchange, A1 will provide a cognitive architecture with representations of the current context in form of situation models.
The project will also contribute to the development of motor-level alignment processes in collaboration with the project "C1 - Interaction space" . The representations and methods developed in C1 will inform dynamic partner modes investigated in project A1.

Publications:

Demos:

see: VR Lab Showcase