B6 investigates alignment processes by focusing on task conditions under which alignment breaks down. In the first phase of the project (2010-2014), our research yielded the following main results:
- We found that the higher children’s working memory capacity, the more they aligned with their interlocutor.
- We found that lexical alignment, but not syntactic alignment, increased over the course of a pick-and-place task with basic geometric shapes and contributed to speedy task completion.
- We developed a computational model which can learn words, syntactic patterns, and their meanings and is biased to prefer recently “heard” lexical units and syntactic frames.
- Advancing clinical technologies: We used machine-learning techniques to distinguish patients with cognitive communicative disorder from healthy controls
In the next funding phase (2014-2018), the focus of B6 will be broadened to the investigation of situational conditions under which alignment breaks down. Our main assumption is that lexical and pragmatic alignment processes are driven by “conceptual pacts” as described by Brennan and Clark (1996). These conceptual pacts can be seen as pragmatic routines that are established over the course of the interaction and which interlocutors rely on when interpreting verbal and non-verbal behavior. We will focus on children, who do not yet have many such routines at their disposal and have to learn them in order to align with the interaction partner. We hypothesize that routines emerge from specific cooperative goal-directed actions and reduce the cognitive load within an interaction. Since they are dynamic constructs, they need to be continuously verified during an interaction. We investigate under which conditions conceptual pacts are transferred across situations, causing extra cognitive load and misalignment between partners. With our approach, we challenge extant theories of alignment (postulating a common ground of mutually shared knowledge between interacting partners). Instead, we propose a lightweight and non-recursive alignment mechanism, i.e. the dynamic and interactive establishment of situation-specific conceptual pacts. Further, we seek to elucidate the role that memory plays in the process of alignment and how it drives the incremental analysis of a situation and allows interacting partners to establish a context- and situation-specific conceptual pact and to infer what a word or linguistic construction means in the given context.
In particular, we pose the following research questions:
- Do recurring routine foster the emergence of long termed representations?
- Changes in which aspects of a situation can lead to misalignment and how does this affect learning?
- How does familiarity with a particular interactional structure reduce cognitive load and foster conceptual alignment in communication?
To answer these questions, we will use a variety of research methods. We will conduct experimental studies with children to investigate the effect of recurring routines on word learning performance. In addition, we intend to develop computational models that demonstrate how contextualized word memories emerge and how they facilitate the interpretation of language in a context. These experiments and models will contribute to explain how conceptual pacts are established, how they are bound to specific contextual conditions and under which circumstances they can be transferred or even overridden.