Staffan Larsson: Grounding as a Side-effect of Grounding

Steffan Larsson’s talk sought to connect communicative grounding, and in particular semantic co-ordination, with a formal model of perception and meaning based on symbolic grounding. Faced with the challenge of integrating perceptual meanings and low-level perceptual data into formal semantics, they were attempting to address the problem using a Type Theory with Records (TTR) perceptron classifier.

The talk detailed efforts to develop a formal model of perception and meaning based on symbolic grounding.  Symbol grounding is the process by which we connect symbols to the world, in such a way that supports composition and learning.

Through the process of semantic co-ordination we attempt to e stablish the meanings of linguistic expression in interaction. This is achieved through a number of mechanisms such as corrective feedback, implicit and explicit word meaning negotiation and litigation, etc. Clarification requests and feedback allow us to communicate to an interlocutor that their interpretation is aligned with our own. This iterative process of semantic co-ordination is an integral aspect of communicative grounding, but, as Larsson suggests, can also lead to symbolic grounding.

As such, the development of a formal semantics using TTR, which aims to account for the connection between communicative grounding and symbolic grounding was undertaken. In interaction, symbolic grounding can result from communicative grounding of concrete referring expressions in a shared visual environment. However, developing a computational cognitive model that solves the binding problem of symbolic and non-symbolic representations in the brain is no mean feat.

Artificial Neural Networks are a form of connectionist model in which data is classified through a parallel computational process. Due to the distributed nature of processing, it has been argued that symbolic understanding cannot be directly constructed within connectionist networks and that the two models are diametrically opposed. The challenge of instantiating the sources of power of symbolic computation within a fully connectionist system has been the concerns of many leading cognitive scientists, with many adopting various hybrid approaches.

He described a study they had conducted using a TTR perceptron classifier to identify left from right and act as a proof of concept for the system.

Summary by Shauna Concannon