Atelier Théorique AT1

Connections and symbols

Version 2014- mise à jour 2015

Séance 1: setting up the debate

What is thought? From its inception, Cognitive Science explored two kinds of answers:

The first kind of answer is rooted in the study of logic as the foundation of (mathematical) reasoning, which gave rise to formal systems, formal grammars and automata theory. These theories were applied in Cognitive Science to the modeling of common-sense reasoning (symbolic AI) and language (formal & computational linguisics).

The second kind of answer is rooted into the study of the brain and of the property of large networks of simple analogical devices. It gave rise to the field of computational neuroscience which is applied to the modeling of many cognitive functions (perception, memory, action, etc) as well as of functional neurophysiological data.

These two lines of research gave rise in the 80s to a large scale confrontation, largely focussed on the study of language. The symbolic versus connectionist debate was ignited by the publication of two landmark volumes (Rumelhart & McClelland, 1986; McClelland & Rumelhard, 1986). Paralleled Distributed Processing was described as a new theory of cognition challenging the idea of symbolic computation that was at the center of debate in theoretical discussions about the mind. The response took the shape of a special cognition issue Connections and Symbols (Pinker & Mehler, 1988), as well as a wealth of empirical and theoretical results that have profoundly modified the field of Cognitive Science.

In this course, we will review some of the theoretical background to understand this debate, and discuss some of positions and arguments.

Topics covered:

PDF of session 1

Reference books

Séance 2: The devil is in the details

In this course, we will focus on papers that were central to the Connections vs Symbol debate. The Fodor & Pylyshyn (1988) paper raised a principled objection to Connectionnism in terms of the expressive and computational capacity of these models in order to adequately address language and thought processes. The two other papers: the recurrent networks (Elman, 1990), and the tensor products (Smolensky, 1990) are technical contribution which address some of the criticism of Fodor and Pylyshyn.

In this course, the students will present two of these papers; a small group will present the central argument of the paper, and the rest of the studentw will prepare questions.

To read and to prepare:

After the discussion of the papers, we will briefly review some of the more recent theoretical development on the question "what is thought?" as well as the status of the connection/symbol debate today.

Download PDF of session 2

Further readings