Does Connectionism undermine Fodor’s Language of Thought Hypothesis?

Abstract

In 1975, Fodor hypothesised that thought is structured in much the same way as language^{. 1} Thoughts have semantics, a combinatorial syntax, and store information symbolically. In the 1980s, Connectionism looked to undermine his view. It suggested that mental information is stored non-symbolically in neural nets; it was considered a “paradigm shift” for cognitive theories. ² In the 1990s, further work by Chalmers and Rowlands undermined Fodor’s Language of Thought Hypothesis. ³⁴⁵ Modern cognitive research into Deep Learning uses an inherently Connectionist framework.

This paper separates Fodor’s hypothesis from his arguments in its support. It argues that Fodor’s Language of Thought Hypothesis is still a legitimate theory of cognition. However, it accepts that Fodor’s arguments in favour of his hypothesis are fallacious. The paper examines three of Fodor’s arguments for a language of thought: the only game in town argument, the argument from systematicity and productivity, and the argument from isomorphism. ⁶⁷⁸⁹ It shows each to be flawed.

Further, this paper dismisses the dilemma Fodor and Pylyshyn present the Connectionist: that they must either merely implement his Language of Thought Hypothesis or concede that it is an inadequate theory of cognition. ¹⁰ he paper uses Chalmers’ Backpropagation Model, a system that encodes grammatical information without using symbols, to escape the dilemma. ¹¹

Throughout, I argue that despite successfully undermining his arguments, Connectionism does not undermine Fodor’s Language of Thought Hypothesis. I provide two positive reasons to upholding the Language of Thought Hypothesis. This paper concludes that – at present – neither Connectionism nor Fodor’s Language of Thought Hypothesis has undermined the other.