Date of Degree


Document Type


Degree Name





David M. Rosenthal

Committee Members

Jakob Hohwy

Eric Mandelbaum

Subject Categories

Cognitive Psychology | Computational Neuroscience | Philosophy of Language | Philosophy of Mind


connectionism, deep learning, generative models, mental representation, natural language processing


Connectionism is an approach to neural-networks-based cognitive modeling that encompasses the recent deep learning movement in artificial intelligence. It came of age in the 1980s, with its roots in cybernetics and earlier attempts to model the brain as a system of simple parallel processors. Connectionist models center on statistical inference within neural networks with empirically learnable parameters, which can be represented as graphical models. More recent approaches focus on learning and inference within hierarchical generative models. Contra influential and ongoing critiques, I argue in this dissertation that the connectionist approach to cognitive science possesses in principle (and, as is becoming increasingly clear, in practice) the resources to model even the most rich and distinctly human cognitive capacities, such as abstract, conceptual thought and natural language comprehension and production.

Consonant with much previous philosophical work on connectionism, I argue that a core principle—that proximal representations in a vector space have similar semantic values—is the key to a successful connectionist account of the systematicity and productivity of thought, language, and other core cognitive phenomena. My work here differs from preceding work in philosophy in several respects: (1) I compare a wide variety of connectionist responses to the systematicity challenge and isolate two main strands that are both historically important and reflected in ongoing work today: (a) vector symbolic architectures and (b) (compositional) vector space semantic models; (2) I consider very recent applications of these approaches, including their deployment on large-scale machine learning tasks such as machine translation; (3) I argue, again on the basis mostly of recent developments, for a continuity in representation and processing across natural language, image processing and other domains; (4) I explicitly link broad, abstract features of connectionist representation to recent proposals in cognitive science similar in spirit, such as hierarchical Bayesian and free energy minimization approaches, and offer a single rebuttal of criticisms of these related paradigms; (5) I critique recent alternative proposals that argue for a hybrid Classical (i.e. serial symbolic)/statistical model of mind; (6) I argue that defending the most plausible form of a connectionist cognitive architecture requires rethinking certain distinctions that have figured prominently in the history of the philosophy of mind and language, such as that between word- and phrase-level semantic content, and between inference and association.