We develop a connectionist approach to processing in quasi-regular domains, as exemplified by English word reading. A consideration of the shortcomings of a previous implementation (Seidenberg & McClelland, 1989, Psych. Rev.) in reading nonwords leads to the development of orthographic and phonological representations that capture better the relevant structure among the written and spoken forms of words. In a number of simulation experiments, networks using the new representations learn to read both regular and exception words, including low-frequency exception words, and yet are still able to read pronounceable nonwords as well as skilled readers. A mathematical analysis of the effects of word frequency and spelling-sound consistency in a related but simpler system serves to clarify the close relationship of these factors in influencing naming latencies. These insights are verified in subsequent simulations, including an attractor network that reproduces the naming latency data directly in its time to settle on a response. Further analyses of the network's ability to reproduce data on impaired reading in surface dyslexia support a view of the reading system that incorporates a graded division-of-labor between semantic and phonological processes. Such a view is consistent with the more general Seidenberg and McClelland framework and has some similarities with---but also important differences from---the standard dual-route account.

Deep dyslexia is an acquired reading disorder marked by the occurrence of semantic errors (e.g., reading RIVER as "ocean"). In addition, patients exhibit a number of other symptoms, including visual and morphological effects in their errors, a part-of-speech effect, and an advantage for concrete over abstract words. Deep dyslexia poses a distinct challenge for cognitive neuropsychology because there is little understanding of why such a variety of symptoms should co-occur in virtually all known patients. Hinton and Shallice (1991) replicated the co-occurrence of visual and semantic errors by lesioning a recurrent connectionist network trained to map from orthography to semantics. While the success of their simulations is encouraging, there is little understanding of what underlying principles are responsible for them. In this paper we evaluate and, where possible, improve on the most important design decisions made by Hinton and Shallice, relating to the task, the network architecture, the training procedure, and the testing procedure. We identify four properties of networks that underly their ability to reproduce the deep dyslexic symptom-complex: distributed orthographic and semantic representations, gradient descent learning, attractors for word meanings, and greater richness of concrete vs. abstract semantics. The first three of these are general connectionist principles and the last is based on earlier theorizing. Taken together, the results demonstrate the usefulness of a connectionist approach to understanding deep dyslexia in particular, and the viability of connectionist neuropsychology in general.

Distributed representations are attractive for a number of reasons. They offer the possibility of representing concepts in a continuous space, they degrade gracefully with noise, and they can be processed in a parallel network of simple processing elements. However, the problem of representing nested structure in distributed representations has been for some time a prominent concern of both proponents and critics of connectionism [Fodor and Pylyshyn 1988; Smolensky 1990; Hinton 1990]. The lack of connectionist representations for complex structure has held back progress in tackling higher-level cognitive tasks such as language understanding and reasoning. In this thesis I review connectionist representations and propose a method for the distributed representation of nested structure, which I call "Holographic Reduced Representations " (HRRs). HRRs provide an implementation of Hinton's [1990] "reduced descriptions". HRRs use circular convolution to associate atomic items, which are rep...

This paper presents a survey of connectionist inference systems.

We show that by introducing an appropriate schema concept and exploiting the higher-level features of a resonance function in a neural network it is possible to define a form of nonmonotonic inference relation between the input and the output of the network. This inference relation satisfies some of the most fundamental postulates for nonmonotonic logics. The construction presented in the paper is an example of how symbolic features can emerge from the subsymbolic level of a neural network.

A puzzle in the philosophy of mind concerns the gap between symbolic and subsymbolic (neuron-like) modes of processing (e.g. Smolensky 1988). The aim of this paper is to overcome this gap by viewing symbolism as a highlevel description of the properties of (a class of) neural networks. Combining methods of algebraic semantics and nonmonotonic logic, the possibility of integrating both modes of viewing cognition is demonstrated. The main results are (I) that certain activities of connectionist networks can be interpreted as nonmonotonic inferences, and (II) that there is a strict correspondence between the coding of knowledge in Hopfield networks and the knowledge representation in weight-annotated Poole systems. These results (a) show the usefulness of nonmonotonic logic as a descriptive and analytic tool for analyzing emerging properties of connectionist networks, (b) single out certain logical systems by giving them a "deeper justification", and (c) pave the way for using connection...

We study the problem of learning to infer hidden state sequences of processes whose states and observations are propositionally or relationally factored. Unfortunately, standard exact inference techniques such as Viterbi and graphical model inference exhibit exponential complexity for these processes. The main motivation behind our work is to identify a restricted space of models, which facilitate efficient inference, yet are expressive enough to remain useful in many applications. In particular, we present the penalty-logic simpletransition model, which utilizes a very simple-transition structure where the transition cost between any two states is constant. While not appropriate for all complex processes, we argue that it is often rich enough in many applications of interest, and when it is applicable there can be inference and learning advantages compared to more general models. In particular, we show that sequential inference for this model, that is, finding a minimumcost state sequence, efficiently reduces to a single-state minimization (SSM) problem. We then show how to define atemporal cost models in terms of penalty logic, or weighted logical constraints, and how to use this representation for practically efficient SSM computation. We present a method for learning the weights of our model from labeled training data based on Perceptron updates. Finally, we give experiments in both propositional and relational video-interpretation domains showing advantages compared to more general models. 1.