Learning orthographic and phonological representations in models of monosyllabic and bisyllabic naming

Parallel Distributed Processing Theory in the Age of Deep Networks

Parallel distributed processing (PDP) models in psychology are the precursors of deep networks used in computer science. However, only PDP models are associated with two core psychological claims, namely that all knowledge is coded in a distributed format and cognition is mediated by non-symbolic computations. These claims have long been debated in cognitive science, and recent work with deep networks speaks to this debate. Specifically, single-unit recordings show that deep networks learn units that respond selectively to meaningful categories, and researchers are finding that deep networks need to be supplemented with symbolic systems to perform some tasks. Given the close links between PDP and deep networks, it is surprising that research with deep networks is challenging PDP theory.

Modelling reading development through phonological decoding and self-teaching: implications for dyslexia

The most influential theory of learning to read is based on the idea that children rely on phonological decoding skills to learn novel words. According to the self-teaching hypothesis, each successful decoding encounter with an unfamiliar word provides an opportunity to acquire word-specific orthographic information that is the foundation of skilled word recognition. Therefore, phonological decoding acts as a self-teaching mechanism or ‘built-in teacher’. However, all previous connectionist models have learned the task of reading aloud through exposure to a very large corpus of spelling–sound pairs, where an ‘external’ teacher supplies the pronunciation of all words that should be learnt. Such a supervised training regimen is highly implausible. Here, we implement and test the developmentally plausible phonological decoding self-teaching hypothesis in the context of the connectionist dual process model. In a series of simulations, we provide a proof of concept that this mechanism works. The model was able to acquire word-specific orthographic representations for more than 25 000 words even though it started with only a small number of grapheme–phoneme correspondences. We then show how visual and phoneme deficits that are present at the outset of reading development can cause dyslexia in the course of reading development.

Models of visual word recognition

•

I review models of visual word recognition and data used to evaluate them.

•

I focus on recent IA and mathematical/Bayesian models.

•

I explain how models process and represent letter order.

•

I suggest how competing models should be evaluated.

Reading is a complex process that draws on a remarkable number of diverse perceptual and cognitive processes. In this review, I provide an overview of computational models of reading, focussing on models of visual word recognition–how we recognise individual words. Early computational models had ‘toy’ lexicons, could simulate only a narrow range of phenomena, and frequently had fundamental limitations, such as being able to handle only four-letter words. The most recent models can use realistic lexicons, can simulate data from a range of tasks, and can process words of different lengths. These models are the driving force behind much of the empirical work on reading. I discuss how the data have guided model development and, importantly, I also provide guidelines to help interpret and evaluate the contribution the models make to our understanding of how we read.

Left- and right-hemisphere forms of phonological alexia

We studied the ability of patients with lesions arising from operation for an anterior or posterior (left or right) brain tumour to read a set of words and pronounceable nonwords. In line with previous works, we observed that damage to the left posterior or left anterior cortex can give rise to phonological alexia, where the reading performance of nonwords is affected more than that of words. More surprisingly, similar effects were found in the right posterior group. However, there were significant differences in the error types, for both complex and positional errors, between phonological alexic patients in the three location groups. The findings present difficulties for the position held by theorists of the triangle model that phonological alexia arises from impairments in the language production system or in a general-purpose orthographic-phonological translation system. They also pose new questions about the possible role of the right posterior cortex in letter sequence representation.