Simulating single word processing in the classic aphasia syndromes based on the Wernicke–Lichtheim–Geschwind theory

An unsupervised learning method for representing simple sentences

“…This involves the reception pathway (discussed herein), a cognitive facility for manipulating meanings gleaned from the input pathway, and a production pathway to allow the model to "speak". After such a model is completed and successfully trained, we intend to study the effects of "lesioning" the model [1]. The data obtained from these tests will be directly comparable to data from human aphasia patients, which will either validate the model or direct further research in modifying it to more closely mimic the behavior of the human language faculty.…”

“…They are thus recurrent neural networks that learn via competitive, unsupervised Hebbian training rules (described below). This type of layer has been found to mimic some organizational properties of biological cortex [1], [6].…”

“…1) begins in the input layer, a vector of binary neural elements upon which input phonemes are imposed in sequence. Each phoneme is represented as a 34-bit feature vector, following previous work [1], [2]; see Fig. 10 in [1] for the full list of phonemes and associated feature vectors.…”

“…Each phoneme is represented as a 34-bit feature vector, following previous work [1], [2]; see Fig. 10 in [1] for the full list of phonemes and associated feature vectors. After a phoneme is presented to the input layer, activity propagates via the weighted connections to the word layer, a special type of self-recurrent layer, which will be discussed below.…”

“…A recent neuroanatomically grounded model of the cerebral language centers [1], based on the WernickeLichtheim-Geschwind model, used one-shot multi-winner self-organizing maps to great effect. This model learned to associate 50 concrete nouns (presented as phoneme sequences) with representative visual images, such that it could repeat heard words and provide the name of seen pictures.…”