Using Semantic Information from Neural Networks to Detect Context-Sensitive Spelling Errors

This paper proposes a means of using the internal representations of an artificial neural network to represent the semantic contexts in which a word can appear. Once the network has been trained, its hidden layer activations are recorded as a representation of the average context in which a word can appear. This context can then be compared to the contexts in which a word appears in novel text to detect context-sensitive spelling errors. While no significant results are found in the trials described here, several modifications of the system are proposed that might prove promising in future