Computational stylistics using artificial neural networks
SG Waugh · Literary and Linguistic Computing · 2000
Previous work in using Artificial Neural Networks for computational stylistics has concentrated on using large, arbitrary network structures. This paper examines the use of the Cascade-Correlation algorithm for the construction of minimal networks. We find that a number of problems in computational stylistics with a large number of variables, but a limited number of training examples may be solved successfully without resorting to large networks. The issue of redundancy in the data is also considered. 1. Introduction to Artificial Neural Networks One recent addition to the tools available for computational stylistics, or stylometry, is that of the Artificial Neural Network (ANN). These are computational methods loosely based on the concept of the neurons within the brain, the idea being that simple, trained processing elements will result in much more complicated behaviour when used in combination. Fig. 1 shows an example of the structure of a typical ANN. The majority of ANNs may b...