Computationally discriminating literary from non-literary texts

Three computational linguistic methods are presented to discriminate literary from non-literary texts. In the first study, a hierarchical clustering technique of results obtained from Latent Semantic Analysis showed a clustering of literary versus non-literary texts. The second study used the frequencies of shared bigrams across the text, resulting in a 100% correct classification of literary versus non-literary texts. The third study used unigrams yielding a 94% correct classification into literary versus non-literary texts. The final two studies using a larger sample of texts showed that the high classification performance cannot be attributed to specific texts. These findings provide evidence that distinguishing literature from non-literature can be done with high accuracy and with relatively simple computational linguistic techniques.