Dalvean, Michael2014-02-252014-02-250268-1145http://hdl.handle.net/1885/11422In this article I extend recent work on the application of computational linguistics to the analysis of poetry. The dataset consist of 85 canonical English poems and a matched control group of obscure poems. I use several content analysis dictionaries to create over 500 linguistic variables and then use machine learning to develop a classifier. The classifier, consisting of 7 linguistic variables is tested using 10-fold crossover validation. Classifier accuracy is 69%. Of the 7 variables, three confirm previous findings about the nature of how “successful” poems differ from less successful poems while four of the variables give further insight into the phenomenon. I then go on to rank the poems using the probability scores of the classifier and find that Blake's “A poison Tree” scores highest. I explain the ranking method as being a means of distilling the “literary” appeal from the “popular” appeal of the poems in the sample. Finally, I discuss the implications for the theory of poetry in general.43 pageshttp://www.sherpa.ac.uk/romeo/issn/0268-1145/author can archive pre-print (ie pre-refereeing);pre-print must be accompanied by set statementpoetrymachine learningcomputational linguistics2014-02-25