“Joy” and “Fear” in Thomas Bernhard’s autobiographies: Aspects of a Computational Sentiment Analysis

Abstract

This pilot-study of a computational analysis of literary texts presents the results of aspects of a “sentiment analysis”. The data of analysis are the autobiographies of the Austrian novelist Thomas Bernhard. The primary object of attention are the sentiments “joy” and “fear”. We elaborate on and demonstrate the impact of several preprocessing procedures, describe the characteristics of the dictionary and the annotations of its entries conceived and used for analysis. We specify the general methodology and the steps involved for quantifying of its result by the use of the functions of the R-package “Quanteda”. The descriptive output of the procedures is examined with several statistical measures to compare the counts of “joy” vs “fear” that were found in the texts individually, contrastively and in combination as a corpus. We conclude that there is a proportional and relative difference between the frequencies of the sentiments of the individual texts, but that this observation is insignificant if interpreted on the basis of the non-parametric Wilcoxon rank-sum test. A “goodness of fit” test, on the other hand, shows that the two sentiments show a homogeneous distribution across the corpus