Greimassian semiotics in authorship atribution: a contribution to forensic linguistics

Abstract

During criminal investigations and judicial processes, authorship analysis has been performed in order to answer the question “who wrote this text?”. Therefore, several studies on the topic have emerged, in the field known as forensic linguistics. Such studies are based on the idea that every author has a particular style. Yet, such an idea is problematic in cases in which the questioned text and the known texts gathered for comparison are very different in nature (they are generally different in terms of audience, genre, register, etc.); the issue is that sociolinguistics has long been demonstrating the presence of intra-speaker variation (e.g. Labov, 1966). Trying to overcome that, this paper considers that if the deeper levels in the generative trajectory are more abstract, presenting a smaller set of categories, they would be less variable. The idea is that the simpler the level, the lower will be the options available for the author to chose from, which will result in a greater chance to always choose the same thing. With this great chance to always make the same choices, if two different individuals repeatedly choose the same options, differently from one another, this means that each one's option has great discriminatory power. So, the aim of this study was to verify if the categories examined in semiotic analysis could distinguish authors in forensic contexts. Texts written by 4 authors were analyzed in software that generates and stores "tags" that can be extracted later, which allows the counting of data. Afterwards, similarities and differences between the authors were statistically measured using two coefficients: Jaccard and Yule, used to measure the level of similarity between samples. The results indicate that the hypothesis is confirmed.