Year: 2019
Type: article
Abstract: This study belongs to the domain of authorship analysis (AA), a discipline under the umbrella of forensic linguistics in which writing style is analysed as a means of authorship identification. Due to advances in natural language processing and machine learning in recent years, interest in computational methods of AA is gaining over traditional stylistic analysis by human experts. It may only be a matter of time before the software will assist, if not replace, a forensic examiner. But can we trust its verdict? The existing computational methods of AA receive critique for the lack of theoretical motivation, black box methodologies and controversial results, and ultimately, many argue that these are unable to deliver viable forensic evidence. The study replicates a popular algorithm of computational AA in order to open one of the existing black boxes. It takes a closer look at the so-called “bag-of-words” (BoW) approach – a word distributions method used in the majority of AA models, evaluates the parameters that the algorithm bases its conclusions on and offers detailed linguistic explanations for the statistical results these discriminators produce. The framework behind the design of this study draws on multidimensional analysis – a multivariate analytical approach to linguistic variation. By building on the theory of systemic functional linguistics and variationist sociolinguistics, the study takes steps toward solving the existing problem of the theoretical validity of computational AA.
Author: Alesia Locker
Institution: Aarhus University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W3006456934