A more interpretable method for the analysis of an author’s language

Description

Computational authorship analysis methods have a track record of success but two of their significant limitations are that they do not have a plausible linguistic explanation, at least on the surface, and that it is hard for an analyst to interpret the features that they use. A significant advance in addressing these two problems has been recently made by Nini et al. (2024), where a new algorithm called LambdaG is introduced. This new algorithm is compatible with Nini’s (2023) Theory of Linguistic Individuality, which, in turn, is based on Cognitive Linguistic theories of language processing. The algorithm, which is also fully embedded within the Likelihood Ratio Framework for forensic sciences (Ishihara & Carne 2022), exploits the notion of a language model to build the grammar of a specific individual. Then, the LambdaG score is calculated as the ratio of the likelihoods of the disputed text being ‘grammatical’ for the candidate author vs. a reference population. Because the algorithm assigns a score for each token, it is possible to fully visualise the output of the analysis. This talk will explain how this method works and then report on its application to a corpus of 19th century authors, thus also demonstrating the use of LambdaG in literary contexts to study and understand the language of a particular author.

References
Ishihara, Shunichi & Michael Carne. 2022. Likelihood ratio estimation for authorship text evidence: An empirical comparison of score- and feature-based methods. Forensic Science International 334. 111268. https://doi.org/10.1016/j.forsciint.2022.111268.
Nini, Andrea. 2023. A Theory of Linguistic Individuality for Authorship Analysis (Elements in Forensic Linguistics). Cambridge, UK: Cambridge University Press.
Nini, Andrea, Oren Halvani, Lukas Graner, Valerio Gherardi & Shunichi Ishihara. 2024. Authorship Verification based on the Likelihood Ratio of Grammar Models. arXiv. https://doi.org/10.48550/arXiv.2403.08462.

Period	7 Jun 2024
Event title	Corpus Linguistics Symposium: Style and Authorship
Event type	Other
Location	Leeds, United KingdomShow on map
Degree of Recognition	International