Blog

Yesterday I had the pleasure to deliver a Keynote talk at the Interdisciplinary Perspectives: Bridging Sociological Studies in the Digital Age conference, organised by the amazing Digital Humanities PhD students at King’s College London.

In the talk I argued that performance in a Natural Language Processing task, namely authorship analysis, can be significantly increased if insight and knowledge from (Cognitive) Linguistics is taken in consideration. Even though the field of NLP is now at a stage of wondering to what extent Linguistics is still needed, I make a case that the often quoted statement that “Every time I fire a linguist, the performance of the system goes up” tends to apply to certain strand of Linguistics that do not engage with the study of real-life language usage. Instead, Cognitive Linguistic Usage-Based approaches to the study of language are very compatible with modern advances in NLP. As shown in this case study, the application of Cognitive Linguistics theoretical frameworks can lead to better trade offs between computational complexity and performance, reducing the number of computational steps, processing time, and model parameters.

The slides of my talk are available here: https://doi.org/10.5281/zenodo.11583694.

Tomorrow afternoon I’ll give a keynote talk at the Corpus Linguistics Symposium: Style and Authorship at the University of Leeds. My talk will be on how our new authorship verification method, LambdaG, can be more transparent in visualising the important features that identify an author than the state of the art and I will use Dickens as a case study. The event is hybrid and the talk is going to be recorded. For more information, the webpage of the event is: https://www.latl.leeds.ac.uk/research-satellites/corpus-linguistics/clstyleauthorship/.