Sunoikisis Digital Classics: Computational Linguistics

Sunoikisis Digital Classics: Computational Linguistics
2 December 2021, 4.15pm - 5.45pm
Research Training

Convenors: Alek Keersmaekers (KU Leiven), Martina Astrid Rodda (Oxford) 

Youtube link:

All welcome

This session will introduce some questions and approaches in Computational Linguistics. We will start by discussing how the approaches we will examine today build on what we saw in the previous sessions on search tools, text analysis, treebanking and translation alignment. We will also give a broad overview of Computational Linguistics as applied to ancient languages: what are the main questions it tries to address and the work that has been done for Greek and Latin, an introduction to the most important concepts and a discussion of the challenges that Greek and Latin present. We will also include two case studies: the first will illustrate how a computational approach can be used to study literary features, specifically the behaviour of formulae in early Greek epic poetry (Homer, Hesiod, and the Homeric hymns). Quantitative data shows that the behaviour of recurring set phrases in early epic poetry is measurably different from both set phrases in non-epic material and recurring expressions in later epic. The second case study discusses so-called ‘transformer’-based approaches to natural language processing, which use neural networks trained on a large corpus to obtain detailed mathematical representations about the usage of a given word. It will show how Electra, a transformer model tailored to languages with a smaller corpus, can considerably improve the state of the art for natural language processing in Greek through a number of examples.


Valerie James
020 7862 8716