A computational approach to lexical polysemy in Ancient Greek

Abstract

Language is a complex and dynamic system. If we consider word meaning, which is the scope of lexical semantics, we observe that some words have several meanings, thus displaying lexical polysemy. In this article, we present the first phase of a project that aims at computationally modelling Ancient Greek semantics over time. Our system is based on Bayesian learning and on the Diorisis Ancient Greek corpus, which we have built for this purpose. We illustrate preliminary results in light of expert annotation, and take this opportunity to discuss the role of computational systems and human analysis in a complex research area like historical semantics. On the one hand, computational approaches allow us to model large corpora of texts. On the other hand, a long and rich scholarly tradition in Ancient Greek has provided us with valuable insights into the mechanisms of semantic change (cf. e.g. Leiwo, M. (2012). Introduction: variation with multiple faces. In Leiwo, M., Halla-aho, H., and Vierros, M. (eds), Variation and Change in Greek and Latin, Helsinki: Suomen Ateenan-instituutin säätiö, pp. 1–11.). In this article, we show that these qualitative analyses can be leveraged to support and complement the computational modelling.

Room to Glo: A Systematic Comparison of Semantic Change Detection Approaches with Word Embeddings

“…Many studies rely on hand-picked examples (e.g., Wijaya and Yeniterzi, 2011;Rodda et al, 2017) or human judgements (e.g., Tredici et al, 2018). Some studies have performed evaluations based on dictionary data (e.g., Cook et al, 2014;Basile and McGillivray, 2018), manual annotation of dictionary senses in corpora (McGillivray et al, 2019), and manual annotation of word types (Kenter et al, 2015), but this approach is not well-suited for recent, yet-to-berecorded changes.…”

The GLAUx corpus: methodological issues in designing a long-term, diverse, multi-layered corpus of Ancient Greek

“…data that differs considerably from the training data. Not all these factors might be equally problematic: for the computational modelling of Greek lexical meaning, for example, McGillivray et al (2019) found that genre is a more important factor than time, and argue that "literary Classical Greek is conservative when it comes to lexical semantics" (I also found similar results in my own experiments with meaning processing: see Keersmaekers 2020b: 119). As a complicating factor, there is a complex interplay between genre, diachrony and dialectal variation in literary Greek: some examples include Atticistic tendencies in postclassical Greek texts (i.e.…”

When Time Makes Sense: A Historically-Aware Approach to Targeted Sense Disambiguation

“…Recent years have seen an increased interest on the application of such methods for modelling semantic change to DH research, primarily using unsupervised methods (McGillivray et al, 2019;Soni et al, 2021). However, in order to be useful, analyses for DH research require a high degree of granularity on highly complex datasets and this has not yet been achieved by state-of-the-art methods.…”