AGILe: The First Lemmatizer for Ancient Greek Inscriptions

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5334–5344 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5334 AGILe: The First Lemmatizer for Ancient Greek Inscriptions Evelien de Graaf∗, Silvia Stopponi∗, Jasper Bos, Saskia Peels-Matthey and Malvina Nissim Centre for Language and Cognition Groningen, University of Groningen, The Netherlands {e.de.graaf.6,j.k.bos.1}@student.rug.nl {s.stopponi,s.peels,m.nissim}@rug.nl Abstract To facilitate corpus searches by classicists as well as to reduce data sparsity when training models, we focus on the automatic lemmatization of ancient Greek inscriptions, which have not received as much attention in this sense as literary text data has. We show that existing lemmatizers for ancient Greek, trained on literary data, are not performant on epigraphic data, due to major language differences between the two types of texts. We thus train the first inscription-specific lemmatizer achieving above 80% accuracy, and make both the models and the lemmatized data available to the community. We also provide a detailed error analysis highlighting peculiarities of inscriptions which again highlights the importance of a lemmatizer dedicated to inscriptions. Keywords: Digital Classics, lemmatization, Ancient Greek inscriptions 1 Introduction Lemmatization is a basic preprocessing step for automatic linguistic analysis. It is particularly helpful in the case of morphologically complex languages: by reconducting a large variety of wordforms to a single lemma, one can en- able meaningful generalizations, especially at the semantic level, which could otherwise remain opaque due to the spar- sity of each single wordform. This is even truer in the case of low-resource languages or varieties, where the amount of available data is limited, and possibly more sophisticated processing tools that might not need to rely on lemmatiza- tion are not (yet) available. We focus here on ancient Greek, a morphologically com- plex language, with rich nominal and verbal inflection. An- cient Greek is also relatively low-resource. While substan- tial effort has gone into collecting and digitizing all avail- able data, such as for the big projects Thesaurus Linguae Graecae (Pantelia, 2001) and the Perseus Digital Library (Smith et al., 2000), the existing collection can be consid- ered finite: apart of occasional discoveries of texts, often on stones and papyri, the corpus is not enlargeable. Another characteristic of the existing ancient Greek data is that the texts are written in different language varieties, sometimes very distant from each other, both synchronically and di- achronically. For example, the language of ancient Greek inscriptions,1 namely all the texts which were written on durable materials, such as stone, ceramic, metal and other materials, can differ substantially from that of literary texts, especially when it comes to orthography, morphology and dialectal variation. Most of the existing tools developed for ancient Greek at large are built on (and for) literary texts, and because of such differences they are not necessarily suitable for other kinds of ancient Greek texts. For literary ancient Greek several lemmatizers have been developed and made available, and the most popular on- line corpora of literary texts, such as the Thesaurus Lin- …

Content truncated to 500 words