Data-driven Approaches to Ancient Languages

This workshop invites contributions centered around digital approaches for ancient languages. The focal point of the workshop is the presence or absence of data of historically attested languages. While the current landscape of computational linguistics is dominated by large-language models, such as the generative GPT model underlying ChatGPT, which is trained on an inconceivable amount of data, these systems, despite their apparent infallibility, do face challenges. Particularly, their performance is contingent on the availability of substantial training data, a condition not typically met by historically attested languages. Notably, our recent findings highlight that more traditional machine learning approaches, particularly into processing languages like Medieval Greek, yield superior outcomes compared to contemporary transformer-based techniques. This underscores the importance of linguistic expertise in computational approaches to lower-resourced languages. Profound knowledge of the corpus at hand is imperative, emphasizing the necessity for researchers of such lower-resourced languages to engage in knowledge exchange.

With the DAAL workshop, we aim to bring together a broad field of researchers engaged in digital approaches to pre-modern languages. This way, we hope to provide a forum to further advance the thriving research domain of NLP for ancient languages, where researchers and practitioners can meet and discuss their latest work. We hope to foster discussions and exchange ideas on shared challenges in language processing across various ancient languages.