CLASSLA-express 2.0 workshop

After the initial strong reactions from the broader community, including linguists, to the release of ChatGPT as an easily accessible chat interface, several studies have been conducted to examine how this AI tool performs in lexicography-related tasks (De Schryver 2023; Rundell 2023; Jakubicek and Rundell 2023; Fuertes-Olivera 2024; Lew 2024). These studies primarily focus on its success rates in executing linguistic tasks, its applicability in applied fields such as lexicography, and insights gained from prompt engineering to optimise results. Lexicographers have long relied on corpus data, valuing its reliability and the transparency of sources that provide insights into actual language use. However, corpus search tools have become increasingly sophisticated, enabling greater automation in lexicographic work by generating high-quality dictionary examples, identifying typical collocations, extracting lemma lists, and much more. These tools also ensure direct interlinking between lexicographic resources and corpora. This situation highlights a persistent problem with digital technologies: the under-representation of low-resource languages. AI tools, as anticipated, are less effective with these languages. This challenge also impacts South Slavic languages, though some have recently made progress toward mid-resource classification. A major step forward in this regard was the release of the CLASSLA web corpora for South Slavic languages in 2024 (Ljubešić and Kuzman, 2024). In addition, AI tools have also been used in studies on South Slavic languages (Filipović Petrović and Beliga 2024; Gantar 2024; Gapsa et al. 2024; Kosem et al. 2024). In 2024, CLARIN Slovenia organised seven workshops across five countries (Ljubešić et al. 2024), with the main goal of sharing knowledge using corpus querying tools. Based on insights into which linguistic and lexicographic tasks corpora handle effectively and which still require manual human work, we have chosen to launch a new series of workshops that will combine these strengths with AI tools. The aim is to test the capabilities of corpora and LLMs in performing various tasks relevant to lexicographers and linguists. The linguistic data to be examined includes phraseological units, i.e., multi-word expressions, in South Slavic languages. In doing so, we aim to contribute to two potentially challenging topics: examining how language technologies perform on low and middle-resource languages such as South Slavic languages and exploring how far they have advanced in handling the ever-challenging multi-word expressions and the ambiguity they carry. This workshop, associated with the eLex conference, will specifically focus on tasks and discussions related to the dictionary-making process. The searches will be conducted on CLASSLA corpora using the NoSketch Engine concordancer and AI-driven interfaces such as ChatGPT 3.5 or Gemini. Each participant will work with tools in their chosen language (one of the CLASSLA languages: Slovenian, Croatian, Serbian, Macedonian, or Bulgarian). Participants will indicate which South Slavic language they want to work with when registering for the workshop. They will be divided into groups based on their language preference, each completing exercises and tracking results in their selected language. This workshop aims to establish a methodologically sound approach to integrating two major tool sets, corpora and LLMs, in language research and lexicography.