The HR-XT-XTEND is one of eight FSTP subprojects of a larger Horizon Europe funded project Unified transcription and translation for extended reality (UTTER). HR-XT-XTEND project aim is to develop a large language model (LLM) for the Croatian language that will be trained on a massive dataset of Croatian text. The project aims to build resources for XR models, extend XR models to a new language, and evaluate the LLM. The project goals are to collect at least 6 billion tokens of Croatian text and prepare that data for LLM training, create a LLM for the Croatian language using monolingual data only, and evaluate the LLM for downstream tasks. The experimental phase will focus on developing and evaluating the model architecture and training process. The training phase will be used to train the LLM. The integration phase will involve integrating the LLM into the UTTER platform. The project results will be accessible under permissive licenses to the research community and the public from the HR-CLARIN repository.
The results of the HR-XR-XTEND project were presented at the international linguistic conference MetaCol – “Collocations in the Light of Metaphors and Metonymies” organized by the MetaCol research team and held on 3 and 4 October 2024 at the Inter-University Centre in Dubrovnik. (2024-10-04)
HR-XR-XTEND project and its main result - HR-GPT Beta - was presented at the 35th Central European Conference on Information and Intelligent Systems (CECIIS2024) held in Varaždin, Croatia, 18-20 September 2024. (2024-09-20)
HR-XR-XTEND project was presented at the second edition of the International Conference ‘New Trends in Translation and Technology’ (NeTTT’2024) held in Varna, Bulgaria, 3 - 6 July 2024. (2024-07-07)
The research on application of the existing LLMs to the downstream task of HR-XR-XTEND project -- sentiment analysis, has been presented at the Main conference of the LREC-COLING2024. The conference was held in Turin, Italy and the paper "M2SA: Multimodal and Multilingual Model for Sentiment Analysis of Tweets" can be found here. (2024-05-27)
AddressUniversity of Zagreb, Faculty of Humanities and Social Sciences, Institute of Linguistics
Phone+385 1 4092142
E-maildasa.farkas [at] ffzg.unizg.hr
AcknowledgmentsThis work was funded from the European Union’s Horizon Europe Research and Innovation Programme under Grant Agreement No 101070631 "Unified transcription and translation for extended reality (UTTER)" and from the UK Research and Innovation (UKRI) under the UK government’s Horizon Europe funding guarantee (Grant No 10039436).