I benefited from an STSM in England, where I developed an Irish resources targeting noun compounds. The impact of these resources is manifold. They are used in assessing the quality of machine translation engines and large language models (LLMs). They underlie a PhD project exploring rare and diverse phenomena that remain a challenge with current NLP models. They pave the way towards new collaboration between Ireland and France concerning LLM evaluation. They allowed me personally to build the research line of my post-doctoral project and to connect with many successful senior researchers in this field.

Coordinating COST Actions, both UniDive and, previously, IC1207 PARSEME (2013-2017) has crucial importance for my career. Between the two Actions, I obtained a full professor position at a major French university, in probably the biggest NLP research group in France, and my COST experience was a major factor in this success. Currently UniDive is one of important international visibility factors in my institution (the LISN lab at Paris-Saclay University). In November 2024, a 6-year evaluation of our university (by the national HCERES certification body) took place. UniDive was selected as one of 2 activities to be highlighted during oral presentations of my research group in this evaluation. I'm also in the process of setting up an ERC proposal, which directly builds upon UniDive assets. In case of success, the impact on my career will be invaluable.

I went for an STSM to Bulgaria in September 2024. The visit was very useful for me to understand new annotation methods. Hopefully, there will be future collaboration between the Bulgarian Academy of Sciences and the LexiCon research group at my university.

Being part of UniDive has been an exciting and enriching journey for me, especially in advancing the representation of the Albanian language in Universal Dependencies and multilingual linguistic research. Collaborating with colleagues from diverse backgrounds has helped us tackle complex annotation challenges, turning dilemmas into learning opportunities. Seeing Albanian take its place in a global linguistic framework is both rewarding and motivating! Beyond technical advancements, UniDive has connected me with an inspiring community of researchers passionate about low-resource languages. The exchange of ideas and methodologies has broadened my perspective and reinforced the importance of linguistic diversity in both research and technology. Moving forward, I am excited to apply these insights to further develop Albanian linguistic resources, strengthen interdisciplinary collaborations, and contribute to innovative language technology solutions.

Since the beginning of UniDive, the Universal Dependencies collection of treebanks was extended by 32 new languages (22 co-created by UniDive members), all of them low resourced: Azerbaijani, Bavarian, Bororo, Cappadocian, Egyptian, Georgian, Gujarati, Haitian Creole, Hausa, Highland Puebla Nahuatl, Latgalian, Macedonian, Malayalam, Middle French, Old Irish, Ottoman Turkish, Paumari, Sinhala, Western Sierra Puebla Nahuatl, Xavante, Zaar, Abaza, Abkhaz, Classical Armenian, Gheg, Kyrgyz, Luxembourgish, and Maghrebi Arabic-French, Nheengatu, Telugu-English, Tswana, and Veps.

My involvement in UniDive has created opportunities to work on dependency parsing for low-resource languages, specifically Amharic and Gedeo. This experience has expanded my professional network and shifted my career focus from pure linguistics to a hybrid field integrating linguistics and computer science. My contribution to Universal Dependencies (UD) began with treebanks for Amharic and Gedeo, and I have drafted a project titled Parsing Amharic and Gedeo UD Treebanks (using MaChAmp, Trankit, and UDPipe). This project is a direct result of my involvement in UniDive and demonstrates how the Action has shaped my research direction and professional growth.

Serving as a Task leader in this Action is being one of the most significant networking opportunities of my career so far, not least due to the large number of actively engaged participants (21 people at the last Task meeting). If successful, the Task will lead to at least one highly collaborative publication with high visibility due to the large number of authors, and to a set of guidelines that will be used by treebank creators for years to come.

The participation of the members of the Institute of Humanities in this COST action has contributed greatly to involvement of both early-career and more experienced researchers into the scientific activities of the department. This will also enhance internationalization of the research at our university, which is in line with our institutional strategic goals. The researchers have expanded their network of cooperation and developed their research skills, which helps the accreditation of the study programmes we deliver.

For me personally, the participation in UniDive COST Action already had, and will continue to have a great impact on my career. Being involved in the organization of the shared task expanded my network of close collaborators to include some of the leading researchers in NLP, and it will be much easier for me to engage in new collaborations even beyond the lifetime of the Action. In addition, coordinating a large scale project that involves a network of dozens of contributors is another piece of valuable experience that is available to only few in my career stage (last year PhD student).

UniDive has given me the opportunity to meet scholars specialised in computational linguistics and to introduce concepts and methods from this discipline into Egyptian philology. Firstly, I have started to develop the first morphosyntactic treebank for pre-Coptic Egyptian. Secondly, I have opened a new line of research in Egyptian philology dedicated to the identification and analysis of multiword expressions in Egyptian. These two assets will contribute to the development of automatic morphosyntactic analysis of Egyptian texts for philologists and automatic translation of Egyptian inscriptions for archaeologists and visitors to Egypt. This will also help me move forward in my academic career after completing the ongoing junior professorship program.

The ILSP & Archimedes/Athena RC team is working in developing speech corpora, STT models, UD treebanks and models for morphosyntactic annotation for several, both living and endangered dialects of Modern Greek that are located in a geographical area spreading from Italy to Syria. UniDive has provided an accommodating scientific environment with the ongoing discussions, community meetings and STSM tools.

I could personally benefit from an STSM to go to the University of Sheffield and work on nominal MWEs. I used this experience to involve students of MSc courses in NLP in the production and evaluation of multiword expression resources for Modern Greek with LLMs. This is a contribution to the students’ careers and to the issue of providing resources to Modern Greek that has no substantive lexica and only medium-size general purpose corpora.

A post-doctoral fellow from my team, Stavros Bompolas, works on Greek dialects with AI tools. UD treebanking is crucial for his research. He attended the UniDive 2024 Chisinau Training School and found it instrumental in shaping his research trajectory. The training provided essential skills and insights, particularly in the Universal Dependencies framework. This knowledge has been vital in building dialectal treebanks for Greek varieties. It has also enabled him to examine how the concept of diversity is applied in NLP, leading to more accurate and comprehensive analyses.

I am an Iranian researcher working on Persian language. It is more than 2 years that I have been working with UniDive teams and It has helped me improve my vision of the similarities and differences among languages. Working with language specialists from different countries helps me have a more accurate analysis in my work. I hope someday we can have a UniDive meeting in Iran.

We presented jointly a tutorial on the annotation of multiword expressions as part of the international conference CLIB-2024 (Computational Linguistics in Bulgaria), drawing on our joint work within UniDive (as well as previous COST Action PARSEME). Joint activities facilitate the long-term collaboration between the individual researchers, like us, as well as between our institutions, both from inclusiveness target countries. The tutorial participants were from Bulgaria, Albania, Romania, Croatia, Latvia, and Spain, most of them ITCs, and working on low-resourced languages.

I work on bridging the gap between under-resourced language varieties and reasonably resourced ones by exploiting the power of LLMs. With UniDive, I went for an STSM to the University of Turin in September 2024. The team has an expertise in offensive speech detection and in Universal Dependencies. I explored the aspect of offensive VMWEs in offensive speech detection, which is an understudied aspect in this domain.