Universality, diversity and idiosyncrasy
in language technology
CA21167 COST Action

10.45–11.00 Introduction to WG3 (slides)

11.00–11.30 Brainstorming on ideas and expectations

• What is most important for you in multilingual and cross-lingual NLP?
• What activities do you think we should prioritize?
• How can we work together to make progress towards our goals?

• Large language models are most important
• Articulating linguistic theories underlying tools
• Defining idiosyncrasy and diversity
• The user perspective is important
• Supporting low-resource languages through cross-lingual technology
• Supporting low-resource languages through annotation tools
• Supporting low-resource languages through data collection
• Supporting low-resource languages with semantics
• Tools for all languages – start with morphology
• Low-resource language is not a homogeneous concept
• Building resources for specific languages (Serbian)
• Linking corpus resources between languages
• Standardized tools applicable to different languages
• Evaluation of tools – coordinate with other WGs
• Tracking evaluation status for different types of tools
• Improved benchmarking and experimental design
• Organize shared tasks

11.30–12.00 Initial discussion on documentation of tools

• Which types of tools do we want to include?
• Where do we want to keep the documentation?
• How do we create this documentation/inventory?

• A huge multidimensional matrix
• A shared repository
• Tools shared between typologically similar languages
• Consider end users
• Too many languages have nothing – document what is missing rather than what exists
• Connect to CLARIN
• Flagship project on MWE
• Include all tools or be selective?
• What about commercial tools?
• What about tools without documentation?

• Define multidimensional taxonomy of tools for documentation
• Define infrastructure and procedure for creating documentation