ExpectedOutcome:
Projects are expected to contribute to the following outcomes:
- Development of natural language understanding and interaction in advanced language technologies based on context-aware language models able to further integrate long-term general knowledge and derive meaning in order to develop automated reasoning and enhanced interaction skills;
- Effective multilingual and bias-controlled language models, capable of learning from smaller language corpora, efficient in computing and respectful of European values (i.e., privacy, non-discrimination, robustness in legal, ethical and technical terms, reliability and trustworthiness, interpretability and explainability, security and safety);
- AI systems and solutions based on novel multilingual pre-trained language models that have assimilated cross-language and cross-cultural knowledge through textual and speech input;
- Higher uptake of innovative language technology solutions by European companies, providing extensive language coverage[1] of AI-enabled applications and services in Europe.
Scope:
As AI becomes increasingly more performant, there is growing potential for humans to directly use and benefit from smarter systems. Effective AI-based human-machine interaction and collaboration relies on grasping real meaning from natural languages, recognising gestures and activities, understanding intention, creating and maintaining shared mental models and designing multi-step interactions. Reciprocally, truly natural interaction between people and machines is essential for future AI-enabled systems across all application areas and domains.
Envisaged AI solutions should address one or both of the following areas:
- Improve context-aware human-machine interaction to increase understanding and exploitation of the interaction context and content in multimodal settings, thus increasing responsiveness of interactive AI solutions, such as smart assistants, conversational and dialogue systems, content generation models, etc.
- Support and enhance seamless human-to-human communication across languages e.g. by means of automatic translation or interpretation (incl. automatic subtitling) in real time with a greater understanding of the communication context and the meaning involved in it.
Multidisciplinary research activities should address at least one of the following:
- Developing novel methods and techniques for producing context-aware models, which incorporate factual-based structured and unstructured knowledge in broader situational and temporal information, and continual learning to achieve natural behaviour and reasoning in all intended settings.
- Improving large pre-trained multilingual language models to cover a large set of languages[2], with a high level of natural language understanding and the ability to efficiently add more languages, including low-resource ones, via transfer or language-independent learning methods.
- Improving language-independent and bias-controlling algorithms and methods for language model training and usage efficiency in terms of data, time and energy consumption while retaining performance, accuracy and general usability.
- Developing language representations, encompassing an effective combination of multilingual, symbolic and sub-symbolic knowledge and allowing systems to perform cross-cultural reasoning in various contextual tasks.
Proposals should involve appropriate expertise in all the relevant disciplines, such as data science, computer science, computational linguistics, machine learning and natural language processing. Particular attention should be paid to control gender or other biases in language models.
Research should build on existing standards, contribute to standardisation and result in findable, accessible, interoperable and reusable research data including metadata schemas and ontologies.
All proposals are expected to embed mechanisms to assess and demonstrate progress (with qualitative and quantitative KPIs, benchmarking and progress monitoring, as well as illustrative application use-cases demonstrating concrete potential added value), and share communicable results with the European R&D community, through the AI-on-demand platform, Common European Data Spaces (especially the dedicated Language Data Space) and other relevant Member States’ initiatives, such as Open GPT-X, and if necessary other relevant digital resource platforms in order to enhance the European AI, Data and Robotics ecosystem through the sharing of results and best practice.
Proposals are also expected to dedicate tasks and resources to collaborate with and provide input to the open innovation challenge under HORIZON-CL4-2023-HUMAN-01-04 addressing natural language understanding and interaction. Research teams involved in the proposals are expected to participate in the respective Innovation Challenges. This topic implements the co-programmed European Partnership on AI, data and robotics.
Specific Topic Conditions:
Activities are expected to start at TRL 2 and achieve TRL 5 by the end of the project – see General Annex B.
[1]All official EU as well as socially and commercially relevant languages.
[2]Focus on all official EU as well as socially and commercially relevant languages.
