AMALEU: a universal representation of language based on automatic learning

Why is automatic translation between English and Portuguese significantly better than automatic translation between Dutch and Spanish? Why does voice recognition work better in German than in Finnish? The main problem is the insufficient amount of labelled data for training in both cases. Although the world is multimodal and highly multilingual, speech and language technology provide a satisfactory response for all languages. We need better learning methods that take advantage of the advances in some modalities and languages to benefit others.

The aim of AMALEU is to automatically learn a universal representation of language, whether it is with voice or text. This can be used in artificial intelligence applications for different languages. The project will use unlabelled information sources and language information. The project focuses on the challenge of learning from few resources and an approach to automatic multilingual translation.

AMALEU will have an impact on highly multidisciplinary communities of specialists in computer sciences, mathematics, engineering and linguistics who work with natural language understanding applications, natural language and speech processing.

AMALEU is funded by the Spanish Ministry of Economy and Competitiveness (MINECO), as part of the Europe Excellence programme. The project lasts two years (January 2019 – December 2020).