Institute of Formal and Applied Linguistics
Charles University, Czech Republic
Faculty of Mathematics and Physics
The project aims to efficiently create derivational networks for multiple languages at once, using machine learning and cross-lingual information transfer.
Unlike inflectional morphology, which has long been studied by many research teams, linguistic resources for derivational morphology are only being created in the last few years, with significant manual effort. We are currently aware of more than 50 lexical derivational networks (networks containing word-formation relations) for over 22 languages.
However, all projects known to us are creating them for each language individually (CELEX network contains multiple languages, but it was created independently for each one), not using the partial language independence of lexical semantics and the fact that related languages express identical ideas through similar means.
We believe that transfer of information between languages can help us decrease the amount of manual work required for creating derivational networks, and that this will help us cover more languages. This approach was successfully used for transferring inflectional and syntactic analysis (Rosa; Žabokrtský, 2017).