We explore the task of multi-source morphological reinflection, whichgeneralizes the standard, single-source version. The input consists of (i) atarget tag and (ii) multiple pairs of source form and source tag for a lemma.The motivation is that it is beneficial to have access to more than one sourceform since different source forms can provide complementary information, e.g.,different stems. We further present a novel extension to the encoder- decoderrecurrent neural architecture, consisting of multiple encoders, to better solvethe task. We show that our new architecture outperforms single-sourcereinflection models and publish our dataset for multi-source morphologicalreinflection to facilitate future research.
展开▼