Morphology Without Borders: Clause-Level Morphology

Abstract Morphological tasks use large multi-lingual datasets that organize words into inflection tables, which then serve as training and evaluation data for various tasks. However, a closer inspection of these reveals profound cross-linguistic inconsistencies, arise from the lack clear linguistic operational definition what is word, severely impair universality derived To overcome this deficiency, we propose to view morphology clause-level phenomenon, rather than word-level. It anchored in fixed yet inclusive set features, encapsulates all functions realized saturated clause. We deliver MightyMorph, novel dataset covering 4 typologically different languages: English, German, Turkish, Hebrew. derive 3 morphological tasks: inflection, reinflection analysis. Our experiments show are substantially harder respective word-level tasks, while having comparable complexity across languages. Furthermore, redefining provides neat interface with contextualized language models (LMs) allows assessing knowledge encoded their usability Taken together, work opens up new horizons study computational morphology, leaving ample space studying neural cross-linguistically.