What do character-level models learn about morphology? The case of dependency parsing
This addresses the problem of parsing morphologically-rich languages for NLP researchers, showing incremental improvements by combining character-level models with explicit morphology.
The paper investigates whether character-level neural models learn morphology by comparing them to an oracle with explicit morphological analysis across twelve languages, finding that while they have strengths, they struggle with disambiguating words, especially in case syncretism, and that explicitly modeling morphological case improves performance.
When parsing morphologically-rich languages with neural models, it is beneficial to model input at the character level, and it has been claimed that this is because character-level models learn morphology. We test these claims by comparing character-level models to an oracle with access to explicit morphological analysis on twelve languages with varying morphological typologies. Our results highlight many strengths of character-level models, but also show that they are poor at disambiguating some words, particularly in the face of case syncretism. We then demonstrate that explicitly modeling morphological case improves our best model, showing that character-level models can benefit from targeted forms of explicit morphological modeling.