Modular Approach to Automating Morphological Components in Grammar Engineering

Abstract

Creating formal grammars is a time-consuming and complex task. We present a method to automatically create the morphological components of a formal grammar in Grammatical Framework. Our method is linguistically interpretable and modular, consisting of three stages: paradigm construction, extraction of inflectional classes, and prediction of inflectional classes. The modular structure allows human interventions after each stage. Moreover, our method supports encoding pre-existing language knowledge in form of Python APIs. Experiments show that automatically extracted morphological rules yield results comparable with manual grammars and that incorporating prior linguistic knowledge leads to improvement in low-resourced scenarios. Our findings show that our method simplifies the process of grammar development while preserving quality and interpretability.

Resources

Details