31 octobre 2018
info:eu-repo/semantics/OpenAccess
Mathieu Dehouck et al., « A Framework for Understanding the Role of Morphology in Universal Dependency Parsing », HAL-SHS : linguistique, ID : 10670/1.xl49ni
This paper presents a simple framework forcharacterizing morphological complexity andhow it encodes syntactic information. In particular,we propose a new measure of morphosyntacticcomplexity in terms of governordependentpreferential attachment that explainsparsing performance. Through experimentson dependency parsing with datafrom Universal Dependencies (UD), we showthat representations derived from morphologicalattributes deliver important parsing performanceimprovements over standard wordform embeddings when trained on the samedatasets. We also show that the new morphosyntacticcomplexity measure is predictive ofthe gains provided by using morphological attributesover plain forms on parsing scores,making it a tool to distinguish languages usingmorphology as a syntactic marker from others.