Toward More Rigorous Criteria for Semantic Role Identification

Abstract

Linguists have established semantic role classification systems of varying granularity, some of which have been applied to semantic knowledge engineering for natural language processing (NLP). However, due to their foundations in diverse linguistic theories, these semantic resources exhibit significant discrepancies in both the categories and quantities of semantic roles they propose, making information sharing and cross-system compatibility particularly challenging.
This study returns to the ontological study of linguistics, addressing fundamental disagreements in semantic role research through three key aspects: (1) the definition and nomenclature of different semantic roles, (2) dual semantic roles, and (3) the "one-role-per-clause" principle. We argue that defining semantic roles should simultaneously consider both form and meaning, proposing that their identification and differentiation be grounded in the basic syntactic constructions of predicates. By integrating syntactic patterns with event-based relational functions, this paper aims to provide a more rigorous framework for resolving existing ambiguities in semantic role annotation.