Proto-Role Semantic Role Labeling for Khmer: A New Approach for Argument Classification

Abstract

This study examines the applicability of Dowty’s (1991) Proto-role theory to Khmer, a morphologically impoverished and under-resourced language, with the aim of exploring its potential contribution to semantic role labeling (SRL) in languages lacking overt case marking and agreement. Based on a manually annotated dataset of 81 Khmer sentences drawn from literary sources, the study analyzes how Proto-Agent and Proto-Patient entailments are distributed across diverse constructions, including serial verb constructions, passive clauses, psychological predicates, and pro-drop structures.
The findings suggest that Proto-role entailments can provide a flexible and empirically useful framework for argument classification in Khmer, even in the absence of explicit morphosyntactic cues. The study aims to make three limited but targeted contributions: (i) it presents the first systematic Proto-role analysis for Khmer, (ii) it offers empirical observations indicating that Proto-role mappings remain largely stable across complex predicate structures, and (iii) it proposes a pilot annotation strategy that may inform the future development of Khmer semantic role resources for computational applications. While exploratory in scope, these results point to the potential of Proto-role-based SRL as a foundation for semantic annotation in under-resourced languages.