A Novel Lightweight Tibetan Semantic Role Labeling Specification

Semantic role labeling is a fundamental research issue in natural language understanding, which is beneficial for various tasks such as information extraction, question answering, machine reading comprehension, and machine translation. The dataset is the driving force and support for achieving automatic semantic role labeling, while the development of corresponding specifications is the premise of constructing the semantic role labeling dataset. Currently the description and analysis of the Tibetan semantic role classification system and annotation specifications are not complete and clear enough. Built upon the existing Tibetan semantic role annotation specifications, this paper develops a novel lightweight Tibetan semantic role annotation specification. To deal with various complex language expressions, the specification firstly defines the selection range of predicates, and then categorizes 28 Tibetan semantic role types along with their detailed introduction. It also provides clear argument word priority in annotation and difficult case illustrations. Finally, the annotation is conducted based on the developed specification, and the feasibility of the specification is demonstrated by the consistency of annotation results.