engleski

Towards Semantic Role Labeling in Slovene and Croatian

In the paper, the semantic role labeling framework is presented, which was developed within the project Semantic Role Labeling in Slovene and Croatian. The main goal of the project was the development of an annotated corpus to be used as training data for supervised machine learning systems. In building this framework we follow the path of previous SRL endeavors such as PDT, Vallex, FrameNET, Propbank etc. In compiling the list of semantic roles and their respective formal descriptions, we follow the approach developed by Prague Dependency Treebank, PDT. The paper describes both corpora used for semantic role annotation, as well as tools used in manual annotation tasks. Special attention is directed towards the description of the experimental automatic semantic role labeling based on supervised machine learning methods, and to its possible improvements. A preliminary quantitative analysis is performed for both languages (in terms of verbs range and frequencies, semantic roles, and typical syntactic-semantic patterns for the most frequent verbs).

semantic role labeling, Slovene, Croatian

nije evidentirano