A novel method is introduced for deriving FrameNet mappings from parsed corpus annotation. By meshing an annotation schema with a semantic calculation, a rich range of dependencies can be established without recourse to indexing. By associating grammatical relations (dependencies) to specific tree structures, basic local and non-local dependencies (argument/predicate, modifier/head, antecedent/pronoun, etc.) are exhaustively described. This index-less annotation can be subsequently processed by means of a mechanism of semantic calculation that reads dependencies from the structure and records them in predicate logic expressions. These logical expressions can then be used to generate derived indices. The result is a grammar-driven, exhaustive analysis of text into sets of target words and grammatically related dependents. Each set can serve as the co-domain for mapping FrameNet roles onto grammatical structures. The technique yields robust and flexible target/dependent pairings, and can be adapted to many different languages, illustrated here with an example from Japanese.
@InProceedings{HORN18.1, author = {Stephen Wright Horn ,Alastair Butler ,Iku Nagasaki and Kei Yoshimoto}, title = {Deriving mappings for FrameNet construction from a parsed corpus of Japanese}, booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)}, year = {2018}, month = {may}, date = {7-12}, location = {Miyazaki, Japan}, editor = {Tiago Timponi Torrent and Lars Borin and Collin F. Baker}, publisher = {European Language Resources Association (ELRA)}, address = {Paris, France}, isbn = {979-10-95546-04-7}, language = {english} }
Summary of the paper