Syntactico-semantic parsing is a process of uncovering the internal structure and informational content of sentences. In particular, it is a process of articulating what the entities and events in a given narrative are and what relationships are between them. System Text2Drs takes English narrative as an input and outputs a discourse representation structure in Neo-Davidsonian style. Text2DRS relies on lexical resource VerbNet for annotating the specific relations between relevant entities and events mentioned in the given narrative using the verb classes and thematic roles of VerbNet.
For example, given a narrative:
John travelled to the hallway. Sandra journeyed to the hallway.
Text2DRS generates output:
% r1, r2, r3, e1, e2
entity(r1). entity(r2). entity(r3).
property(r1, "John"). property(r2, "hallway"). property(r3, "Sandra").
eventType(e1, "51.3.2-1"). eventType(e2, "51.3.2-1").
eventTime(e1, 0). eventTime(e2, 1).
eventArgument(e1, "Theme", r1). eventArgument(e1, "Location", r2). eventArgumen(e2, "Theme", r3). eventArgument(e2, "Location", r2).
The software is available for download here (under MIT license):
MS project report and technical abstract on the system posted below provide the details:
Gang Ling. From Narrative Text to VerbNet-Based DRSes: System Text2DRS (PDF). Master Project Report, University of Nebraska Omaha, USA, Spring 2018
Gang Ling. From Narrative Text to VerbNet-Based DRSes (PDF). Presented as a poster in the North American Summer School on Logic, Language, and Information (NASSLLI-2018), Pittsburgh, USA, 2018
The project has been inspired by the ideas discussed in:
Yuliya Lierler, Daniela Inclezan and Michael Gelfond. "Action Languages and Question Answering" Proceeding of the 12th International Conference on Computational Semantics (2017)