EventRelationDataset

Data and scripts for building a dataset of relation between events from texts, covering different types of relations: (Causality, Enabling, Prevention, Intention, Not causality).

Dataset

The Event Relation dataset is released under the TimeML format, extending it with a new RLINK tag for expressing several type of relations, specified in the relType field in this way:

<RLINK eventInstanceID="ei264" lid="l42" relType="prevention" relatedEventInstance="ei268" />

It has been realised by re-annotating two existing datasets:

• TimeBank 1.2
• The EventCausality Dataset

Outline of the repository

Scripts are provided in Jupyter Notebooks, in which all required function to run the whole code are in the same file.

• datasets includes the final dataset in TimeML format;
• generation includes the scripts for realising the dataset:
  • Get_relation_*.ipynb: to extract relations from datasets and save them in CSV files (Raw data without annotation) with different configurations for different datasets formats.
  • generate LINKS.ipynb: extract all pairs of events and the relation between them from CSV annotated files.
• annotation_csv: the csv files used during the annotation process