Welcome to the hub for the PropBank-Wikidata incorporation project. This repository contains the DWD Overlay (see documentation below), which is no longer being updated, in favor of migrating all information in the overlay to Wikidata itself. We have several thousand annotated mappings between PropBank rolesets and Wikidata Items that are ready to be migrated onto Wikidata. See below for more information.
- Join the discussion on Wikidata and help us come to an agreement on how to get our data onto Wikidata.
- Read: PropBank goes Public: Incorporation into Wikidata
- Read: The DARPA Wikidata Overlay: Wikidata as an ontology for natural language processing
The DWD overlay is a subset of the DWD along with mappings to PropBank rolesets, their argument structures, and LDC tagsets. The overlay, which is in JSON format, is split into four sub-dictionaries: events, entities, relations, and temporal relations. Each key is the identifier for the DWD node, and each value contains a dictionary with various fields giving information for that node.
| ERE Type | Total | Annotated |
|---|---|---|
| Entities | 280 | 280 |
| Events | 5764 | 2224 |
| Relations | 216 | 216 |
| Temporal Relations | 8 | 8 |
Additionally, each of the 5,764 events has an event template, which are also undergoing annotation. There are currently 2,461 annotated templates.
Finally, the PropBank-Wikidata mapping annotators record every time they find a roleset that has no viable mapping in Wikidata, There are currently 9,011 rolesets that annotators have marked as "gaps."
type- the type of node. one ofevent_type,entity_type, orrelation_typewd_node- Q or P identifier from Wikidataname- Qnode or Pnode label from Wikidatawd_description- description from Wikidatacurated_by- how was the node added to the overlay? If the value iscmu, the node was added semi-automatically along with a PropBank roleset. (As such, those entries may be imperfect, so please submit any errors you find as an issue.) If the value isxpo, the node was added after a manual curation process and at least double-annotated. If the value isxpo_partial, the mapping is from one annotator's single pass.pb_rolesetandadditional_rolesets- the PropBank roleset mappings for the node.additional_rolesetsgives additional mappings that are as compatible with the mainpb_rolesetmapping, but added to the overlay later.arguments- a list of arguments with their names and slot constraints. Note that the slot constraints should be taken as a suggestion rather than a strict constraint, unlessmapping_typessays otherwise.- Relations have a
wd_slotand possibly apb_mappingwithmapping_types. Thewd_slotgives the Pnode slot in Wikidata (either subject or object);pb_mappinggives the PropBank argument name corresponding the the WD slot; andmapping_typesgives a list of flags warning users when a mapping may not work. (The only type right now isstrict_constraints, which means a PropBank event cannot be used to generate a Pnode with this mapping unless the constraints are met.)
- Relations have a
overlay_parents- direct superclasses of the node within the sub-ontology of the overlay (ie, in Wikidata, the parent may not be a direct superclass, but an ancestor further up the tree)ldc_types- a list of LDC types that are mapped to the DWD node. One DWD node can have several LDC types. This dictionary also contains arguments for the LDC event types which can be cross-referenced to the DWD arguments using the fielddwd_arg_namesimilar_nodes- a list of similar Qnodes. The type of similarity can beSS(semantic similarity) orNN(nearest neighbor)related_qnodes- for Pnodes, the "Wikidata item of this property" Qnode can be found in this sectiontemplate- A template which allows users to automatically generate a natural language sentence with an event and its argument instantiation. Only for events.template_curation- The curation status of the template. Either manually vetted by XPO (xpo) or automatically generated, and possibly an unnatural or incorrect sentence (auto).
Full changelog in CHANGELOG.md.
- 602 event nodes with PB mappings, see tab 1 here
- New field
additional_rolesetsto accommodate a one-Qnode-to-many-rolesets mapping