Constarium
← Search

Table · dataset · 2026

Parallel Atomic and Hyper-Relational Reference Triple Annotations for European Portuguese Football Match Reports

Listed in INESC TEC Research Data Repository

Description

This dataset contains manually annotated reference triples derived from 50 professionally written Liga Portugal football match reports, created in support of the paper “A Writer–Editor–Critic Pipeline with Deterministic Fact Verification for Football Match Report Generation”, accepted at the 19th International Natural Language Generation Conference (INLG 2026). The annotations are provided in two parallel representations: an atomic representation, comprising 2,380 units, and a hyper-relational representation, comprising 1,568 units.

Both representations cover the same 50 matches and encode the same annotated facts, although they are not aligned row by row. The hyper-relational representation was obtained through deterministic conversion of the atomic annotations. Each record includes a match identifier, subject, predicate, object, a reference to the corresponding span of the source report, and an Extractable_from_JSON flag indicating whether the annotated fact is directly derivable from the structured match data.

Read the rest (2 more)

The dataset captures information selected by professional journalists for inclusion in match reports and was used in the accompanying study to validate fact extraction. The underlying match data and professional reports were provided by ZOS Lda., operator of the zerozero.pt platform, under a research collaboration. The full source reports are not redistributed in this dataset and remain hosted on zerozero.pt; only the text spans associated with individual annotations are included.

Code, automatic evaluation outputs, and human evaluation materials from the accompanying study are not distributed.

Links

Get the data

Where it is published

Catalogue records · 1

Topics

Inferred from text
Text 75%
Provenance · 1 source records, 10 field assertions
SourceKeyLast seenRaw
INESC TEC Research Data Repositoryf3d224d7-28f4-4c60-9743-7761357f59945 d agoJSON v1
FieldAssertionExtractorEvidence
concepts[field].local:field:computer-science-aimapping · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0
concepts[field].local:field:engineeringmapping · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0
concepts[field].local:field:humanitiesmapping · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0
concepts[modality].local:modality:textenrichment · rdm inesctec ptkeyword-concept-rules@1.0.0title+description (75%)
created_datesource · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0
descriptionsource · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0/notes
license_textsource · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0
publication_datesource · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0
titlesource · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0/title
updated_datesource · rdm inesctec ptconnector:rdm_inesctec_pt@1.0.0