Event Coreference and Relations (MAVEN-ERE)
Unified event relation extraction covering coreference, temporal, causal, and subevent relations. Annotators identify event mentions and link them with typed relations (coreference, temporal_before/after/overlap, causal, subevent). Based on the MAVEN-ERE dataset which provides large-scale, unified annotations for four types of event relations.
About this dataset
MAVEN-ERE is a dataset for event relation extraction, the task of finding events in text and labeling how they relate. It covers four relation types at once: coreference (two mentions of the same event), temporal order, causal links, and subevent (part-of) structure. Xiaozhi Wang and colleagues at Tsinghua University (THU-KEG) introduced it at EMNLP 2022, building on the earlier MAVEN event-detection dataset.
The corpus is built from 4,480 English Wikipedia documents. Annotators first mark event triggers, then go through pairs of events and assign a relation label where one holds. Because a single document can contain many events, the relation annotations are dense: the paper reports an average of 272 temporal links per document.
In total the dataset has 103,193 event coreference chains, 1,216,217 temporal relations, 57,992 causal relations, and 15,841 subevent relations, which the authors note is at least an order of magnitude larger than prior event-relation datasets on every relation type.
The Potato config below reproduces this workflow in two steps: a span scheme for tagging event mentions with eleven coarse type labels (action, state change, movement, and so on), then a span-link scheme that connects two tagged events with one of six relation labels (coreference, temporal before/after/overlap, causal, subevent). The coarse type set is a simplified stand-in for MAVEN's finer event ontology. It suits annotating discourse-level event structure over full documents.
- Documents
- 4,480 (English Wikipedia)
- Event coreference chains
- 103,193
- Temporal relations
- 1,216,217
- Causal relations
- 57,992
- Subevent relations
- 15,841
- Relation types
- 4 (coref / temporal / causal / subevent)
Configuration Fileconfig.yaml
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# Event Coreference and Relations (MAVEN-ERE)
# Based on Wang et al., EMNLP 2022
# Paper: https://aclanthology.org/2022.emnlp-main.60/
# Dataset: https://github.com/THU-KEG/MAVEN-ERE
#
# MAVEN-ERE provides unified annotations for four types of event relations:
# - Coreference: Two event mentions refer to the same real-world event
# - Temporal: Ordering between events (before, after, overlap)
# - Causal: One event causes or enables another
# - Subevent: One event is a component of another
#
# Annotation process:
# 1. Identify event mentions (triggers) in the text
# 2. For each pair of events, determine if a relation exists
# 3. Label the relation with its type
annotation_task_name: "MAVEN-ERE Event Coreference and Relations"
task_dir: "."
data_files:
- sample-data.json
item_properties:
id_key: "id"
text_key: "text"
output_annotation_dir: "annotation_output/"
output_annotation_format: "json"
annotation_schemes:
# Step 1: Event mention identification
- annotation_type: span
name: event_mentions
description: "Highlight all event mentions (triggers) in the text. Select the event type for each mention."
labels:
- "Action"
- "State_change"
- "Movement"
- "Communication"
- "Conflict"
- "Life"
- "Transaction"
- "Natural_disaster"
- "Creation"
- "Perception"
- "Other"
label_colors:
"Action": "#3b82f6"
"State_change": "#22c55e"
"Movement": "#f59e0b"
"Communication": "#8b5cf6"
"Conflict": "#ef4444"
"Life": "#ec4899"
"Transaction": "#06b6d4"
"Natural_disaster": "#f97316"
"Creation": "#14b8a6"
"Perception": "#a855f7"
"Other": "#9ca3af"
tooltips:
"Action": "Deliberate action performed by an agent (e.g., 'arrested', 'built', 'signed')"
"State_change": "Change in state or condition (e.g., 'collapsed', 'increased', 'dissolved')"
"Movement": "Physical movement or transportation (e.g., 'moved', 'traveled', 'deployed')"
"Communication": "Speech, writing, or reporting (e.g., 'announced', 'said', 'published')"
"Conflict": "Violent or adversarial event (e.g., 'attacked', 'protested', 'clashed')"
"Life": "Birth, death, injury (e.g., 'born', 'died', 'injured')"
"Transaction": "Exchange or transfer (e.g., 'purchased', 'sold', 'donated')"
"Natural_disaster": "Natural event (e.g., 'earthquake', 'flood', 'erupted')"
"Creation": "Making or producing something (e.g., 'created', 'founded', 'invented')"
"Perception": "Sensing or observing (e.g., 'detected', 'observed', 'discovered')"
"Other": "Event type not covered by the above categories"
keyboard_shortcuts:
"Action": "1"
"State_change": "2"
"Movement": "3"
"Communication": "4"
"Conflict": "5"
allow_overlapping: false
# Step 2: Event relation linking
- annotation_type: span_link
name: event_relations
description: "Link pairs of event mentions with their relation type. Select two previously annotated event spans and specify the relation."
source_scheme: event_mentions
target_scheme: event_mentions
labels:
- "Coreference"
- "Temporal_before"
- "Temporal_after"
- "Temporal_overlap"
- "Causal"
- "Subevent"
label_colors:
"Coreference": "#3b82f6"
"Temporal_before": "#f59e0b"
"Temporal_after": "#f97316"
"Temporal_overlap": "#eab308"
"Causal": "#ef4444"
"Subevent": "#22c55e"
tooltips:
"Coreference": "Both mentions refer to the same real-world event"
"Temporal_before": "The source event occurs before the target event"
"Temporal_after": "The source event occurs after the target event"
"Temporal_overlap": "The two events overlap in time"
"Causal": "The source event causes or enables the target event"
"Subevent": "The source event is a component or subevent of the target event"
keyboard_shortcuts:
"Coreference": "c"
"Temporal_before": "b"
"Temporal_after": "a"
"Temporal_overlap": "o"
"Causal": "u"
"Subevent": "s"
html_layout: |
<div style="margin-bottom: 10px;">
<h4>Document: {{document_id}}</h4>
<p style="line-height: 1.8; font-size: 15px;">{{text}}</p>
</div>
<div style="background: #f0f0f0; padding: 10px; border-radius: 5px; margin-top: 10px;">
<strong>Instructions:</strong>
<ol>
<li>Highlight all event mentions and classify their type.</li>
<li>Link event pairs with relation labels: coreference, temporal, causal, or subevent.</li>
</ol>
</div>
allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false
Sample Datasample-data.json
[
{
"id": "maven_ere_001",
"text": "A massive earthquake struck the coastal city of Valdivia on May 22, 1960. The tremor devastated the region, destroying thousands of buildings and displacing over two million people. The seismic event triggered a series of tsunamis that hit the coastlines of Chile, Hawaii, and Japan. Scientists later determined that the quake measured 9.5 on the Richter scale, making it the most powerful earthquake ever recorded.",
"document_id": "doc_001"
},
{
"id": "maven_ere_002",
"text": "The company announced a major restructuring plan on Monday. As part of the reorganization, over 3,000 employees will be laid off across multiple divisions. The layoffs are expected to begin next quarter. CEO Martha Reynolds said the cuts were necessary to ensure the firm's long-term survival. The restructuring follows a series of quarterly losses that began in early 2021.",
"document_id": "doc_002"
}
]
// ... and 8 more itemsGet This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git cd potato-showcase/text/coreference/maven-ere-event-coreference potato start config.yaml
Dataset & paper
Wang et al., EMNLP 2022
Citation (BibTeX)
@inproceedings{wang-etal-2022-maven,
title = "{MAVEN}-{ERE}: A Unified Large-scale Dataset for Event Coreference, Temporal, Causal, and Subevent Relation Extraction",
author = "Wang, Xiaozhi and Chen, Yulin and Ding, Ning and Peng, Hao and Wang, Zimu and Lin, Yankai and Han, Xu and Hou, Lei and Li, Juanzi and Liu, Zhiyuan and Li, Peng and Zhou, Jie",
booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2022",
address = "Abu Dhabi, United Arab Emirates",
publisher = "Association for Computational Linguistics",
pages = "926--941",
doi = "10.18653/v1/2022.emnlp-main.60"
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
CrossRE: Cross-Domain Relation Extraction
Cross-domain relation extraction across 6 domains (news, politics, science, music, literature, AI). Annotators identify entities and label 17 relation types between entity pairs, enabling study of domain transfer in relation extraction.
Legal Event Coreference (LegalCore)
Event coreference resolution in legal documents including court opinions, contracts, and statutes. Annotators identify legal event mentions such as actions, states, obligations, and violations, then link coreferent events that refer to the same real-world event. Based on the LegalCore dataset for domain-specific event coreference in the legal domain.
Multilingual Coreference Resolution (CorefUD)
Multilingual coreference resolution across 17 languages using Universal Dependencies-style annotations. Annotators identify entity mentions (names, nominals, pronouns) and link them into coreference chains. Based on the CorefUD dataset and the CRAC 2023 Shared Task on Multilingual Coreference Resolution.