Showcase/Event Argument Extraction (MAVEN-Arg)
advancedtext

Event Argument Extraction (MAVEN-Arg)

Document-level event argument extraction based on MAVEN-Arg (Wang et al., ACL 2024). Annotates event triggers with their argument roles including Agent, Patient, Location, Time, Instrument, and more. Supports both entity and non-entity arguments across document context.

📝

text annotation

Configuration Fileconfig.yaml

# Event Argument Extraction (MAVEN-Arg style)
# Based on Wang et al., ACL 2024
# Paper: https://aclanthology.org/2024.acl-long.224/
# Dataset: https://github.com/THU-KEG/MAVEN-Argument
#
# MAVEN-Arg provides:
# - 162 event types with expert-written definitions
# - 612 argument roles across all event types
# - Document-level annotation (arguments can be anywhere in document)
# - Both entity and non-entity arguments
#
# This showcase demonstrates a subset of common event types and argument roles.
# Full schema available in the dataset download.
#
# Annotation process:
# 1. Identify event trigger (word/phrase indicating event occurrence)
# 2. Classify event type
# 3. Identify arguments filling semantic roles for that event
# 4. Arguments may be entities (linked) or non-entity spans

port: 8000
server_name: localhost
task_name: "Event Argument Extraction"

data_files:
  - sample-data.json
id_key: id
text_key: text

output_file: annotations.json

annotation_schemes:
  # Step 1: Event trigger identification and typing
  - annotation_type: span
    name: event_trigger
    description: "Highlight the event trigger and select its event type"
    labels:
      # Life events
      - "Life:Birth"
      - "Life:Death"
      - "Life:Injury"
      # Movement events
      - "Movement:Transport"
      - "Movement:Travel"
      # Conflict events
      - "Conflict:Attack"
      - "Conflict:Protest"
      - "Conflict:War"
      # Transaction events
      - "Transaction:Purchase"
      - "Transaction:Sale"
      - "Transaction:Transfer"
      # Business events
      - "Business:Start"
      - "Business:End"
      - "Business:Merge"
      # Communication events
      - "Communication:Statement"
      - "Communication:Meeting"
      - "Communication:Announcement"
      # Justice events
      - "Justice:Arrest"
      - "Justice:Trial"
      - "Justice:Sentence"
      # Personnel events
      - "Personnel:Hire"
      - "Personnel:Fire"
      - "Personnel:Elect"
      # Other
      - "Other"
    label_colors:
      "Life:Birth": "#22c55e"
      "Life:Death": "#6b7280"
      "Life:Injury": "#ef4444"
      "Movement:Transport": "#3b82f6"
      "Movement:Travel": "#06b6d4"
      "Conflict:Attack": "#dc2626"
      "Conflict:Protest": "#f97316"
      "Conflict:War": "#991b1b"
      "Transaction:Purchase": "#8b5cf6"
      "Transaction:Sale": "#a855f7"
      "Transaction:Transfer": "#c084fc"
      "Business:Start": "#14b8a6"
      "Business:End": "#64748b"
      "Business:Merge": "#0ea5e9"
      "Communication:Statement": "#eab308"
      "Communication:Meeting": "#facc15"
      "Communication:Announcement": "#fde047"
      "Justice:Arrest": "#b91c1c"
      "Justice:Trial": "#7c3aed"
      "Justice:Sentence": "#4c1d95"
      "Personnel:Hire": "#10b981"
      "Personnel:Fire": "#f43f5e"
      "Personnel:Elect": "#06b6d4"
      "Other": "#9ca3af"
    tooltips:
      "Life:Birth": "Birth of a person"
      "Life:Death": "Death of a person (natural, accident, or killing)"
      "Life:Injury": "Physical harm or injury to a person"
      "Movement:Transport": "Moving objects or goods from one place to another"
      "Movement:Travel": "Person traveling from one location to another"
      "Conflict:Attack": "Violent physical act causing harm"
      "Conflict:Protest": "Public demonstration or protest"
      "Conflict:War": "Armed conflict between groups/nations"
      "Transaction:Purchase": "Buying something with money"
      "Transaction:Sale": "Selling something for money"
      "Transaction:Transfer": "Transfer of ownership or possession"
      "Business:Start": "Starting a business or organization"
      "Business:End": "Closing or ending a business"
      "Business:Merge": "Combining two or more organizations"
      "Communication:Statement": "Formal statement or declaration"
      "Communication:Meeting": "Meeting or gathering of people"
      "Communication:Announcement": "Public announcement of information"
      "Justice:Arrest": "Law enforcement taking someone into custody"
      "Justice:Trial": "Legal court proceeding"
      "Justice:Sentence": "Legal judgment or sentencing"
      "Personnel:Hire": "Hiring or employing someone"
      "Personnel:Fire": "Terminating someone's employment"
      "Personnel:Elect": "Electing someone to a position"
      "Other": "Event type not covered by above categories"
    allow_overlapping: false

  # Step 2: Argument identification
  - annotation_type: span
    name: arguments
    description: "Highlight arguments (participants/properties) of the identified events"
    labels:
      # Core arguments
      - "Agent"
      - "Patient"
      - "Theme"
      # Participants
      - "Victim"
      - "Attacker"
      - "Buyer"
      - "Seller"
      - "Giver"
      - "Recipient"
      # Circumstantial
      - "Location"
      - "Origin"
      - "Destination"
      - "Time"
      - "Duration"
      # Other
      - "Instrument"
      - "Cause"
      - "Purpose"
      - "Content"
      - "Value"
    label_colors:
      "Agent": "#3b82f6"
      "Patient": "#22c55e"
      "Theme": "#8b5cf6"
      "Victim": "#ef4444"
      "Attacker": "#dc2626"
      "Buyer": "#06b6d4"
      "Seller": "#0891b2"
      "Giver": "#14b8a6"
      "Recipient": "#10b981"
      "Location": "#f59e0b"
      "Origin": "#fbbf24"
      "Destination": "#f97316"
      "Time": "#a855f7"
      "Duration": "#c084fc"
      "Instrument": "#6b7280"
      "Cause": "#ec4899"
      "Purpose": "#db2777"
      "Content": "#eab308"
      "Value": "#84cc16"
    tooltips:
      "Agent": "Entity performing or causing the event (doer)"
      "Patient": "Entity affected by or undergoing the event"
      "Theme": "Entity being moved, transferred, or described"
      "Victim": "Person harmed in the event"
      "Attacker": "Person/entity causing harm"
      "Buyer": "Entity purchasing something"
      "Seller": "Entity selling something"
      "Giver": "Entity giving or transferring something"
      "Recipient": "Entity receiving something"
      "Location": "Where the event takes place"
      "Origin": "Starting point of movement"
      "Destination": "End point of movement"
      "Time": "When the event occurs"
      "Duration": "How long the event lasts"
      "Instrument": "Object used to perform the event"
      "Cause": "What caused the event"
      "Purpose": "Goal or intention of the event"
      "Content": "What is communicated or expressed"
      "Value": "Monetary or quantitative value involved"
    allow_overlapping: true

  # Step 3: Link argument to event
  - annotation_type: radio
    name: argument_event_link
    description: "Which event does the most recently annotated argument belong to? (Enter event number)"
    labels:
      - "Event 1"
      - "Event 2"
      - "Event 3"
      - "Event 4"
      - "Multiple events"
    tooltips:
      "Event 1": "Argument belongs to the first event in the text"
      "Event 2": "Argument belongs to the second event"
      "Event 3": "Argument belongs to the third event"
      "Event 4": "Argument belongs to the fourth event"
      "Multiple events": "Argument is shared across multiple events"

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false

Sample Datasample-data.json

[
  {
    "id": "eae_001",
    "text": "Apple Inc. announced on Tuesday that it will acquire the startup Acme Technologies for $2 billion. The deal is expected to close by the end of March. CEO Tim Cook stated that the acquisition will strengthen Apple's AI capabilities."
  },
  {
    "id": "eae_002",
    "text": "A powerful earthquake struck central Turkey early Monday morning, killing at least 50 people and injuring hundreds more. Rescue teams from neighboring countries have been deployed to the affected regions. The 7.2 magnitude quake damaged thousands of buildings in the city of Adana."
  }
]

// ... and 8 more items

Get This Design

View on GitHub

Clone or download from the repository

Quick start:

git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/event-argument-extraction
potato start config.yaml

Details

Annotation Types

spanradio

Domain

NLPInformation Extraction

Use Cases

Event ExtractionInformation ExtractionKnowledge Graphs

Tags

event-extractionargument-extractioneaemavenacl2024document-level

Found an issue or want to improve this design?

Open an Issue

Related Designs

Dialogue Relation Extraction (DialogRE)

Extract relations between entities in dialogue. Based on Yu et al., ACL 2020. Identify 36 relation types between speakers and entities mentioned in conversations.

spanradio

Coreference Resolution (OntoNotes)

Link pronouns and noun phrases to the entities they refer to in text. Based on the OntoNotes coreference annotation guidelines and CoNLL shared tasks. Identify mention spans and cluster coreferent mentions together.

spanradio

Emotion Cause Extraction (RECCON)

Extract emotion causes from conversational text based on RECCON (Poria et al., EMNLP 2020). Identify which utterances and specific spans caused an emotion expressed in dialogue.

radiospan