Annotate events from historical text, with a specific focus on 19th-century industrialisation-related events.

Event Extraction Annotation Guidelines

1. Overview

This guideline defines how to annotate events from historical text, with a specific focus on 19th-century industrialisation-related events (e.g., railways, strikes, infrastructure development, industrial accidents, and social movements).

Target Datasets

The annotation will be conducted on the historical British Library Newspaper datasets (1802-1920). These datasets contain OCR-derived text with varying quality, requiring careful interpretation during annotation.

Annotation Goal

Our objective is to extract structured event information to support:

• Historical analysis of industrialisation trends
• Knowledge graph construction
• Downstream tasks such as search, question answering, and spatial-temporal analysis

Event Schema Design

We adopt a refined event schema designed based on:

• ACE 2005 (for well-defined core event types)
• MAVEN (for broader and more flexible event coverage)

This hybrid schema is extended and adapted to better capture domain-specific events in historical industrial contexts, such as:

• Railway construction and operation
• Industrial disasters (e.g., fires, explosions)
• Labour movements and strikes

2. Key Concepts

Overall, an annotated event includes:

• Trigger: the word/phrase indicating the event
• Event Type: category of the event (See seciton Event Types for more details)
• Arguments: entities participating in the event with roles

2.1 What is an event

An event is a specific occurrence involving participants. Here, we will only annnotate events which were explicitly indicated as a real occurence or non-occurrence, and can be mapped to our defined event types (introduced in section Event Types). For example,

• He traveled to Edinburgh in late August.
• Anna said he did not travel to Edinburgh in late August.
• Anna said he traveled to Edinburgh by train.

Examples of events which can not be decided if it happended or not are:

• Rumors of arrests circulated in Vancouver.
• Should he not pay the money, they would kill him.
• He was ordered to return to Moscow.
• They wanted to acquire the company last year.
• He said he would leave town.

We will annotate an occured events as POSITVE, and NEGETIVE otherwise. This feature is called polarity.

2.2 What is an event trigger

An event must have a single trigger, and a sepecific type. A trigger is the word(s) that most clearly expressed the occureence of a taggable event. Mostly, triggers being the main verb are eaisiler to identify. For exmaple:

• He traveled to Edinburgh in late August.
• He was killed in a car accident.

Triggers can also be a noun or pronoun:

• The explosion clamed at least 30 lives.
• The meeting was held in Edinburgh, and it was a great success.

Sometimes, the modifier can also be annotated as the trigger for an event:

• The injured solider was very tall.
• The rioting crowd approached the train station.

2.3 What is an event argument

An event can have none or multiple arguments. An event argument is the entity that is involved in that event. Each argument has a specific role playing in that event, the types of roles vary from event types, which is detailed in section Event Schema. In general the argument can be the person, time, and location where an event happens. For example:

• Damon traveled to Edinburgh in late August.

Note that, we only annotate the closest syntactic proximity to the event trigger as the argument, especially when multiple entities refer to the same thing:

• Damon was in holiday, and he traveled to Edinburgh in late August.

Also, same arguments can be shared in multiple events.

We only annotate arguments which are clearly indicated participated or not participated in the event. Similarly, we use NEGATIVE, and POSITIVE polarity to mark them.

3. Annotation Format and Example

Text:

David died in his 70s.

1
2
3
4
5
6
7
8

{
  "event_type": "Life:Die",
  "trigger": "died",
  "arguments": [
    {"role": "Person", "text": "David"},
    {"role": "Age", "text": "his 70s"}
  ]
}

4. Event Schema

Accident - Scenario:Incident:Accident

Definition: Something bad that happens that is not expected or intended and that often damages something or injures someone

Potential triggers:

• accident
• crash
• collision
• explosion
• fire

Arguments:

• Victim — person injured or killed
• Cause — person or factor causing the accident, if known
• Source - The location, object, or entity from which the event originates or begins.
• Artifact — train, engine, cart, machine, ship, building, etc.
• Place — where it happened
• Time — when it happened
• Damage — what was damaged or destroyed

Building - Creating:Building

Definition: An event in which a physical structure or infrastructure is built, assembled, or created over time, such as railways, canals, bridges, buildings, or industrial facilities.

• Agent - who builds it
• Artifact* - what is being built
• Place - where construction occurs
• Purpose - Why it is built
• Time - When it occurs

Removing - CauseToBeHidden:Removing (Transport:Removing)

Definition: An event in which an agent causes a object to move away from its original location. The removed object has to be explictly expressed.

Arguments:

• Agent - who removed the object
• Purpose - why it is removed
• Artifact* - what is removed
• Time - When it occurs
• Place - Where it occurs
• Origin - Where the artifact was before it was removed

Transport - Movement:Transport

Definition: An event occurs whenever an artifact or a person is moved from one place to another. The moved artifact or person has to be explictly expressed. Either the origin or destination must be explicitly expressed.

Note that, if an event can be annotated as Removing event, then it should not be Transport event.

Arguments:

• Agent - who explicitly directs the movement, such as pilots and drivers.
• Purpose - why it occurs
• Artifact* - any artifact other than the vechicle doing the transporting, such as goods, passenger.
• Vehicle - any vehicle used
• Price - the price of transporting the person or artifact
• Time - When it occurs
• Origin(*) - where the transporting originated
• Destination(*) - where the transporting is directed
• Area - where the movement takes place excludeing origin and destintation

Flooding - Catastrophe:Flooding

Definition: a situation in which an area is covered with water, especially from rain.

Arguments:

• Place - Where the flooding takes place
• Time - When it occurs
• Cause - what causes the flooding
• Source - Where the water came from
• Target - what has been flooded

Damaging - AlterBadState:Damaging

Definition: an event in which an Agent affects an Artifact in such a way that the Artifact (or some part of the Artifact) ends up in a non-canonical state. Often this non-canonical state is undesirable.

Potential triggers:

• damange
• deface
• dent
• tear
• scratch

Arguments:

• Agent - who or what performs the intentional action that results in the damage to the Artifact
• Artifact* - what is being damaged
• Place - where it occured
• Cause - what event leads to the damage
• Time - When it occured
• Result - result of the damage

Delay (Optional as it is for crossreference)

Definition: the situation in which you have to wait longer than expected for something to happen.

Arguments:

• Event - what event has been delayed
• Time - when it happened
• Duration - how long it has beed delayed
• Cause - why it happened

Arbitration

Definition: the process of solving an argument between people by helping them to agree to an acceptable solution.

Arguments:

• Agent - who need solution for the target issue
• Stage
• Adjudicator - the judge or court helped to solve the issue.
• Place - where it occured
• Time - When it occured
• Target - the issue to be solved

Transfer-Money - Transcation:Transfer-Money

Definition: TRANSFER-MONEY Events refer to the giving, receiving, borrowing, or lending money when it is not in the context of purchasing something. The canonical examples are: (1) people giving money to organizations (and getting nothing tangible in return); and (2) organizations lending money to people or other orgs.

Arguments:

• Giver - the donating agent
• Recipient - the recipient agent
• Beneficiary - the agent that benefits from the transfer
• Money - the amount given, donated or loaned
• Place - where it occured
• Time - when it occured
• Purpose - why it occured

Cause_to_make_progress - Change:Cause_to_make_progress

Definition: An Agent works on somehthing so that it reaches a more advanced and desirable state.

Arguments:

• Entity - what has been advanced (an event, an object or an idea)
• Agent - who did it
• Place - where it occured
• Time - when it occured

Openness - Change:Openness

Definition: A place (location, facility)

5. Argument Roles

Arguments depend on event type. Common roles:

Example

Text:

The train passed through the smoke safely.

1
2
3
4
5
6
7
8

{
  "event_type": "Movement:Transport",
  "trigger": "passed",
  "arguments": [
    {"role": "Vehicle", "text": "train"},
    {"role": "Path", "text": "through the smoke"}
  ]
}

6. Special Cases

6.1 Implicit Arguments

• Include if clearly inferable

Text:

The fire spread rapidly.

1
2
3
4
5
6
7

{
  "event_type": "Disaster:FireSpread",
  "trigger": "spread",
  "arguments": [
    {"role": "Agent", "text": "fire"}
  ]
}

6.2 Multiple Events in One Sentence

Text:

He jumped and was injured.

→ Two events:

• Movement: jump
• Life: injure

6.4 Non-Events (Do NOT Annotate)

7. Temporal Expression Normalization

Normalize relative expressions:

Use standard formats like ISO 8601.

8. Output Format

Return results as JSON:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

{
  "events": [
    {
      "event_type": "...",
      "trigger": "...",
      "arguments": [
        {"role": "...", "text": "..."}
      ]
    }
  ]
}

9. Annotation Workflow

1. Read full sentence/document
2. Identify triggers
3. Assign event types
4. Extract arguments
5. Apply minimal span rule
6. Validate consistency

10. Quality Control

• Ensure type consistency
• Avoid duplicate events
• Check span correctness
• Verify argument-role alignment

11. Notes for Historical Texts

• Handle OCR noise carefully

• Normalize archaic language when interpreting

• Pay attention to:

  • industrial events (railways, strikes)
  • disasters (fires, floods)
  • social gatherings (rallies, meetings)