Worksheet: Coreference Resolution

Background

Coreference resolution identifies when different expressions in text refer to the same real-world entity.

Coreference: Two or more expressions (mentions) refer to the same entity in the world.

Mention: A noun phrase, pronoun, or name that refers to an entity.

Cluster: A group of mentions that all refer to the same entity.

Types of Referring Expressions

Pronouns: he, she, it, they, his, her
Named entities: Barack Obama, Apple Inc.
Definite descriptions: the president, the company
Demonstratives: this, that, these

Part 1: Basic Coreference

[Barack Obama] was the 44th president. [He] served two terms. [Michelle Obama] was [his] wife. [She] was also a lawyer.

Cluster 1

Cluster 2

Question 1

The example above shows two coreference clusters. List the mentions in each:

Cluster 1 (Barack Obama):

Cluster 2 (Michelle Obama):

Part 2: Ambiguous Pronouns

[John] told [Bill] that [he] would be late.

Question 2

Who does "he" refer to?

John Bill Ambiguous (could be either) Someone else not mentioned

How should ambiguous pronouns be handled in annotation?

Part 3: Generic vs. Specific Reference

[Dogs] are loyal animals. [My dog] certainly is. [He] follows me everywhere.

Question 3

Does "Dogs" (generic reference to all dogs) corefer with "My dog" (specific dog)?

Yes, they're related No, generic ≠ specific Partial coreference

What's the difference between generic and specific reference?

Part 4: Bridging and Near-Coreference

I visited [a restaurant] last night. [The food] was excellent, but [the waiter] was rude.

Question 4

"The food" and "the waiter" are related to "a restaurant" but don't refer to the same entity. This is called bridging. Should these be linked?

Yes, annotate bridging relations No, only strict coreference Annotate separately as different relation type

What makes bridging different from coreference?

Part 5: Events and Abstract Entities

[The company announced layoffs]. [This] shocked employees. [The decision] was made by [the board].

Question 5

Do "The company announced layoffs", "This", and "The decision" all refer to the same thing?

Yes, same event/entity No, different concepts Complex relationship

What challenges arise when annotating event coreference?

Part 6: Winograd Schema Challenge

[The city councilmen] refused to give [the demonstrators] a permit because [they] feared violence.

Question 6a

Who does "they" refer to?

The city councilmen The demonstrators

[The city councilmen] refused to give [the demonstrators] a permit because [they] advocated violence.

Question 6b

Now who does "they" refer to?

The city councilmen The demonstrators

What changed the interpretation? What knowledge is required?

Part 7: Annotate This Passage

[Apple] released [a new iPhone] yesterday. [The company] said [it] would be available next month. [Tim Cook] presented [the device] at [their headquarters]. [He] called [it] "revolutionary."

Question 7

Identify all coreference clusters:

Cluster 1:

Cluster 2:

Cluster 3:

Cluster 4:

What was the hardest linking decision?

Part 8: Group Discussion

Question 8

Where did your group disagree?

Mention boundaries Pronoun resolution Generic vs. specific Bridging vs. coreference Event coreference

Describe one specific disagreement:

Part 9: Reflection

Question 9

Why is coreference resolution difficult?

Requires world knowledge Pronouns are ambiguous Mention boundaries are unclear Identity itself is complex All of the above

Key Takeaway

Coreference is about reference, not similarity—and reference depends on context, knowledge, and inference.

Two expressions can be very different but refer to the same entity
Resolution often requires reasoning about the world, not just text
The boundary between coreference and related concepts (bridging, generics) is fuzzy