Step by Step
C
Coreference — identifying same-entity references
Coreference resolution identifies which different expressions in a text (pronouns, noun phrases, proper names) all refer to the same underlying real-world entity.
Example: recognizing that "John," "he," and "the manager" might all refer to the exact same person within a single passage of text.
T
Types — pronouns, noun phrases, proper names
Coreference can involve pronouns (he, she, it), definite noun phrases (the president), or proper names (John) — all potentially referring back to the same entity mentioned elsewhere in the text.
Example: "the president" and "he" both referring back to the same specific person named earlier in an article.
W
World knowledge required — the Winograd Schema Challenge
Correctly resolving coreference often requires genuine world knowledge and reasoning, not just grammatical rules — a classic example being "The trophy didn't fit in the suitcase because it was too big," which requires understanding that a trophy being too big (not the suitcase) is what would prevent it from fitting.
Example: "The city council refused the protesters a permit because they feared violence" — correctly resolving "they" to mean the city council (not the protesters) requires understanding the underlying logic of who would plausibly "fear violence" in this context, not just grammar.
Applied Walkthrough
1
A reading comprehension system encounters the sentence: "The trophy didn't fit in the suitcase because it was too big."
2
Ask: does purely grammatical analysis tell you whether "it" refers to the trophy or the suitcase? No — grammatically, "it" could refer to either noun; genuine world knowledge about size and fitting is required to correctly resolve this.
3
Applying real-world reasoning (a trophy that's too big wouldn't fit; a suitcase being too big doesn't prevent something else from fitting inside it) correctly resolves "it" to mean the trophy.
4
This exact type of sentence is the basis of the Winograd Schema Challenge, a well-known benchmark specifically designed to test whether AI systems can perform this kind of world-knowledge-dependent coreference resolution.
Exam Application
Exams test whether you understand that coreference resolution often requires genuine world knowledge and reasoning, not just grammatical pattern matching, and whether you're familiar with the Winograd Schema Challenge as the classic benchmark illustrating this requirement.
⚠ Common Trap
The most common trap is assuming coreference resolution can always be solved with simple grammatical rules (like matching pronoun gender or proximity to the nearest noun). Classic Winograd Schema-style examples specifically demonstrate that genuine world knowledge and reasoning are often required, since the correct answer can't be determined from grammar alone.
✓ Quick Self-Check
1. What does coreference resolution identify?
Which different expressions in a text (pronouns, noun phrases, proper names) refer to the same real-world entity.
Tap to reveal / hide
2. Name the three types of expressions coreference resolution typically deals with.
Pronouns, definite noun phrases, and proper names.
Tap to reveal / hide
3. In "The trophy didn't fit in the suitcase because it was too big," what does "it" refer to, and why?
The trophy — because a trophy being too big would prevent it from fitting, whereas a suitcase being too big wouldn't have that same effect.
Tap to reveal / hide
4. What is the Winograd Schema Challenge designed to test?
Whether AI systems can correctly resolve coreference cases that require genuine world knowledge and reasoning, not just grammar.
Tap to reveal / hide
5. Can coreference resolution always be solved using purely grammatical rules?
No — some cases specifically require world knowledge and reasoning beyond grammar alone.
Tap to reveal / hide