Scribd
Upload
27 views24 pages

Chapter 5.2

Uploaded by

bhelravisoni
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
27 views24 pages

Chapter 5.2

Uploaded by

bhelravisoni
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Discourse:

Coreference
Deep Processing Techniques for NLP
Ling 571
March 1, 2017
Roadmap
— Coreference
— Referring expressions

— Syntactic & semantic constraints


— Syntactic & semantic preferences

— Reference resolution:
— Hobbs Algorithm: Baseline
— Machine learning approaches
— Sieve models

— Challenges
Entity-based Coherence
— John went to his favorite music store to buy a piano.
— He had frequented the store for many years.
— He was excited that he could finally buy a piano.
— VS
— John went to his favorite music store to buy a piano.
— It was a store John had frequented for many years.
— He was excited that he could finally buy a piano.
— It was closing just as John arrived.

— Which is better? Why?


— ‘about’ one entity vs two, focuses on it for coherence
Reference Resolution
— Match referring expressions to referents
— Syntactic & semantic constraints
— Syntactic & semantic preferences

— Reference resolution algorithms


Reference
— Queen Elizabeth set about transforming her
husband, King George VI, into a viable monarch.
Logue, a renowned speech therapist, was
summoned to help the King overcome his speech
impediment...

Referring expression: (refexp)


Linguistic form that picks out entity in some model
That entity is the “referent”
When introduces entity, “evokes” it
Set up later reference, “antecedent”
2 refexps with same referent “co-refer”
Reference (terminology)
— Queen Elizabeth set about transforming her
husband, King George VI, into a viable monarch.
Logue, a renowned speech therapist, was
summoned to help the King overcome his speech
impediment...

— Anaphor:
— Abbreviated linguistic form interpreted in context
— Her, his, the King
— Refers to previously introduced item (“accesses”)
— Referring expression is then anaphoric
Referring Expressions
— Many alternatives:
— Queen Elizabeth, she, her, the Queen, etc
— Possible correct forms depend on discourse context
— E.g. she, her presume prior mention, or presence in world

— Interpretation (and generation) requires:


— Discourse Model with representations of:
— Entities referred to in the discourse
— Relationships of these entities
— Need way to construct, update model
— Need way to map refexp to hearer’s beliefs
Reference and Model
Reference Resolution
— Queen Elizabeth set about transforming her
husband, King George VI, into a viable monarch.
Logue, a renowned speech therapist, was
summoned to help the King overcome his speech
impediment...

Coreference resolution:
Find all expressions referring to same entity, ‘corefer’
Colors indicate coreferent sets
Pronominal anaphora resolution:
Find antecedent for given pronoun
Referring Expressions
— Indefinite noun phrases (NPs): e.g. “a cat”
— Introduces new item to discourse context
— Definite NPs: e.g. “the cat”
— Refers to item identifiable by hearer in context
— By verbal, pointing, or environment availability; implicit

— Pronouns: e.g. “he”,”she”, “it”


— Refers to item, must be “salient”
— Demonstratives: e.g. “this”, “that”
— Refers to item, sense of distance (literal/figurative)
— Names: e.g. “Miss Woodhouse”,”IBM”
— New or old entities
Information Status
— Some expressions (e.g. indef NPs) introduce new info
— Others refer to old referents (e.g. pronouns)
— Theories link form of refexp to given/new status

— Accessibility:
— More salient elements easier to call up, can be shorter
Correlates with length: more accessible, shorter refexp
Complicating Factors
— Inferrables:
— Refexp refers to inferentially related entity
— I bought a car today, but the door had a dent, and the engine
was noisy.
— E.g. car à door, engine

— Generics:
— I want to buy a Mac. They are very stylish.
— General group evoked by instance.
— Non-referential cases:
— It’s raining.
Syntactic Constraints for
Reference Resolution
— Some fairly rigid rules constrain possible referents
— Agreement:
— Number: Singular/Plural

— Person: 1st: I,we; 2nd: you; 3rd: he, she, it, they

— Gender: he vs she vs it
Syntactic & Semantic
Constraints
— Binding constraints:
— Reflexive (x-self): corefers with subject of clause
— Pronoun/Def. NP: can’t corefer with subject of clause

— “Selectional restrictions”:
— “animate”: The cows eat grass.
— “human”: The author wrote the book.
— More general: drive: John drives a car….
Syntactic & Semantic
Preferences
— Recency: Closer entities are more salient
— The doctor found an old map in the chest. Jim found an
even older map on the shelf. It described an island.

— Grammatical role: Saliency hierarchy of roles


— e.g. Subj > Object > I. Obj. > Oblique > AdvP
— Billy Bones went to the bar with Jim Hawkins. He called
for a glass of rum. [he = Billy]
— Jim Hawkins went to the bar with Billy Bones. He called
for a glass of rum. [he = Jim]
Syntactic & Semantic
Preferences
— Repeated reference: Pronouns more salient
— Once focused, likely to continue to be focused
— Billy Bones had been thinking of a glass of rum. He hobbled
over to the bar. Jim Hawkins went with him. He called for a
glass of rum. [he=Billy]

— Parallelism: Prefer entity in same role


— Silver went with Jim to the bar. Billy Bones went with him to
the inn. [him = Jim]
— Overrides grammatical role

— Verb roles: “implicit causality”, thematic role match,...


— John telephoned Bill. He lost the laptop. [He=John]
— John criticized Bill. He lost the laptop. [He=Bill]
Reference Resolution
Approaches
— Common features
— “Discourse Model”
— Referents evoked in discourse, available for reference
— Structure indicating relative salience
— Syntactic & Semantic Constraints
— Syntactic & Semantic Preferences

— Differences:
— Which constraints/preferences? How combine?
Rank?
Hobbs’ Resolution
Algorithm
— Requires:
— Syntactic parser
— Gender and number checker
— Input:
— Pronoun
— Parse of current and previous sentences

— Captures:
— Preferences: Recency, grammatical role
— Constraints: binding theory, gender, person, number
Hobbs Algorithm
— Intuition:
— Start with target pronoun
— Climb parse tree to S root
— For each NP or S
— Do breadth-first, left-to-right search of children
— Restricted to left of target
— For each NP, check agreement with target
— Repeat on earlier sentences until matching NP found
Hobbs Algorithm Detail
— Begin at NP immediately dominating pronoun
— Climb tree to NP or S: X=node, p = path
— Traverse branches below X, and left of p: BF, LR
— If find NP, propose as antecedent
— If separated from X by NP or S
— Loop: If X highest S in sentence, try previous sentences.
— If X not highest S, climb to next NP or S: X = node
— If X is NP, and p not through X’s nominal, propose X
— Traverse branches below X, left of p: BF,LR
— Propose any NP
— If X is S, traverse branches of X, right of p: BF, LR
— Do not traverse NP or S; Propose any NP
— Go to Loop
Hobbs Example

Lyn’s mom is a gardener. Craige likes her.


Another Hobbs Example
— The castle in Camelot remained the residence of the
King until 536 when he moved it to London.

— What is it?
— residence
Another Hobbs Example

Hobbs, 1978
Hobbs Algorithm
— Results: 88% accuracy ; 90+% intrasentential
— On perfect, manually parsed sentences
— Useful baseline for evaluating pronominal anaphora
— Issues:
— Parsing:
— Not all languages have parsers
— Parsers are not always accurate
— Constraints/Preferences:
— Captures: Binding theory, grammatical role, recency
— But not: parallelism, repetition, verb semantics, selection

You might also like