This document discusses algorithms for coreference resolution, which is the task of linking mentions of entities in a text to their corresponding real-world entities. It presents Hobbs' algorithm, which proposes potential antecedents for anaphora by traversing the parse tree, and Lappin and Leass' algorithm, which applies salience factors to mentions before applying syntactic filters and moving to the previous sentence if no antecedent is found. The document also discusses different types of anaphora like pronouns and proper nouns, as well as agreement constraints and anaphor binding.
2. Coreference Resolution
2
Anaphora Resolution
Link noun phrase to its antecedents.
↑
Mostly noun phrases.
The Tin Woodman went to Emerald City to see the
Wizard of Oz and ask for a heart. After he asked for it,
the Woodman waited there for the master’s response.
Types of noun phrases?
PronounsProper nouns Common nouns
3. Anaphora Resolution
3
Proper noun ← Pronoun
Jinho bought a car. He was happy.
He bought a car for himself.
Challenges
Jinho ate a cookie. He also ate a cake. It was delicious.
Pronoun ← Pronoun
Common noun ← Pronoun
Jinho bought a car for his wife. She was happy.
Jinho ate a cookie. He also ate a cake. They were delicious.
4. Anaphora Resolution
4
Proper noun ← Proper noun
Jinho Choi bought a car. Dr. Choi was happy.
While he was studying, Jinho was singing.
Pronoun ← Proper noun
Common noun ← Proper noun
My professor’s name is Jinho Choi.
Proper noun
Pronoun
Common noun
← Common noun
5. Why Coreference Resolution?
5
In 2004, Obama received national attention during his
campaign to represent Illinois in the United States Senate
with his victory in the March Democratic Party primary, his
keynote address at the Democratic National Convention in
July, and his election to the Senate in November. He began
his presidential campaign in 2007 and, after a close primary
campaign against Hillary Rodham Clinton in 2008, he won
sufficient delegates in the Democratic Party primaries to
receive the presidential nomination. He then defeated
Republican nominee John McCain in the general election,
and was inaugurated as president on January 20, 2009. Nine
months after his inauguration, Obama was named the 2009
Nobel Peace Prize laureate.
7. Agreements
7
Syntactic Agreement
Jinho bought a car for him.
Jinho bought a car for himself.
Jinho?
Selectional Agreement
Jinho bought a book from Amazon. He read it over night.
Which?
8. Hobb’s Algorithm
8
1.Begin at NP.
2.Go up tree to first NP or S. Call this X, and the path p.
3.Traverse all branches below X to the left of p. Propose as
antecedent any NP that has a NP or S between it and X.
4.If X is the highest S in the sentence, traverse the parse
trees of the previous sentences in the order of recency.
Traverse left-to-right, breadth first. When a NP is
encountered, propose as antecedent. If not the highest
node, go to step 5.
9. Hobb’s Algorithm
9
5.From node X, go up the tree to the first NP or S. Call it
X, and the path p.
6.If X is an NP and the path to X did not pass through the
nominal that X dominates, propose X as antecedent.
7.Traverse all branches below X to the right of the path, in a
left-to-right, breadth first manner. Propose any NP
encountered as the antecedent.
8.If X is an S node, traverse all branches of X to the right of
the path but do not go below any NP or S encountered.
Propose any NP as the antecedent.
10. Lappin and Leass’ Algorithm
10
Collect all mentions (NPs) in the sentence.
Apply salience factors to the mentions.
Personal or Possessive
Pronouns
Reciprocal or Reflective
Pronouns
Syntactic Filter Anaphor Binding
If not found, move onto the previous sentence.
11. Pleonastic-It
11
It is A that S
It is A (for NP) toVP
It is C that S
It R that S
NP S it A (for NP) toVP
It is time toVP
It is thanks to NP that S
Modal adjectives A
necessary, possible, certain, likely, important, good, useful, advisable,
convenient, sufficient, economical, easy, desirable, difficult, legal
Cognitive verbs C
recommend, think, believe, know, anticipate, assume, expect
Raising verbs R
seem, appear, mean, follow
Causal verbs S
make, find
12. Syntactic Filter
12
A pronoun P is non-coreferential with a noun phrase N if:
P and N have incompatible agreement features.
The woman said that he is funny.
P is in the argument domain of N.
She likes her. John seems to want to see him.
P is in the adjunct domain of N.
She sat near her.
P is an argument of a head H, N is not a pronoun, and N is contained in H.
He believes that the man is amusing.
P is in the NP domain of N.
John’s portrait of him is interesting.
P is a determiner of a noun Q, and N is contained in Q.
His portrait of John is interesting.
His description of the portrait by John is interesting.
13. Anaphor Binding
13
A noun phrase N is a possible antecedent binder for R iff:
R is in the argument domain of N, and N > R.
They wanted to see themselves.
Mary knows the people who John introduced to each other.
R is in the adjunct domain of N.
He worked by himself.
Which friends plan to travel with each other?
R is in the NP domain of N.
John likes Bill’s portrait of himself.
Argument slot hierarchy
subj > agent > obj > (iobjlpobj)
Lexical anaphor R
reciprocal or reflective pronoun
N and R do not have incompatible agreement features. &
14. Anaphor Binding
14
A noun phrase N is a possible antecedent binder for R iff:
N is an argument of a verbV,
∃ an NP Q in the argument domain or the adjunct domain of N s.t.
(Q has no noun determiner) and
[(R is an argument of Q) or
(R is an argument of a preposition P and P is an adjunct of Q)].
They told stories about themselves.
Argument slot hierarchy
subj > agent > obj > (iobjlpobj)
Lexical anaphor R
reciprocal or reflective pronoun
(R is a determiner of a noun Q) and
[(Q is in the argument domain of N and N > Q) or
(Q is in the adjunct domain of N)].
John and Mary like each other’s portraits.
15. Salience Weighting
15
Salience Factor Weight Example
Sentence recency 100
Subject emphasis 80 John likes the dog.
Head noun emphasis 80 The dog in the house is here.
Existential emphasis 70 There is a dog at John’s place.
Accusative emphasis 50 John bought the dog last night.
Non-adverbial emphasis 50 According to John, Mary likes the dog.
Indirect object and
oblique comp. emphasis
40 John bought the dog to Mary.
16. Lappin and Leass’ Algorithm
16
John is a good professor.
Tony hated him until he talked with him.