EasyFirst Coreference Resolution Veselin Stoyanov and Jason Eisner

Coreference Resolution • Given a peace of text, identify all noun phrases that refer

Coreference Resolution • Typical pairwise approaches • But, inherently global task Former Boxing Champion

Coreference Resolution • Inherently global task Former Boxing Champion Mike Tyson revealed that he

Global Approaches to Coreference Resolution • A difficult search problem – Exponentially many possible

A Different Approach • A Multi-Pass Sieve [Raghunathan et al. , 2010] – Build

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] Former Boxing Champion Mike

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] • Sieves: 1. Exact

Easy-First Coreference Resolution • Use the same idea: – Greedily build clusters – Using

Easy-First Coreference Resolution • Incrementally and greedily build clusters – Using machine learning •

Easy-First Coreference: Talk Overview • Test-Time Inference • Training Algorithm • Experimental results

Test-Time Inference • Begin with each NP in a separate cluster Former Boxing Champion

Test-Time Inference • Begin with each NP in a separate cluster Cluster 1 Sem.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1

Test-Time Inference • Compute a weight for each join action Cluster 1 Sem. Type:

Test-Time Inference • Perform the action with the highest weight Cluster 1 Sem. Type:

Test-Time Inference • Perform the action with the highest weight Cluster 15 Sem. Type:

Test-Time Inference • Perform the action with the highest weight Former Boxing Champion Mike

Test-Time Inference • Until the action with the highest weight is the special action

Time complexity • Let n = the number of mentions in a document •

Time complexity • Each step requires: – Popping the top action -- O(logn) (priority

Learning • Learn weight vector w that leads the Easy-First algorithm to good clusterings

Reinforcement Learning • A classic reinforcement learning setting • Act in the world and

Perceptron-Style Training • A version of the early update structured perceptron [Collins and Roark,

Training algorithm • Perform N iterations: – For each document d in the training

Experimental Setup • Five datasets: – MUC 6, MUC 7, ACE 03, ACE 04,

Baselines • A pairwise baseline [Stoyanov et al. 2009] • Reimplementation of the pairwise

Results: Automatic Mentions Reconcile MUC 6 MUC 7 ACE 03 ACE 04 ACE 05

Results: Automatic Mentions Reconcile Pairwise baseline MUC 6 MUC 7 ACE 03 ACE 04

Results: Automatic Mentions Reconcile Pairwise baseline Easy-First percep MUC 6 MUC 7 ACE 03

Results: Automatic Mentions Reconcile Pairwise baseline Easy-First percep Easy-First struct MUC 6 MUC 7

Results: Gold Mentions MUC 6 B^3 ACE 04 MUC B^3 MUC Haghighi and Klein

Conclusions • A new approach to coreference using greedy incremental clustering – More accurate

Slides: 78

Download presentation

Easy-First Coreference Resolution Veselin Stoyanov and Jason Eisner Johns Hopkins University

Coreference Resolution • Given a peace of text, identify all noun phrases that refer to the same real-world entity

Coreference Resolution • Given a peace of text, identify all noun phrases that refer to the same real-world entity Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Coreference Resolution • Typical pairwise approaches • But, inherently global task Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Coreference Resolution • Inherently global task Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Global Approaches to Coreference Resolution • A difficult search problem – Exponentially many possible clusterings • Recent approaches have been successful: – [Culotta et al. , 2007] – train a global coherence measure; use greedy inference at test time – [Poon and Domingos, 2008] – use an unsupervised approach based on MLNs to enforce global properties

A Different Approach • A Multi-Pass Sieve [Raghunathan et al. , 2010] – Build coreference clusters incrementally – Applying sieves (hand-specified rules) of decreasing precision and increasing recall – Winner of the Co. NLL-2011 shared task

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] • Sieves: 1. Exact match Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] • Sieves: 1. Exact match 2. Precise Constructs Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] • Sieves: 1. Exact match 2. Precise Constructs 3. Strict Head match Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

A Multi-Pass Sieve for Coreference Resolution [Raghunathan et al. 2010] • Sieves: 1. Exact match 2. Precise Constructs 3. Strict Head match …………. 7. Pronoun Matching Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Easy-First Coreference Resolution • Use the same idea: – Greedily build clusters – Using more information for the less confident decisions • But learn automatically what constitutes an “easy” (confident) decision – Leads to empirical improvement over the Multi. Pass Sieve approach and a competitive baseline

Easy-First Coreference Resolution • Incrementally and greedily build clusters – Using machine learning • Learn policy for the test time inference procedure

Easy-First Coreference: Talk Overview • Test-Time Inference • Training Algorithm • Experimental results

TEST-TIME INFERENCE

Test-Time Inference • Begin with each NP in a separate cluster Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Begin with each NP in a separate cluster Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Cluster 3 Sem. Type: person Gender: male Number: single Names: NIL Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Sem. Type. Same = False Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Sem. Type. Same = False Sem. Type. Comp = True Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Sem. Type. Same = False Sem. Type. Comp = True Number. Same = True Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Sem. Type. Same = False Sem. Type. Comp = True Number. Same = True Role. Appositive = True Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Sem. Type. Same = False Sem. Type. Comp = True Number. Same = True Role. Appositive = True Sent. Distance = 0 Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson Sem. Type. Same = False Sem. Type. Comp = True Number. Same = True Role. Appositive = True Sent. Distance = 0 Global features Local features Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a feature vector for each pair of clusters Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson f(joinij) Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a weight for each join action Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson w. f(joinij) Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Compute a weight for each join action Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson weightij= w. f(joinij) Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Perform the action with the highest weight Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson aij=argmaxij w. f(joinij) Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Perform the action with the highest weight Cluster 1 Sem. Type: uknwn Gender: uknwn Number: single Names: NIL Cluster 2 Sem. Type: person Gender: male Number: single Names: Mike Tyson a 12=argmaxij w. f(joinij) Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Perform the action with the highest weight Cluster 15 Sem. Type: person Gender: male Number: single Names: Mike Tyson a 12=argmaxij w. f(joinij) Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Perform the action with the highest weight Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Test-Time Inference • Until the action with the highest weight is the special action HALT Former Boxing Champion Mike Tyson revealed that he had caught Hollywood heartthrob Brad Pitt with his ex-wife Robin Givens. In an interview Tyson said that he was amidst the process of divorce from Givens in the late 1980 s when he found her sleeping with the Hollywood star.

Time complexity • Let n = the number of mentions in a document • At most n clusters at any time • s = number of joins performed by the algorithm • s<=n at most n joins before HALT (usually much less)

Time complexity • Each step requires: – Popping the top action -- O(logn) (priority queue) – Performing the join – O(1) – Deleting O(n) old actions – O(nlogn) – Computing scores for O(n) new actions -- O(n) – Inserting those -- O(nlogn) or O(n) • Total runtime = O(n 2 logn)+O(snlogn)

TRAINING ALGORITHM

Learning • Learn weight vector w that leads the Easy-First algorithm to good clusterings • At each step we are given a state s and a set of possible actions {joinij}∪{HALT} • We want to learn what is “the best” action in each state • But the current state s depends on the action taken so far

Reinforcement Learning • A classic reinforcement learning setting • Act in the world and learn “good” actions in each state • We tried several reinforcement learning algorithms – Policy gradient [Sutton et al. , 2000] – Q-learning [Watkins and Dayan, 1992] – No success: distant supervision; too many possible states

Perceptron-Style Training • A version of the early update structured perceptron [Collins and Roark, 2004] – Inspired by the success of easy-first parsing [Goldberg and Elhadad, 2010] • The algorithm: – Keep acting in the world until the next action will result in an error (reduction in score) – Update weights away from the top action and towards the highest scoring positive action

Training algorithm • Perform N iterations: – For each document d in the training set do: • Initialize C so that each mention in its own cluster • Initialize A = {joinij}∪{HALT} • repeat until amax != HALT: – for k steps do: » amax =A. peek. Max() » if amax. is. Positive() break » apos =A. peek. Max. Positive() » w+=(feat(apos, C)-feat(amax , C)) » Recompute scores in A using new w update weight

Training algorithm • Perform N iterations: – For each document d in the training set do: • Initialize C so that each mention in its own cluster • Initialize A = {joinij}∩{HALT} • repeat until amax != HALT: – Update weights – amax =A. pop. Max() – if amax is a join » perform amax

EXPERIMENTAL EVALUATION

Experimental Setup • Five datasets: – MUC 6, MUC 7, ACE 03, ACE 04, ACE 05 • Two evaluation metrics: – MUC score – B^3 score • Implemented in the Reconcile research platform [Stoyanov et al. 2010] – Code available upon request • Two settings: – Automatic mentions – Gold (manual) mentions

Baselines • A pairwise baseline [Stoyanov et al. 2009] • Reimplementation of the pairwise baseline using different features • Easy-First coreference resolution – Easy-Firstpercep – weights from the pairwise baseline – Easy-Firststruct – weights from our training algorithm • Only for the gold mentions data – Entity-centered approach [Haghighi and Klein, 2009] – Multi-pass sieve [Raghunathan et al. , 2010]

Results: Automatic Mentions Reconcile MUC 6 MUC 7 ACE 03 ACE 04 ACE 05 B^3 MUC B^3 MUC 70. 9 68. 5 65. 9 62. 8 79. 4 67. 9 76. 5 62. 0 73. 7 67. 4

Results: Automatic Mentions Reconcile Pairwise baseline MUC 6 MUC 7 ACE 03 ACE 04 ACE 05 B^3 MUC B^3 MUC 70. 9 68. 5 65. 9 62. 8 79. 4 67. 9 76. 5 62. 0 73. 7 67. 4 70. 7 67. 9 62. 2 57. 9 81. 7 73. 5 79. 4 68. 0 75. 7 70. 2

Results: Automatic Mentions Reconcile Pairwise baseline Easy-First percep MUC 6 MUC 7 ACE 03 ACE 04 ACE 05 B^3 MUC B^3 MUC 70. 9 68. 5 65. 9 62. 8 79. 4 67. 9 76. 5 62. 0 73. 7 67. 4 70. 7 67. 9 62. 2 57. 9 81. 7 73. 5 79. 4 68. 0 75. 7 70. 2 71. 5 68. 4 64. 4 58. 1 82. 8 74. 1 80. 3 68. 4 76. 0 70. 0

Results: Automatic Mentions Reconcile Pairwise baseline Easy-First percep Easy-First struct MUC 6 MUC 7 ACE 03 ACE 04 ACE 05 B^3 MUC B^3 MUC 70. 9 68. 5 65. 9 62. 8 79. 4 67. 9 76. 5 62. 0 73. 7 67. 4 70. 7 67. 9 62. 2 57. 9 81. 7 73. 5 79. 4 68. 0 75. 7 70. 2 71. 5 68. 4 64. 4 58. 1 82. 8 74. 1 80. 3 68. 4 76. 0 70. 0 72. 1 68. 2 64. 6 57. 7 83. 1 74. 5 80. 9 68. 8 75. 9 70. 9

Results: Gold Mentions MUC 6 B^3 ACE 04 MUC B^3 MUC Haghighi and Klein 75. 0 81. 9 76. 5 Raghunathan et al. 73. 2 77. 7 78. 9 78. 1

Results: Gold Mentions MUC 6 B^3 ACE 04 MUC B^3 MUC Haghighi and Klein 75. 0 81. 9 76. 5 Raghunathan et al. 73. 2 77. 7 78. 9 78. 1 Stoyanov et al. 76. 1 88. 2 77. 8 76. 2

Results: Gold Mentions MUC 6 B^3 ACE 04 MUC B^3 MUC Haghighi and Klein 75. 0 81. 9 76. 5 Raghunathan et al. 73. 2 77. 7 78. 9 78. 1 Stoyanov et al. 76. 1 88. 2 77. 8 76. 2 Pairwise baseline 72. 7 88. 2 75. 9 78. 9

Conclusions • A new approach to coreference using greedy incremental clustering – More accurate clustering by utilizing global information as it becomes available for less confident decisions • Learning policy for the test-time inference procedure leads to improvements

THANK YOU!