Unsupervised Learning of Narrative Schemas and their Participants
Chambers, Nathanael and Jurafsky, Dan

Article Structure

Abstract

We describe an unsupervised system for learning narrative schemas, coherent sequences or sets of events (arrested(POLICE,SUSPECT), convicted( JUDGE, SUSPECT)) whose arguments are filled with participant semantic roles defined over words (JUDGE = {judge, jury, court}, POLICE = {police, agent, authorities}).

Introduction

This paper describes a new approach to event semantics that jointly learns event relations and their participants from unlabeled corpora.

Background

This paper addresses two areas of work in event semantics, narrative event chains and semantic role labeling.

Narrative Schemas

The next sections introduce typed narrative chains and chain merging, extensions that allow us to jointly learn argument roles with event structure.

Sample Narrative Schemas

Figures 3 and 4 show two criminal schemas learned completely automatically from the NYT portion of the Gigaword Corpus (Graff, 2002).

Frames and Roles

Most previous work on unsupervised semantic role labeling assumes that the set of possible

Evaluation: Cloze

The previous section compared our learned knowledge to current work in event and role semantics.

Discussion

Our significant improvement in the cloze evaluation shows that even though narrative cloze does not evaluate argument types, jointly modeling the arguments with events improves event clustering.

Topics

semantic role

Appears in 16 sentences as: Semantic Role (1) semantic role (8) semantic roles (8)
In Unsupervised Learning of Narrative Schemas and their Participants
  1. We describe an unsupervised system for learning narrative schemas, coherent sequences or sets of events (arrested(POLICE,SUSPECT), convicted( JUDGE, SUSPECT)) whose arguments are filled with participant semantic roles defined over words (JUDGE = {judge, jury, court}, POLICE = {police, agent, authorities}).
    Page 1, “Abstract”
  2. Unlike most previous work in event structure or semantic role learning, our system does not use supervised techniques, hand-built knowledge, or predefined classes of events or roles.
    Page 1, “Abstract”
  3. By jointly addressing both tasks, we improve on previous results in narrative/frame learning and induce rich frame-specific semantic roles .
    Page 1, “Abstract”
  4. Instead, modern work on understanding has focused on shallower representations like semantic roles , which express at least one aspect of the semantics of events and have proved amenable to supervised learning from corpora like PropBank (Palmer et al., 2005) and Framenet (Baker et al., 1998).
    Page 1, “Introduction”
  5. Even unsupervised attempts to learn semantic roles have required a predefined set of roles (Grenager and Manning, 2006) and often a hand-labeled seed corpus (Swier and Stevenson, 2004; He and Gildea, 2006).
    Page 1, “Introduction”
  6. This paper shows that verbs in distinct narrative chains can be merged into an improved single narrative schema, while the shared arguments across verbs can provide rich information for inducing semantic roles .
    Page 1, “Introduction”
  7. This paper addresses two areas of work in event semantics, narrative event chains and semantic role labeling.
    Page 2, “Background”
  8. 2.2 Semantic Role Labeling
    Page 3, “Background”
  9. The task of semantic role learning and labeling is to identify classes of entities that fill predicate slots; semantic roles seem like they’d be a good model for the kind of argument types we’d like to learn for narratives.
    Page 3, “Background”
  10. Most work on semantic role labeling, however, is supervised, using Propbank (Palmer et al., 2005), FrameNet (Baker et al., 1998) or VerbNet (Kipper et al., 2000) as gold standard roles and training data.
    Page 3, “Background”
  11. As just described, Chambers and Jurafsky (2008) offers an unsupervised approach to event learning (goal 1), but lacks semantic role
    Page 3, “Background”

See all papers in Proc. ACL 2009 that mention semantic role.

See all papers in Proc. ACL that mention semantic role.

Back to top.

coreference

Appears in 8 sentences as: coreference (5) coreferences (1) coreferent (1) coreferential (2)
In Unsupervised Learning of Narrative Schemas and their Participants
  1. The Chambers and Jurafsky (2008) model learns chains completely unsupervised, (albeit after parsing and resolving coreference in the text) by counting pairs of verbs that share corefer-ring arguments within documents and computing the pointwise mutual information (PMI) between these verb-argument pairs.
    Page 2, “Background”
  2. Even more telling is that these arguments are jointly shared (the same or coreferent ) across all three events.
    Page 2, “Background”
  3. As mentioned above, narrative chains are learned by parsing the text, resolving coreference , and extracting chains of events that share participants.
    Page 3, “Narrative Schemas”
  4. In our new model, argument types are learned simultaneously with narrative chains by finding salient words that represent coreferential arguments.
    Page 3, “Narrative Schemas”
  5. We record counts of arguments that are observed with each pair of event slots, build the referential set for each word from its coreference chain, and then represent each observed argument by the most frequent head word in its referential set (ignoring pronouns and mapping entity mentions with person pronouns to a constant PERSON identifier).
    Page 3, “Narrative Schemas”
  6. The four bolded terms are coreferential and (hopefully) identified by coreference .
    Page 4, “Narrative Schemas”
  7. We parse the text into dependency graphs and resolve coreferences .
    Page 5, “Sample Narrative Schemas”
  8. We use the OpenNLP1 coreference engine to resolve entity mentions.
    Page 8, “Evaluation: Cloze”

See all papers in Proc. ACL 2009 that mention coreference.

See all papers in Proc. ACL that mention coreference.

Back to top.

role labeling

Appears in 5 sentences as: Role Labeling (1) role labeling (4)
In Unsupervised Learning of Narrative Schemas and their Participants
  1. This paper addresses two areas of work in event semantics, narrative event chains and semantic role labeling .
    Page 2, “Background”
  2. 2.2 Semantic Role Labeling
    Page 3, “Background”
  3. Most work on semantic role labeling , however, is supervised, using Propbank (Palmer et al., 2005), FrameNet (Baker et al., 1998) or VerbNet (Kipper et al., 2000) as gold standard roles and training data.
    Page 3, “Background”
  4. Most previous work on unsupervised semantic role labeling assumes that the set of possible
    Page 5, “Frames and Roles”
  5. Our argument learning algorithm not only performs unsupervised induction of situation-specific role classes, but the resulting roles and linking structures may also offer the possibility of (unsupervised) FrameNet-style semantic role labeling .
    Page 8, “Discussion”

See all papers in Proc. ACL 2009 that mention role labeling.

See all papers in Proc. ACL that mention role labeling.

Back to top.

semantic role labeling

Appears in 5 sentences as: Semantic Role Labeling (1) semantic role labeling (4)
In Unsupervised Learning of Narrative Schemas and their Participants
  1. This paper addresses two areas of work in event semantics, narrative event chains and semantic role labeling .
    Page 2, “Background”
  2. 2.2 Semantic Role Labeling
    Page 3, “Background”
  3. Most work on semantic role labeling , however, is supervised, using Propbank (Palmer et al., 2005), FrameNet (Baker et al., 1998) or VerbNet (Kipper et al., 2000) as gold standard roles and training data.
    Page 3, “Background”
  4. Most previous work on unsupervised semantic role labeling assumes that the set of possible
    Page 5, “Frames and Roles”
  5. Our argument learning algorithm not only performs unsupervised induction of situation-specific role classes, but the resulting roles and linking structures may also offer the possibility of (unsupervised) FrameNet-style semantic role labeling .
    Page 8, “Discussion”

See all papers in Proc. ACL 2009 that mention semantic role labeling.

See all papers in Proc. ACL that mention semantic role labeling.

Back to top.

PDF available from www.aclweb.org.
Send us an email