Precise-first models for incorporating global constraints in coreference resolution
Abstract/Contents
- Abstract
- Coreference resolution, the task of finding all textual mentions that refer to the same real-world entity, is crucial for many natural language processing applications, including information extraction, question answering, and summarization. Previous research has shown the importance of a rich feature space that models lexical, syntactic, semantic, pragmatic, and discourse phenomena. Incorporating this varied information during the resolution process is a key factor in high performance coreference resolution. This dissertation introduces a simple scaffolding framework that deploys strong features through tiers of models that perform significantly better than a single-pass model. The sieve architecture applies a battery of deterministic coreference models one at a time from highest to lowest precision, each model building on the previous model's cluster output. The approach makes use of global information through an entity-centric model that encourages the sharing of features across all mentions that point to the same real-world entity. I also show that this precise-first architecture can be applied to a statistical as well as rule-based classification. Replacing the rule-based sieves with a hybrid system, using a precise-first sequence of (supervised) statistical coreference resolution models in the same framework leads to state-of-the-art performance in coreference resolution. Finally, I show how this hybrid classifier can be made faster and with a smaller memory footprint to widen its practical applicability. Compared to the extensive work on entity coreference, the related problem of event coreference remains relatively under-explored, with minimal work on how entity and event coreference can be considered jointly on an open domain. I introduce a novel coreference resolution system that models entities and events jointly. The iterative method cautiously constructs clusters of entity and event mentions using linear regression to model cluster merge operations. As clusters are built, information flows between entity and event clusters through features that model semantic role dependencies. My system handles nominal and verbal events as well as entities, and the joint formulation allows information from event coreference to help entity coreference, and vice versa. All the approaches presented in the dissertation rely on two high-level insights: incorporating both local and global information effectively in coreference resolution, and using this information in a precise-first manner, allowing high-precision information to propagate throughout the system.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2015 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Lee, Heeyoung |
|---|---|
| Associated with | Stanford University, Department of Electrical Engineering. |
| Primary advisor | Jurafsky, Dan, 1962- |
| Thesis advisor | Jurafsky, Dan, 1962- |
| Thesis advisor | Garcia-Molina, Hector |
| Thesis advisor | Manning, Christopher D |
| Thesis advisor | Widrow, Bernard, 1929- |
| Advisor | Garcia-Molina, Hector |
| Advisor | Manning, Christopher D |
| Advisor | Widrow, Bernard, 1929- |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Heeyoung Lee. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis (Ph.D.)--Stanford University, 2015. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2015 by Heeyoung Lee
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...