Pairwise embedding for event coreference resolution

Event coreference resolution is an important part in information extraction research and natural language understanding areas. Recently, the pre-trained language models emerging in modern Natural Language Processing (NLP) community provide a new perspective of solving classical NLP tasks. This thesis presents a novel, extensible, and error-tolerant model for within-document event coreference resolution. The model takes events from the same document pair wisely and extracts the features from both events. These features include but are not limited to sentence embeddings, trigger embeddings, and argument type representations. The extracted features are then used to fine-tune a pre-trained language model, BERT (Bidirectional Encoder Representations from Transformers), to perform event coreference resolution. The coreference results are evaluated by different metrics including B^3, MUC, and CEAF. Experimental results show that our pro-posed model outperforms baseline models with improvements of 0.009 on B^3 and 0.035 on MUC over the state-of-the-art models.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Hu, Yanda

Contributors dc:contributor

• Ji, Heng

• event coreference
• embedding

Statement dc:rights

• Copyright 2020 Yanda Hu

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108341

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108341