Character language models for generalization of multilingual named entity recognition

Yu, Xiaodong

Permalink

https://hdl.handle.net/2142/104934 Copy

Description

Title

Character language models for generalization of multilingual named entity recognition

Author(s)

Yu, Xiaodong

Issue Date

2019-04-25

Director of Research (if dissertation) or Advisor (if thesis)

Roth, Dan

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Character Language Models
• Named Entity Recognition
• Generalization
• Multilingual
• Multilingual Named Entity Recognition
• NER

Abstract

"State-of-the-art Named Entity Recognition (NER) models usually achieve high performance on entities that they have seen in training data, but a significantly lower performance on unseen entities. This is one of the key reasons in performance degradation observed when NER models are evaluated on new domains. Motivated by this observation, quantified for the first time in this thesis, we study an improved, multi-domain and multi-lingual, capability for identifying \what is a name"". Character-level patterns have been widely used as features in English Named Entity Recognition (NER) systems. However, to date there has been no direct investigation of the inherent differences between name and non-name tokens in text, nor whether this property holds across multiple languages. The key contribution of this thesis is to develop a Character-level Language Model (CLM) that, as we show, allow us to better learn \what is a name"". We analyze the capabilities of corpus-agnostic Character-level Language Models (CLMs) in the binary task of distinguishing name tokens from non-name tokens and demonstrate that CLMs provide a simple yet powerful model for capturing these differences. Specifically, we show that it can identify named entity tokens in a diverse set of languages at close to the performance of full NER systems. Moreover, by adding very simple CLM-based features we can significantly improve the performance of an o -the-shelf NER system for multiple languages."

Graduation Semester

2019-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/104934

Copyright and License Information

Copyright 2019 Xiaodong Yu

Owning Collections