Fine-Grained Named Entity Recognition

This is one of the suggested topics for the course Language Technology Project 2005.

Introduction

A named entity recognizer is a system which identifies names in texts and classifies them according to broad categories like person, organization and location. For some applications, like question answering, such a broad classification is insufficient and a more fine-grained categorization is required.

Task

Develop and build a fine-grained named entity recognizer for a language other than English (preferably for Dutch) which identifies names in text and assigns to them one or more specific categories

The suggested corpus for extracting information is the free online Wikipedia encyclopedia. The proposed implementation method for the named entity recognition system is table lookup in combination with hand-crafted or learned patterns.

Modules

The named entity recognition system will contain several modules which will be realized by different people in the project groups:

• Interface module: reads a text, sends text to retrieval module, sends queries to database, deals with answers and returns the text with XML tags around identified names with specifications of their classes (two persons).
  Relevant background knowledge: natural language processing plus basic programming skills.
• Retrieval module: collect documents from Wikipedia and extract relevant name-category pairs from the web pages (two persons).
  Relevant background knowledge: retrieval and experience in programming of web interfaces.
• Database module: store name-category pairs, deal with incoming queries and return relevant answers (one person).
  Relevant background knowledge: databases, SQL and basic programming skills.

The system will be evaluated with a text to be supplied by the teachers in the final week of the project.

Literature and tools

• PhD thesis Andrew Borthwick (1999): chapters 1 and 2 and bibliography.
• Gideon S. Mann, Fine-Grained Proper Noun Ontologies for Question Answering Systems. In: "Proceedings of SemaNet'02", Taipei, Taiwan, 2002. [pdf]
• Maynard et al. paper: NE recognition without training data on a language you don't speak
• Demo of basic named entity recognition at the University of Illinois at Urbana-Champaign
• Tools and example scripts for common tasks