Introduction to the CoNLL-2003 Shared Task: Language-Independent Named Entity Recognition

TL;DR

The paper introduces the CoNLL-2003 shared task on language-independent named entity recognition, focusing on English and German datasets with four entity types (PER, LOC, ORG, MISC) and an emphasis on using resources beyond the training data. It details data sources, preprocessing, IOB tagging format, and an evaluation framework based on F1 with bootstrap significance, while highlighting a spectrum of machine learning approaches and extensive feature sets. The study shows that ensemble methods combining Maximum Entropy, HMMs, and other learners, augmented by gazetteers and externally trained NER outputs, yield the best results, with English and German gains demonstrating the value of external resources. The findings underscore the potential and challenges of leveraging unannotated data and external tools to improve multilingual NER performance, establishing a benchmark for future research.

Abstract

We describe the CoNLL-2003 shared task: language-independent named entity recognition. We give background information on the data sets (English and German) and the evaluation method, present a general overview of the systems that have taken part in the task and discuss their performance.