Annotation of chemical named entities

Abstract

We describe the annotation of chemical named entities in scientific text. A set of annotation guidelines defines 5 types of named entities, and provides instructions for the resolution of special cases. A corpus of full-text chemistry papers was annotated, with an inter-annotator agreement score of 93%. An investigation of named entity recognition using LingPipe suggests that scores of 63% are possible without customisation, and scores of 74% are possible with the addition of custom tokenisation and the use of dictionaries.