We describe a simple approach to named-entity recognition (NER), aimed initially at the Dutch language, but potentially applicable to other languages. Our NER system employs a two-stage architecture, with handcrafted but dataset-independent features for both stages, and is on a par with state-of-the-art systems described in the literature. Notably, our approach does not depend on language-specific assets such as gazetteers. The resulting system is quite fast and is implemented in less than 500 lines of code.
展开▼