Name entities made obvious: the participation in the ERD 2014 evaluation

The paper describes NEMO, a system for extracting entity mentions from text and linking them to Wikipedia (and Freebase), which participated in the ERD 2014 challenge. The model employed by the system allows a seamless use of traditional priors and lexical features in conjunction with various types of latent features, which are computed based on the attributes associated with all extractions of entity mentions from an input text and their possible linkage to Wikipedia. Additionally, it allows a unified approach for handling both features computed globally, at document level, and features computed based on the local context, such as syntactic patterns, of each hypothesized entity mention. The model is trained on a large dataset derived from Wikipedia, and achieves state-of-the-art results on the datasets in the ERD evaluation without employing explicitly ERD-specific training data.