Out-of-context fine-grained multi-word entity classification: exploring token, character n-gram and NN-based models for multilingual entity classification.