A Model for High-coverage Lexical Semantic Annotation Generation

AI applications often receive their input in the form of natural language text, or as the transcription of spoken text. A commonsense inference system should transform such input to a formal representation with limited vocabulary in order to be able to process them. In this paper, we present a method based on neural word embeddings that automatically assigns semic features to words of natural language. These features either describe the ontological category of a given word or provide some characterization or additional information. We show that our method has high coverage and performs well for English and Hungarian, and can easily be extended to other languages as well.