ÃÛÌÒ´«Ã½Æƽâ°æÏÂÔØ

Our goal was richer and more accurate representations of utterances in English, Chinese, Hindi/Urdu, and Arabic. Our principle approach involved the application of supervised machine learning to data with linguistic annotation. There were several different layers of annotation, and correspondingly several individual NLP components, many of which were trained on a single layer. We began by describing several different end-to-end systems we were building which incorporated these components, then described the individual components. Next we described the lexical resources which informed the linguistic annotation, and then the individual layers of annotation and the different domains and genres they had been applied to. Finally we described CLEARTK - the CLEAR NLP Toolkit - that was being used by some of the end-to-end systems.