Summary report for project Exploiting Language Information for Situational Awareness (ELISA) For year 2017
automatic speech recognition, topic detection, incidents, situational awareness
The report summarizes the results of project "Exploiting Language Information for Situational Awareness (ELISA)" performed for the University of Southern California in L.A. in 2016.
Cross lingual topic id in latent space: Initial experiments were conducted on a multi-lingual (EN, FR, DE, IT) corpus of political texts comprising of 7 topics and approximately 4400 samples per language. Classification using uncertainties in the learned representations: We proposed a model that learns the document / sentence embeddings along with the uncertainties.