TRAnslation and Transcription Assisted by Humans on the Internet


The goal of this exploratory project, in cooperation with Carnegie Mellon University, is to design a web-based tool for man-in-the-loop joint robust Automatic Speech Recognition (ASR) and Machine Translation (MT). For this purpose research will be centered on representing the uncertainty of ASR and MT outputs expressed as a lattice, incorporating user feedback to the lattice and sharing information among MT and ASR lattices for improved performance.