Rapid Adaptation of Machine Translation to Current Topics and Events


New events occur and current topics change at a rapid pace. Machine translation systems have traditionally not adapted well, if at all, to meet these changes – usually requiring a time intensive human managed and assisted data collection, preparation and training process that causes too much of a delay between re-trainings of the engines to be useful in the current context.

This presentation will demonstrate a set of tools and processes that enable unsupervised self-learning and adaptation of machine translation engines via data manufacturing so as to better adapt to the context of what is being discussed. This is further assisted by an automated cross language analysis of named entities between two languages to learn new names in the source language. Through this rapid learning and training of the machine translation engines, translation quality improves and adapts in time with the topics of the day, enabling more relevant and accurate analysis.