Proactive Human Translation by Adaptively Predicting Eye Gazes and Keystrokes

The exploration of hidden patterns within different types of data streams arising from psycholinguistic experiments is of growing interest in the area of translation process research. In order to support psycholinguists experts in quantitatively discovering the non-self-explanatory cognitive processes within the data, we present the StrPMiner algorithm for mining and generating sequential patterns hidden within multimodal streaming data. We propose additionally a prediction framework that uses the rules generated by the StrPMiner to evaluate the latest items from the data stream and tries to predict items that are likely to appear in the next few time steps. The predictor aims at proactively helping the translators to focus on the translation task, by forwarding the predicted next action to some translation assistance tool.
We show through extensive experimental evaluation over real datasets that our proposed predictor manages to predict the next item accurately in the real time. They show that our predictor is able to exactly predict the next action in almost 60% of the cases.

Authors: Hassani M., Beecks C., Töws D., Seidl T.
Published in: Proceedings of the ProActIR workshop @ECIR (to appear)
Language: EN
Year: 2016
Conference: ECIR
Type: Conference papers (peer reviewed)
Research topic: Exploration of Multimedia Databases