Abstract:

The current major Machine Translation (MT) evaluation effort, funded by the Advanced Research Projects Agency (ARPA), shows that when compared to expert human translators, MT systems perform only about 65% as well on the average. In this paper it is argued that despite their overall poor performance, MT software can be used as a powerful focal point to improve second language (L2) skills. The paper describes the evaluation of Computronics Corporation's Targumatik (Hebrew-->English), a PC-based MT system running under DOS, and shows how each problem and potential obstruction to learning can be overcome by means of discovery procedures using a set of tools and procedures called the 'learning algorithm.'

Author: Don D. Anderson