Executive Summary

We present the results of three evaluations of the performance of the ModernMT translation platform. First, we evaluate the performance of three ModernMT translation engines — phrase-based, neural without adaptation, and neural with adaptation — on the Benchmark 2.6 collection of data sets in 14 translation directions. Next, we report the results of an evaluation that focuses on the translation of terminology from English into Italian. Finally, we assess the effect of data cleaning on test sets for en→es , en→de , and en→ru.

Download the PDF

MMT - D5.4 - Second Evaluation ReportMMT – D5.4 – Second Evaluation Report (PDF, 558 KB)