Automatic Improvement of Machine Translation Using Mutamorphic Relation: Invited Talk Paper

Abstract

This paper introduces Mutamorphic Relation for Machine Learning Testing. Mutamorphic Relation combines data mutation and metamorphic relations as test oracles for machine learning systems. These oracles can help achieve fully automatic testing as well as automatic repair of the machine learning models. The paper takes TransRepair as an example to show the effectiveness of Mutamorphic Relation in automatically testing and improving machine translators, TransRepair detects inconsistency bugs without access to human oracles. It then adopts probability-reference or cross-reference to post-process the translations, in a grey-box or black-box manner, to repair the inconsistencies. Manual inspection indicates that the translations repaired by TransRepair improve consistency in 87% of cases (degrading it in 2%), and that the repairs of have better translation acceptability in 27% of the cases (worse in 8%).