Machine learning may enable the automated generation of test oracles. We have\ncharacterized emerging research in this area through a systematic literature\nreview examining oracle types, researcher goals, the ML techniques applied, how\nthe generation process was assessed, and the open research challenges in this\nemerging field.\n Based on a sample of 22 relevant studies, we observed that ML algorithms\ngenerated test verdict, metamorphic relation, and - most commonly - expected\noutput oracles. Almost all studies employ a supervised or semi-supervised\napproach, trained on labeled system executions or code metadata - including\nneural networks, support vector machines, adaptive boosting, and decision\ntrees. Oracles are evaluated using the mutation score, correct classifications,\naccuracy, and ROC. Work-to-date show great promise, but there are significant\nopen challenges regarding the requirements imposed on training data, the\ncomplexity of modeled functions, the ML algorithms employed - and how they are\napplied - the benchmarks used by researchers, and replicability of the studies.\nWe hope that our findings will serve as a roadmap and inspiration for\nresearchers in this field.\n