This paper provides an evaluation of a wide range of advanced sentence-level Quality Estimation models, including Support Vector Regression, Ride Regression, Neural Networks, Gaussian Processes, Bayesian Neural Networks, Deep Kernel Learning and Deep Gaussian Processes. Beside the accurateness, our main concerns are also the robustness of Quality Estimation models. Our work raises the difficulty in building strong models. Specifically, we show that Quality Estimation models often behave differently in Quality Estimation feature space, depending on whether the scale of feature space is small, medium or large. We also show that Quality Estimation models often behave differently in evaluation settings, depending on whether test data come from the same domain as the training data or not. Our work suggests several strong candidates to use in different circumstances.

Existing work on domain adaptation for statistical machine translation has consistently assumed access to a small sample from the test distribution (target domain) at training time. In practice, however, the target domain may not be known at training time or it may change to match user needs. In such situations, it is natural to push the system to make safer choices, giving higher preference to domain-invariant translations, which work well across domains, over risky domain-specific alternatives. We encode this intuition by (1) inducing latent subdomains from the training data only; (2) introducing features which measure how specialized phrases are to individual induced sub-domains; (3) estimating feature weights on out-of-domain data (rather than on the target domain). We conduct experiments on three language pairs and a number of different domains. We observe consistent improvements over a baseline which does not explicitly reward domain invariance.