We introduce a curriculum learning approach to adapt generic neural machine translation models to a specific domain. Samples are grouped by their similarities to the domain of interest and each group is fed to the training algorithm with a particular schedule. This approach is simple to implement on top of any neural framework or architecture, and consistently outperforms both unadapted and adapted baselines in experiments with two distinct domains and two language pairs.

No abstract

This task combines the labeling of multiword expressions and supersenses (coarse-grained classes) in an explicit, yet broad-coverage paradigm for lexical semantics. Nine systems participated; the best scored 57.7% F 1 in a multi-domain evaluation setting, indicating that the task remains largely unresolved. An error analysis reveals that a large number of instances in the data set are either hard cases, which no systems get right, or easy cases, which all systems correctly solve.

We describe the system built by the National Research Council Canada for the "Discriminating between similar languages" (DSL) shared task. Our system uses various statistical classifiers and makes predictions based on a two-stage process: we first predict the language group, then discriminate between languages or variants within the group. Language groups are predicted using a generative classifier with 99.99% accuracy on the five target groups. Within each group (except English), we use a voting combination of discriminative classifiers trained on a variety of feature spaces, achieving an average accuracy of 95.71%, with per-group accuracy between 90.95% and 100% depending on the group. This approach turns out to reach the best performance among all systems submitted to the open and closed tasks.

Despite impressive progress in highresource settings, Neural Machine Translation (NMT) still struggles in lowresource and out-of-domain scenarios, often failing to match the quality of phrasebased translation. We propose a novel technique that combines back-translation and multilingual NMT to improve performance in these difficult cases. Our technique trains a single model for both directions of a language pair, allowing us to back-translate source or target monolingual data without requiring an auxiliary model. We then continue training on the augmented parallel data, enabling a cycle of improvement for a single model that can incorporate any source, target, or parallel data to improve both translation directions. As a byproduct, these models can reduce training and deployment costs significantly compared to uni-directional models. Extensive experiments show that our technique outperforms standard backtranslation in low-resource scenarios, improves quality on cross-domain tasks, and effectively reduces costs across the board.