Translationese as a Language in "Multilingual" NMT
This addresses the issue for machine translation practitioners and evaluators by improving output naturalness and exposing biases in automated metrics, though it is incremental as it builds on existing multilingual NMT methods.
The paper tackled the problem of machine translation producing undesirable 'translationese' artifacts that inflate BLEU scores but reduce human preference, by modeling translationese and original text as separate languages in a multilingual NMT model to enable zero-shot translation between original source and target texts. The result was a model that produced more natural outputs, yielding gains in human evaluation scores for accuracy and fluency, and demonstrated the ability to manipulate BLEU scores by biasing towards translationese.
Machine translation has an undesirable propensity to produce "translationese" artifacts, which can lead to higher BLEU scores while being liked less by human raters. Motivated by this, we model translationese and original (i.e. natural) text as separate languages in a multilingual model, and pose the question: can we perform zero-shot translation between original source text and original target text? There is no data with original source and original target, so we train sentence-level classifiers to distinguish translationese from original target text, and use this classifier to tag the training data for an NMT model. Using this technique we bias the model to produce more natural outputs at test time, yielding gains in human evaluation scores on both accuracy and fluency. Additionally, we demonstrate that it is possible to bias the model to produce translationese and game the BLEU score, increasing it while decreasing human-rated quality. We analyze these models using metrics to measure the degree of translationese in the output, and present an analysis of the capriciousness of heuristically-based train-data tagging.