Chau Tran

NLLB Team, Marta R. Costa-jussà, James Cross et al. · meta-ai, stanford

Driven by the goal of eradicating language barriers on a global scale, machine translation has solidified itself as a key focus of artificial intelligence research today. However, such efforts have coalesced around a small subset of languages, leaving behind the vast majority of mostly low-resource languages. What does it take to break the 200 language barrier while ensuring safe, high quality results, all while keeping ethical considerations in mind? In No Language Left Behind, we took on this challenge by first contextualizing the need for low-resource language translation support through exploratory interviews with native speakers. Then, we created datasets and models aimed at narrowing the performance gap between low and high-resource languages. More specifically, we developed a conditional compute model based on Sparsely Gated Mixture of Experts that is trained on data obtained with novel and effective data mining techniques tailored for low-resource languages. We propose multiple architectural and training improvements to counteract overfitting while training on thousands of tasks. Critically, we evaluated the performance of over 40,000 different translation directions using a human-translated benchmark, Flores-200, and combined human evaluation with a novel toxicity benchmark covering all languages in Flores-200 to assess translation safety. Our model achieves an improvement of 44% BLEU relative to the previous state-of-the-art, laying important groundwork towards realizing a universal translation system. Finally, we open source all contributions described in this work, accessible at https://github.com/facebookresearch/fairseq/tree/nllb.

Roland Olsson, Chau Tran, Lars Magnusson

We present a new class of neurons, ARNs, which give a cross entropy on test data that is up to three times lower than the one achieved by carefully optimized LSTM neurons. The explanations for the huge improvements that often are achieved are elaborate skip connections through time, up to four internal memory states per neuron and a number of novel activation functions including small quadratic forms. The new neurons were generated using automatic programming and are formulated as pure functional programs that easily can be transformed. We present experimental results for eight datasets and found excellent improvements for seven of them, but LSTM remained the best for one dataset. The results are so promising that automatic programming to generate new neurons should become part of the standard operating procedure for any machine learning practitioner who works on time series data such as sensor signals.

Chau Tran, Kaylea Champion, Benjamin Mako Hill et al.

Many online communities rely on postpublication moderation where contributors, even those that are perceived as being risky, are allowed to publish material immediately and where moderation takes place after the fact. An alternative arrangement involves moderating content before publication. A range of communities have argued against prepublication moderation by suggesting that it makes contributing less enjoyable for new members and that it will distract established community members with extra moderation work. We present an empirical analysis of the effects of a prepublication moderation system called FlaggedRevs that was deployed by several Wikipedia language editions. We used panel data from 17 large Wikipedia editions to test a series of hypotheses related to the effect of the system on activity levels and contribution quality. We found that the system was very effective at keeping low-quality contributions from ever becoming visible. Although there is some evidence that the system discouraged participation among users without accounts, our analysis suggests that the system's effects on contribution volume and quality were moderate at most. Our findings imply that concerns regarding the major negative effects of prepublication moderation systems on contribution quality and project productivity may be overstated.

Simeng Sun, Angela Fan, James Cross et al.

Recent work in multilingual machine translation (MMT) has focused on the potential of positive transfer between languages, particularly cases where higher-resourced languages can benefit lower-resourced ones. While training an MMT model, the supervision signals learned from one language pair can be transferred to the other via the tokens shared by multiple source languages. However, the transfer is inhibited when the token overlap among source languages is small, which manifests naturally when languages use different writing systems. In this paper, we tackle inhibited transfer by augmenting the training data with alternative signals that unify different writing systems, such as phonetic, romanized, and transliterated input. We test these signals on Indic and Turkic languages, two language families where the writing systems differ but languages still share common features. Our results indicate that a straightforward multi-source self-ensemble -- training a model on a mixture of various signals and ensembling the outputs of the same model fed with different signals during inference, outperforms strong ensemble baselines by 1.3 BLEU points on both language families. Further, we find that incorporating alternative inputs via self-ensemble can be particularly effective when training set is small, leading to +5 BLEU when only 5% of the total training data is accessible. Finally, our analysis demonstrates that including alternative signals yields more consistency and translates named entities more accurately, which is crucial for increased factuality of automated systems.

Chau Tran, Shruti Bhosale, James Cross et al.

We describe Facebook's multilingual model submission to the WMT2021 shared task on news translation. We participate in 14 language directions: English to and from Czech, German, Hausa, Icelandic, Japanese, Russian, and Chinese. To develop systems covering all these directions, we focus on multilingual models. We utilize data from all available sources --- WMT, large-scale data mining, and in-domain backtranslation --- to create high quality bilingual and multilingual baselines. Subsequently, we investigate strategies for scaling multilingual model size, such that one system has sufficient capacity for high quality representations of all eight languages. Our final submission is an ensemble of dense and sparse Mixture-of-Expert multilingual translation models, followed by finetuning on in-domain news data and noisy channel reranking. Compared to previous year's winning submissions, our multilingual system improved the translation quality on all language directions, with an average improvement of 2.0 BLEU. In the WMT2021 task, our system ranks first in 10 directions based on automatic evaluation.