For many languages, machine translation progress is hindered by the lack of reliable training data. Models are trained on whatever pre-existing datasets may be available and then augmented with synthetic data, because it is often not economical to pay for the creation of large-scale datasets. But for the case of low-resource languages, would the creation of a few thousand professionally translated sentence pairs give any benefit? In this paper, we show that it does. We describe a broad data collection effort involving around 6k professionally translated sentence pairs for each of 39 low-resource languages, which we make publicly available. We analyse the gains of models trained on this small but high-quality data, showing that it has significant impact even when larger but lower quality pre-existing corpora are used, or when data is augmented with millions of sentences through backtranslation.

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.

A multilingual tokenizer is a fundamental component of multilingual neural machine translation. It is trained from a multilingual corpus. Since a skewed data distribution is considered to be harmful, a sampling strategy is usually used to balance languages in the corpus. However, few works have systematically answered how language imbalance in tokenizer training affects downstream performance. In this work, we analyze how translation performance changes as the data ratios among languages vary in the tokenizer training corpus. We find that while relatively better performance is often observed when languages are more equally sampled, the downstream performance is more robust to language imbalance than we usually expected. Two features, UNK rate and closeness to the character level, can warn of poor downstream performance before performing the task. We also distinguish language sampling for tokenizer training from sampling for model training and show that the model is more sensitive to the latter.

Obtaining meaningful quality scores for machine translation systems through human evaluation remains a challenge given the high variability between human evaluators, partly due to subjective expectations for translation quality for different language pairs. We propose a new metric called XSTS that is more focused on semantic equivalence and a cross-lingual calibration method that enables more consistent assessment. We demonstrate the effectiveness of these novel contributions in large scale evaluation studies across up to 14 language pairs, with translation both into and out of English.

We aim to investigate the performance of current OCR systems on low resource languages and low resource scripts. We introduce and make publicly available a novel benchmark, OCR4MT, consisting of real and synthetic data, enriched with noise, for 60 low-resource languages in low resource scripts. We evaluate state-of-the-art OCR systems on our benchmark and analyse most common errors. We show that OCR monolingual data is a valuable resource that can increase performance of Machine Translation models, when used in backtranslation. We then perform an ablation study to investigate how OCR errors impact Machine Translation performance and determine what is the minimum level of OCR quality needed for the monolingual data to be useful for Machine Translation.

We present MLQE-PE, a new dataset for Machine Translation (MT) Quality Estimation (QE) and Automatic Post-Editing (APE). The dataset contains annotations for eleven language pairs, including both high- and low-resource languages. Specifically, it is annotated for translation quality with human labels for up to 10,000 translations per language pair in the following formats: sentence-level direct assessments and post-editing effort, and word-level binary good/bad labels. Apart from the quality-related scores, each source-translation sentence pair is accompanied by the corresponding post-edited sentence, as well as titles of the articles where the sentences were extracted from, and information on the neural MT models used to translate the text. We provide a thorough description of the data collection and annotation process as well as an analysis of the annotation distribution for each language pair. We also report the performance of baseline systems trained on the MLQE-PE dataset. The dataset is freely available and has already been used for several WMT shared tasks.

One of the biggest challenges hindering progress in low-resource and multilingual machine translation is the lack of good evaluation benchmarks. Current evaluation benchmarks either lack good coverage of low-resource languages, consider only restricted domains, or are low quality because they are constructed using semi-automatic procedures. In this work, we introduce the Flores-101 evaluation benchmark, consisting of 3001 sentences extracted from English Wikipedia and covering a variety of different topics and domains. These sentences have been translated in 101 languages by professional translators through a carefully controlled process. The resulting dataset enables better assessment of model quality on the long tail of low-resource languages, including the evaluation of many-to-many multilingual translation systems, as all translations are fully aligned. By publicly releasing such a high-quality and high-coverage dataset, we hope to foster progress in the machine translation community and beyond.

We present the results of the WMT’22 SharedTask on Large-Scale Machine Translation Evaluation for African Languages. The shared taskincluded both a data and a systems track, alongwith additional innovations, such as a focus onAfrican languages and extensive human evaluation of submitted systems. We received 14system submissions from 8 teams, as well as6 data track contributions. We report a largeprogress in the quality of translation for Africanlanguages since the last iteration of this sharedtask: there is an increase of about 7.5 BLEUpoints across 72 language pairs, and the average BLEU scores went from 15.09 to 22.60.

The scarcity of parallel data is a major obstacle for training high-quality machine translation systems for low-resource languages. Fortunately, some low-resource languages are linguistically related or similar to high-resource languages; these related languages may share many lexical or syntactic structures. In this work, we exploit this linguistic overlap to facilitate translating to and from a low-resource language with only monolingual data, in addition to any parallel data in the related high-resource language. Our method, NMT-Adapt, combines denoising autoencoding, back-translation and adversarial objectives to utilize monolingual data for low-resource adaptation. We experiment on 7 languages from three different language families and show that our technique significantly improves translation into low-resource language compared to other translation baselines.

Multilingual neural machine translation has shown the capability of directly translating between language pairs unseen in training, i.e. zero-shot translation. Despite being conceptually attractive, it often suffers from low output quality. The difficulty of generalizing to new translation directions suggests the model representations are highly specific to those language pairs seen in training. We demonstrate that a main factor causing the language-specific representations is the positional correspondence to input tokens. We show that this can be easily alleviated by removing residual connections in an encoder layer. With this modification, we gain up to 18.5 BLEU points on zero-shot translation while retaining quality on supervised directions. The improvements are particularly prominent between related languages, where our proposed model outperforms pivot-based translation. Moreover, our approach allows easy integration of new languages, which substantially expands translation coverage. By thorough inspections of the hidden layer outputs, we show that our approach indeed leads to more language-independent representations.

Quality estimation aims to measure the quality of translated content without access to a reference translation. This is crucial for machine translation systems in real-world scenarios where high-quality translation is needed. While many approaches exist for quality estimation, they are based on supervised machine learning requiring costly human labelled data. As an alternative, we propose a technique that does not rely on examples from human-annotators and instead uses synthetic training data. We train off-the-shelf architectures for supervised quality estimation on our synthetic data and show that the resulting models achieve comparable performance to models trained on human-annotated data, both for sentence and word-level prediction.

We present an approach based on multilingual sentence embeddings to automatically extract parallel sentences from the content of Wikipedia articles in 96 languages, including several dialects or low-resource languages. We do not limit the extraction process to alignments with English, but we systematically consider all possible language pairs. In total, we are able to extract 135M parallel sentences for 16720 different language pairs, out of which only 34M are aligned with English. This corpus is freely available. To get an indication on the quality of the extracted bitexts, we train neural MT baseline systems on the mined data only for 1886 languages pairs, and evaluate them on the TED corpus, achieving strong BLEU scores for many language pairs. The WikiMatrix bitexts seem to be particularly interesting to train MT systems between distant languages without the need to pivot through English.

Sentence-level Quality estimation (QE) of machine translation is traditionally formulated as a regression task, and the performance of QE models is typically measured by Pearson correlation with human labels. Recent QE models have achieved previously-unseen levels of correlation with human judgments, but they rely on large multilingual contextualized language models that are computationally expensive and make them infeasible for real-world applications. In this work, we evaluate several model compression techniques for QE and find that, despite their popularity in other NLP tasks, they lead to poor performance in this regression setting. We observe that a full model parameterization is required to achieve SoTA results in a regression task. However, we argue that the level of expressiveness of a model in a continuous range is unnecessary given the downstream applications of QE, and show that reframing QE as a classification problem and evaluating QE models using classification metrics would better reflect their actual performance in real-world applications.

Cross-lingual named-entity lexica are an important resource to multilingual NLP tasks such as machine translation and cross-lingual wikification. While knowledge bases contain a large number of entities in high-resource languages such as English and French, corresponding entities for lower-resource languages are often missing. To address this, we propose Lexical-Semantic-Phonetic Align (LSP-Align), a technique to automatically mine cross-lingual entity lexica from mined web data. We demonstrate LSP-Align outperforms baselines at extracting cross-lingual entity pairs and mine 164 million entity pairs from 120 different languages aligned with English. We release these cross-lingual entity pairs along with the massively multilingual tagged named entity corpus as a resource to the NLP community.

NLP models are vulnerable to data poisoning attacks. One type of attack can plant a backdoor in a model by injecting poisoned examples in training, causing the victim model to misclassify test instances which include a specific pattern. Although defences exist to counter these attacks, they are specific to an attack type or pattern. In this paper, we propose a generic defence mechanism by making the training process robust to poisoning attacks through gradient shaping methods, based on differentially private training. We show that our method is highly effective in mitigating, or even eliminating, poisoning attacks on text classification, with only a small cost in predictive accuracy.

We present the results of the first task on Large-Scale Multilingual Machine Translation. The task consists on the many-to-many evaluation of a single model across a variety of source and target languages. This year, the task consisted on three different settings: (i) SMALL-TASK1 (Central/South-Eastern European Languages), (ii) the SMALL-TASK2 (South-East Asian Languages), and (iii) FULL-TASK (all 101 x 100 language pairs). All the tasks used the FLORES-101 dataset as the evaluation benchmark. To ensure the longevity of the dataset, the test sets were not publicly released and the models were evaluated in a controlled environment on Dynabench. There were a total of 10 participating teams for the tasks, with a total of 151 intermediate model submissions and 13 final models. This year’s result show a significant improvement over the known base-lines with +17.8 BLEU for SMALL-TASK2, +10.6 for FULL-TASK and +3.6 for SMALL-TASK1.

Predicting the quality of machine translation has traditionally been addressed with language-specific models, under the assumption that the quality label distribution or linguistic features exhibit traits that are not shared across languages. An obvious disadvantage of this approach is the need for labelled data for each given language pair. We challenge this assumption by exploring different approaches to multilingual Quality Estimation (QE), including using scores from translation models. We show that these outperform single-language models, particularly in less balanced quality label distributions and low-resource settings. In the extreme case of zero-shot QE, we show that it is possible to accurately predict quality for any given new language from models trained on other languages. Our findings indicate that state-of-the-art neural QE models based on powerful pre-trained representations generalise well across languages, making them more applicable in real-world settings.

Document alignment aims to identify pairs of documents in two distinct languages that are of comparable content or translations of each other. Such aligned data can be used for a variety of NLP tasks from training cross-lingual representations to mining parallel data for machine translation. In this paper we develop an unsupervised scoring function that leverages cross-lingual sentence embeddings to compute the semantic distance between documents in different languages. These semantic distances are then used to guide a document alignment algorithm to properly pair cross-lingual web documents across a variety of low, mid, and high-resource language pairs. Recognizing that our proposed scoring function and other state of the art methods are computationally intractable for long web documents, we utilize a more tractable greedy algorithm that performs comparably. We experimentally demonstrate that our distance metric performs better alignment than current baselines outperforming them by 7% on high-resource language pairs, 15% on mid-resource language pairs, and 22% on low-resource language pairs.

Reliably evaluating Machine Translation (MT) through automated metrics is a long-standing problem. One of the main challenges is the fact that multiple outputs can be equally valid. Attempts to minimise this issue include metrics that relax the matching of MT output and reference strings, and the use of multiple references. The latter has been shown to significantly improve the performance of evaluation metrics. However, collecting multiple references is expensive and in practice a single reference is generally used. In this paper, we propose an alternative approach: instead of modelling linguistic variation in human reference we exploit the MT model uncertainty to generate multiple diverse translations and use these: (i) as surrogates to reference translations; (ii) to obtain a quantification of translation variability to either complement existing metric scores or (iii) replace references altogether. We show that for a number of popular evaluation metrics our variability estimates lead to substantial improvements in correlation with human judgements of quality by up 15%.

Recent advances in pre-trained multilingual language models lead to state-of-the-art results on the task of quality estimation (QE) for machine translation. A carefully engineered ensemble of such models won the QE shared task at WMT19. Our in-depth analysis, however, shows that the success of using pre-trained language models for QE is over-estimated due to three issues we observed in current QE datasets: (i) The distributions of quality scores are imbalanced and skewed towards good quality scores; (iii) QE models can perform well on these datasets while looking at only source or translated sentences; (iii) They contain statistical artifacts that correlate well with human-annotated QE labels. Our findings suggest that although QE models might capture fluency of translated sentences and complexity of source sentences, they cannot model adequacy of translations effectively.

This paper shows that pretraining multilingual language models at scale leads to significant performance gains for a wide range of cross-lingual transfer tasks. We train a Transformer-based masked language model on one hundred languages, using more than two terabytes of filtered CommonCrawl data. Our model, dubbed XLM-R, significantly outperforms multilingual BERT (mBERT) on a variety of cross-lingual benchmarks, including +14.6% average accuracy on XNLI, +13% average F1 score on MLQA, and +2.4% F1 score on NER. XLM-R performs particularly well on low-resource languages, improving 15.7% in XNLI accuracy for Swahili and 11.4% for Urdu over previous XLM models. We also present a detailed empirical analysis of the key factors that are required to achieve these gains, including the trade-offs between (1) positive transfer and capacity dilution and (2) the performance of high and low resource languages at scale. Finally, we show, for the first time, the possibility of multilingual modeling without sacrificing per-language performance; XLM-R is very competitive with strong monolingual models on the GLUE and XNLI benchmarks. We will make our code and models publicly available.

Cross-lingual document alignment aims to identify pairs of documents in two distinct languages that are of comparable content or translations of each other. In this paper, we exploit the signals embedded in URLs to label web documents at scale with an average precision of 94.5% across different language pairs. We mine sixty-eight snapshots of the Common Crawl corpus and identify web document pairs that are translations of each other. We release a new web dataset consisting of over 392 million URL pairs from Common Crawl covering documents in 8144 language pairs of which 137 pairs include English. In addition to curating this massive dataset, we introduce baseline methods that leverage cross-lingual representations to identify aligned documents based on their textual content. Finally, we demonstrate the value of this parallel documents dataset through a downstream task of mining parallel sentences and measuring the quality of machine translations from models trained on this mined data. Our objective in releasing this dataset is to foster new research in cross-lingual NLP across a variety of low, medium, and high-resource languages.

Pre-training text representations have led to significant improvements in many areas of natural language processing. The quality of these models benefits greatly from the size of the pretraining corpora as long as its quality is preserved. In this paper, we describe an automatic pipeline to extract massive high-quality monolingual datasets from Common Crawl for a variety of languages. Our pipeline follows the data processing introduced in fastText (Mikolov et al., 2017; Grave et al., 2018), that deduplicates documents and identifies their language. We augment this pipeline with a filtering step to select documents that are close to high quality corpora like Wikipedia.

Quality Estimation (QE) is an important component in making Machine Translation (MT) useful in real-world applications, as it is aimed to inform the user on the quality of the MT output at test time. Existing approaches require large amounts of expert annotated data, computation, and time for training. As an alternative, we devise an unsupervised approach to QE where no training or access to additional resources besides the MT system itself is required. Different from most of the current work that treats the MT system as a black box, we explore useful information that can be extracted from the MT system as a by-product of translation. By utilizing methods for uncertainty quantification, we achieve very good correlation with human judgments of quality, rivaling state-of-the-art supervised QE models. To evaluate our approach we collect the first dataset that enables work on both black-box and glass-box approaches to QE.

We report the findings of the second edition of the shared task on improving robustness in Machine Translation (MT). The task aims to test current machine translation systems in their ability to handle challenges facing MT models to be deployed in the real world, including domain diversity and non-standard texts common in user generated content, especially in social media. We cover two language pairs – English-German and English-Japanese and provide test sets in zero-shot and few-shot variants. Participating systems are evaluated both automatically and manually, with an additional human evaluation for ”catastrophic errors”. We received 59 submissions by 11 participating teams from a variety of types of institutions.

Following two preceding WMT Shared Task on Parallel Corpus Filtering (Koehn et al., 2018, 2019), we posed again the challenge of assigning sentence-level quality scores for very noisy corpora of sentence pairs crawled from the web, with the goal of sub-selecting the highest-quality data to be used to train ma-chine translation systems. This year, the task tackled the low resource condition of Pashto–English and Khmer–English and also included the challenge of sentence alignment from document pairs.

We report the results of the WMT20 shared task on Quality Estimation, where the challenge is to predict the quality of the output of neural machine translation systems at the word, sentence and document levels. This edition included new data with open domain texts, direct assessment annotations, and multiple language pairs: English-German, English-Chinese, Russian-English, Romanian-English, Estonian-English, Sinhala-English and Nepali-English data for the sentence-level subtasks, English-German and English-Chinese for the word-level subtask, and English-French data for the document-level subtask. In addition, we made neural machine translation models available to participants. 19 participating teams from 27 institutions submitted altogether 1374 systems to different task variants and language pairs.

This paper presents our submission to the WMT2020 Shared Task on Quality Estimation (QE). We participate in Task and Task 2 focusing on sentence-level prediction. We explore (a) a black-box approach to QE based on pre-trained representations; and (b) glass-box approaches that leverage various indicators that can be extracted from the neural MT systems. In addition to training a feature-based regression model using glass-box quality indicators, we also test whether they can be used to predict MT quality directly with no supervision. We assess our systems in a multi-lingual setting and show that both types of approaches generalise well across languages. Our black-box QE models tied for the winning submission in four out of seven language pairs inTask 1, thus demonstrating very strong performance. The glass-box approaches also performed competitively, representing a light-weight alternative to the neural-based models.

For machine translation, a vast majority of language pairs in the world are considered low-resource because they have little parallel data available. Besides the technical challenges of learning with limited supervision, it is difficult to evaluate methods trained on low-resource language pairs because of the lack of freely and publicly available benchmarks. In this work, we introduce the FLORES evaluation datasets for Nepali–English and Sinhala– English, based on sentences translated from Wikipedia. Compared to English, these are languages with very different morphology and syntax, for which little out-of-domain parallel data is available and for which relatively large amounts of monolingual data are freely available. We describe our process to collect and cross-check the quality of translations, and we report baseline performance using several learning settings: fully supervised, weakly supervised, semi-supervised, and fully unsupervised. Our experiments demonstrate that current state-of-the-art methods perform rather poorly on this benchmark, posing a challenge to the research community working on low-resource MT. Data and code to reproduce our experiments are available at https://github.com/facebookresearch/flores.

Following the WMT 2018 Shared Task on Parallel Corpus Filtering, we posed the challenge of assigning sentence-level quality scores for very noisy corpora of sentence pairs crawled from the web, with the goal of sub-selecting 2% and 10% of the highest-quality data to be used to train machine translation systems. This year, the task tackled the low resource condition of Nepali-English and Sinhala-English. Eleven participants from companies, national research labs, and universities participated in this task.

In this paper, we describe our submission to the WMT19 low-resource parallel corpus filtering shared task. Our main approach is based on the LASER toolkit (Language-Agnostic SEntence Representations), which uses an encoder-decoder architecture trained on a parallel corpus to obtain multilingual sentence representations. We then use the representations directly to score and filter the noisy parallel sentences without additionally training a scoring function. We contrast our approach to other promising methods and show that LASER yields strong results. Finally, we produce an ensemble of different scoring methods and obtain additional gains. Our submission achieved the best overall performance for both the Nepali-English and Sinhala-English 1M tasks by a margin of 1.3 and 1.4 BLEU respectively, as compared to the second best systems. Moreover, our experiments show that this technique is promising for low and even no-resource scenarios.

In this article, we explore the potential of using sentence-level discourse structure for machine translation evaluation. We first design discourse-aware similarity measures, which use all-subtree kernels to compare discourse parse trees in accordance with the Rhetorical Structure Theory (RST). Then, we show that a simple linear combination with these measures can help improve various existing machine translation evaluation metrics regarding correlation with human judgments both at the segment level and at the system level. This suggests that discourse information is complementary to the information used by many of the existing evaluation metrics, and thus it could be taken into account when developing richer evaluation metrics, such as the WMT-14 winning combined metric DiscoTKparty. We also provide a detailed analysis of the relevance of various discourse elements and relations from the RST parse trees for machine translation evaluation. In particular, we show that (i) all aspects of the RST tree are relevant, (ii) nuclearity is more useful than relation type, and (iii) the similarity of the translation RST tree to the reference RST tree is positively correlated with translation quality.

Evaluation of machine translation (MT) into morphologically rich languages (MRL) has not been well studied despite posing many challenges. In this paper, we explore the use of embeddings obtained from different levels of lexical and morpho-syntactic linguistic analysis and show that they improve MT evaluation into an MRL. Specifically we report on Arabic, a language with complex and rich morphology. Our results show that using a neural-network model with different input representations produces results that clearly outperform the state-of-the-art for MT evaluation into Arabic, by almost over 75% increase in correlation with human judgments on pairwise MT evaluation quality task. More importantly, we demonstrate the usefulness of morpho-syntactic representations to model sentence similarity for MT evaluation and address complex linguistic phenomena of Arabic.

This paper presents the AMARA corpus of on-line educational content: a new parallel corpus of educational video subtitles, multilingually aligned for 20 languages, i.e. 20 monolingual corpora and 190 parallel corpora. This corpus includes both resource-rich languages such as English and Arabic, and resource-poor languages such as Hindi and Thai. In this paper, we describe the gathering, validation, and preprocessing of a large collection of parallel, community-generated subtitles. Furthermore, we describe the methodology used to prepare the data for Machine Translation tasks. Additionally, we provide a document-level, jointly aligned development and test sets for 14 language pairs, designed for tuning and testing Machine Translation systems. We provide baseline results for these tasks, and highlight some of the challenges we face when building machine translation systems for educational content.

We describe the Arabic-English and English-Arabic statistical machine translation systems developed by the Qatar Computing Research Institute for the IWSLT’2013 evaluation campaign on spoken language translation. We used one phrase-based and two hierarchical decoders, exploring various settings thereof. We further experimented with three domain adaptation methods, and with various Arabic word segmentation schemes. Combining the output of several systems yielded a gain of up to 3.4 BLEU points over the baseline. Here we also describe a specialized normalization scheme for evaluating Arabic output, which was adopted for the IWSLT’2013 evaluation campaign.

In this paper, we introduce a new parallel corpus of subtitles of educational videos: the AMARA corpus for online educational content. We crawl a multilingual collection community generated subtitles, and present the results of processing the Arabic–English portion of the data, which yields a parallel corpus of about 2.6M Arabic and 3.9M English words. We explore different approaches to align the segments, and extrinsically evaluate the resulting parallel corpus on the standard TED-talks tst-2010. We observe that the data can be successfully used for this task, and also observe an absolute improvement of 1.6 BLEU when it is used in combination with TED data. Finally, we analyze some of the specific challenges when translating the educational content.