Back-translation is a well established approach to improve the performance of Neural Machine Translation (NMT) systems when large monolingual corpora of the target language and domain are available. Recently, diverse approaches have been proposed to get better automatic evaluation results of NMT models using back-translation, including the use of sampling instead of beam search as decoding algorithm for creating the synthetic corpus. Alternatively, it has been proposed to append a tag to the back-translated corpus for helping the NMT system to distinguish the synthetic bilingual corpus from the authentic one. However, not all the combinations of the previous approaches have been tested, and thus it is not clear which is the best approach for developing a given NMT system. In this work, we empirically compare and combine existing techniques for back-translation in a real low resource setting: the translation of clinical notes from Basque into Spanish. Apart from automatically evaluating the MT systems, we ask bilingual healthcare workers to perform a human evaluation, and analyze the different synthetic corpora by measuring their lexical diversity (LD). For reproducibility and generalizability, we repeat our experiments for German to English translation using public data. The results suggest that in lower resource scenarios tagging only helps when using sampling for decoding, in contradiction with the previous literature using bigger corpora from the news domain. When fine-tuning with a few thousand bilingual in-domain sentences, one of our proposed method (tagged restricted sampling) obtains the best results both in terms of automatic and human evaluation. We will publish the code upon acceptance.

Round-trip Machine Translation (MT) is a popular choice for paraphrase generation, which leverages readily available parallel corpora for supervision. In this paper, we formalize the implicit similarity function induced by this approach, and show that it is susceptible to non-paraphrase pairs sharing a single ambiguous translation. Based on these insights, we design an alternative similarity metric that mitigates this issue by requiring the entire translation distribution to match, and implement a relaxation of it through the Information Bottleneck method. Our approach incorporates an adversarial term into MT training in order to learn representations that encode as much information about the reference translation as possible, while keeping as little information about the input as possible. Paraphrases can be generated by decoding back to the source from this representation, without having to generate pivot translations. In addition to being more principled and efficient than round-trip MT, our approach offers an adjustable parameter to control the fidelity-diversity trade-off, and obtains better results in our experiments.

Document-level Neural Machine Translation aims to increase the quality of neural translation models by taking into account contextual information. Properly modelling information beyond the sentence level can result in improved machine translation output in terms of coherence, cohesion and consistency. Suitable corpora for context-level modelling are necessary to both train and evaluate context-aware systems, but are still relatively scarce. In this work we describe TANDO, a document-level corpus for the under-resourced Basque-Spanish language pair, which we share with the scientific community. The corpus is composed of parallel data from three different domains and has been prepared with context-level information. Additionally, the corpus includes contrastive test sets for fine-grained evaluations of gender and register contextual phenomena on both source and target language sides. To establish the usefulness of the corpus, we trained and evaluated baseline Transformer models and context-aware variants based on context concatenation. Our results indicate that the corpus is suitable for fine-grained evaluation of document-level machine translation systems.

In this work, we present the work that has been carried on in the MT4All CEF project and the resources that it has generated by leveraging recent research carried out in the field of unsupervised learning. In the course of the project 18 monolingual corpora for specific domains and languages have been collected, and 12 bilingual dictionaries and translation models have been generated. As part of the research, the unsupervised MT methodology based only on monolingual corpora (Artetxe et al., 2017) has been tested on a variety of languages and domains. Results show that in specialised domains, when there is enough monolingual in-domain data, unsupervised results are comparable to those of general domain supervised translation, and that, at any rate, unsupervised techniques can be used to boost results whenever very little data is available.

Relation extraction systems require large amounts of labeled examples which are costly to annotate. In this work we reformulate relation extraction as an entailment task, with simple, hand-made, verbalizations of relations produced in less than 15 min per relation. The system relies on a pretrained textual entailment engine which is run as-is (no training examples, zero-shot) or further fine-tuned on labeled examples (few-shot or fully trained). In our experiments on TACRED we attain 63% F1 zero-shot, 69% with 16 examples per relation (17% points better than the best supervised system on the same conditions), and only 4 points short to the state-of-the-art (which uses 20 times more training data). We also show that the performance can be improved significantly with larger entailment models, up to 12 points in zero-shot, allowing to report the best results to date on TACRED when fully trained. The analysis shows that our few-shot systems are specially effective when discriminating between relations, and that the performance difference in low data regimes comes mainly from identifying no-relation cases.

Recent research on cross-lingual word embeddings has been dominated by unsupervised mapping approaches that align monolingual embeddings. Such methods critically rely on those embeddings having a similar structure, but it was recently shown that the separate training in different languages causes departures from this assumption. In this paper, we propose an alternative approach that does not have this limitation, while requiring a weak seed dictionary (e.g., a list of identical words) as the only form of supervision. Rather than aligning two fixed embedding spaces, our method works by fixing the target language embeddings, and learning a new set of embeddings for the source language that are aligned with them. To that end, we use an extension of skip-gram that leverages translated context words as anchor points, and incorporates self-learning and iterative restarts to reduce the dependency on the initial dictionary. Our approach outperforms conventional mapping methods on bilingual lexicon induction, and obtains competitive results in the downstream XNLI task.

We review motivations, definition, approaches, and methodology for unsupervised cross-lingual learning and call for a more rigorous position in each of them. An existing rationale for such research is based on the lack of parallel data for many of the world’s languages. However, we argue that a scenario without any parallel data and abundant monolingual data is unrealistic in practice. We also discuss different training signals that have been used in previous work, which depart from the pure unsupervised setting. We then describe common methodological issues in tuning and evaluation of unsupervised cross-lingual models and present best practices. Finally, we provide a unified outlook for different types of research in this area (i.e., cross-lingual word embeddings, deep multilingual pretraining, and unsupervised machine translation) and argue for comparable evaluation of these models.

Existing models of multilingual sentence embeddings require large parallel data resources which are not available for low-resource languages. We propose a novel unsupervised method to derive multilingual sentence embeddings relying only on monolingual data. We first produce a synthetic parallel corpus using unsupervised machine translation, and use it to fine-tune a pretrained cross-lingual masked language model (XLM) to derive the multilingual sentence representations. The quality of the representations is evaluated on two parallel corpus mining tasks with improvements of up to 22 F1 points over vanilla XLM. In addition, we observe that a single synthetic bilingual corpus is able to improve results for other language pairs.

Both human and machine translation play a central role in cross-lingual transfer learning: many multilingual datasets have been created through professional translation services, and using machine translation to translate either the test set or the training set is a widely used transfer technique. In this paper, we show that such translation process can introduce subtle artifacts that have a notable impact in existing cross-lingual models. For instance, in natural language inference, translating the premise and the hypothesis independently can reduce the lexical overlap between them, which current models are highly sensitive to. We show that some previous findings in cross-lingual transfer learning need to be reconsidered in the light of this phenomenon. Based on the gained insights, we also improve the state-of-the-art in XNLI for the translate-test and zero-shot approaches by 4.3 and 2.8 points, respectively.

In this paper we describe the systems developed at Ixa for our participation in WMT20 Biomedical shared task in three language pairs, en-eu, en-es and es-en. When defining our approach, we have put the focus on making an efficient use of corpora recently compiled for training Machine Translation (MT) systems to translate Covid-19 related text, as well as reusing previously compiled corpora and developed systems for biomedical or clinical domain. Regarding the techniques used, we base on the findings from our previous works for translating clinical texts into Basque, making use of clinical terminology for adapting the MT systems to the clinical domain. However, after manually inspecting some of the outputs generated by our systems, for most of the submissions we end up using the system trained only with the basic corpus, since the systems including the clinical terminologies generated outputs shorter in length than the corresponding references. Thus, we present simple baselines for translating abstracts between English and Spanish (en/es); while for translating abstracts and terms from English into Basque (en-eu), we concatenate the best en-es system for each kind of text with our es-eu system. We present automatic evaluation results in terms of BLEU scores, and analyse the effect of including clinical terminology on the average sentence length of the generated outputs. Following the recent recommendations for a responsible use of GPUs for NLP research, we include an estimation of the generated CO2 emissions, based on the power consumed for training the MT systems.

While machine translation has traditionally relied on large amounts of parallel corpora, a recent research line has managed to train both Neural Machine Translation (NMT) and Statistical Machine Translation (SMT) systems using monolingual corpora only. In this paper, we identify and address several deficiencies of existing unsupervised SMT approaches by exploiting subword information, developing a theoretically well founded unsupervised tuning method, and incorporating a joint refinement procedure. Moreover, we use our improved SMT system to initialize a dual NMT model, which is further fine-tuned through on-the-fly back-translation. Together, we obtain large improvements over the previous state-of-the-art in unsupervised machine translation. For instance, we get 22.5 BLEU points in English-to-German WMT 2014, 5.5 points more than the previous best unsupervised system, and 0.5 points more than the (supervised) shared task winner back in 2014.

Recent research in cross-lingual word embeddings has almost exclusively focused on offline methods, which independently train word embeddings in different languages and map them to a shared space through linear transformations. While several authors have questioned the underlying isomorphism assumption, which states that word embeddings in different languages have approximately the same structure, it is not clear whether this is an inherent limitation of mapping approaches or a more general issue when learning cross-lingual embeddings. So as to answer this question, we experiment with parallel corpora, which allows us to compare offline mapping to an extension of skip-gram that jointly learns both embedding spaces. We observe that, under these ideal conditions, joint learning yields to more isomorphic embeddings, is less sensitive to hubness, and obtains stronger results in bilingual lexicon induction. We thus conclude that current mapping methods do have strong limitations, calling for further research to jointly learn cross-lingual embeddings with a weaker cross-lingual signal.

A recent research line has obtained strong results on bilingual lexicon induction by aligning independently trained word embeddings in two languages and using the resulting cross-lingual embeddings to induce word translation pairs through nearest neighbor or related retrieval methods. In this paper, we propose an alternative approach to this problem that builds on the recent work on unsupervised machine translation. This way, instead of directly inducing a bilingual lexicon from cross-lingual embeddings, we use them to build a phrase-table, combine it with a language model, and use the resulting machine translation system to generate a synthetic parallel corpus, from which we extract the bilingual lexicon using statistical word alignment techniques. As such, our method can work with any word embedding and cross-lingual mapping technique, and it does not require any additional resource besides the monolingual corpus used to train the embeddings. When evaluated on the exact same cross-lingual embeddings, our proposed method obtains an average improvement of 6 accuracy points over nearest neighbor and 4 points over CSLS retrieval, establishing a new state-of-the-art in the standard MUSE dataset.

Following the recent success of word embeddings, it has been argued that there is no such thing as an ideal representation for words, as different models tend to capture divergent and often mutually incompatible aspects like semantics/syntax and similarity/relatedness. In this paper, we show that each embedding model captures more information than directly apparent. A linear transformation that adjusts the similarity order of the model without any external resource can tailor it to achieve better results in those aspects, providing a new perspective on how embeddings encode divergent linguistic information. In addition, we explore the relation between intrinsic and extrinsic evaluation, as the effect of our transformations in downstream tasks is higher for unsupervised systems than for supervised ones.

We describe the first experimental results in neural machine translation for Basque. As a synthetic language featuring agglutinative morphology, an extended case system, complex verbal morphology and relatively free word order, Basque presents a large number of challenging characteristics for machine translation in general, and for data-driven approaches such as attentionbased encoder-decoder models in particular. We present our results on a large range of experiments in Basque-Spanish translation, comparing several neural machine translation system variants with both rule-based and statistical machine translation systems. We demonstrate that significant gains can be obtained with a neural network approach for this challenging language pair, and describe optimal configurations in terms of word segmentation and decoding parameters, measured against test sets that feature multiple references to account for word order variability.

Recent work has managed to learn cross-lingual word embeddings without parallel data by mapping monolingual embeddings to a shared space through adversarial training. However, their evaluation has focused on favorable conditions, using comparable corpora or closely-related languages, and we show that they often fail in more realistic scenarios. This work proposes an alternative approach based on a fully unsupervised initialization that explicitly exploits the structural similarity of the embeddings, and a robust self-learning algorithm that iteratively improves this solution. Our method succeeds in all tested scenarios and obtains the best published results in standard datasets, even surpassing previous supervised systems. Our implementation is released as an open source project at https://github.com/artetxem/vecmap.

While modern machine translation has relied on large parallel corpora, a recent line of work has managed to train Neural Machine Translation (NMT) systems from monolingual corpora only (Artetxe et al., 2018c; Lample et al., 2018). Despite the potential of this approach for low-resource settings, existing systems are far behind their supervised counterparts, limiting their practical interest. In this paper, we propose an alternative approach based on phrase-based Statistical Machine Translation (SMT) that significantly closes the gap with supervised systems. Our method profits from the modular architecture of SMT: we first induce a phrase table from monolingual corpora through cross-lingual embedding mappings, combine it with an n-gram language model, and fine-tune hyperparameters through an unsupervised MERT variant. In addition, iterative backtranslation improves results further, yielding, for instance, 14.08 and 26.22 BLEU points in WMT 2014 English-German and English-French, respectively, an improvement of more than 7-10 BLEU points over previous unsupervised systems, and closing the gap with supervised SMT (Moses trained on Europarl) down to 2-5 BLEU points. Our implementation is available at https://github.com/artetxem/monoses.

Most methods to learn bilingual word embeddings rely on large parallel corpora, which is difficult to obtain for most language pairs. This has motivated an active research line to relax this requirement, with methods that use document-aligned corpora or bilingual dictionaries of a few thousand words instead. In this work, we further reduce the need of bilingual resources using a very simple self-learning approach that can be combined with any dictionary-based mapping technique. Our method exploits the structural similarity of embedding spaces, and works with as little bilingual evidence as a 25 word dictionary or even an automatically generated list of numerals, obtaining results comparable to those of systems that use richer resources.

This paper presents a method to improve the translation of Verb-Noun Combinations (VNCs) in a rule-based Machine Translation (MT) system for Spanish-Basque. Linguistic information about a set of VNCs is gathered from the public database Konbitzul, and it is integrated into the MT system, leading to an improvement in BLEU, NIST and TER scores, as well as the results being evidently better according to human evaluators.

This paper presents how an state-of-the-art SMT system is enriched by using an extra in-domain parallel corpora extracted from Wikipedia. We collect corpora from parallel titles and from parallel fragments in comparable articles from Wikipedia. We carried out an evaluation with a double objective: evaluating the quality of the extracted data and evaluating the improvement due to the domain-adaptation. We think this can be very useful for languages with limited amount of parallel corpora, where in-domain data is crucial to improve the performance of MT sytems. The experiments on the Spanish-English language pair improve a baseline trained with the Europarl corpus in more than 2 points of BLEU when translating in the Computer Science domain.

We present a linguistic analysis of a set of English and Spanish verb+noun combinations (VNCs), and a method to use this information to improve VNC identification. Firstly, a sample of frequent VNCs are analysed in-depth and tagged along lexico-semantic and morphosyntactic dimensions, obtaining satisfactory inter-annotator agreement scores. Then, a VNC identification experiment is undertaken, where the analysed linguistic data is combined with chunking information and syntactic dependencies. A comparison between the results of the experiment and the results obtained by a basic detection method shows that VNC identification can be greatly improved by using linguistic information, as a large number of additional occurrences are detected with high precision.

This work compares the post-editing productivity of professional translators and lay users. We integrate an English to Basque MT system within Bologna Translation Service, an end-to-end translation management platform, and perform a producitivity experiment in a real working environment. Six translators and six lay users translate or post-edit two texts from English into Basque. Results suggest that overall, post-editing increases translation throughput for both translators and users, although the latter seem to benefit more from the MT output. We observe that translators and users perceive MT differently. Additionally, a preliminary analysis seems to suggest that familiarity with the domain, source text complexity and MT quality might affect potential productivity gain.

We present a new morphological processor for Biscayan, a dialect of Basque, developed on the description of the morphology of standard Basque. The database for the standard morphology has been extended for dialects and an open-source tool for morphological description named foma is used for building the processor. Biscayan is a dialect of the Basque language spoken mainly in Biscay, a province on the western of the Basque Country. The description of the lexicon and the morphotactics (or word grammar) for the standard Basque was carried out using a relational database and the database has been extended in order to include dialectal variants linked to the standard entries. XuxenB, a spelling checker/corrector for this dialect, is the first application of this work. Additionally to the basic analyzer used for spelling, a new transducer is included. It is an enhanced analyzer for linking standard form with the corresponding standard ones. It is used in correction for generation of proposals when in the input text appear standard forms which we want to replace with dialectal forms.

We present our initial strategy for Spanish-to-Basque MultiEngine Machine Translation, a language pair with very different structure and word order and with no huge parallel corpus available. This hybrid proposal is based on the combination of three different MT paradigms: Example-Based MT, Statistical MT and Rule- Based MT. We have evaluated the system, reporting automatic evaluation metrics for a corpus in a test domain. The first results obtained are encouraging.

We present the current status of development of an open architecture for the translation from Spanish into Basque. The machine translation architecture uses an open source analyser for Spanish and new modules mainly based on finite-state transducers. The project is integrated in the OpenTrad initiative, a larger government funded project shared among different universities and small companies, which will also include MT engines for translation among the main languages in Spain. The main objective is the construction of an open, reusable and interoperable framework. This paper describes the design of the engine, the formats it uses for the communication among the modules, the modules reused from other project named Matxin and the new modules we are building.