Explore the vast range of titles available.

We hypothesize that parallel corpora as well as machine translation outputs contain many literal translations that are the result of transferring the constructions of the source language to the target language. When translating passive expressions from English to Spanish, there are several constructions available, however, both automatic and human (if of low quality) translations tend to select the periphrastic structure, which is the literal construction. The objective of this article is to make use of strategies trained on monolingual corpora to translate English passive expressions into Spanish so as to verify whether unsupervised translation with monolingual corpora benefits syntactic diversity. Special attention will be given to the monolingual-based strategy relying on dependency-based contextualization. The results of the experiments carried out show that the methods relying on monolingual corpora tend to offer more non-literal translations (middle-voice) than those trained on parallel corpora.

Multiword Expressions (MWEs) are idiosyncratic combinations of words which pose important challenges to Natural Language Processing. Some kinds of MWEs, such as verbal ones, are particularly hard to identify in corpora, due to their high degree of morphosyntactic flexibility. This paper describes a linguistically motivated method to gather detailed information about verb+noun MWEs (VNMWEs) from corpora. Although the main focus of this study is Spanish, the method is easily adaptable to other languages. Monolingual and parallel corpora are used as input, and data about the morphosyntactic variability of VNMWEs is extracted. This information is then tested in an identification task, obtaining an F score of 0.52, which is considerably higher than related work.

This article presents UnsupNMT, a 3-year project of which the first year has already been completed. UnsupNMT proposes a radically different approach to machine translation: unsupervised translation, that is, translation based on monolingual data alone with no need for bilingual resources. This method is based on deep learning of temporal sequences and uses cutting-edge interlingual word representations in the form of cross-lingual word embeddings. This project is not only a highly innovative proposal but it also opens a new paradigm in machine translation which branches out to other disciplines, such us transfer learning. Despite the current limitations of unsupervised machine translation, the techniques developed are expected to have great repercussions in areas where machine translation achieves worse results, such as translation between languages which have little contact, e.g. German and Russian.

This work explores the feasibility of a crowd-based pair-wise comparison evaluation to get feedback on machine translation progress for under-resourced languages. Specifically, we propose a task based on simple work units to compare the outputs of five English-to-Basque systems, which we implement in a web application. In our design, we put forward two key aspects that we believe community collaboration initiatives should consider in order to attract and maintain participants, that is, providing both a community challenge and a personal challenge. We describe how these aspects can comply with a strict methodology to ensure research validity. In particular, we consider the evaluation set size and the characteristics of the test sentences, the number of evaluators per comparison pair, and a mechanism to identify dishonest participation (or participants with insufficient linguistic knowledge). We also describe our dissemination effort, which targeted both general users and interest groups. Over 500 people participated actively in the Ebaluatoia campaign and we were able to collect over 35,000 evaluations in a short period of 10 days. From the results, we complete the ranking of the systems under evaluation and establish whether the difference in quality between the systems is significant.

This paper presents a successful domain adaptation of a general neural machine translation (NMT) system using a bilingual corpus created with captions for images in Wikimedia Commons for the Spanish-Basque and English-Irish pairs.

The goal of the project is to develop, test and disseminate new innovative linguistic resources, tools and solutions for a better digitalization level of the Aragonian, Basque and Occitan languages. As a result, we will obtain, among others, (1) a road map of Aragonian Digitalization, (2) new monolingual and bilingual lexicons and morphosyntactic and syntactic analysers for Occitan, (3) a Northern Basque speech recognition system, and several linguistic tools as well as (4) new innovative solutions for Aragonian, Basque and Occitan.

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.

Back-translation provides a simple yet effective approach to exploit monolingual corpora in Neural Machine Translation (NMT). Its iterative variant, where two opposite NMT models are jointly trained by alternately using a synthetic parallel corpus generated by the reverse model, plays a central role in unsupervised machine translation. In order to start producing sound translations and provide a meaningful training signal to each other, existing approaches rely on either a separate machine translation system to warm up the iterative procedure, or some form of pre-training to initialize the weights of the model. In this paper, we analyze the role that such initialization plays in iterative back-translation. Is the behavior of the final system heavily dependent on it? Or does iterative back-translation converge to a similar solution given any reasonable initialization? Through a series of empirical experiments over a diverse set of warmup systems, we show that, although the quality of the initial system does affect final performance, its effect is relatively small, as iterative back-translation has a strong tendency to convergence to a similar solution. As such, the margin of improvement left for the initialization method is narrow, suggesting that future research should focus more on improving the iterative mechanism itself.

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.

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.