In this paper, we present the MULTILINGTOOL project, led by the Elhuyar Foundation and funded by the European Commission under the CREA-MEDIA2022-INNOVBUSMOD call. The aim of the project is to develop an advanced platform for automatic multilingual subtitling and dubbing. It will provide support for Spanish, English, and French, as well as the co-official languages of Spain, namely Basque, Catalan, and Galician.

Terminology-constrained NMT systems facilitate the forced translation of domain-specific vocabulary. A notable method in this context is the “copy-and-inflect” approach, which appends the target term lemmas of constraints to their corresponding source terms in the input sentence. In this work, we propose a novel adaptation of the “copy-and-inflect” method, referred to as “morph-masking”. Our method involves masking the source terms of the constraints from the input sentence while retaining essential grammatical information. Our approach is based on the hypothesis that “copy-and-inflect” systems have access to both source and target terms, allowing them to generate the correct surface form of the constraint by either translating the source term itself or properly inflecting the target term lemma. Through extensive validation of our method in two translation directions with different levels of source morphological complexity, Basque to Spanish and English to German, we have demonstrated that “morph-masking” is capable of providing a harder constraint signal, resulting in a notable improvement over the “copy-and-inflect” method (up to 38% in term accuracy), especially in challenging constraint scenarios.

This work investigates the acquisition of formal linguistic competence by neural language models, hypothesizing that languages with complex grammar, such as Basque, present substantial challenges during the pre-training phase. Basque is distinguished by its complex morphology and flexible word order, potentially complicating grammar extraction. In our analysis, we evaluated the grammatical knowledge of BERT models trained under various pre-training configurations, considering factors such as corpus size, model size, number of epochs, and the use of lemmatization. To assess this grammatical knowledge, we constructed the BL2MP (Basque L2 student-based Minimal Pairs) test set. This test set consists of minimal pairs, each containing both a grammatically correct and an incorrect sentence, sourced from essays authored by students at different proficiency levels in the Basque language. Additionally, our analysis explores the difficulties in learning various grammatical phenomena, the challenges posed by flexible word order, and the influence of the student’s proficiency level on the difficulty of correcting grammar errors.

XNLI is a popular Natural Language Inference (NLI) benchmark widely used to evaluate cross-lingual Natural Language Understanding (NLU) capabilities across languages. In this paper, we expand XNLI to include Basque, a low-resource language that can greatly benefit from transfer-learning approaches. The new dataset, dubbed XNLIeu, has been developed by first machine-translating the English XNLI corpus into Basque, followed by a manual post-edition step. We have conducted a series of experiments using mono- and multilingual LLMs to assess a) the effect of professional post-edition on the MT system; b) the best cross-lingual strategy for NLI in Basque; and c) whether the choice of the best cross-lingual strategy is influenced by the fact that the dataset is built by translation. The results show that post-edition is necessary and that the translate-train cross-lingual strategy obtains better results overall, although the gain is lower when tested in a dataset that has been built natively from scratch. Our code and datasets are publicly available under open licenses.

In recent years, pre-trained transformer-based language models (LM) have become a key resource for implementing most NLP tasks. However, pre-training such models demands large text collections not available in most languages. In this paper, we study the use of machine-translated corpora for pre-training LMs. We answer the following research questions: RQ1: Is MT-based data an alternative to real data for learning a LM?; RQ2: Can real data be complemented with translated data and improve the resulting LM? In order to validate these two questions, several BERT models for Basque have been trained, combining real data and synthetic data translated from Spanish.The evaluation carried out on 9 NLU tasks indicates that models trained exclusively on translated data offer competitive results. Furthermore, models trained with real data can be improved with synthetic data, although further research is needed on the matter.

Large language models are very resource intensive, both financially and environmentally, and require an amount of training data which is simply unobtainable for the majority of NLP practitioners. Previous work has researched the scaling laws of such models, but optimal ratios of model parameters, dataset size, and computation costs focused on the large scale. In contrast, we analyze the effect those variables have on the performance of language models in constrained settings, by building three lightweight BERT models (16M/51M/124M parameters) trained over a set of small corpora (5M/25M/125M words).We experiment on four languages of different linguistic characteristics (Basque, Spanish, Swahili and Finnish), and evaluate the models on MLM and several NLU tasks. We conclude that the power laws for parameters, data and compute for low-resource settings differ from the optimal scaling laws previously inferred, and data requirements should be higher. Our insights are consistent across all the languages we study, as well as across the MLM and downstream tasks. Furthermore, we experimentally establish when the cost of using a Transformer-based approach is worth taking, instead of favouring other computationally lighter solutions.

In the news, statements from information sources are often quoted, made by individuals who interact in the news. Detecting those quotes and the gender of their sources is a key task when it comes to media analysis from a gender perspective. It is a challenging task: the structure of the quotes is variable, gender marks are not present in many languages, and quote authors are often omitted due to frequent use of coreferences. This paper proposes a strategy to measure the presence of women and men as information sources in news. We approach the problem of detecting sentences including quotes and the gender of the speaker as a joint task, by means of a supervised multiclass classifier of sentences. We have created the first datasets for Spanish and Basque by manually annotating quotes and the gender of the associated sources in news items. The results obtained show that BERT based approaches are significantly better than bag-of-words based classical ones, achieving accuracies close to 90%. We also analyse a bilingual learning strategy and generating additional training examples synthetically; both provide improvements up to 3.4% and 5.6%, respectively.

Natural Language Understanding (NLU) technology has improved significantly over the last few years and multitask benchmarks such as GLUE are key to evaluate this improvement in a robust and general way. These benchmarks take into account a wide and diverse set of NLU tasks that require some form of language understanding, beyond the detection of superficial, textual clues. However, they are costly to develop and language-dependent, and therefore they are only available for a small number of languages. In this paper, we present BasqueGLUE, the first NLU benchmark for Basque, a less-resourced language, which has been elaborated from previously existing datasets and following similar criteria to those used for the construction of GLUE and SuperGLUE. We also report the evaluation of two state-of-the-art language models for Basque on BasqueGLUE, thus providing a strong baseline to compare upon. BasqueGLUE is freely available under an open license.

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.

It has been found that NMT systems have a strong preference towards social defaults and biases when translating certain occupations, which due to their widespread use, can unintentionally contribute to amplifying and perpetuating these patterns. In that sense, this work focuses on sentence-level gender agreement between gendered entities and occupations when translating from genderless languages to languages with grammatical gender. Specifically, we address the Basque to Spanish translation direction for which bias mitigation has not been addressed. Gender information in Basque is explicit in neither the grammar nor the morphology. It is only present in a limited number of gender specific common nouns and person proper names. We propose a template-based fine-tuning strategy with explicit gender tags to provide a stronger gender signal for the proper inflection of occupations. This strategy is compared against systems fine-tuned on real data extracted from Wikipedia biographies. We provide a detailed gender bias assessment analysis and perform a template ablation study to determine the optimal set of templates. We report a substantial gender bias mitigation (up to 50% on gender bias scores) while keeping the original translation quality.

This papers presents a platform for monitoring press narratives with respect to several social challenges, including gender equality, migrations and minority languages. As narratives are encoded in natural language, we have to use natural processing techniques to automate their analysis. Thus, crawled news are processed by means of several NLP modules, including named entity recognition, keyword extraction,document classification for social challenge detection, and sentiment analysis. A Flask powered interface provides data visualization for a user-based analysis of the data. This paper presents the architecture of the system and describes in detail its different components. Evaluation is provided for the modules related to extraction and classification of information regarding social challenges.

This paper presents an approach for developing a task-oriented dialog system for less-resourced languages in scenarios where training data is not available. Both intent classification and slot filling are tackled. We project the existing annotations in rich-resource languages by means of Neural Machine Translation (NMT) and posterior word alignments. We then compare training on the projected monolingual data with direct model transfer alternatives. Intent Classifiers and slot filling sequence taggers are implemented using a BiLSTM architecture or by fine-tuning BERT transformer models. Models learnt exclusively from Basque projected data provide better accuracies for slot filling. Combining Basque projected train data with rich-resource languages data outperforms consistently models trained solely on projected data for intent classification. At any rate, we achieve competitive performance in both tasks, with accuracies of 81% for intent classification and 77% for slot filling.

Word embeddings and pre-trained language models allow to build rich representations of text and have enabled improvements across most NLP tasks. Unfortunately they are very expensive to train, and many small companies and research groups tend to use models that have been pre-trained and made available by third parties, rather than building their own. This is suboptimal as, for many languages, the models have been trained on smaller (or lower quality) corpora. In addition, monolingual pre-trained models for non-English languages are not always available. At best, models for those languages are included in multilingual versions, where each language shares the quota of substrings and parameters with the rest of the languages. This is particularly true for smaller languages such as Basque. In this paper we show that a number of monolingual models (FastText word embeddings, FLAIR and BERT language models) trained with larger Basque corpora produce much better results than publicly available versions in downstream NLP tasks, including topic classification, sentiment classification, PoS tagging and NER. This work sets a new state-of-the-art in those tasks for Basque. All benchmarks and models used in this work are publicly available.

This article describes the systems submitted by Elhuyar to the 2020 Biomedical Translation Shared Task, specifically the systems presented in the subtasks of terminology translation for English-Basque and abstract translation for English-Basque and English-Spanish. In all cases a Transformer architecture was chosen and we studied different strategies to combine open domain data with biomedical domain data for building the training corpora. For the English-Basque pair, given the scarcity of parallel corpora in the biomedical domain, we set out to create domain training data in a synthetic way. The systems presented in the terminology and abstract translation subtasks for the English-Basque language pair ranked first in their respective tasks among four participants, achieving 0.78 accuracy for terminology translation and a BLEU of 0.1279 for the translation of abstracts. In the abstract translation task for the English-Spanish pair our team ranked second (BLEU=0.4498) in the case of OK sentences.

This paper presents the evaluation of the translation quality and Cross-Lingual Information Retrieval (CLIR) performance when using session information as the context of queries. The hypothesis is that previous queries provide context that helps to solve ambiguous translations in the current query. We tested several strategies on the TREC 2010 Session track dataset, which includes query reformulations grouped by generalization, specification, and drifting types. We study the Basque to English direction, evaluating both the translation quality and CLIR performance, with positive results in both cases. The results show that the quality of translation improved, reducing error rate by 12% (HTER) when using session information, which improved CLIR results 5% (nDCG). We also provide an analysis of the improvements across the three kinds of sessions: generalization, specification, and drifting. Translation quality improved in all three types (generalization, specification, and drifting), and CLIR improved for generalization and specification sessions, preserving the performance in drifting sessions.

Polarity lexicons are a basic resource for analyzing the sentiments and opinions expressed in texts in an automated way. This paper explores three methods to construct polarity lexicons: translating existing lexicons from other languages, extracting polarity lexicons from corpora, and annotating sentiments Lexical Knowledge Bases. Each of these methods require a different degree of human effort. We evaluate how much manual effort is needed and to what extent that effort pays in terms of performance improvement. Experiment setup includes generating lexicons for Basque, and evaluating them against gold standard datasets in different domains. Results show that extracting polarity lexicons from corpora is the best solution for achieving a good performance with reasonable human effort.

Bilingual dictionaries are key resources in several fields such as translation, language learning or various NLP tasks. However, only major languages have such resources. Automatically built dictionaries by using pivot languages could be a useful resource in these circumstances. Pivot-based bilingual dictionary building is based on merging two bilingual dictionaries which share a common language (e.g. LA-LB, LB-LC) in order to create a dictionary for a new language pair (e.g LA-LC). This process may include wrong translations due to the polisemy of words. We built Basque-Chinese (Mandarin) dictionaries automatically from Basque-English and Chinese-English dictionaries. In order to prune wrong translations we used different methods adequate for less resourced languages. Inverse Consultation and Distributional Similarity methods are used because they just depend on easily available resources. Finally, we evaluated manually the quality of the built dictionaries and the adequacy of the methods. Both Inverse Consultation and Distributional Similarity provide good precision of translations but recall is seriously damaged. Distributional similarity prunes rare translations more accurately than other methods.

This paper deals with the main problems that arise in the query translation process in dictionary-based Cross-lingual Information Retrieval (CLIR): translation selection, presence of Out-Of-Vocabulary (OOV) terms and translation of Multi-Word Expressions (MWE). We analyse to what extent each problem affects the retrieval performance for the Basque-English pair of languages, and the improvement obtained when using parallel corpora free methods to address them. To tackle the translation selection problem we provide novel extensions of an already existing monolingual target co-occurrence-based method, the Out-Of Vocabulary terms are dealt with by means of a cognate detection-based method and finally, for the Multi-Word Expression translation problem, a naïve matching technique is applied. The error analysis shows significant differences in the deterioration of the performance depending on the problem, in terms of Mean Average Precision (MAP), the translation selection problem being the cause of most of the errors. Otherwise, the proposed combined strategy shows a good performance to tackle the three above-mentioned main problems.

This project combines linguistic and statistical information to develop a term extraction tool for Basque. Being Basque an agglutinative and highly inflected language, the treatment of morphosyntactic information is vital. In addition, due to late unification process of the language, texts present more elevated term dispersion than in a highly normalized language. The result is a semi-automatic terminology extraction tool based on XML, for its use in technical and scientific information managing.