Sentence-level alignment systems have been developed and evaluated primarily on parallel data, leaving their behaviour across the broader parallel-comparable spectrum of real web content poorly understood. We present a stratified empirical study of alignment quality for Catalan-English using 300 document pairs across three parallelism bands defined by mean-max LaBSE cosine similarity. We compare four systems: a hierarchical alignment pipeline (DocAlign), an ablation with paragraph pre-filtering disabled (DocAlign-NoFilter), the flat aligner Vecalign, and a flat LaBSE greedy baseline. Evaluation uses human-annotated sentence pairs and coverage-weighted quality. Quality degrades at different rates by system type: hierarchical systems maintain usable-pair rates ranging from 25% to 51% on comparable data while flat systems collapse to 2-7%. Paragraph pre-filtering reduces output volume on comparable data while raising pair quality relative to the unfiltered ablation. Vecalign is statistically indistinguishable from the greedy baseline at all parallelism levels, suggesting that LaBSE embedding discrimination is the binding constraint on flat alignment quality. Failure mode analysis of 550 low-rated pairs identifies topical mismatch as the dominant failure mode, with structural noise concentrated in flat systems.

Large Language Models (LLMs) achieve strong performance in machine translation (MT) but often encode gender bias, particularly when translating from non-gendered into gendered languages. This paper introduces a fine-tuning strategy to mitigate such bias in English-Spanish and English-Catalan translation. Using parameter-efficient LoRA fine-tuning, we apply linguistic knowledge infusion—a reasoning-based method that trains models to identify gendered referents and syntactic cues before generating translations. Experiments with Mistral–7B and Salamandrata–7B on MT-GenEval show that linguistically infused models improve gender accuracy by 15 percentage points and reduce gender gaps by 27 points in English-Spanish translation, with comparable trends for Catalan. Gains are strongest for Mistral, suggesting that explicit linguistic reasoning particularly benefits general-purpose LLMs. Overall, these results demonstrate that structured linguistic priors can enhance fairness and referential consistency in multilingual machine translation.

We present ACADATA, a high-quality parallel dataset for academic translation, that consists of two subsets: ACAD-TRAIN, which contains approximately 1.5 million author-generated paragraph pairs across 96 language directions and ACAD-BENCH, a curated evaluation set of almost 6,000 translations covering 12 directions. To validate its utility, we fine-tune two Large Language Models (LLMs) on ACAD-TRAIN and benchmark them on ACAD-BENCH against specialized machine-translation systems, general-purpose, open-weight LLMs, and several large-scale proprietary models. Experimental results demonstrate that fine-tuning on ACAD-TRAIN leads to improvements in academic translation quality by +6.1 and +12.4 d-BLEU points on average for 7B and 2B models respectively, while also improving long-context translation in a general domain by up to 24.9% when translating out of English. The fine-tuned top-performing model surpasses the best propietary and open-weight models on academic translation domain. By releasing ACAD-TRAIN, ACAD-BENCH and the fine-tuned models, we provide the community with a valuable resource to advance research in academic domain and long-context translation.

In recent years, Large Language Models (LLMs) have demonstrated exceptional proficiency across a broad spectrum of Natural Language Processing (NLP) tasks, including Machine Translation. However, previous methods predominantly relied on iterative processes such as instruction fine-tuning or continual pre-training, leaving unexplored the challenges of training LLMs solely on parallel data. In this work, we introduce Plume (Parallel Language Model), a collection of three 2B LLMs featuring varying vocabulary sizes (32k, 128k, and 256k) trained exclusively on Catalan-centric parallel examples. These models perform comparably to previous encoder-decoder architectures on 16 supervised translation directions and 56 zero-shot ones. Utilizing this set of models, we conduct a thorough investigation into the translation capabilities of LLMs, probing their performance, the role of vocabulary size, the impact of the different elements of the prompt, and their cross-lingual representation space. We find that larger vocabulary sizes improve zero-shot performance and that different layers specialize in distinct aspects of the prompt, such as language-specific tags. We further show that as the vocabulary size grows, a larger number of attention heads can be pruned with minimal loss in translation quality, achieving a reduction of over 64.7% in attention heads.

High-quality machine translation requires datasets that not only ensure linguistic accuracy but also capture regional and cultural nuances. While many existing benchmarks, such as FLORES-200, rely on English as a pivot language, this approach can overlook the specificity of direct language pairs, particularly for underrepresented combinations like Catalan-Chinese. In this study, we demonstrate that even with a relatively small dataset of approximately 1,000 sentences, we can significantly improve MT localization. To this end, we introduce a dataset specifically designed to enhance Catalan-to-Chinese translation by prioritizing regionally and culturally specific topics. Unlike pivot-based datasets, our data source ensures a more faithful representation of Catalan linguistic and cultural elements, leading to more accurate translations of local terms and expressions. Using this dataset, we demonstrate better performance over the English-pivot FLORES-200 dev set and achieve competitive results on the FLORES-200 devtest set when evaluated with neural-based metrics. We release this dataset as both a human-preference resource and a benchmark for Catalan-Chinese translation. Additionally, we include Spanish translations for each sentence, facilitating extensions to Spanish-Chinese translation tasks.

This paper presents a comprehensive evaluation of gender bias in English-Catalan machine translation, encompassing the creation of a novel language resource and an analysis of translation quality across four different tokenization models. The study introduces a new dataset derived from the MuST-SHE corpus, focusing on gender-neutral terms that necessitate gendered translations in Catalan. The results reveal noteworthy gender bias across all translation models, with a consistent preference for masculine forms. Notably, the study finds that when context is available, BPE and Sentencepiece Unigram tokenization methods outperform others, achieving higher accuracy in gender translation. However, when no context is provided, Morfessor outputs more feminine forms than other tokenization methods, albeit still a small percentage. The study also reflects that stereotypes present in the data are amplified in the translation output. Ultimately, this work serves as a valuable resource for addressing and mitigating gender bias in machine translation, emphasizing the need for improved awareness and sensitivity to gender issues in natural language processing applications.

Generative Pre-trained Transformers (GPTs) have recently been scaled to unprecedented sizes in the history of machine learning. These models, solely trained on the language modeling objective, have been shown to exhibit outstanding zero, one, and few-shot learning capabilities in a number of different tasks. Nevertheless, aside from anecdotal experiences, little is known regarding their multilingual capabilities, given the fact that the pre-training corpus is almost entirely composed of English text. In this work, we investigate its potential and limits in three tasks: extractive question-answering, text summarization and natural language generation for five different languages, as well as the effect of scale in terms of model size. Our results show that GPT-3 can be almost as useful for many languages as it is for English, with room for improvement if optimization of the tokenization is addressed.

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.

The de-identification of sensible data, also known as automatic textual anonymisation, is essential for data sharing and reuse, both for research and commercial purposes. The first step for data anonymisation is the detection of sensible entities. In this work, we present four new datasets for named entity detection in Spanish in the legal domain. These datasets have been generated in the framework of the MAPA project, three smaller datasets have been manually annotated and one large dataset has been automatically annotated, with an estimated error rate of around 14%. In order to assess the quality of the generated datasets, we have used them to fine-tune a battery of entity-detection models, using as foundation different pre-trained language models: one multilingual, two general-domain monolingual and one in-domain monolingual. We compare the results obtained, which validate the datasets as a valuable resource to fine-tune models for the task of named entity detection. We further explore the proposed methodology by applying it to a real use case scenario.

We describe the MAPA project, funded under the Connecting Europe Facility programme, whose goal is the development of an open-source de-identification toolkit for all official European Union languages. It will be developed since January 2020 until December 2021.

The Neural Translation for the European Union (NTEU) project aims to build a neural engine farm with all European official language combinations for eTranslation, without the necessity to use a high-resourced language as a pivot. NTEU started in September 2019 and will run until August 2021.

The European MAPA (Multilingual Anonymisation for Public Administrations) project aims at developing an open-source solution for automatic de-identification of medical and legal documents. We introduce here the context, partners and aims of the project, and report on preliminary results.

We describe the European Language Resources Infrastructure project, whose main aim is the provision of an infrastructure to help collect, prepare and share language resources that can in turn improve translation services in Europe.

Abstract We present research aiming to build tools for the normalization of User-Generated Content (UGC). We argue that processing this type of text requires the revisiting of the initial steps of Natural Language Processing, since UGC (micro-blog, blog, and, generally, Web 2.0 user-generated texts) presents a number of nonstandard communicative and linguistic characteristics – often closer to oral and colloquial language than to edited text. We present a corpus of UGC text in Spanish from three different sources: Twitter, consumer reviews, and blogs, and describe its main characteristics. We motivate the need for UGC text normalization by analyzing the problems found when processing this type of text through a conventional language processing pipeline, particularly in the tasks of lemmatization and morphosyntactic tagging. Our aim with this paper is to seize the power of already existing spell and grammar correction engines and endow them with automatic normalization capabilities in order to pave the way for the application of standard Natural Language Processing tools to typical UGC text. Particularly, we propose a strategy for automatically normalizing UGC by adding a module on top of a pre-existing spell-checker that selects the most plausible correction from an unranked list of candidates provided by the spell-checker. To build this selector module we train four language models, each one containing a different type of linguistic information in a trade-off with its generalization capabilities. Our experiments show that the models trained on truecase and lowercase word forms are more discriminative than the others at selecting the best candidate. We have also experimented with a parametrized combination of the models by both optimizing directly on the selection task and doing a linear interpolation of the models. The resulting parametrized combinations obtain results close to the best performing model but do not improve on those results, as measured on the test set. The precision of the selector module in ranking number one the expected correction proposal on the test corpora reaches 82.5% for Twitter text (baseline 57%) and 88% for non-Twitter text (baseline 64%).

This paper presents an overview of the broad-coverage, application-independent natural language generation component of the NLP system being developed at Microsoft Research. It demonstrates how this component functions within a multilingual Machine Translation system (MSR-MT), using the languages that we are currently working on (English, Spanish, Japanese, and Chinese). Section 1 provides a system description of MSR-MT. Section 2 focuses on the generation component and its set of core rules. Section 3 describes an additional layer of generation rules with examples that address issues specific to MT. Section 4 presents evaluation results in the context of MSR-MT. Section 5 addresses generation issues outside of MT.