This paper studies gender bias in machine translation through the lens of Large Language Models (LLMs). Four widely-used test sets are employed to benchmark various base LLMs, comparing their translation quality and gender bias against state-of-the-art Neural Machine Translation (NMT) models for English to Catalan (En → Ca) and English to Spanish (En → Es) translation directions. Our findings reveal pervasive gender bias across all models, with base LLMs exhibiting a higher degree of bias compared to NMT models.To combat this bias, we explore prompting engineering techniques applied to an instruction-tuned LLM. We identify a prompt structure that significantly reduces gender bias by up to 12% on the WinoMT evaluation dataset compared to more straightforward prompts. These results significantly reduce the gender bias accuracy gap between LLMs and traditional NMT systems.

Transformer models often demand a vast amount of training data to achieve the desired level of performance. However, this data requirement poses a major challenge for low-resource languages seeking access to high-quality systems, particularly in tasks like Machine Translation. To address this issue, we propose adding Dropout to Transformer’s Residual Connections. Our experimental results demonstrate that this modification effectively mitigates overfitting during training, resulting in substantial performance gains of over 4 BLEU points on a dataset consisting of merely 10 thousand examples.

Our proposed method, RESETOX (REdoSEarch if TOXic), addresses the issue ofNeural Machine Translation (NMT) gener-ating translation outputs that contain toxicwords not present in the input. The ob-jective is to mitigate the introduction oftoxic language without the need for re-training. In the case of identified addedtoxicity during the inference process, RE-SETOX dynamically adjusts the key-valueself-attention weights and re-evaluates thebeam search hypotheses. Experimental re-sults demonstrate that RESETOX achievesa remarkable 57% reduction in added tox-icity while maintaining an average trans-lation quality of 99.5% across 164 lan-guages. Our code is available at: https://github.com

This paper describes the BSC’s submission to the AmericasNLP 2024 Shared Task. We participated in the Spanish to Quechua and Spanish to Guarani tasks. In this paper we show that by using LoRA adapters we can achieve similar performance as a full parameter fine-tuning by only training 14.2% of the total number of parameters. Our systems achieved the highest ChrF++ scores and ranked first for both directions in the final results outperforming strong baseline systems in the provided development and test datasets.

In this paper, we present the two strategies employed for the WMT24 Shared Task on Translation into Low-Resource Languages of Spain. We participated in the language pairs of Spanish-to-Aragonese, Spanish-to-Aranese, and Spanish-to-Asturian, developing neural-based translation systems and moving away from rule-based approaches for these language directions. To create these models, two distinct strategies were employed. The first strategy involved a thorough cleaning process and curation of the limited provided data, followed by fine-tuning the multilingual NLLB-200-600M model (Constrained Submission). The other strategy involved training a transformer from scratch using a vast amount of synthetic data (Open Submission). Both approaches relied on generated synthetic data and resulted in high ChrF and BLEU scores. However, given the characteristics of the task, the strategy used in the Constrained Submission resulted in higher scores that surpassed the baselines across the three translation directions, whereas the strategy employed in the Open Submission yielded slightly lower scores than the highest baseline.

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.

This paper describes the submission of the TALP-UPC team to the Problem List Summarization task from the BioNLP 2023 workshop. This task consists of automatically extracting a list of health issues from the e-health medical record of a given patient. Our submission combines additional steps of data annotationwith finetuning of BERT pre-trained language models. Our experiments focus on the impact of finetuning on different datasets as well as the addition of data augmentation techniques to delay overfitting.

Machine Translation systems can produce different types of errors, some of which are characterized as critical or catastrophic due to the specific negative impact that they can have on users. In this paper we focus on one type of critical error: added toxicity. We evaluate and analyze added toxicity when translating a large evaluation dataset (HOLISTICBIAS, over 472k sentences, covering 13 demographic axes) from English into 164 languages. An automatic toxicity evaluation shows that added toxicity across languages varies from 0% to 5%. The output languages with the most added toxicity tend to be low-resource ones, and the demographic axes with the most added toxicity include sexual orientation, gender and sex, and ability. We also perform human evaluation on a subset of 8 translation directions, confirming the prevalence of true added toxicity. We use a measurement of the amount of source contribution to the translation, where a low source contribution implies hallucination, to interpret what causes toxicity. Making use of the input attributions allows us to explain toxicity, because the source contributions significantly correlate with toxicity for 84% of languages studied. Given our findings, our recommendations to reduce added toxicity are to curate training data to avoid mistranslations, mitigate hallucination and check unstable translations.

In Neural Machine Translation (NMT), each token prediction is conditioned on the source sentence and the target prefix (what has been previously translated at a decoding step). However, previous work on interpretability in NMT has mainly focused solely on source sentence tokens’ attributions. Therefore, we lack a full understanding of the influences of every input token (source sentence and target prefix) in the model predictions. In this work, we propose an interpretability method that tracks input tokens’ attributions for both contexts. Our method, which can be extended to any encoder-decoder Transformer-based model, allows us to better comprehend the inner workings of current NMT models. We apply the proposed method to both bilingual and multilingual Transformers and present insights into their behaviour.

This paper describes the submissions of the UPC Machine Translation group to the IWSLT 2022 Offline Speech Translation and Speech-to-Speech Translation tracks. The offline task involves translating English speech to German, Japanese and Chinese text. Our Speech Translation systems are trained end-to-end and are based on large pretrained speech and text models. We use an efficient fine-tuning technique that trains only specific layers of our system, and explore the use of adapter modules for the non-trainable layers. We further investigate the suitability of different speech encoders (wav2vec 2.0, HuBERT) for our models and the impact of knowledge distillation from the Machine Translation model that we use for the decoder (mBART). For segmenting the IWSLT test sets we fine-tune a pretrained audio segmentation model and achieve improvements of 5 BLEU compared to the given segmentation. Our best single model uses HuBERT and parallel adapters and achieves 29.42 BLEU at English-German MuST-C tst-COMMON and 26.77 at IWSLT 2020 test. By ensembling many models, we further increase translation quality to 30.83 BLEU and 27.78 accordingly. Furthermore, our submission for English-Japanese achieves 15.85 and English-Chinese obtains 25.63 BLEU on the MuST-C tst-COMMON sets. Finally, we extend our system to perform English-German Speech-to-Speech Translation with a pretrained Text-to-Speech model.

Machine Translation is highly impacted by social biases present in data sets, indicating that it reflects and amplifies stereotypes. In this work, we study mitigating gender bias by jointly learning the translation, the part-of-speech, and the gender of the target language with different morphological complexity. This approach has shown improvements up to 6.8 points in gender accuracy without significantly impacting the translation quality.

Introducing factors, that is to say, word features such as linguistic information referring to the source tokens, is known to improve the results of neural machine translation systems in certain settings, typically in recurrent architectures. This study proposes enhancing the current state-of-the-art neural machine translation architecture, the Transformer, so that it allows to introduce external knowledge. In particular, our proposed modification, the Factored Transformer, uses linguistic factors that insert additional knowledge into the machine translation system. Apart from using different kinds of features, we study the effect of different architectural configurations. Specifically, we analyze the performance of combining words and features at the embedding level or at the encoder level, and we experiment with two different combination strategies. With the best-found configuration, we show improvements of 0.8 BLEU over the baseline Transformer in the IWSLT German-to-English task. Moreover, we experiment with the more challenging FLoRes English-to-Nepali benchmark, which includes both extremely low-resourced and very distant languages, and obtain an improvement of 1.2 BLEU

This paper describes the submission to the IWSLT 2021 offline speech translation task by the UPC Machine Translation group. The task consists of building a system capable of translating English audio recordings extracted from TED talks into German text. Submitted systems can be either cascade or end-to-end and use a custom or given segmentation. Our submission is an end-to-end speech translation system, which combines pre-trained models (Wav2Vec 2.0 and mBART) with coupling modules between the encoder and decoder, and uses an efficient fine-tuning technique, which trains only 20% of its total parameters. We show that adding an Adapter to the system and pre-training it, can increase the convergence speed and the final result, with which we achieve a BLEU score of 27.3 on the MuST-C test set. Our final model is an ensemble that obtains 28.22 BLEU score on the same set. Our submission also uses a custom segmentation algorithm that employs pre-trained Wav2Vec 2.0 for identifying periods of untranscribable text and can bring improvements of 2.5 to 3 BLEU score on the IWSLT 2019 test set, as compared to the result with the given segmentation.

State-of-the-art multilingual machine translation relies on a universal encoder-decoder, which requires retraining the entire system to add new languages. In this paper, we propose an alternative approach that is based on language-specific encoder-decoders, and can thus be more easily extended to new languages by learning their corresponding modules. So as to encourage a common interlingua representation, we simultaneously train the N initial languages. Our experiments show that the proposed approach outperforms the universal encoder-decoder by 3.28 BLEU points on average, while allowing to add new languages without the need to retrain the rest of the modules. All in all, our work closes the gap between shared and language-specific encoderdecoders, advancing toward modular multilingual machine translation systems that can be flexibly extended in lifelong learning settings.

This paper describes the submission to the WMT 2021 news translation shared task by the UPC Machine Translation group. The goal of the task is to translate German to French (De-Fr) and French to German (Fr-De). Our submission focuses on fine-tuning a pre-trained model to take advantage of monolingual data. We fine-tune mBART50 using the filtered data, and additionally, we train a Transformer model on the same data from scratch. In the experiments, we show that fine-tuning mBART50 results in 31.69 BLEU for De-Fr and 23.63 BLEU for Fr-De, which increases 2.71 and 1.90 BLEU accordingly, as compared to the model we train from scratch. Our final submission is an ensemble of these two models, further increasing 0.3 BLEU for Fr-De.

In this article, we describe the TALP-UPC participation in the WMT20 news translation shared task for Tamil-English. Given the low amount of parallel training data, we resort to adapt the task to a multilingual system to benefit from the positive transfer from high resource languages. We use iterative backtranslation to fine-tune the system and benefit from the monolingual data available. In order to measure the effectivity of such methods, we compare our results to a bilingual baseline system.

Multilingual Neural Machine Translation approaches are based on the use of task specific models and the addition of one more language can only be done by retraining the whole system. In this work, we propose a new training schedule that allows the system to scale to more languages without modification of the previous components based on joint training and language-independent encoder/decoder modules allowing for zero-shot translation. This work in progress shows close results to state-of-the-art in the WMT task.

The main alternatives nowadays to deal with sequences are Recurrent Neural Networks (RNN) architectures and the Transformer. In this context, Both RNN’s and Transformer have been used as an encoder-decoder architecture with multiple layers in each module. Far beyond this, these architectures are the basis for the contextual word embeddings which are revolutionizing most natural language downstream applications. However, intermediate representations in either the RNN or Transformer architectures can be difficult to interpret. To make these layer representations more accessible and meaningful, we introduce a web-based tool that visualizes them both at the sentence and token level. We present three use cases. The first analyses gender issues in contextual word embeddings. The second and third are showing multilingual intermediate representations for sentences and tokens and the evolution of these intermediate representations along with the multiple layers of the decoder and in the context of multilingual machine translation.

In this article, we describe the TALP-UPC research group participation in the WMT19 news translation shared task for Kazakh-English. Given the low amount of parallel training data, we resort to using Russian as pivot language, training subword-based statistical translation systems for Russian-Kazakh and Russian-English that were then used to create two synthetic pseudo-parallel corpora for Kazakh-English and English-Kazakh respectively. Finally, a self-attention model based on the decoder part of the Transformer architecture was trained on the two pseudo-parallel corpora.

In this article we describe the TALP-UPC research group participation in the WMT18 news shared translation task for Finnish-English and Estonian-English within the multi-lingual subtrack. All of our primary submissions implement an attention-based Neural Machine Translation architecture. Given that Finnish and Estonian belong to the same language family and are similar, we use as training data the combination of the datasets of both language pairs to paliate the data scarceness of each individual pair. We also report the translation quality of systems trained on individual language pair data to serve as baseline and comparison reference.

This paper presents experiments comparing character-based and byte-based neural machine translation systems. The main motivation of the byte-based neural machine translation system is to build multi-lingual neural machine translation systems that can share the same vocabulary. We compare the performance of both systems in several language pairs and we see that the performance in test is similar for most language pairs while the training time is slightly reduced in the case of byte-based neural machine translation.