Papers by Mikel Artetxe
Unsupervised Multilingual Sentence Embeddings for Parallel Corpus Mining (2020.acl-srw)
Copied to clipboard
| Challenge: | Existing models of multilingual sentence embeddings require large parallel data resources which are not available for low-resource languages. |
| Approach: | They propose an unsupervised method to derive multilingual sentence embeddings using monolingual data. |
| Outcome: | The proposed method improves on two parallel corpus mining tasks and for other languages. |
On the Cross-lingual Transferability of Monolingual Representations (2020.acl-main)
Copied to clipboard
| Challenge: | State-of-the-art unsupervised multilingual models generalize in zero-shot cross-lingual setting . generalization ability attributed to shared subword vocabulary and joint training across multiple languages . |
| Approach: | They propose an approach that transfers a monolingual model to new languages at the lexical level. |
| Outcome: | The proposed approach is competitive with multilingual BERT on cross-lingual classification benchmarks and on a new cross-linguistic question answering dataset. |
Does Corpus Quality Really Matter for Low-Resource Languages? (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing work on multilingual pre-training has relied on automatically filtered versions of CommonCrawl. |
| Approach: | They propose to use tailored crawling to identify and scrape websites with high-quality content to improve representation learning in Basque. |
| Outcome: | The proposed corpus, called EusCrawl, has a much higher quality according to native annotators than the Basque portion of popular multilingual corpora like CC100 and mC4. |
Analyzing the Limitations of Cross-lingual Word Embedding Mappings (P19-1)
Copied to clipboard
| Challenge: | Existing methods for cross-lingual word embeddings have limited results . existing methods require little or no cross-linguistic signal to work . |
| Approach: | They compare offline mapping methods to an extension of skip-gram that jointly learns both embedding spaces. |
| Outcome: | The proposed method yields more isomorphic embeddings, is less sensitive to hubness, and achieves stronger results in bilingual lexicon induction. |
Prompting ELECTRA: Few-Shot Learning with Discriminative Pre-Trained Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained masked language models perform few-shot learning, but discriminative models like ELECTRA do not fit into the paradigm. |
| Approach: | They propose to use ELECTRA to train pre-trained models to score originality of target options without introducing new parameters. |
| Outcome: | The proposed model outperforms masked language models in a wide range of tasks without adding new parameters. |
Principled Paraphrase Generation with Parallel Corpora (2022.acl-long)
Copied to clipboard
| Challenge: | a popular method for paraphrase generation is round-trip machine translation (MT). |
| Approach: | They propose a method that formalizes the implicit similarity function and relaxes it by requiring the entire translation distribution to match . they propose MT can be used to generate paraphrases by decoding back to the source without having to generate pivot translations. |
| Outcome: | The proposed approach is more principled and efficient than round-trip machine translation (MT) and offers an adjustable parameter to control the fidelity-diversity trade-off. |
Mini-Model Adaptation: Efficiently Extending Pretrained Models to New Languages via Aligned Shallow Training (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to pretrain Masked Language Models (MLMs) are expensive and require a full forward and backward pass over the entire model. |
| Approach: | They propose to learn a shallow mini-model from a fraction of a large model's parameters and plug it into a larger model for rapid cross-lingual transfer. |
| Outcome: | Experiments on XNLI, MLQA and PAWS-X show that mini-model adaptation matches the standard approach using up to 2.3x less compute on average. |
CombLM: Adapting Black-Box Language Models through Small Fine-Tuned Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Methods for adapting language models to new tasks and domains have traditionally assumed white-box access to the model and work by modifying its parameters. |
| Approach: | They propose a method for adapting large language models to new domains and tasks . they fine-tune a small white-box LM and combine it with a large black-box model at the probability level through a network, learned on a smaller validation set. |
| Outcome: | The proposed method improves performance in all cases, while using a domain expert 23x smaller. |
Margin-based Parallel Corpus Mining with Multilingual Sentence Embeddings (P19-1)
Copied to clipboard
| Challenge: | Traditional parallel corpus mining methods focus on the textual content instead of the size and quality of training data. |
| Approach: | They propose a method for machine translation based on multilingual sentence embeddings. |
| Outcome: | The proposed method outperforms the best published methods on the BUCC mining task and the UN reconstruction task by more than 10 F1 and 30 precision points. |
WiCkeD: A Simple Method to Make Multiple Choice Benchmarks More Challenging (2025.acl-short)
Copied to clipboard
| Challenge: | Multiple choice question (MCQ) benchmarks are widely used to evaluate Large Language Models (LLMs). |
| Approach: | They propose a method to increase the complexity of existing multiple-choice benchmarks by randomly replacing a choice with “None of the above”. |
| Outcome: | The proposed method can be applied to 6 popular benchmarks and evaluate 18 open-weight LLMs. |
PARADISE: Exploiting Parallel Data for Multilingual Sequence-to-Sequence Pretraining (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing methods for multilingual sequence-to-sequence pretraining rely on monolingual corpora and do not use strong cross-lingual signal contained in parallel data. |
| Approach: | They propose a method that replaces monolingual words with a bilingual dictionary and predicts the reference translation according to a parallel corpus instead of recovering the original sequence. |
| Outcome: | The proposed method improves machine translation and cross-lingual natural language inference by 2.0 BLEU points and 6.7 accuracy points over existing methods at a fraction of their computational cost. |
Multilingual Machine Translation: Closing the Gap between Shared and Language-specific Encoder-Decoders (2021.eacl-main)
Copied to clipboard
| Challenge: | State-of-the-art multilingual machine translation relies on a universal encoder-decoder, which requires retraining the entire system to add new languages. |
| Approach: | They propose an encoder-decoder approach that can be extended to new languages by learning their corresponding modules. |
| Outcome: | The proposed approach outperforms the universal encoder-decoder by 3.28 BLEU points on average while allowing to add new languages without retraining the rest of the modules. |
Efficient Large Scale Language Modeling with Mixtures of Experts (2022.emnlp-main)
Copied to clipboard
Mikel Artetxe, Shruti Bhosale, Naman Goyal, Todor Mihaylov, Myle Ott, Sam Shleifer, Xi Victoria Lin, Jingfei Du, Srinivasan Iyer, Ramakanth Pasunuru, Giridharan Anantharaman, Xian Li, Shuohui Chen, Halil Akin, Mandeep Baines, Louis Martin, Xing Zhou, Punit Singh Koura, Brian O’Horo, Jeffrey Wang, Luke Zettlemoyer, Mona Diab, Zornitsa Kozareva, Veselin Stoyanov
| Challenge: | Mixture of Experts layers (MoEs) enable efficient scaling of language models . large autoregressive language models such as GPT-3 can be adapted to a wide range of tasks . |
| Approach: | They propose to use Mixture of Experts layers to enable efficient scaling of language models . they find that MoEs are substantially more compute efficient than dense models compared to MoE models - but only when they are more modestly trained . |
| Outcome: | The proposed model outperforms dense models in a wide range of tasks and domains. |
Few-shot Learning with Multilingual Generative Language Models (2022.emnlp-main)
Copied to clipboard
Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O’Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li
| Challenge: | Large-scale generative language models such as GPT-3 are competitive few-shot learners. |
| Approach: | They train multilingual generative language models on a corpus covering a diverse set of languages and study their few- and zero-shot learning capabilities. |
| Outcome: | The proposed model outperforms GPT-3 on 171 out of 182 directions with 32 training examples and surpasses the official supervised baseline in 45 directions. |
Do Multilingual Language Models Think Better in English? (2024.naacl-short)
Copied to clipboard
| Challenge: | Existing studies show that translation-test improves performance of multilingual models by translating the input into English using an external machine translation system. |
| Approach: | They propose a new approach that leverages the few-shot translation capabilities of multilingual language models. |
| Outcome: | The proposed approach outperforms direct inference on 5 tasks. |
Multilingual Machine Translation with Hyper-Adapters (2022.emnlp-main)
Copied to clipboard
| Challenge: | Multilingual machine translation suffers from negative interference across languages. |
| Approach: | They propose a rescaling fix that reduces the number of parameters and enables training larger hyper-networks. |
| Outcome: | The proposed approach outperforms regular adapters and achieves the same performance with 12 times less parameters. |
On the Role of Bidirectionality in Language Model Pre-Training (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Prior work on language model pre-training explored different architectures and learning objectives, but differences in data, hyperparameters and evaluation make a principled comparison difficult. |
| Approach: | They propose a framework that generalizes prior approaches to pre-training language models by focusing on bidirectionality and controlling each of them separately. |
| Outcome: | The proposed framework generalizes prior approaches including fully unidirectional models like GPT, fully bidirectional models and hybrid models like CM3 and prefix LM. |
An Effective Approach to Unsupervised Machine Translation (P19-1)
Copied to clipboard
| Challenge: | a recent research line has managed to train both unsupervised and unsupervised machine translation systems using monolingual corpora only. |
| Approach: | They propose to use monolingual corpora to train both unsupervised and unsupervised machine translation systems. |
| Outcome: | The proposed system achieves 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 in the (supervised) shared task winner back in 2014. |
Lifting the Curse of Multilinguality by Pre-training Modular Transformers (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent work on multilingual pre-trained models has focused on pre-training transformers on concatenated corpora of a large number of languages. |
| Approach: | They propose a language-specific module approach that allows for more languages to be trained post-hoc. |
| Outcome: | The proposed model can be pre-trained on multiple languages with no drop in performance . |
On the Role of Parallel Data in Cross-lingual Transfer Learning (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing multilingual models do not exploit the full potential of monolingual data, a new study finds . prior work has shown that parallel data is beneficial for cross-lingual learning, but it is unclear if it is the data itself or the modeling of parallel interactions that matters. |
| Approach: | They compare unsupervised machine translation to supervised machine translator and gold parallel data to generate synthetic parallel data. |
| Outcome: | The proposed model generated parallel data is better than supervised machine translation and gold parallel data in both general and task-specific settings. |
The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants (2024.acl-long)
Copied to clipboard
Lucas Bandarkar, Davis Liang, Benjamin Muller, Mikel Artetxe, Satya Narayan Shukla, Donald Husa, Naman Goyal, Abhinandan Krishnan, Luke Zettlemoyer, Madian Khabsa
| Challenge: | Existing benchmarks for text comprehension only cover 30 languages, but lack of labeled data is a major obstacle to building functional systems in most languages. |
| Approach: | They present a multiple-choice machine reading comprehension dataset spanning 122 languages . they use it to evaluate the capabilities of multilingual masked language models and large language models . |
| Outcome: | The proposed dataset enables the evaluation of text models in high-, medium- and low-resource languages. |
BOUQuET : dataset, Benchmark and Open initiative for Universal Quality Evaluation in Translation (2025.emnlp-main)
Copied to clipboard
Pierre Andrews, Mikel Artetxe, Mariano Coria Meglioli, Marta R. Costa-jussà, Joe Chuang, David Dale, Mark Duppenthaler, Nathanial Paul Ekberg, Cynthia Gao, Daniel Edward Licht, Jean Maillard, Alexandre Mourachko, Christophe Ropers, Safiyyah Saleem, Eduardo Sánchez, Ioannis Tsiamas, Arina Turkatenko, Albert Ventayol-Boada, Shireen Yates
| Challenge: | BOUQUET is a multi-way, multicentric and multi-register/domain dataset and benchmark . the dataset is handcrafted in 8 non-English languages . |
| Approach: | They propose to use BOUQuET to collect a multi-way, multicentric and multi-register/domain dataset and benchmark in 8 non-English languages. |
| Outcome: | The proposed dataset is available at https://huggingface.co/datasets/facebook/bouquet. |
Multilingual Autoregressive Entity Linking (2022.tacl-1)
Copied to clipboard
Nicola De Cao, Ledell Wu, Kashyap Popat, Mikel Artetxe, Naman Goyal, Mikhail Plekhanov, Luke Zettlemoyer, Nicola Cancedda, Sebastian Riedel, Fabio Petroni
| Challenge: | mGENRE is a sequence-to-sequence system for multilingual entity linking . mGenRE is used to solve language-specific mentions to a multilingual Knowledge Base . |
| Approach: | They propose a sequence-to-sequence system for multilingual entity linking . they match language-specific mentions against a multilingual Knowledge Base (KB) mGENRE is a sequential system that predicts the name of the target entity token-by-token . |
| Outcome: | The proposed system improves on three popular MEL benchmarks and shows improvements in accuracy. |
Improving Factuality in Clinical Abstractive Multi-Document Summarization by Guided Continued Pre-training (2024.naacl-short)
Copied to clipboard
| Challenge: | Existing methods for enhancing the factual accuracy of abstractive summarization models are not effective in fact-critical domains like clinical document summarizing. |
| Approach: | They propose a guided continued pre-training stage for encoder-decoder models followed by supervised fine-tuning on summarization. |
| Outcome: | The proposed approach improves the quality and factuality of the summaries and achieves the best-published results on the clinical document summarization task. |
A Call for More Rigor in Unsupervised Cross-lingual Learning (2020.acl-main)
Copied to clipboard
| Challenge: | Existing research on unsupervised cross-lingual learning has focused on purely unsupervised learning without any parallel data for most of the world's languages. |
| Approach: | They propose to define "multilingual learning" as learning a common model for two or more languages from raw text, without any downstream task labels. |
| Outcome: | The proposed model is based on a model with no parallel data and abundant monolingual data. |
A robust self-learning method for fully unsupervised cross-lingual mappings of word embeddings (P18-1)
Copied to clipboard
| Challenge: | Existing methods to learn cross-lingual word embeddings have failed in more realistic scenarios . a fully unsupervised initialization and a robust self-learning algorithm are needed to improve the existing methods. |
| Approach: | They propose an unsupervised initialization method that exploits structural similarity of embeddings and a robust self-learning algorithm that iteratively improves it. |
| Outcome: | The proposed method achieves the best published results in standard datasets even surpassing previous supervised systems. |
Don’t Prompt, Search! Mining-based Zero-Shot Learning with Language Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent work has obtained strong zero-shot results by prompting language models. |
| Approach: | They propose a mining-based approach that uses regular expressions to mine labeled examples from unlabeled corpora and fine tune a pretrained model. |
| Outcome: | The proposed method outperforms prompting on a wide range of tasks when using comparable templates. |
Translation Artifacts in Cross-lingual Transfer Learning (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing cross-lingual transfer learning techniques involve human and machine translations. |
| Approach: | They propose to use machine translation to translate test set or training set to introduce subtle artifacts that have a notable impact in existing cross-lingual models. |
| Outcome: | The proposed translation process reduces the lexical overlap between the premise and hypothesis by 4.3 and 2.8 points . the proposed translation-test and zero-shot approaches improve on previous work . |
Revisiting Machine Translation for Cross-lingual Classification (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent work in cross-lingual learning has pivoted around multilingual models, which are typically pretrained on unlabeled corpora in multiple languages using some form of language modeling objective. |
| Approach: | They propose to use a stronger machine translation system to mitigat mismatch between training on original text and running inference on machine translated text. |
| Outcome: | The proposed approach is highly task dependent and calls into question the dominance of multilingual models for cross-lingual classification. |
PoeLM: A Meter- and Rhyme-Controllable Language Model for Unsupervised Poetry Generation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for generating formal verse poetry use existing poems for supervision, which are difficult to obtain for most languages and poetic forms. |
| Approach: | They propose an unsupervised approach to generate formal verse poetry without supervision . they use control codes to describe meter and rhyme scheme constraints, and train a transformer language model . |
| Outcome: | The proposed method generates poems that follow any given meter and rhyme scheme without training . it is comparable to those written by humans and generates comparable quality poems . |
Beyond Offline Mapping: Learning Cross-lingual Word Embeddings through Context Anchoring (2021.acl-long)
Copied to clipboard
| Challenge: | Recent research on cross-lingual word embeddings has been dominated by unsupervised mapping approaches that align monolingual embedders. |
| Approach: | They propose an unsupervised mapping approach that fixes fixed embeddings and learns new ones for the source language that are aligned with them. |
| Outcome: | The proposed method outperforms conventional mapping methods on bilingual lexicon induction and obtains competitive results in the downstream XNLI task. |
Unsupervised Statistical Machine Translation (D18-1)
Copied to clipboard
| Challenge: | Neural Machine Translation (NMT) systems can be trained from monolingual corpora without supervision. |
| Approach: | They propose a phrase-based approach that trains from monolingual corpora . their method is based on phrase-driven Statistical Machine Translation (SMT) they propose to train NMT systems without supervision from monolinguistic corpors . |
| Outcome: | The proposed approach improves on the existing supervised systems by combining a phrase table with an n-gram language model and fine-tuning hyperparameters through an unsupervised MERT variant. |
Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? (2022.emnlp-main)
Copied to clipboard
Sewon Min, Xinxi Lyu, Ari Holtzman, Mikel Artetxe, Mike Lewis, Hannaneh Hajishirzi, Luke Zettlemoyer
| Challenge: | Large language models can in-context learn by conditioning on a few input-label pairs and making predictions for new inputs. |
| Approach: | They propose to use ground truth demonstrations to replace labels in demonstrations . they also show that other aspects of the demonstrations are key drivers of endtask performance . |
| Outcome: | The proposed model outperforms zeroshot inference on a wide range of tasks using ground truth demonstrations. |
Bilingual Lexicon Induction through Unsupervised Machine Translation (P19-1)
Copied to clipboard
| Challenge: | Existing methods for bilingual lexicon induction use nearest neighbor or related retrieval methods to induce word translation pairs. |
| Approach: | They propose a method that aligns word embeddings in two languages and uses them to build a phrase-table and a language model to extract the bilingual lexicon. |
| Outcome: | The proposed method improves accuracy 6 points over nearest neighbor and 4 points over CSLS retrieval on the same cross-lingual embeddings. |
Training Trajectories of Language Models Across Scales (2023.acl-long)
Copied to clipboard
Mengzhou Xia, Mikel Artetxe, Chunting Zhou, Xi Victoria Lin, Ramakanth Pasunuru, Danqi Chen, Luke Zettlemoyer, Veselin Stoyanov
| Challenge: | Scaling up language models has led to unprecedented performance gains, but little is understood about how the training dynamics change as models get larger. |
| Approach: | They analyze the training checkpoints of different-sized OPT models on next-token prediction, sequence-level generation and downstream tasks. |
| Outcome: | The results show that language models of different sizes learn more during training . small models halt at hallucinations, larger ones learn to assign lower probabilities . |
Emergent Abilities of Large Language Models under Continued Pre-training for Language Adaptation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing large language models are notoriously English-centric, and their performance has been reported to drop significantly in lessresourced languages. |
| Approach: | They propose a language-agnostic benchmark for in-context learning that reveals catastrophic forgetting early on CPT when English is not included. |
| Outcome: | The proposed method does not impact validation perplexity but is critical for emergence of downstream capabilities in the target language. |
Instructing Large Language Models for Low-Resource Languages: A Systematic Study for Basque (2025.emnlp-main)
Copied to clipboard
Oscar Sainz, Naiara Perez, Julen Etxaniz, Joseba Fernandez de Landa, Itziar Aldabe, Iker García-Ferrero, Aimar Zabala, Ekhi Azurmendi, German Rigau, Eneko Agirre, Mikel Artetxe, Aitor Soroa
| Challenge: | Instructing language models with user intent requires large instruction datasets limited to a limited set of languages. |
| Approach: | They propose to use existing LLMs and synthetically generated instructions to train models with user intent. |
| Outcome: | The proposed model outperforms base non-instructed models on Basque without Basque instructions. |