In recent years, the use of large language models (LLMs) to generate music content, particularly lyrics, has gained in popularity. These advances provide valuable tools for artists and enhance their creative processes, but they also raise concerns about copyright violations, consumer satisfaction, and content spamming. Previous research has explored content detection in various domains. However, no work has focused on the text modality, lyrics, in music. To address this gap, we curated a diverse dataset of real and synthetic lyrics from multiple languages, music genres, and artists. The generation pipeline was validated using both humans and automated methods. We performed a thorough evaluation of existing synthetic text detection approaches on lyrics, a previously unexplored data type. We also investigated methods to adapt the best-performing features to lyrics through unsupervised domain adaptation. Following both music and industrial constraints, we examined how well these approaches generalize across languages, scale with data availability, handle multilingual language content, and perform on novel genres in few-shot settings. Our findings show promising results that could inform policy decisions around AI-generated music and enhance transparency for users.

Large Language Models (LLMs) have demonstrated remarkable versatility in recent years, offering potential applications across specialized domains such as healthcare and medicine. Despite the availability of various open-source LLMs tailored for health contexts, adapting general-purpose LLMs to the medical domain presents significant challenges.In this paper, we introduce BioMistral, an open-source LLM tailored for the biomedical domain, utilizing Mistral as its foundation model and further pre-trained on PubMed Central. We conduct a comprehensive evaluation of BioMistral on a benchmark comprising 10 established medical question-answering (QA) tasks in English. We also explore lightweight models obtained through quantization and model merging approaches. Our results demonstrate BioMistral’s superior performance compared to existing open-source medical models and its competitive edge against proprietary counterparts. Finally, to address the limited availability of data beyond English and to assess the multilingual generalization of medical LLMs, we automatically translated and evaluated this benchmark into 7 other languages. This marks the first large-scale multilingual evaluation of LLMs in the medical domain. Datasets, multilingual evaluation benchmarks, scripts, and all the models obtained during our experiments are freely released.

L’édition 2024 du DÉfi Fouille de Textes (DEFT) met l’accent sur le développement de méthodes pour la sélection automatique de réponses pour des questions à choix multiples (QCM) en français. Les méthodes sont évaluées sur un nouveau sous-ensemble du corpus FrenchMedMCQA, comprenant 3 105 questions fermées avec cinq options chacune, provenant des archives d’examens de pharmacie. Dans la première tâche, les participants doivent se concentrer sur des petits modèles de langue (PML) avec moins de 3 milliards de paramètres et peuvent également utiliser les corpus spécifiques au domaine médical NACHOS et Wikipedia s’ils souhaitent appliquer des approches du type Retrieval-Augmented Generation (RAG). La second tâche lève la restriction sur la taille des modèles de langue. Les résultats, mesurés par l’Exact Match Ratio (EMR), varient de 1,68 à 11,74 tandis que les performances selon le score de Hamming vont de 28,75 à 49,15 pour la première tâche. Parmi les approches proposées par les cinq équipes participantes, le meilleur système utilise une chaîne combinant un classifieur CamemBERT-bio pour identifier le type de question et un système RAG fondé sur Apollo 2B, affiné avec la méthode d’adaptation LoRA sur les données de l’année précédente.

The recent emergence of Large Language Models (LLMs) has enabled significant advances in the field of Natural Language Processing (NLP). While these new models have demonstrated superior performance on various tasks, their application and potential are still underexplored, both in terms of the diversity of tasks they can handle and their domain of application. In this context, we evaluate four state-of-the-art instruction-tuned LLMs (ChatGPT, Flan-T5 UL2, Tk-Instruct, and Alpaca) on a set of 13 real-world clinical and biomedical NLP tasks in English, including named-entity recognition (NER), question-answering (QA), relation extraction (RE), and more. Our overall results show that these evaluated LLMs approach the performance of state-of-the-art models in zero- and few-shot scenarios for most tasks, particularly excelling in the QA task, even though they have never encountered examples from these tasks before. However, we also observe that the classification and RE tasks fall short of the performance achievable with specifically trained models designed for the medical field, such as PubMedBERT. Finally, we note that no single LLM outperforms all others across all studied tasks, with some models proving more suitable for certain tasks than others.

The biomedical domain has sparked a significant interest in the field of Natural Language Processing (NLP), which has seen substantial advancements with pre-trained language models (PLMs). However, comparing these models has proven challenging due to variations in evaluation protocols across different models. A fair solution is to aggregate diverse downstream tasks into a benchmark, allowing for the assessment of intrinsic PLMs qualities from various perspectives. Although still limited to few languages, this initiative has been undertaken in the biomedical field, notably English and Chinese. This limitation hampers the evaluation of the latest French biomedical models, as they are either assessed on a minimal number of tasks with non-standardized protocols or evaluated using general downstream tasks. To bridge this research gap and account for the unique sensitivities of French, we present the first-ever publicly available French biomedical language understanding benchmark called DrBenchmark. It encompasses 20 diversified tasks, including named-entity recognition, part-of-speech tagging, question-answering, semantic textual similarity, or classification. We evaluate 8 state-of-the-art pre-trained masked language models (MLMs) on general and biomedical-specific data, as well as English specific MLMs to assess their cross-lingual capabilities. Our experiments reveal that no single model excels across all tasks, while generalist models are sometimes still competitive.

Subword tokenization has become the prevailing standard in the field of natural language processing (NLP) over recent years, primarily due to the widespread utilization of pre-trained language models. This shift began with Byte-Pair Encoding (BPE) and was later followed by the adoption of SentencePiece and WordPiece. While subword tokenization consistently outperforms character and word-level tokenization, the precise factors contributing to its success remain unclear. Key aspects such as the optimal segmentation granularity for diverse tasks and languages, the influence of data sources on tokenizers, and the role of morphological information in Indo-European languages remain insufficiently explored. This is particularly pertinent for biomedical terminology, characterized by specific rules governing morpheme combinations. Despite the agglutinative nature of biomedical terminology, existing language models do not explicitly incorporate this knowledge, leading to inconsistent tokenization strategies for common terms. In this paper, we seek to delve into the complexities of subword tokenization in French biomedical domain across a variety of NLP tasks and pinpoint areas where further enhancements can be made. We analyze classical tokenization algorithms, including BPE and SentencePiece, and introduce an original tokenization strategy that integrates morpheme-enriched word segmentation into existing tokenization methods.

In recent years, pre-trained language models (PLMs) achieve the best performance on a wide range of natural language processing (NLP) tasks. While the first models were trained on general domain data, specialized ones have emerged to more effectively treat specific domains. In this paper, we propose an original study of PLMs in the medical domain on French language. We compare, for the first time, the performance of PLMs trained on both public data from the web and private data from healthcare establishments. We also evaluate different learning strategies on a set of biomedical tasks. In particular, we show that we can take advantage of already existing biomedical PLMs in a foreign language by further pre-train it on our targeted data. Finally, we release the first specialized PLMs for the biomedical field in French, called DrBERT, as well as the largest corpus of medical data under free license on which these models are trained.

Ces dernières années, les modèles de langage pré-entraînés ont obtenu les meilleures performances sur un large éventail de tâches de traitement automatique du langage naturel (TALN). Alors que les premiers modèles ont été entraînés sur des données issues de domaines généraux, des modèles spécialisés sont apparus pour traiter plus efficacement des domaines spécifiques. Dans cet article, nous proposons une étude originale de modèles de langue dans le domaine médical en français. Nous comparons pour la première fois les performances de modèles entraînés sur des données publiques issues du web et sur des données privées issues d’établissements de santé. Nous évaluons également différentes stratégies d’apprentissage sur un ensemble de tâches biomédicales. Enfin, nous publions les premiers modèles spécialisés pour le domaine biomédical en français, appelés DrBERT, ainsi que le plus grand corpus de données médicales sous licence libre sur lequel ces modèles sont entraînés.

Cet article présente MORFITT, le premier corpus multi-labels en français annoté en spécialités dans le domaine médical. MORFITT est composé de 3 624 résumés d’articles scientifiques issus de PubMed, annotés en 12 spécialités pour un total de 5 116 annotations. Nous détaillons le corpus, les expérimentations et les résultats préliminaires obtenus à l’aide d’un classifieur fondé sur le modèle de langage pré-entraîné CamemBERT. Ces résultats préliminaires démontrent la difficulté de la tâche, avec un F-score moyen pondéré de 61,78%.

L’édition 2023 du DÉfi Fouille de Textes (DEFT) s’est concentrée sur le développement de méthodes permettant de choisir automatiquement des réponses dans des questions à choix multiples (QCMs) en français. Les approches ont été évaluées sur le corpus FrenchMedMCQA, intégrant un ensemble de QCMs avec, pour chaque question, cinq réponses potentielles, dans le cadre d’annales d’examens de pharmacie.Deux tâches ont été proposées. La première consistait à identifier automatiquement l’ensemble des réponses correctes à une question. Les résultats obtenus, évalués selon la métrique de l’Exact Match Ratio (EMR), variaient de 9,97% à 33,76%, alors que les performances en termes de distance de Hamming s’échelonnaient de 24,93 à 52,94. La seconde tâche visait à identifier automatiquement le nombre exact de réponses correctes. Les résultats, quant à eux, étaient évalués d’une part avec la métrique de F1-Macro, variant de 13,26% à 42,42%, et la métrique (Accuracy), allant de 47,43% à 68,65%. Parmi les approches variées proposées par les six équipes participantes à ce défi, le meilleur système s’est appuyé sur un modèle de langage large de type LLaMa affiné en utilisant la méthode d’adaptation LoRA.

Cet article présente les systèmes développés par l’équipe LIA-LS2N dans le cadre de la campagne d’évaluation DEFT 2022 (Grouin & Illouz, 2022). Nous avons participé à la première tâche impliquant la correction automatique de copies d’étudiants à partir de références existantes. Nous proposons trois systèmes de classification reposant sur des caractéristiques extraites de plongements de mots contextuels issus d’un modèle BERT (CamemBERT). Nos approches reposent sur les concepts suivants : extraction de mesures de similarité entre les plongements de mots, attention croisée bidirectionnelle entre les plongements et fine-tuning (affinage) des plongements de mots. Les soumissions finales comprenaient deux systèmes fusionnés combinant l’attention croisée bidirectionnelle avec nos classificateurs basés sur BERT et celui sur les mesures de similarité. Notre meilleure soumission obtient une précision de 72,6 % en combinant le classifieur basé sur un modèle CamemBERT affiné et le mécanisme d’attention croisée bidirectionnelle. Ces résultats sont proches de ceux obtenus par le meilleur système de cette édition (75,6 %).

This paper introduces FrenchMedMCQA, the first publicly available Multiple-Choice Question Answering (MCQA) dataset in French for medical domain. It is composed of 3,105 questions taken from real exams of the French medical specialization diploma in pharmacy, mixing single and multiple answers. Each instance of the dataset contains an identifier, a question, five possible answers and their manual correction(s). We also propose first baseline models to automatically process this MCQA task in order to report on the current performances and to highlight the difficulty of the task. A detailed analysis of the results showed that it is necessary to have representations adapted to the medical domain or to the MCQA task: in our case, English specialized models yielded better results than generic French ones, even though FrenchMedMCQA is in French. Corpus, models and tools are available online.