We release Pantagruel models, a new family of self-supervised encoder models for French text and speech. Instead of predicting modality-tailored targets such as textual tokens or speech units, Pantagruel learns contextualized target representations in the feature space, allowing modality-specific encoders to capture linguistic and acoustic regularities more effectively. Separate models are pre-trained on large-scale French corpora, including Wikipedia, OSCAR and CroissantLLM for text, together with MultilingualLibriSpeech, LeBenchmark, and INA-100k for speech. INA-100k is a newly introduced 100,000-hour corpus of French audio derived from the archives of the Institut National de l’Audiovisuel (INA), the national repository of French radio and television broadcasts, providing highly diverse audio data. We evaluate Pantagruel across a broad range of downstream tasks spanning both modalities, including those from the standard French benchmarks such as FLUE or LeBenchmark. Across these tasks, Pantagruel models show competitive or superior performance compared to strong French baselines such as CamemBERT, FlauBERT, and LeBenchmark2.0, while maintaining a shared architecture that can seamlessly handle either speech or text inputs. These results confirm the effectiveness of feature-space self-supervised objectives for French representation learning and highlight Pantagruel as a robust foundation for multimodal speech-text understanding.

This study introduces a new ASR-transcribed coreference corpus for French and explores the transferability of coreference resolution models from human-transcribed to ASR-transcribed data. Given the challenges posed by differences in text characteristics and errors introduced by ASR systems, we evaluate model performance using newly constructed parallel human-ASR silver training and gold validation datasets. Our findings show a decline in performance on ASR data for models trained on manual transcriptions. However, combining silver ASR data with gold manual data enhances model robustness. Through detailed error analysis, we observe that models emphasizing recall are more resilient to ASR-induced errors compared to those focusing on precision. The resulting ASR corpus, along with all related materials, is freely available under the CC BY-NC-SA 4.0 license at: https://github.com/ina-foss/french-asr-coreference.

The ever-growing number of people suffering from mental distress has motivated significant research initiatives towards automated depression estimation. Despite the multidisciplinary nature of the task, very few of these approaches include medical professionals in their research process, thus ignoring a vital source of domain knowledge. In this paper, we propose to bring the domain experts back into the loop and incorporate their knowledge within the gold-standard DAIC-WOZ dataset. In particular, we define a novel transformer-based architecture and analyse its performance in light of our expert annotations. Overall findings demonstrate a strong correlation between the psychological tendencies of medical professionals and the behavior of the proposed model, which additionally provides new state-of-the-art results.

This paper addresses the quality of annotations in mental health datasets used for NLP-based depression level estimation from social media texts. While previous research relies on social media-based datasets annotated with binary categories, i.e. depressed or non-depressed, recent datasets such as D2S and PRIMATE aim for nuanced annotations using PHQ-9 symptoms. However, most of these datasets rely on crowd workers without the domain knowledge for annotation. Focusing on the PRIMATE dataset, our study reveals concerns regarding annotation validity, particularly for the lack of interest or pleasure symptom. Through reannotation by a mental health professional, we introduce finer labels and textual spans as evidence, identifying a notable number of false positives. Our refined annotations, to be released under a Data Use Agreement, offer a higher-quality test set for anhedonia detection. This study underscores the necessity of addressing annotation quality issues in mental health datasets, advocating for improved methodologies to enhance NLP model reliability in mental health assessments.

This paper explores the impact of incorporating sentiment, emotion, and domain-specific lexicons into a transformer-based model for depression symptom estimation. Lexicon information is added by marking the words in the input transcripts of patient-therapist conversations as well as in social media posts. Overall results show that the introduction of external knowledge within pre-trained language models can be beneficial for prediction performance, while different lexicons show distinct behaviours depending on the targeted task. Additionally, new state-of-the-art results are obtained for the estimation of depression level over patient-therapist interviews.

This paper presents our system for the MEDIQA-Chat 2023 shared task on medical conversation summarization. Our approach involves finetuning a LongT5 model on multiple tasks simultaneously, which we demonstrate improves the model’s overall performance while reducing the number of factual errors and hallucinations in the generated summary. Furthermore, we investigated the effect of augmenting the data with in-text annotations from a clinical named entity recognition model, finding that this approach decreased summarization quality. Lastly, we explore using different text generation strategies for medical note generation based on the length of the note. Our findings suggest that the application of our proposed approach can be beneficial for improving the accuracy and effectiveness of medical conversation summarization.

We propose a novel hybrid approach to lemmatization that enhances the seq2seq neural model with additional lemmas extracted from an external lexicon or a rule-based system. During training, the enhanced lemmatizer learns both to generate lemmas via a sequential decoder and copy the lemma characters from the external candidates supplied during run-time. Our lemmatizer enhanced with candidates extracted from the Apertium morphological analyzer achieves statistically significant improvements compared to baseline models not utilizing additional lemma information, achieves an average accuracy of 97.25% on a set of 23 UD languages, which is 0.55% higher than obtained with the Stanford Stanza model on the same set of languages. We also compare with other methods of integrating external data into lemmatization and show that our enhanced system performs considerably better than a simple lexicon extension method based on the Stanza system, and it achieves complementary improvements w.r.t. the data augmentation method.

We explore how well a sequence labeling approach, namely, recurrent neural network, is suited for the task of resource-poor and POS tagging free word stress detection in the Russian, Ukranian, Belarusian languages. We present new datasets, annotated with the word stress, for the three languages and compare several RNN models trained on three languages and explore possible applications of the transfer learning for the task. We show that it is possible to train a model in a cross-lingual setting and that using additional languages improves the quality of the results.

In this study we address the problem of automated word stress detection in Russian using character level models and no part-speech-taggers. We use a simple bidirectional RNN with LSTM nodes and achieve accuracy of 90% or higher. We experiment with two training datasets and show that using the data from an annotated corpus is much more efficient than using only a dictionary, since it allows to retain the context of the word and its morphological features.