Laurent Romary

Francis Kulumba, Guillaume Vimont, Laurent Romary et al.

Authorship attribution models fine-tuned with the same pretrained encoder, data, and loss can differ four-fold in performance depending only on their scoring mechanism. We use mechanistic interpretability tools to explain this gap. Stylistic features such as word length, punctuation density, and function-word frequency are equally available at every layer in every model, including in an off-the-shelf control encoder, hence the gap not coming from representation quality. Instead, causal intervention shows that the scorer determines where the encoder consolidates authorship signal. Mean pooling forces consolidation by early to mid layers, while late interaction defers it to later layers. We further derive this difference from the gradient structure of each scorer, and training dynamics reveal distinct learning trajectories that follow from that difference.

Francis Kulumba, Wissam Antoun, Guillaume Vimont et al.

Authorship attribution in natural language processing traditionally struggles to distinguish genuine stylistic signals from topical confounds. While contrastive learning approaches have addressed this by maximizing semantic overlap between positive pairs, creating large-scale datasets under strict topic constraints remains challenging. We introduce HALvest, a 17-billion-token multilingual corpus harvested from 778k open-access academic papers, and HALvest-Contrastive, a derived dataset designed to isolate stylometric signals through controlled topic variation. Unlike prior work that minimizes lexical overlap, we exploit natural topic drift between papers by the same author, treating residual lexical patterns as authorial fingerprints rather than noise. Comparing lexical baselines (BM25) against neural models trained on unrestricted (topic-rich) versus base (topic-decoupled) triplets, we demonstrate that models trained exclusively on topic-decoupled data achieve superior performance across all test conditions, outperforming both retrieval baselines and models exposed to topic-rich training data. Our analysis reveals that while lexical signals provide substantial performance gains for keyword-driven methods, neural architectures learn robust stylometric representations that plateau with moderate context length, suggesting they capture distributional style beyond surface-level tokens. Both datasets and code are publicly available.

Biswesh Mohapatra, Giovanni Duca, Laurent Romary et al.

Situated dialogue requires speakers to maintain a reliable representation of shared context rather than reasoning only over isolated utterances. Current conversational agents often struggle with this requirement, especially when the common ground must be preserved beyond the immediate context window. In such settings, fine-grained distinctions are frequently compressed into purely textual representations, leading to a critical failure mode we call \emph{representational blur}, in which similar but distinct entities collapse into interchangeable descriptions. This semantic flattening creates an illusion of grounding, where agents appear locally coherent but fail to track shared context persistently over time. Inspired by the role of mental imagery in human reasoning, and based on the increased availability of multimodal models, we explore whether conversational agents can be given an analogous ability to construct some depictive intermediate representations during dialogue to address these limitations. Thus, we introduce an active visual scaffolding framework that incrementally converts dialogue state into a persistent visual history that can later be retrieved for grounded response generation. Evaluation on the IndiRef benchmark shows that incremental externalization itself improves over full-dialog reasoning, while visual scaffolding provides additional gains by reducing representational blur and enforcing concrete scene commitments. At the same time, textual representations remain advantageous for non-depictable information, and a hybrid multimodal setting yields the best overall performance. Together, these findings suggest that conversational agents benefit from an explicitly multimodal representation of common ground that integrates depictive and propositional information.

Rian Touchent, Laurent Romary, Eric de la Clergerie

Clinical data in hospitals are increasingly accessible for research through clinical data warehouses. However these documents are unstructured and it is therefore necessary to extract information from medical reports to conduct clinical studies. Transfer learning with BERT-like models such as CamemBERT has allowed major advances for French, especially for named entity recognition. However, these models are trained for plain language and are less efficient on biomedical data. Addressing this gap, we introduce CamemBERT-bio, a dedicated French biomedical model derived from a new public French biomedical dataset. Through continual pre-training of the original CamemBERT, CamemBERT-bio achieves an improvement of 2.54 points of F1-score on average across various biomedical named entity recognition tasks, reinforcing the potential of continual pre-training as an equally proficient yet less computationally intensive alternative to training from scratch. Additionally, we highlight the importance of using a standard evaluation protocol that provides a clear view of the current state-of-the-art for French biomedical models.

Biswesh Mohapatra, Théo Charlot, Giovanni Duca et al.

Common ground plays a critical role in situated spoken dialogues, where interlocutors must establish and maintain shared references to entities, events, and relations to sustain coherent interaction. For dialog systems, the ability to correctly ground conversational content in order to refer back to it later is particularly important. Prior studies have demonstrated that LLMs are capable of performing grounding acts such as requesting clarification or producing acknowledgments, yet relatively little work has investigated how common ground can be explicitly represented and stored for later use. Without such mechanisms, it remains unclear whether acknowledgment or clarification behaviors truly reflect a grounded understanding. In this work, we evaluate a model's ability to establish and exploit common ground through relational references to entities within the shared context in a situational dialogue. We test multiple methods for representing common ground in situated dialogues and further propose approaches to improve both the establishment of common ground and its subsequent use in the conversation.

Biswesh Mohapatra, Seemab Hassan, Laurent Romary et al.

Successful conversations often rest on common understanding, where all parties are on the same page about the information being shared. This process, known as conversational grounding, is crucial for building trustworthy dialog systems that can accurately keep track of and recall the shared information. The proficiencies of an agent in grounding the conveyed information significantly contribute to building a reliable dialog system. Despite recent advancements in dialog systems, there exists a noticeable deficit in their grounding capabilities. Traum provided a framework for conversational grounding introducing Grounding Acts and Grounding Units, but substantial progress, especially in the realm of Large Language Models, remains lacking. To bridge this gap, we present the annotation of two dialog corpora employing Grounding Acts, Grounding Units, and a measure of their degree of grounding. We discuss our key findings during the annotation and also provide a baseline model to test the performance of current Language Models in categorizing the grounding acts of the dialogs. Our work aims to provide a useful resource for further research in making conversations with machines better understood and more reliable in natural day-to-day collaborative dialogs.

Petr Knoth, Laurent Romary, Patrice Lopez et al.

A key issue hindering discoverability, attribution and reusability of open research software is that its existence often remains hidden within the manuscript of research papers. For these resources to become first-class bibliographic records, they first need to be identified and subsequently registered with persistent identifiers (PIDs) to be made FAIR (Findable, Accessible, Interoperable and Reusable). To this day, much open research software fails to meet FAIR principles and software resources are mostly not explicitly linked from the manuscripts that introduced them or used them. SoFAIR is a 2-year international project (2024-2025) which proposes a solution to the above problem realised over the content available through the global network of open repositories. SoFAIR will extend the capabilities of widely used open scholarly infrastructures (CORE, Software Heritage, HAL) and tools (GROBID) operated by the consortium partners, delivering and deploying an effective solution for the management of the research software lifecycle, including: 1) ML-assisted identification of research software assets from within the manuscripts of scholarly papers, 2) validation of the identified assets by authors, 3) registration of software assets with PIDs and their archival.

Thibault Clérice, Juliette Janes, Hugo Scheithauer et al.

We present a novel, open-access dataset designed for semantic layout analysis, built to support document recreation workflows through mapping with the Text Encoding Initiative (TEI) standard. This dataset includes 7,254 annotated pages spanning a large temporal range (1600-2024) of digitised and born-digital materials across diverse document types (magazines, papers from sciences and humanities, PhD theses, monographs, plays, administrative reports, etc.) sorted into modular subsets. By incorporating content from different periods and genres, it addresses varying layout complexities and historical changes in document structure. The modular design allows domain-specific configurations. We evaluate object detection models on this dataset, examining the impact of input size and subset-based training. Results show that a 1280-pixel input size for YOLO is optimal and that training on subsets generally benefits from incorporating them into a generic model rather than fine-tuning pre-trained weights.

Julien Abadji, Pedro Ortiz Suarez, Laurent Romary et al.

The need for raw large raw corpora has dramatically increased in recent years with the introduction of transfer learning and semi-supervised learning methods to Natural Language Processing. And while there have been some recent attempts to manually curate the amount of data necessary to train large language models, the main way to obtain this data is still through automatic web crawling. In this paper we take the existing multilingual web corpus OSCAR and its pipeline Ungoliant that extracts and classifies data from Common Crawl at the line level, and propose a set of improvements and automatic annotations in order to produce a new document-oriented version of OSCAR that could prove more suitable to pre-train large generative language models as well as hopefully other applications in Natural Language Processing and Digital Humanities.

Pedro Javier Ortiz Suárez, Laurent Romary, Benoît Sagot

We use the multilingual OSCAR corpus, extracted from Common Crawl via language classification, filtering and cleaning, to train monolingual contextualized word embeddings (ELMo) for five mid-resource languages. We then compare the performance of OSCAR-based and Wikipedia-based ELMo embeddings for these languages on the part-of-speech tagging and parsing tasks. We show that, despite the noise in the Common-Crawl-based OSCAR data, embeddings trained on OSCAR perform much better than monolingual embeddings trained on Wikipedia. They actually equal or improve the current state of the art in tagging and parsing for all five languages. In particular, they also improve over multilingual Wikipedia-based contextual embeddings (multilingual BERT), which almost always constitutes the previous state of the art, thereby showing that the benefit of a larger, more diverse corpus surpasses the cross-lingual benefit of multilingual embedding architectures.

Pedro Javier Ortiz Suárez, Yoann Dupont, Benjamin Muller et al.

The French TreeBank developed at the University Paris 7 is the main source of morphosyntactic and syntactic annotations for French. However, it does not include explicit information related to named entities, which are among the most useful information for several natural language processing tasks and applications. Moreover, no large-scale French corpus with named entity annotations contain referential information, which complement the type and the span of each mention with an indication of the entity it refers to. We have manually annotated the French TreeBank with such information, after an automatic pre-annotation step. We sketch the underlying annotation guidelines and we provide a few figures about the resulting annotations.

Louis Martin, Benjamin Muller, Pedro Javier Ortiz Suárez et al.

Pretrained language models are now ubiquitous in Natural Language Processing. Despite their success, most available models have either been trained on English data or on the concatenation of data in multiple languages. This makes practical use of such models --in all languages except English-- very limited. In this paper, we investigate the feasibility of training monolingual Transformer-based language models for other languages, taking French as an example and evaluating our language models on part-of-speech tagging, dependency parsing, named entity recognition and natural language inference tasks. We show that the use of web crawled data is preferable to the use of Wikipedia data. More surprisingly, we show that a relatively small web crawled dataset (4GB) leads to results that are as good as those obtained using larger datasets (130+GB). Our best performing model CamemBERT reaches or improves the state of the art in all four downstream tasks.

Laurent Romary, Mohamed Khemakhem, Fahad Khan et al.

Lexical Markup Framework (LMF) or ISO 24613 [1] is a de jure standard that provides a framework for modelling and encoding lexical information in retrodigitised print dictionaries and NLP lexical databases. An in-depth review is currently underway within the standardisation subcommittee , ISO-TC37/SC4/WG4, to find a more modular, flexible and durable follow up to the original LMF standard published in 2008. In this paper we will present some of the major improvements which have so far been implemented in the new version of LMF.

Jack Bowers, Laurent Romary

This paper aims to provide a comprehensive modeling and representation of etymological data in digital dictionaries. The purpose is to integrate in one coherent framework both digital representations of legacy dictionaries, and also born-digital lexical databases that are constructed manually or semi-automatically. We want to propose a systematic and coherent set of modeling principles for a variety of etymological phenomena that may contribute to the creation of a continuum between existing and future lexical constructs, where anyone interested in tracing the history of words and their meanings will be able to seamlessly query lexical resources.Instead of designing an ad hoc model and representation language for digital etymological data, we will focus on identifying all the possibilities offered by the TEI guidelines for the representation of lexical information.

Laurent Romary, Mike Mertens, Anne Baillot

This paper provides both an update concerning the setting up of the European DARIAH infrastructure and a series of strong action lines related to the development of a data centred strategy for the humanities in the coming years. In particular we tackle various aspect of data management: data hosting, the setting up of a DARIAH seal of approval, the establishment of a charter between cultural heritage institutions and scholars and finally a specific view on certification mechanisms for data.

Laurent Romary

This paper provides an overview of the various projects carried out within ISO committee TC 37/SC 4 dealing with the management of language (digital) resources. On the basis of the technical experience gained in the committee and the wider standardization landscape the paper identifies some possible trends for the future.

Laurent Romary, Andreas Witt

In recent years, new developments in the area of lexicography have altered not only the management, processing and publishing of lexicographical data, but also created new types of products such as electronic dictionaries and thesauri. These expand the range of possible uses of lexical data and support users with more flexibility, for instance in assisting human translation. In this article, we give a short and easy-to-understand introduction to the problematic nature of the storage, display and interpretation of lexical data. We then describe the main methods and specifications used to build and represent lexical data. This paper is targeted for the following groups of people: linguists, lexicographers, IT specialists, computer linguists and all others who wish to learn more about the modelling, representation and visualization of lexical knowledge. This paper is written in two languages: French and German.

Laurent Romary

This paper presents an attempt to customise the TEI (Text Encoding Initiative) guidelines in order to offer the possibility to incorporate TBX (TermBase eXchange) based terminological entries within any kind of TEI documents. After presenting the general historical, conceptual and technical contexts, we describe the various design choices we had to take while creating this customisation, which in turn have led to make various changes in the actual TBX serialisation. Keeping in mind the objective to provide the TEI guidelines with, again, an onomasiological model, we try to identify the best comprise in maintaining both the isomorphism with the existing TBX Basic standard and the characteristics of the TEI framework.

Laurent Romary

The present paper explores various arguments in favour of making the Text Encoding Initia-tive (TEI) guidelines an appropriate serialisation for ISO standard 24613:2008 (LMF, Lexi-cal Mark-up Framework) . It also identifies the issues that would have to be resolved in order to reach an appropriate implementation of these ideas, in particular in terms of infor-mational coverage. We show how the customisation facilities offered by the TEI guidelines can provide an adequate background, not only to cover missing components within the current Dictionary chapter of the TEI guidelines, but also to allow specific lexical projects to deal with local constraints. We expect this proposal to be a basis for a future ISO project in the context of the on going revision of LMF.

Kais Haddar, Héla Fehri, Laurent Romary

The comparative evaluation of Arabic HPSG grammar lexica requires a deep study of their linguistic coverage. The complexity of this task results mainly from the heterogeneity of the descriptive components within those lexica (underlying linguistic resources and different data categories, for example). It is therefore essential to define more homogeneous representations, which in turn will enable us to compare them and eventually merge them. In this context, we present a method for comparing HPSG lexica based on a rule system. This method is implemented within a prototype for the projection from Arabic HPSG to a normalised pivot language compliant with LMF (ISO 24613 - Lexical Markup Framework) and serialised using a TEI (Text Encoding Initiative) based representation. The design of this system is based on an initial study of the HPSG formalism looking at its adequacy for the representation of Arabic, and from this, we identify the appropriate feature structures corresponding to each Arabic lexical category and their possible LMF counterparts.