We unveil that internal representations in large language models (LLMs) serve as reliable proxies of learned knowledge, and propose **RECALL**, a novel representation-aware model merging framework for continual learning without access to historical data. RECALL computes inter-model similarity from layer-wise hidden representations over clustered typical samples, and performs adaptive, hierarchical parameter fusion to align knowledge across models. This design enables the preservation of domain-general features in shallow layers while allowing task-specific adaptation in deeper layers. Unlike prior methods that require task labels or incur performance trade-offs, RECALL achieves seamless multi-domain integration and strong resistance to catastrophic forgetting. Extensive experiments across five NLP tasks and multiple continual learning scenarios show that RECALL outperforms baselines in both knowledge retention and generalization, providing a scalable and data-free solution for evolving LLMs.

BIOASQ Task B Phase B challenge focuses on extracting answers from snippets for a given question. The dataset provided by the organizers contains answers, but not all their variants. Henceforth a manual annotation was performed to extract all forms of correct answers. This article shows the impact of using all occurrences of correct answers for training on the evaluation scores which are improved significantly.

Ontology-based semantic annotation aims at putting fragments of a text in correspondence with proper elements of an ontology such that the formal semantics encoded by the ontology can be exploited to represent text interpretation. In this paper, we formalize a resource for this goal. The main difficulty in achieving good semantic annotations consists in identifying fragments to be annotated and labels to be associated with them. To this end, our approach takes advantage of standard web ontology languages as well as rich linguistic annotation platforms. This in turn is concerned with how to formalize the combination of the ontological and linguistical information, which is a topical issue that has got an increasing discussion recently. Different from existing formalizations, our purpose is to extend ontologies by semantic annotation rules whose complexity increases along two dimensions: the linguistic complexity and the rule syntactic complexity. This solution allows reusing best NLP tools for the production of various levels of linguistic annotations. It also has the merit to distinguish clearly the process of linguistic analysis and the ontological interpretation.

L’annotation sémantique a pour objectif d’apporter au texte une représentation explicite de son interprétation sémantique. Dans un précédent article, nous avons proposé d’étendre les ontologies par des règles d’annotation sémantique. Ces règles sont utilisées pour l’annotation sémantique d’un texte au regard d’une ontologie dans le cadre d’une plate-forme d’annotation linguistique automatique. Nous présentons dans cet article une mesure, basée sur la valeur de Shapley, permettant d’identifier les règles qui sont sources de contradiction dans l’annotation sémantique. Par rapport aux classiques mesures de précision et de rappel, l’intérêt de cette mesure est de ne pas nécessiter de corpus manuellement annoté, d’être entièrement automatisable et de permettre l’identification des règles qui posent problème.