In this work, we introduce a comprehensive error typology specifically designed for evaluating two distinct tasks in machine-generated patent texts: claims-to-abstract generation, and the generation of the next claim given previous ones. We have also developed a benchmark, PatentEval, for systematically assessing language models in this context. Our study includes a comparative analysis, annotated by humans, of various models. These range from those specifically adapted during training for tasks within the patent domain to the latest general-purpose large language models (LLMs). Furthermore, we explored and evaluated some metrics to approximate human judgments in patent text evaluation, analyzing the extent to which these metrics align with expert assessments. These approaches provide valuable insights into the capabilities and limitations of current language models in the specialized field of patent text generation.

This paper proposes a novel approach to French patent classification leveraging data-centric strategies. We compare different approaches for the two deepest levels of the IPC hierarchy: the IPC group and subgroups. Our experiments show that while simple ensemble strategies work for shallower levels, deeper levels require more sophisticated techniques such as data augmentation, clustering, and negative sampling. Our research highlights the importance of language-specific features and data-centric strategies for accurate and reliable French patent classification. It provides valuable insights and solutions for researchers and practitioners in the field of patent classification, advancing research in French patent classification.