We compared three machine-learning approaches for segmenting discourse units. The performance of each method (“f_score”) is plotted against the number of training tokens (“size”, in log-scale). “weaksup” means data-programming weak supervision, in which the training set is annotated by an automatic noisy labeller based on a manually written set of labeling rules; “finetune” means fine-tuning an LLM (RoBERTa) with varying amounts of hand-labelled data;  “kshot” means prompting GPT-2 with different numbers of examples. Standard fine-tuning of an LLM emerges as the most effective method. It reaches the same performance as the “weaksup” approach while relying on a more straightforward training procedure. The prompting “kshot” approach was lagging behind. Although the models used with the prompting approach are improving rapidly, their intrinsic opacity makes systematic error analysis almost impossible.

Laurent Prévot and Philippe Muller. 2025.
Dialogue & Discourse 16 (2): 35–73. —  @HAL

To account for human language, we propose a multifaceted and explicitly biocultural framework grounded in empirical investigations spanning a diverse array of empirical and theoretical fields. The framework is illustrated with three facets of language with distinctive evolutionary history, each involving multiple disciplines and several species — e.g., humans, primates, and songbirds. Such case studies highlight the importance of both biological preparedness and cultural processes, as well as the interactions between them, in the emergence of language. The empty gray boxes suggest other facets of language, not discussed, that could be similarly investigated under this framework.

Arnon, Inbal, Liran Carmel, Nicolas Claidière, et al. 2025.
What Enables Human Language? A Biocultural Framework
Science 390 (6775): eadq8303. —  @HAL