A core aspect of compositionality, systematicity is a desirable property in ML models as it enables strong generalization to novel contexts. This has led to numerous studies proposing benchmarks to assess systematic generalization, as well as models and training regimes designed to enhance it. Many of these efforts are framed as addressing the challenge posed by Fodor and Pylyshyn. However, while they argue for systematicity of representations, existing benchmarks and models primarily focus on the systematicity of behaviour. We emphasize the crucial nature of this distinction. Furthermore, building on Hadley’s (1994) taxonomy of systematic generalization, we analyze the extent to which behavioural systematicity is tested by key benchmarks in the literature across language and vision. Finally, we highlight ways of assessing systematicity of representations in ML models as practiced in the field of mechanistic interpretability.

Despite being ubiquitous in natural language, collocations (e.g., kick+habit) incur a unique processing cost, compared to compositional phrases (kick+door) and idioms (kick+bucket). We confirm this cost with behavioural data as well as MINERVA2, a memory model, suggesting that collocations constitute a distinct linguistic category. While the model fails to fully capture the observed human processing patterns, we find that below a specific item frequency threshold, the model’s retrieval failures align with human reaction times across conditions. This suggests an alternative processing mechanism that activates when memory retrieval fails.