Cognitive Neuroscience最新文献

Despite the overtly discrete nature of language, the use of semantic embedding spaces is pervasive in modern computational linguistics and machine learning for natural language. I argue that this is intelligible if language is viewed as an interface into a general-purpose system of concepts, in which metric spaces capture rich relationships. At the same time, language embeddings can be regarded, at least heuristically, as equivalent to parameters of distributions over word-word relationships.

Parr et al., 2025 examines how auto-regressive and deep temporal models differ in their treatment of non-Markovian sequence modelling. Building on this, we highlight the need for dissociating model architectures-i.e., how the predictive distribution factorises-from the computations invoked at inference. We demonstrate that deep temporal computations are mimicked by autoregressive models by structuring context access during iterative inference. Using a transformer trained on next-token prediction, we show that inducing hierarchical temporal factorisation during iterative inference maintains predictive capacity while instantiating fewer computations. This emphasises that processes for constructing and refining predictions are not necessarily bound to their underlying model architectures.

We were grateful for the level of engagement and insight from the commentaries on our discussion article, and for the opportunity to pick up on some of the common themes in what follows. Several commentaries focused upon the degeneracy in the relationship in how an internal model - of the sort that might be used either by an AI system or by our brains - might be formulated and the way in which this degeneracy might be resolved. Further themes were the role of model width as opposed to depth, the phenomenology of non-Markovian time, and a useful reminder that linguistic communication is necessarily a multi-agent, collective, endeavor.

Mougenot and Matheson outline a theoretical approach to cognitive neuroscience that combines the commitments of embodied cognition with a mechanistic approach to scientific explanation. They argue that this theoretical approach provides several general benefits, including enabling researchers to develop more robust theories and ontologies that do not require either neuroscientific reductionism or the complete autonomy of psychology from neuroscience. In this commentary, I argue that the sort of embodied cognitive neuroscience that they envision has a more specific benefit: it has the potential to help resolve internal tensions within 4E cognition.

Ecological and enactive approaches to embodied cognition endorse a concept of constitution that involves dynamical causality. I argue that this is a challenge for new mechanistic accounts which hold to a strict distinction between causality and constitution.

I aim to discuss which constitutive components are essential for explaining how the mind works. Rather than focusing on some specific components, I emphasize their diversity. Thus, I seek to complement the recent mechanistic proposal by underscoring that researchers should remain open-minded about which constitutive components should be investigated.

The review article Theoretical strategies for an embodied cognitive neuroscience proposes that the embodied cognition framework can be applied to develop mechanistic explanations for cognitive neuroscience phenomena. In our commentary we argue that any mechanistic explanation of such phenomena must be able to account for individual differences in cognition that are an inevitable consequence of the varied brain-body-environment experiences that comprise embodied cognition. We propose that, while mechanistic accounts may be able to model individual differences, the definition of mechanistic models may limit their application to the study of individual differences.

Mougenot and Matheson provide an interesting analysis on how some core ideas of the 'New Mechanists' - the proponents of a normative framework for scientific explanations based on the identification and description of mechanisms - might be relevant for the development of an embodied approach to cognitive neuroscience. Although we are highly sympathetic to such an approach, we struggle to identify the benefits of adopting the notion of mechanism for such enterprise.

Mougenot and Matheson propose that mechanistic models can explain behavior by describing the complex interactions among components of the brain, body, and environment as an integrated system, which aligns with embodied cognition. However, we suggest incorporating cognitive ontology theory and addressing degeneracy and neuronal reuse. We also recommend studying natural embodied cognition through artificial systems to develop a comprehensive mechanistic framework.