There is an important challenge in systematically interpreting the internal representations of deep neural networks (DNNs). Existing techniques are often less effective for non-tabular tasks, or they primarily focus on qualitative, ad-hoc interpretations of models. In response, this study introduces a cognitive science-inspired, multi-dimensional quantification and visualization approach that captures two temporal dimensions of model learning: the "information-processing trajectory" and the "developmental trajectory." The former represents the influence of incoming signals on an agent's decision-making, while the latter conceptualizes the gradual improvement in an agent's performance throughout its lifespan. Tracking the learning curves of DNNs enables researchers to explicitly identify the model appropriateness of a given task, examine the properties of the underlying input signals, and assess the model's alignment (or lack thereof) with human learning experiences. To illustrate this method, we conducted 750 runs of simulations on two temporal tasks: gesture detection and sentence classification, showcasing its applicability across different types of deep learning tasks. Using four descriptive metrics to quantify the mapped learning curves-start, end - start, max, tmax-, we identified significant differences in learning patterns based on data sources and class distinctions (all p's < .0001), the prominent role of spatial semantics in gesture learning, and larger information gains in language learning. We highlight three key insights gained from mapping learning curves: non-monotonic progress, pairwise comparisons, and domain distinctions. We reflect on the theoretical implications of this method for cognitive processing, language models and representations from multiple modalities.

Perspectives are central to paradox, from perceptual illusions (Bruno, N. (2001). When does action resist visual illusions? Trends in Cognitive Sciences, 5(9), 379–382), to dialetheism (Dietrich, E. (2008). The Bishop and Priest: Toward a point-of-view based epistemology of true contradictions. Logos Architekton. Journal of Logic and Philosophy of Science, 2, 35–58.; Dietrich & Webber, this issue). But why? We argue that apparent inconsistencies among perspectives are driven by the observer's methods, goals, etc., of inquiry: The manner in which the observer is observing. It is well known in various areas of science that observers inherently disrupt the system that is being studied. But even after we have measurement schemes (or theoretical apparatus) for collecting maximally observer-untainted behavioural data, there remains an inextricable observer-centred influence. We consider the case of cognitive science, in which diverse theoretical frameworks on offer continue to be pitted against one another as if one were going to be churned out as true, and the others falsified. A simulation is developed in which the observer's observer's paradox is demonstrated in a highly idealised computational model of a scientist studying a system. We use a model-building method referred to as ‘ε-machine reconstruction’ (Crutchfield, 1994) and provide readers a basic introduction to it. By using this framework to model the observer–measurement process, we explore how different measurement schemes impact resultant theories. The end product is a kind of existence proof, which we liken to more complicated circumstances of cognitive science. We argue that progress in cognitive science – overcoming apparently paradoxical persistence of several theoretical frameworks – may come from rendering the observer as an active aspect of these cognitive theories. The upshot would be one of two possibilities. Either theories are inconsistent because they ‘access’ divergent levels of organisation in the human cognitive system, or they may be integrated in an as-yet unknown theoretical framework.

A case for a pluralistic approach to cognitive science is sketched. It is argued that cognitive scientists should take seriously the possibility that a single, unified framework for all of cognition is an unrealistic expectation for its diverse interdisciplinary goals and subject matter. A pluralistic approach instead seeks ways of integrating the multiple perspectives that have provided explanatory success in loosely interconnected sub-domains of cognitive phenomena. Research strategies recommended by this approach are discussed, with review of research currently carrying out such strategies and others that may hold promise for the future. The article ends with a discussion of seeking closer integration of the inquirer into consideration of which explanatory framework to choose. A systematic exploration of this transactional approach to cognitive science may grant coherence to pluralism even as it embraces diverse schemes of explanation.

I am surprised to be replying to a series of commentaries that agree with the basic premise of pluralism. Two pairs of commentators were invited to offer evaluations of pluralism from contrasting p...

We introduce symbolic dynamics to cognitive scientists with the aim of furthering constructive debate on representation. Symbolic dynamics is a mathematical framework in which both continuous and discrete states of a system can be considered jointly. We discuss a number of theoretical implications this framework has for cognitive science, and offer some consideration of the way in which it might be employed for comparing or conciliating discrete and continuous representational theories. Symbolic dynamics may thus serve as a common, level playing field for debate in theories of cognitive representation.