The color phi phenomenon: Not so special, after all?

We show how anomalous time reversal of stimuli and their associated responses can exist in very small connectionist models. These networks are built from dynamical toy model neurons which adhere to a minimal set of biologically plausible properties. The appearance of a “ghost” response, temporally and spatially located in between responses caused by actual stimuli, as in the phi phenomenon, is demonstrated in a similar small network, where it is caused by priming and long-distance feedforward paths. We then demonstrate that the color phi phenomenon can be present in an echo state network, a recurrent neural network, without explicitly training for the presence of the effect, such that it emerges as an artifact of the dynamical processing. Our results suggest that the color phi phenomenon might simply be a feature of the inherent dynamical and nonlinear sensory processing in the brain and in and of itself is not related to consciousness.

Formal Models

This chapter introduces formal models of cognition and explains how they are similar to verbal theories but use computer programs and mathematics to avoid the many limitations of human reasoning, thereby adding precision and rigor to their explanations. The chapter discusses Marr’s (1982) levels of analyses and how information-processing systems can be understood and described in terms of the task being performed, the representations and algorithms used to perform the task, and how the latter are implemented by physical systems. This then motivates discussion of three common approaches to modeling human cognition and behavior: process models, production-system models, and connectionist models. Each of these approaches is critiqued, with discussion of its merits and limitations. The three modeling approaches are then further illustrated by showing how each might be used to explain the finding that words can be identified more efficiently if they occur in predictable sentence contexts. The chapter closes with a discussion of how cognitive models are evaluated using their simplicity, theoretical scope, compatibility (e.g., with biology), and their capacity to generate novel predictions for guiding research.

Folk psychology, eliminativism, and the present state of connectionism

Three decades ago, William Ramsey, Steven Stich & Joseph Garon put forward an argument in favor of the following conditional: if connectionist models that implement parallelly distributed processing represent faithfully human cognitive processing, eliminativism about propositional attitudes is true. The corollary of their argument (if it proves to be sound) is that there is no place for folk psychology in contemporary cognitive science. This understanding of connectionism as a hypothesis about cognitive architecture compatible with eliminativism is also endorsed by Paul Churchland, a radical opponent of folk psychology and a prominent supporter of eliminative materialism. I aim to examine whether current connectionist models based on long-short term memory (LSTM) neural networks can back up these arguments in favor of eliminativism. Nonetheless, I will rather put my faith in the eliminativism of the limited domain. This position amount to the following claim: even though that connectionist cognitive science has no need whatsoever for folk psychology qua theory, this does not entail illegitimacy of folk psychology per se in other scientific domains, most notably in humanities, but only if one sees folk psychology as mere heuristics.

The inflectional morphology representation of individual words in the mental lexicon

Models of complex word recognition can be separated into two wide groups: symbolic and connectionist. Symbolic models presume the existence of an explicit morphological representation of individual words; connectionist models do not and consider morphological effects to be a by-product of interaction between phonological, orthographic and semantic information. This study aimed to test whether there are explicit mental representations of inflected lexical units in the mental lexicon. Accordingly, the method of inflected suffix morphological and semantic priming of nouns in the Serbian language was used. In the morphological priming condition, the prime and the target shared the same inflectional suffix. In Experiment 1 overt priming was used, while in Experiment 2, masked priming. The results showed no significant effects of inflected suffix morphological priming, while significant semantic priming effects were recorded. The results obtained in this research are in line with predictions of the connectionist models.

A connectionist approach to analogy. On the modal meaning of periphrastic do in Early Modern English

This paper innovatively charts the analogical influence of the modal auxiliaries on the regulation of periphrastic do in Early Modern English by means of Convolutional Neural Networks (CNNs), a flavour of connectionist models known for their applications in computer vision. CNNs can be harnessed to model the choice between competitors in a linguistic alternation by extracting not only the contexts a construction occurs in, but also the contexts it could have occurred in, but did not. Bearing on the idea that two forms are perceived as similar if they occur in similar contexts, the models provide us with pointers towards potential loci of analogical attraction that would be hard to retrieve otherwise. Our analysis reveals clear functional overlap between do and all modals, indicating not only that analogical pressure was highly likely, but even that affirmative declarative do functioned as a modal auxiliary itself throughout the late 16th century.

Object-Oriented Generative Adversarial Networks

(Performed in 2018 as a class project) Deep learning is a field that has been mainly driven by connectionist models like neural networks, characterized by layered processing of distributed, sub-symbolic and statistical features. However, human high-level thoughts appear to be highly symbolic, focusing on objects and relations.To bridge the gap between perception and symbols, a series of models on "Object Oriented Deep Learning" was proposed [9,8,7]. In this project we further explore this class of models. We implement a generative version of OODL that can generate images instead of performing object recognition, in a similar way to Generative Adversarial Networks (GANs). In comparison to conventional “feature-oriented” deep learning, OODL naturally handles properties of objects by incorporating them as fields. It offers exact equivariance [8] to translation, rotation and scaling. When implementing it as a generative model, one should be able to precisely control such geometric properties of the generated objects.