The commentaries reflect three core themes that pertain not just to our theory, but to the enterprise of connectionist modeling more generally. The first concerns the relationship between a cognitive theory and an implemented computer model. Specifically, how does one determine, when a model departs from the theory it exemplifies, whether the departure is a useful simplification or a critical flaw? We argue that the answer to this question depends partially upon the model's intended function, and we suggest that connectionist (...) models have important functions beyond the commonly accepted goals of fitting data and making predictions. The second theme concerns perceived in-principle limitations of the connectionist approach to cognition, and the specific concerns these perceived limitations raise for our theory. We argue that the approach is not in fact limited in the ways our critics suggest. One common misconception, that connectionist models cannot address abstract or relational structure, is corrected through new simulations showing directly that such structure can be captured. The third theme concerns the relationship between parallel distributed processing (PDP) models and structured probabilistic approaches. In this case we argue that there the difference between the approaches is not merely one of levels. Our PDP approach differs from structured statistical approaches at all of Marr's levels, including the characterization of the goals of cognitive computations, and of the representations and algorithms used. (shrink)

Abstract Since their inception, distributional models of semantics have been criticized as inadequate cognitive theories of human semantic learning and representation. A principal challenge is that the representations derived by distributional models are purely symbolic and are not grounded in perception and action; this challenge has led many to favor feature-based models of semantic representation. We argue that the amount of perceptual and other semantic information that can be learned from purely distributional statistics has been underappreciated. We compare the representations (...) of three feature-based and nine distributional models using a semantic clustering task. Several distributional models demonstrated semantic clustering comparable with clustering-based on feature-based representations. Furthermore, when trained on child-directed speech, the same distributional models perform as well as sensorimotor-based feature representations of children’s lexical semantic knowledge. These results suggest that, to a large extent, information relevant for extracting semantic categories is redundantly coded in perceptual and linguistic experience. Detailed analyses of the semantic clusters of the feature-based and distributional models also reveal that the models make use of complementary cues to semantic organization from the two data streams. Rather than conceptualizing feature-based and distributional models as competing theories, we argue that future focus should be on understanding the cognitive mechanisms humans use to integrate the two sources. (shrink)

Similarity is one of the most important relations humans perceive, arguably subserving category learning and categorization, generalization and discrimination, judgment and decision making, and other cognitive functions. Researchers have proposed a wide range of representations and metrics that could be at play in similarity judgment, yet have not comprehensively compared the power of these representations and metrics for predicting similarity within and across different semantic categories. We performed such a comparison by pairing nine prominent vector semantic representations with seven established (...) similarity metrics that could operate on these representations, as well as supervised methods for dimensional weighting in the similarity function. This approach yields a factorial model structure with 126 distinct representation‐metric pairs, which we tested on a novel dataset of similarity judgments between pairs of cohyponymic words in eight categories. We found that cosine similarity and Pearson correlation were the overall best performing unweighted similarity functions, and that word vectors derived from free association norms often outperformed word vectors derived from text (including those specialized for similarity). Importantly, models that used human similarity judgments to learn category‐specific weights on dimensions yielded substantially better predictions than all unweighted approaches across all types of similarity functions and representations, although dimension weights did not generalize well across semantic categories, suggesting strong category context effects in similarity judgment. We discuss implications of these results for cognitive modeling and natural language processing, as well as for theories of the representations and metrics involved in similarity. (shrink)

The structure of people’s conceptual knowledge of concrete nouns has traditionally been viewed as hierarchical (Collins & Quillian, 1969). For example, superordinate concepts (vegetable) are assumed to reside at a higher level than basic‐level concepts (carrot). A feature‐based attractor network with a single layer of semantic features developed representations of both basic‐level and superordinate concepts. No hierarchical structure was built into the network. In Experiment and Simulation 1, the graded structure of categories (typicality ratings) is accounted for by the flat (...) attractor network. Experiment and Simulation 2 show that, as with basic‐level concepts, such a network predicts feature verification latencies for superordinate concepts (vegetable ). In Experiment and Simulation 3, counterintuitive results regarding the temporal dynamics of similarity in semantic priming are explained by the model. By treating both types of concepts the same in terms of representation, learning, and computations, the model provides new insights into semantic memory. (shrink)

Recent studies of semantic memory have investigated two theories of optimal search adopted from the animal foraging literature: Lévy flights and marginal value theorem. Each theory makes different simplifying assumptions and addresses different findings in search behaviors. In this study, an experiment is conducted to test whether clustering in semantic memory may play a role in evidence for both theories. Labeled magnets and a whiteboard were used to elicit spatial representations of semantic knowledge about animals. Category recall sequences from a (...) separate experiment were used to trace search paths over the spatial representations of animal knowledge. Results showed that spatial distances between animal names arranged on the whiteboard were correlated with inter-response intervals during category recall, and distributions of both dependent measures approximated inverse power laws associated with Lévy flights. In addition, IRIs were relatively shorter when paths first entered animal clusters, and longer when they exited clusters, which is consistent with marginal value theorem. In conclusion, area-restricted searches over clustered semantic spaces may account for two different patterns of results interpreted as supporting two different theories of optimal memory foraging. (shrink)

Two experiments investigated subliminal semantic and form priming in a word-completion task. Visual gap-words with a dominant and a subordinate solution were preceded by form-related or by semantically related words, which were briefly presented and sandwich-masked. Priming of the subordinate solution was assessed in Experiment 1, relative to a neutral condition. Both solutions were primed in Experiment 2. In the absence of conscious prime recognition, both semantic and form primes reliably increased the probability with which the primed solution was given. (...) With our variant of fragment-completion, response priming can be ruled out as explanation. Moreover, effects were already present at first presentation, excluding an interpretation in terms of partial awareness due to massive repetition. The data demonstrate automatic activation at both form and semantic levels in the absence of conscious awareness. (shrink)

Models of category learning have been extensively studied in cognitive science and primarily tested on perceptual abstractions or artificial stimuli. In this paper, we focus on categories acquired from natural language stimuli, that is, words. We present a Bayesian model that, unlike previous work, learns both categories and their features in a single process. We model category induction as two interrelated subproblems: the acquisition of features that discriminate among categories, and the grouping of concepts into categories based on those features. (...) Our model learns categories incrementally using particle filters, a sequential Monte Carlo method commonly used for approximate probabilistic inference that sequentially integrates newly observed data and can be viewed as a plausible mechanism for human learning. Experimental results show that our incremental learner obtains meaningful categories which yield a closer fit to behavioral data compared to related models while at the same time acquiring features which characterize the learned categories. (shrink)

Understanding spoken words involves a rapid mapping from speech to conceptual representations. One distributed feature-based conceptual account assumes that the statistical characteristics of concepts’ features—the number of concepts they occur in and likelihood of co-occurrence —determine conceptual activation. To test these claims, we investigated the role of distinctiveness/sharedness and correlational strength in speech-to-meaning mapping, using a lexical decision task and computational simulations. Responses were faster for concepts with higher sharedness, suggesting that shared features are facilitatory in tasks like lexical decision (...) that require access to them. Correlational strength facilitated responses for slower participants, suggesting a time-sensitive co-occurrence-driven settling mechanism. The computational simulation showed similar effects, with early effects of shared features and later effects of correlational strength. These results support a general-to-specific account of conceptual processing, whereby early activation of shared features is followed by the gradual emergence of a specific target representation. (shrink)

Computational modeling and eye-tracking were used to investigate how phonological and semantic information interact to influence the time course of spoken word recognition. We extended our recent models to account for new evidence that competition among phonological neighbors influences activation of semantically related concepts during spoken word recognition . The model made a novel prediction: Semantic input modulates the effect of phonological neighbors on target word processing, producing an approximately inverted-U-shaped pattern with a high phonological density advantage at an intermediate (...) level of semantic input—in contrast to the typical disadvantage for high phonological density words in spoken word recognition. This prediction was confirmed with a new analysis of the Apfelbaum et al. data and in a visual world paradigm experiment with preview duration serving as a manipulation of strength of semantic input. These results are consistent with our previous claim that strongly active neighbors produce net inhibitory effects and weakly active neighbors produce net facilitative effects. (shrink)

I present a case study of scientific discovery, where building two functional and behavioral approximations of neurons, one physical and the other computational, led to conceptual and implementation breakthroughs in a neural engineering laboratory. Such building of external systems that mimic target phenomena, and the use of these external systems to generate novel concepts and control structures, is a standard strategy in the new engineering sciences. I develop a model of the cognitive mechanism that connects such built external systems with (...) internal models, and I examine how new discoveries, and consensus on discoveries, could arise from this external‐internal coupling and the building process. The model is based on the emerging framework of common coding, which proposes a shared representation in the brain between the execution, perception, and imagination of movement. (shrink)

Artificial intelligence and robots may progressively take a more and more prominent place in our daily environment. Interestingly, in the study of how humans perceive these artificial entities, science has mainly taken an anthropocentric perspective (i.e., how distant from humans are these agents). Considering people’s fears and expectations from robots and artificial intelligence, they tend to be simultaneously afraid and allured to them, much as they would be to the conceptualisations related to the divine entities (e.g., gods). In two experiments, (...) we investigated the proximity of representation between artificial entities (i.e., artificial intelligence and robots), divine entities and natural entities (i.e., humans and other animals) at both an explicit (Study 1) and an implicit level (Study 2). In the first study, participants evaluated these entities explicitly on positive and negative attitudes. Hierarchical clustering analysis showed that participants’ representation of artificial intelligence, robots and divine entities were similar, while the representation of humans tended to be associated with that of animals. In the second study, participants carried out a word/non-word decision task including religious semantic-related words and neutral words after the presentation of a masked prime referring to divine entities, artificial entities and natural entities (or a control prime). Results showed that after divine and artificial entity primes, participants were faster to identify religious words as words compared to neutral words arguing for a semantic activation. We conclude that people make sense of the new entities by relying on already familiar entities and in the case of artificial intelligence and robots, people appear to draw parallels to divine entities. (shrink)

This thesis explores the use of the concept 'realization' in the philosophy of mind. The primary focus is on the role realization plays in assessing or opposing identity theory. The history of the use of the concept of realization in the philosophy of mind is reviewed, and from that a set of desiderata to be used for assessing accounts of realization is extracted. The desiderata are applied to a sample account of realization proposed by Sydney Shoemaker. (2007) Next the application (...) of 'realization' in contemporary contexts is considered, focusing on the idea that mental kinds are, potentially, multiply realized. Based on interpretations of results from research in the relevant sciences this thesis considers two kinds of strategies used to object to multiple realization, (1) arguments against the concept of multiple realization: the Grain Argument (Bechtel and Mundale 1999), and Shapiro's Dilemma (Shapiro 2000), and (2) defeaters for alleged cases/examples of multiple realization proposed by Lawrence Shapiro and Thomas Polger: unification, individual differences, kind splitting, and abstraction and idealization. (Polger and Shapiro 2016) The thesis argues that, ultimately, these arguments hinge on the claim that there is no empirical evidence of multiply realized mental kinds. In response to this claim, the fourth chapter presents novel examples of multiple realization drawn from research in the cognitive neuroscience of language. There are three aspects from the study of language that are presented as examples of multiple realization: (1) language lateralization, (2) reading acquisition, and (3) second/multi- language learning. The analysis of these case studies applies Shoemaker's account of realization as a framework for describing the empirical data as cases of multiple realization. Objections to, and defeaters for, these cases are considered and rejected. The thesis concludes by raising the possibility that mental kinds are realized by mechanisms. This proposal involves drawing connections between traditional questions/views from the philosophy of mind with a current research program from the philosophy of science; the view is particularly important in the philosophy of neuroscience and the philosophy of biology, known as 'new mechanism'. (shrink)