Since the experiments of Saffran et al. (1996a), there has been a great deal of interest in the question of how statistical regularities in the speech stream might be used by infants to begin to identify individual words. In this work, we use computational modeling to explore the effects of different assumptions the learner might make regarding the nature of words – in particular, how these assumptions affect the kinds of words that are segmented from a corpus of transcribed child-directed speech. We develop several models within a Bayesian ideal observer framework, and use them to examine the consequences of assuming either that words are independent units, or units that help to predict other units. We show through empirical and theoretical results that the assumption of independence causes the learner to undersegment the corpus, with many two- and three-word sequences (e.g. what’s that, do you, in the house) misidentified as individual words. In contrast, when the learner assumes that words are predictive, the resulting segmentation is far more accurate. These results indicate that taking context into account is important for a statistical word segmentation strategy to be successful, and raise the possibility that even young infants may be able to exploit more subtle statistical patterns than have usually been considered. 1

Inducing causal relationships from observations is a classic problem in scientific inference, statistics, and machine learning. It is also a central part of human learning, and a task that people perform remarkably well given its notorious difficulties. People can learn causal structure in various settings, from diverse forms of data: observations of the co-occurrence frequencies between causes and effects, interactions between physical objects, or patterns of spatial or temporal coincidence. These different modes of learning are typically thought of as distinct psychological processes and are rarely studied together, but at heart they present the same inductive challenge—identifying the unobservable mechanisms that generate observable relations between variables, objects, or events, given only sparse and limited data. We present a computational-level analysis of this inductive problem and a framework for its solution, which allows us to model all these forms of causal learning in a common language. In this framework, causal induction is the product of domain-general statistical inference guided by domain-specific prior knowledge, in the form of an abstract causal theory. We identify 3 key aspects of abstract prior knowledge—the ontology of entities, properties, and relations that organizes a domain; the plausibility of specific causal relationships; and the functional form of those relationships—and show how they provide the constraints that people need to induce useful causal models from sparse data.

Probabilistic methods are providing new explanatory approaches to fundamental cognitive science questions of how humans structure, process and acquire language. This review examines probabilistic models defined over traditional symbolic structures. Language comprehension and production involve probabilistic inference in such models; and acquisition involves choosing the best model, given innate constraints and linguistic and other input. Probabilistic models can account for the learning and processing of language, while maintaining the sophistication of symbolic models. A recent burgeoning of theoretical developments and online corpus creation has enabled large models to be tested, revealing probabilistic constraints in processing, undermining acquisition arguments based on a perceived poverty

We present a Bayesian model that allows to automatically generate fixations/foveations and that can be suitably exploited for compression purposes. The twofold aim of this work is to investigate how the exploitation of high-level perceptual cues provided by human faces occurring in the video can enhance the compression process without reducing the perceived quality of the video and to validate such assumption with an extensive and principled experimental protocol. To such end, the model integrates top-down and bottom-up cues to choose the fixation point on a video frame: at the highest level, a fixation is driven by prior information and by relevant objects, namely human faces, within the scene; at the same time, local saliency together with novel and abrupt visual events contribute by triggering lower level control. The performance of the resulting video compression system has been evaluated with respect to both the perceived quality of foveated video clips and the compression gain with an extensive evaluation campaign, which has eventually involved 200 subjects.

The brain’s decoding of fast sensory streams is currently impossible to emulate, even approximately, with artificial agents. For example, robust speech recognition is relatively easy for humans but exceptionally difficult for artificial speech recognition systems. In this paper, we propose that recognition can be simplified with an internal model of how sensory input is generated, when formulated in a Bayesian framework. We show that a plausible candidate for an internal or generative model is a hierarchy of ‘stable heteroclinic channels’. This model describes continuous dynamics in the environment as a hierarchy of sequences, where slower sequences cause faster sequences. Under this model, online recognition corresponds to the dynamic decoding of causal sequences, giving a representation of the environment with predictive power on several timescales. We illustrate the ensuing decoding or recognition scheme using synthetic sequences of syllables, where syllables are sequences of phonemes and phonemes are sequences of sound-wave modulations. By presenting anomalous stimuli, we find that the resulting recognition dynamics disclose inference at multiple time scales and are reminiscent of neuronal dynamics seen in the real brain.

In this paper, we propose a computational framework for integrating the physics of motion with the neurobiological basis of perception in order to model and recognize human actions and object activities. The essence, or gist, of an action is intrinsically related to the motion of the scene’s objects. We define the Hamiltonian Energy Signature (HES) and derive the S-Metric to yield a global representation of the motion of the scene’s objects in order to capture the gist of the activity. The HES is a scalar time-series that represents the motion of an object over the course of an activity and the S-Metric is a distance metric which characterizes the global motion of the object, or the entire scene, with a single, scalar value. The neurobiological aspect of activity recognition is handled by casting our analysis within a framework inspired by Neuromorphic Computing (NMC), in which we integrate a Motion Energy model with a Form/Shape model. We employ different Form/Shape representations depending on the video resolution but use our HES and S-Metric for the Motion Energy approach in either case. As the core of our Integration mechanism, we utilize variants of the latest neurobiological models of feature integration and biased competition, which we implement within a Multiple Hypothesis Testing (MHT) framework. Experimental validation of the theory is provided on standard datasets capturing a variety of problem settings: single agent actions (KTH), multi-agent actions, and aerial sequences (VIVID). 1.

Causal learning requires integrating constraints provided by domain-specific theories with domaingeneral statistical learning. In order to investigate the interaction between these factors, the authors presented preschoolers with stories pitting their existing theories against statistical evidence. Each child heard 2 stories in which 2 candidate causes co-occurred with an effect. Evidence was presented in the

Abstract. Recently, Gerd Niestegge developed a new approach to quantum mechanics via conditional probabilities developing the well-known proposal to consider the Lüders-von Neumann measurement as a non-classical extension of probability conditionalization. I will apply his powerful and rigorous approach to the treatment of concepts using a geometrical model of meaning. In this model, instances are treated as vectors of a Hilbert space ℋ. In the present approach there are at least two possibilities to form categories. The first possibility sees categories as a mixture of its instances (described by a density matrix). In the simplest case we get the classical probability theory including the Bayesian formula. The second possibility sees categories formed by a distinctive prototype which is the superposition of the (weighted) instances. The construction of prototypes can be seen as transferring a mixed quantum state into a pure quantum state freezing the probabilistic characteristics of the superposed instances into the structure of the formed prototype. Closely related to the idea of forming concepts by prototypes is the existence of interference effects. Such inference effects are typically found in macroscopic quantum systems and I will discuss them in connection with several puzzles of bounded rationality. The present approach nicely generalizes earlier proposals made by authors such as Diederik Aerts, Andrei Khrennikov, Ricardo Franco, and Jerome Busemeyer. Concluding, I will suggest that an active dialogue between cognitive approaches to logic and semantics and the modern approach of quantum information science is mandatory.

Predicting & parsing language This dissertation asks how prediction might help explain effects of seemingly high-level factors of language processing that occur as early as 100 ms after the presentation of a word. I propose a Sensory Hypothesis to explain such findings, predicting that they reflect mismatches with form-feature predictions derived from contextual predictions for syntactic categories or lexical-semantic representations. Consistent with this hypothesis, the set of experiment presented here are the first to demonstrate that early visual responses to word forms can be influenced by prior linguistic context, affected by both syntactic and lexicalsemantic predictions. An exploratory investigation into the neural correlates of predictive language processing in prestimulus time-windows suggest that the generation of predicted form representations associated with lexical-semantic predictions is initiated in ventromedial prefrontal cortex, flowing downstream via occipito-temporal regions, to (the vicinity of primary) visual cortex. Importantly, the same brain areas were more active for words that violated lexical-semantic expectations than for words that satisfied predictions, but in reverse order. As such, this research might get us a step closer to elucidating the mechanisms by which prediction allows rapid language processing, *Much of this research was conducted in collaboration with Hugh Rabagliati

Abstract not found