Unsupervised Statistical Learning of Higher-Order Spatial Structures From Visual Scenes

Three experiments investigated the ability of human observers to extract the joint and conditional probabilities of shape co-occurrences during passive viewing of complex visual scenes. Results indicated that statistical learning of shape conjunctions was both rapid and automatic, as subjects were not instructed to attend to any particular features of the displays. Moreover, in addition to single-shape frequency, subjects acquired in parallel several different higher-order aspects of the statistical structure of the displays, including absolute shape-position relations in an array, shape-pair arrangements independent of position, and conditional probabilities of shape co-occurrences. Unsupervised learning of these higher-order statistics provides support for Barlow's theory of visual recognition, which posits that detecting "suspicious coincidences" of elements during recognition is a necessary prerequisite for efficient learning of new visual features.     

Across space and time: infants learn from backward and forward visual statistics

This study investigates whether infants are sensitive to backward and forward transitional probabilities within temporal and spatial visual streams. Two groups of 8‐month‐old infants were familiarized with an artificial grammar of shapes, comprising backward and forward base pairs (i.e. two shapes linked by strong backward or forward transitional probability) and part‐pairs (i.e. two shapes with weak transitional probabilities in both directions). One group viewed the continuous visual stream as a temporal sequence, while the other group viewed the same stream as a spatial array. Following familiarization, infants looked longer at test trials containing part‐pairs than base pairs, although they had appeared with equal frequency during familiarization. This pattern of looking time was evident for both forward and backward pairs, in both the temporal and spatial conditions. Further, differences in looking time to part‐pairs that were consistent or inconsistent with the predictive direction of the base pairs (forward or backward) indicated that infants were indeed sensitive to direction when presented with temporal sequences, but not when presented with spatial arrays. These results suggest that visual statistical learning is flexible in infancy and depends on the nature of visual input.     

基于名人面孔视觉特征和语义信息的视觉统计学习

视觉统计学习是指个体依据视觉刺激之间的转接概率来掌握统计规律的过程。本研究通过5个实验探讨了个体基于名人面孔视觉特征和语义信息进行视觉统计学习的加工机制。每个实验均包括熟悉(学习)和测试两个阶段：在熟悉阶段, 让被试观看名人面孔并完成重复图片探测的无关任务; 在测试阶段, 让被试进行二选一迫选任务。其中, 实验1和2分别考察基于名人面孔视觉特征和语义信息的视觉统计学习效果; 实验3分别考察基于名人面孔视觉特征和语义信息视觉进行统计学习的精确性; 实验4进一步考察基于名人面孔视觉特征和语义信息进行视觉统计学习的时间特征; 实验5验证基于名人面孔视觉特征的视觉统计学习具有面孔特异性。结果表明：个体能同时基于名人面孔视觉特征和语义信息进行精确的视觉统计学习; 基于正立名人面孔的视觉统计学习效果显著高于基于倒置名人面孔的视觉统计学习效果; 虽然基于视觉特征和语义信息的统计加工都具有一致的精确性, 但后者需要更多的加工时间。这提示：基于名人面孔视觉特征的视觉统计学习具有面孔特异性, 个体基于名人面孔视觉特征和语义信息的视觉统计学习过程是分离的, 统计运算发生于面孔特征加工完成之后。     

The influence of object size and surface shape on shape constancy from stereo

The failure of shape constancy from stereoscopic information is widely reported in the literature. In this study we investigate how shape constancy is influenced by the size of the object and by the shape of the object's surface. Participants performed a shape-judgment task on objects of five sizes with three different surface shapes. The shapes used were: a frontoparallel rectangle, a triangular ridge surface, and a cylindrical surface, all of which contained the same maximum depth information, but different variations in depth across the surface. The results showed that, generally, small objects appear stretched and large objects appear squashed along the depth dimension. We also found a larger variance in shape judgments for rectangular stimuli than for cylindrical and ridge-shaped stimuli, suggesting that, when performing shape judgments with cylindrical and ridge-shaped stimuli, observers rely on a higher-order shape representation.     

How modality specific is the iambic-trochaic law? Evidence from vision

The iambic-trochaic law has been proposed to account for the grouping of auditory stimuli: Sequences of sounds that differ only in duration are grouped as iambs (i.e., the most prominent element marks the end of a sequence of sounds), and sequences that differ only in pitch or intensity are grouped as trochees (i.e., the most prominent element marks the beginning of a sequence). In 3 experiments, comprising a familiarization and a test phase, we investigated whether a similar grouping principle is also present in the visual modality. During familiarization, sequences of visual stimuli were repeatedly presented to participants, who were asked to memorize their order of presentation. In the test phase, participants were better at remembering fragments of the familiarization sequences that were consistent with the iambic-trochaic law. Thus, they were better at remembering fragments that had the element with longer duration in final position (iambs) and fragments that had the element with either higher temporal frequency or higher intensity in initial position (trochees), as compared with fragments that were inconsistent with the iambic-trochaic law or that never occurred during familiarization.     

Visual statistical learning at basic and subordinate category levels in real-world images

Visual statistical learning (VSL) has been proposed as a powerful mechanism underlying the striking ability of human observers to handle complex visual environments. Previous studies have shown that VSL can occur when statistical information is embedded at multiple levels of abstraction, such as at semantically different category levels. In the present study, we further examined whether statistical regularities at a basic category level (e.g., a regular sequence of a bird, then a car, and then a dog) could influence the ability to extract statistical regularities at the subordinate level (e.g., a regular sequence of a parrot, then a sports car, and then an Eskimo dog). In the familiarization phase, participants were exposed to a stream of real-world images whose semantic categories had temporal regularities. Importantly, the temporal regularities existed at both the basic and subordinate levels, or the regularities existed at only the subordinate level, depending on the experimental condition. After completing the familiarization, participants performed a surprise two-alternative forced choice (2AFC) task for a familiarity judgment between two triplets in which the temporal regularities were either preserved or not preserved. Our results showed that the existence of statistical regularities at the basic level did not influence VSL at the subordinate level. The subsequent experiments showed these results consistently even when the basic-level categories had to be explicitly recognized and when the stimuli were not easily categorized at their subordinate level. Our results suggest that VSL is constrained to learn a particular level of patterns when patterns are presented across multiple levels.     

Role of Attention and Perceptual Grouping in Visual Statistical Learning

Statistical learning has been widely proposed as a mechanism by which observers learn to decompose complex sensory scenes. To determine how robust statistical learning is, we investigated the impact of attention and perceptual grouping on statistical learning of visual shapes. Observers were presented with stimuli containing two shapes that were either connected by a bar or unconnected. When observers were required to attend to both locations at which shapes were presented, the degree of statistical learning was unaffected by whether the shapes were connected or not. However, when observers were required to attend to just one of the shapes' locations, statistical learning was observed only when the shapes were connected. These results demonstrate that visual statistical learning is not just a passive process. It can be modulated by both attention and connectedness, and in natural scenes these factors may constrain the role of stimulus statistics in learning.     

Multidimensional visual statistical learning

Recent studies of visual statistical learning (VSL) have demonstrated that statistical regularities in sequences of visual stimuli can be automatically extracted, even without intent or awareness. Despite much work on this topic, however, several fundamental questions remain about the nature of VSL. In particular, previous experiments have not explored the underlying units over which VSL operates. In a sequence of colored shapes, for example, does VSL operate over each feature dimension independently, or over multidimensional objects in which color and shape are bound together? The studies reported here demonstrate that VSL can be both object-based and feature-based, in systematic ways based on how different feature dimensions covary. For example, when each shape covaried perfectly with a particular color, VSL was object-based: Observers expressed robust VSL for colored-shape sub-sequences at test but failed when the test items consisted of monochromatic shapes or color patches. When shape and color pairs were partially decoupled during learning, however, VSL operated over features: Observers expressed robust VSL when the feature dimensions were tested separately. These results suggest that VSL is object-based, but that sensitivity to feature correlations in multidimensional sequences (possibly another form of VSL) may in turn help define what counts as an object.     

Hearing what the eyes see: auditory encoding of visual temporal sequences

When the senses deliver conflicting information, vision dominates spatial processing, and audition dominates temporal processing. We asked whether this sensory specialization results in cross-modal encoding of unisensory input into the task-appropriate modality. Specifically, we investigated whether visually portrayed temporal structure receives automatic, obligatory encoding in the auditory domain. In three experiments, observers judged whether the changes in two successive visual sequences followed the same or different rhythms. We assessed temporal representations by measuring the extent to which both task-irrelevant auditory information and task-irrelevant visual information interfered with rhythm discrimination. Incongruent auditory information significantly disrupted task performance, particularly when presented during encoding; by contrast, varying the nature of the rhythm-depicting visual changes had minimal impact on performance. Evidently, the perceptual system automatically and obligatorily abstracts temporal structure from its visual form and represents this structure using an auditory code, resulting in the experience of "hearing visual rhythms."     

The role of sequence order in determining view canonicality for novel wire-frame objects

Objects are best recognized from so-called “canonical” views. The characteristics of canonical views of arbitrary objects have been qualitatively described using a variety of different criteria, but little is known regarding how these views might be acquired during object learning. We address this issue, in part, by examining the role of object motion in the selection of preferred views of novel objects. Specifically, we adopt a modeling approach to investigate whether or not the sequence of views seen during initial exposure to an object contributes to observers’ preferences for particular images in the sequence. In two experiments, we exposed observers to short sequences depicting rigidly rotating novel objects and subsequently collected subjective ratings of view canonicality (Experiment 1) and recall rates for individual views (Experiment 2). Given these two operational definitions of view canonicality, we attempted to fit both sets of behavioral data with a computational model incorporating 3-D shape information (object foreshortening), as well as information relevant to the temporal order of views presented during training (the rate of change for object foreshortening). Both sets of ratings were reasonably well predicted using only 3-D shape; the inclusion of terms that capture sequence order improved model performance significantly.     

Perceiving parts and shapes from concave surfaces

“A hole is nothing at all, but it can break your neck.” In a similar fashion to the danger illustrated by this folk paradox, concave regions pose difficulties to theories of visual shape perception. We can readily identify their shapes, but according to principles of how observers determine part boundaries, concavities in a planar surface should have very different figural shapes from the ones that we perceive. In three experiments, we tested the hypothesis that observers perceive local image features differently in simulated 3-D concave and convex regions but use them to arrive at similar shape percepts. Stimuli were shape-from-shading images containing regions that appeared either concave or convex in depth, depending on their orientation in the picture plane. The results show that concavities did not benefit from the same global object-based attention or holistic shape encoding as convexities and that the participants relied on separable spatial dimensions to judge figural shape in concavities. Concavities may exploit a secondary process for shape perception that allows regions composed of perceptually independent features to ultimately be perceived as gestalts.     

The metamorphosis of the statistical segmentation output: Lexicalization during artificial language learning

This study combined artificial language learning (ALL) with conventional experimental techniques to test whether statistical speech segmentation outputs are integrated into adult listeners’ mental lexicon. Lexicalization was assessed through inhibitory effects of novel neighbors (created by the parsing process) on auditory lexical decisions to real words. Both immediately after familiarization and post-one week, ALL outputs were lexicalized only when the cues available during familiarization (transitional probabilities and wordlikeness) suggested the same parsing (Experiments 1 and 3). No lexicalization effect occurred with incongruent cues (Experiments 2 and 4). Yet, ALL differed from chance, suggesting a dissociation between item knowledge and lexicalization. Similarly contrasted results were found when frequency of occurrence of the stimuli was equated during familiarization (Experiments 3 and 4). Our findings thus indicate that ALL outputs may be lexicalized as far as the segmentation cues are congruent, and that this process cannot be accounted for by raw frequency.     

Superordinate shape classification using natural shape statistics

This paper investigates the classification of shapes into broad natural categories such as animal or leaf. We asked whether such coarse classifications can be achieved by a simple statistical classification of the shape skeleton. We surveyed databases of natural shapes, extracting shape skeletons and tabulating their parameters within each class, seeking shape statistics that effectively discriminated the classes. We conducted two experiments in which human subjects were asked to classify novel shapes into the same natural classes. We compared subjects’ classifications to those of a naive Bayesian classifier based on the natural shape statistics, and found good agreement. We conclude that human superordinate shape classifications can be well understood as involving a simple statistical classification of the shape skeleton that has been “tuned” to the natural statistics of shape.     

Rapid acquisition of phonological alternations by infants

We explore whether infants can learn novel phonological alternations on the basis of distributional information. In Experiment 1, two groups of 12-month-old infants were familiarized with artificial languages whose distributional properties exhibited either stop or fricative voicing alternations. At test, infants in the two exposure groups had different preferences for novel sequences involving voiced and voiceless stops and fricatives, suggesting that each group had internalized a different familiarization alternation. In Experiment 2, 8.5-month-olds exhibited the same patterns of preference. In Experiments 3 and 4, we investigated whether infants' preferences were driven solely by preferences for sequences of high transitional probability. Although 8.5-month-olds in Experiment 3 were sensitive to the relative probabilities of sequences in the familiarization stimuli, only 12-month-olds in Experiment 4 showed evidence of having grouped alternating segments into a single functional category. Taken together, these results suggest a developmental trajectory for the acquisition of phonological alternations using distributional cues in the input.     

An interaction between prosody and statistics in the segmentation of fluent speech

Sensitivity to prosodic cues might be used to constrain lexical search. Indeed, the prosodic organization of speech is such that words are invariably aligned with phrasal prosodic edges, providing a cue to segmentation. In this paper we devise an experimental paradigm that allows us to investigate the interaction between statistical and prosodic cues to extract words from a speech stream. We provide evidence that statistics over the syllables are computed independently of prosody. However, we also show that trisyllabic sequences with high transition probabilities that straddle two prosodic constituents appear not to be recognized. Taken together, our findings suggest that prosody acts as a filter, suppressing possible word-like sequences that span prosodic constituents.     

Automatic gaze to the nose region cannot be inhibited during observation of facial expression in Eastern observers

Humans can extract a great deal of information about others very quickly. This is partly because the face automatically captures observers’ attention. Specifically, the eyes can attract overt attention. Although it has been reported that not only the eyes but also the nose can capture initial oculomotor movement in Eastern observers, its generalizability remains unknown. In this study, we applied the “don’t look” paradigm wherein participants are asked not to fixate on a specific facial region (i.e., eyes, nose, and mouth) during an emotion recognition task with upright (Experiment 1) and inverted (Experiment 2) faces. In both experiments, we found that participants were less able to inhibit the initial part of their fixations to the nose, which can be interpreted as the nose automatically capturing attention. Along with previous studies, our overt attention tends to be attracted by a part of the face, which is the nose region in Easterner observers.     

Systematic angular biases in the representation of visual space

Representing spatial information is one of our most foundational abilities. Yet in the present work we find that even the simplest possible spatial tasks reveal surprising, systematic misrepresentations of space—such as biases wherein objects are perceived and remembered as being nearer to the centers of their surrounding quadrants. We employed both a placement task (in which observers see two differently sized shapes, one of which has a dot in it, and then must place a second dot in the other shape so that their relative locations are equated) and a matching task (in which observers see two dots, each inside a separate shape, and must simply report whether their relative locations are matched). Some of the resulting biases were shape specific. For example, when dots appeared in a triangle during the placement task, the dots placed by observers were biased away from certain parts of the symmetry axes. But other systematic biases were not shape specific, and seemed instead to reflect differences in the grain of resolution for different regions of space. For example, with both a circle and even a shapeless configuration (with only a central landmark) in the matching task, observers were better at discriminating angular differences (when a dot changed positions around the circle, as opposed to inward/outward changes) in cardinal versus oblique sectors. These data reveal a powerful angular spatial bias, and highlight how the resolution of spatial representation differs for different regions and dimensions of space itself.

     

Haptic curvature comparison of convex and concave shapes

A sculpture and the mould in which it was formed are typical examples of objects with an identical, but opponent, surface shape: each convex (ie outward pointing) surface part of a sculpture has a concave counterpart in the mould. The question arises whether the object features of opponent shapes can be compared by touch. Therefore, we investigated whether human observers were able to discriminate the curvatures of convex and concave shapes, irrespective of whether the shape was convex or concave. Using a 2AFC procedure, subjects had to compare the curvature of a convex shape to the curvature of a concave shape. In addition, results were also obtained for congruent shapes, when the curvature of either only convex shapes or only concave shapes had to be compared. Psychometric curves were fitted to the data to obtain threshold and bias results. When subjects explored the stimuli with a single index finger, significantly higher thresholds were obtained for the opponent shapes than for the congruent shapes. However, when the stimuli were touched by two index fingers, one finger per surface, we found similar thresholds. Systematic biases were found when the curvature of opponent shapes was compared: the curvature of a more curved convex surface was judged equal to the curvature of a less curved concave surface. We conclude that human observers had the ability to compare the curvature of shapes with an opposite direction, but that their performance decreased when they sensed the opponent surfaces with the same finger. Moreover, they systematically underestimated the curvature of convex shapes compared to the curvature of concave shapes.     

Understanding the structural determinants of object confusion in memory: an assessment of psychophysical approaches to estimating visual similarity

Computer-generated shapes varying on visual dimensions such as curvature, tapering, and thickness have been used to investigate identification deficits in the category-specific visual agnosia (CSVA) Patient E.L.M.. However, whether the implemented variations on each of these dimensions were perceived by novice observers as "similar amounts of change" is unknown. To estimate distance in psychophysical shape space, sets of shapes were developed using two different scaling methods--an objective method based on visual search, and a subjective method based on judgments of similarity--and a third approach that did not involve scaling. How well each method estimated psychophysical shape space was assessed by measuring the confusions within memory among the shapes. The results suggested that, although neither of the approaches perfectly reflected psychophysical shape space, subjective scaling was a better estimator of distance in psychophysical shape space than were other approaches. The number of confusions produced on each set of shapes was used to develop a new set of shapes that accurately estimated distance in psychophysical shape space. These results suggest that a combination of approaches is preferable in order to accurately estimate distance in psychophysical shape space.     

Modulation of brain activity during phonological familiarization

We measured brain activity in 12 adults for the repetition of auditorily presented words and nonwords, before and after repeated exposure to their phonological form. The nonword phoneme combinations were either of high (HF) or low (LF) phonotactic frequency. After familiarization, we observed, for both word and nonword conditions, decreased activation in the left posterior superior temporal gyrus, in the bilateral temporal pole and middle temporal gyri. At the same time, interaction analysis showed that the magnitude of decrease of activity in bilateral posterior temporal lobe was significantly smaller for LF nonwords, relative to words and HF nonwords. Decrease of activity in this area also correlated with the size of behavioral familiarization effects for LF nonwords. The results show that the posterior superior temporal gyrus plays a fundamental role during phonological learning. Its relationship to sublexical and lexical phonological processing as well as to phonological short-term memory is discussed.