Identification and discrimination of handshape in American Sign Language

Lane, Boyes-Braem, and Bellugi (1976) suggested that American Sign Language (ASL) was perceived according to distinctive features, as is the case with speech. They advanced a binary model of distinctive features for one component of ASL signs, handshape. To test the validity of this model, three native users of ASL and three English speakers who knew no ASL participated in two experiments. In the first, subjects identified ASL handshapes obscured by visual noise, and confusion frequencies yielded similarity scores for all possible pairs of handshapes. Pairs that shared more features according to the model produced higher similarity scores for both groups of subjects. In the second experiment, subjects discriminated a subset of all possible handshape pairs in a speeded “same-different” task; discrimination accuracy and reaction times yielded a d’ and a d_s value, respectively, for each pair. Pairs that shared more features according to a slightly revised version of the model produced lower discrimination indices for both groups. While the binary model was supported, a model in which handshape features varied continuously in two dimensions was more consistent with all sets of data. Both models describe hearing and deaf performance equally well, suggesting that linguistic experience does little to alter perception of the visual features germane to handshape identification and discrimination.     

Density versus feature weights as predictors of visual identifications: comment on Appelman and Mayzner

Appleman and Mayzner's application of the distance-density model to confusion data is compared with Keren and Baggen's application of the feature-matching model. In both applications, the distinctive features of two stimuli are predictors of the number of confusion errors. However, the models differ in that the feature-matching model assigns weights to the features and assumes that the shared feature weights also affect the probability of confusion. In contrast, the distance-density model assumes that the number of confusions between two stimuli is affected by the number of stimuli in the entire stimulus set that are similar to the two stimuli (density). The two models are compared in the context of a set of digit identification data.     

Feature lists and confusion matrices

This report compares three feature list sets for capital letters, previously proposed by different investigators, on the ability of each to predict empirical confusion matrices. Toward this end, several variants of assumed information processes in recognition were also compared. The best model incorporated: (1) variable feature retrieval probabilities, (2) a goodness-of-match lower threshold below which guessing determines response, and (3) response bias on guessing trials. This model, when combined with one particular proposed feature list set, produced stress values of less than 9% in comparisons to empirical matrices for each of three different Ss. The feature retrieval probability vectors associated with these minimum-stress predictions were highly correlated ( \(\bar r = .83\) ), suggesting considerable generality of process and feature sets between Ss.     

Application of geometric models to letter recognition: distance and density

This article reviews studies in which a single letter is visually presented under adverse conditions and the subject's task is to identify the letter. The typical results for such studies are (a) certain pairs of letters are more often confused than other pairs of letters; (b) certain letters are more easily recognized than others; and (c) confusion errors for a letter pair are often asymmetric, the number of errors differing depending on which letter of the pair is presented as the stimulus. A geometric model incorporating the properties of distance and spatial density (after Krumhansl) is presented to account for these results. The present application of the distance-density model assumes that each letter is constructed in a typical 5 X 7 dot matrix. Each letter is represented in 35-dimensional space based on its constituent dots. A central idea behind the model, embodied in the property of spatial density, is that an explanation of typical results must take into account the relationship of the entire stimulus set to both the presented letter and the responded letter. Specifically, according to the model, (a) pairs of letters that are close in geometric space are more often confused than pairs of letters that are distant; (b) letters that are in less spatially dense regions are more easily recognized than letters that are in more spatially dense regions; and (c) asymmetric confusion errors result when one member of a letter pair is in a denser region than the other member of the letter pair. The distance-density model is applied to published and unpublished results of the authors as well as published results from two other laboratories. Alternative explanations of the three typical letter recognition results are also considered. The most successful alternative explanations are (a) confusions are an increasing function of the number of dots that two letters share; (b) letters constructed from fewer dots are easier to recognize; and (c) asymmetries arise when one member of a letter pair is more easily recognized, since that letter then has fewer confusion errors to give to the other letter of the pair. The model is discussed in terms of the distinction between template matching and feature analysis. An alternative classification of letter recognition models is proposed based on the global versus local qualities of features and the spatial information associated with each feature. The model is extended to explain reaction time study results. It is suggested that the distance-density model can be used to create optimal letter fonts by minimizing interletter confusions and maximizing letter recognizability.     

Feature binding in visual short term memory: A General Recognition Theory analysis

Creating and maintaining accurate bindings of elementary features (e.g., color and shape) in visual short-term memory (VSTM) is fundamental for veridical perception. How are low-level features bound in memory? The present work harnessed a multivariate model of perception – the General Recognition Theory (GRT) – to unravel the internal representations underlying feature binding in VSTM. On each trial, preview and target colored shapes were presented in succession, appearing in either repeated or altered spatial locations. Participants gave two same/different responses: one with respect to color and one with respect to shape. Converging GRT analyses on the accuracy confusion matrices provided substantial evidence for binding in the form of violations of perceptual independence at the level of the individual stimulus, such that positive correlations were obtained when both features repeated or alternated together, while negative correlations were obtained when one feature repeated and the other alternated. This “cloverleaf” GRT pattern of binding was similar whether the spatial location of the preview and target repeated or altered. The current results are consistent with: (a) the discrete memory “slots” model of VSTM, and (b) the notion that spatial location is not necessary for the formation of “object files.” The GRT approach presented here offers a viable quantitative model for testing various questions regarding feature binding in VSTM.     

Inhibition of feature extraction with multiple instances of the target feature in different orientations

Summary Analyses of alphabetic confusion matrices have produced feature lists defined in terms of confusability of characters thought to contain them. Identical shapes of low confusability (e.g., b, d, p, q) therefore share few or no features. Several researchers have observed mutual inhibition of feature extraction with simultaneously presented confusable shapes. Mutual inhibition observed in the current experiments provides a basis for a definition of features in terms of orientation-independent structural relationships. In the first experiment, four alphanumeric characters were each confused with one other character in this same set; however, when another set was formed by reorienting some of these same shapes to become different characters (e.g., 6 to 9), these confusions disappeared. In the second experiment, characters within each set were exposed parafoveally in pairs to different groups of subjects. If members of a pair were similar in shape (disregarding orientation), identification accuracy was poorer than if they were not. Similar shapes in the same orientation were not signficantly more mutually inhibiting than similar shapes in different orientations.     

Multicomponent theory of perception

Summary In a complete identification experiment the three sides of an equilateral triangle were briefly presented as stimuli either singly or in pairs. The resulting 6×6 confusion matrices obtained with three subjects were analyzed according to Rumelhart's (1970, 1971) multicomponent theory of perception (MCTP) in order to test assumptions about feature extraction and decision processes. In agreement with MCTP the detection of a line segment occurring within a pair was stochastically independent of the detection of the other line segment. Two predictions of MCTP were however not confirmed: different line segments are not detected with equal probability, and the probability of detecting a line segment depends on the presence or absence of others. Several decision assumptions of MCTP were tested. The results are the following: If at most one line segment has been detected, then several responses (the candidate set) are compatible with this sensory state. It was argued that response selection from the candidate set is better described by the choice model (Luce, 1959) than by the matching Bayesian rule assumed by MCTP. For a given sensory state the size of the candidate set appears to vary over trials, whereas MCTP assumes it to be constant. In general, the confusion matrices could be predicted quite accurately by MCTP modified according to the above assumptions. However, more accurate predictions were achieved by assuming holistic perceptual processes as well as single feature extraction.This study was supported by a grant from the Universität Konstanz, Forschungsprojekt 32/72 Visuelle Reize, and the experiment to this study was performed in the laboratory of the Arbeitseinheit Mathematische Psychologie im Fachbereich Psychologie/Soziologie der Universität Konstanz.     

Independent feature processing in the visual system

Summary A recognition experiment with a set of simple visual patterns was performed. The patterns were composed of three binary features: Lines, angles and curve segments. The pattern sets differed with respect to the distance between components and the symmetry of the arrangement. The empirical confusion matrices were analysed assuming two different models: A model assuming independent feature analysis and a model that assumes that one of the features is analysed independently of the other two. The main result was that quite a large distance between features was necessary to achieve independent processing of the features. Moreover, in the asymmetric pattern set the predictions of the independence model were better than in the symmetric pattern set.     

Contrasting effects of feature-based statistics on the categorisation and basic-level identification of visual objects

Conceptual representations are at the heart of our mental lives, involved in every aspect of cognitive functioning. Despite their centrality, a long-standing debate persists as to how the meanings of concepts are represented and processed. Many accounts agree that the meanings of concrete concepts are represented by their individual features, but disagree about the importance of different feature-based variables: some views stress the importance of the information carried by distinctive features in conceptual processing, others the features which are shared over many concepts, and still others the extent to which features co-occur. We suggest that previously disparate theoretical positions and experimental findings can be unified by an account which claims that task demands determine how concepts are processed in addition to the effects of feature distinctiveness and co-occurrence. We tested these predictions in a basic-level naming task which relies on distinctive feature information (Experiment 1) and a domain decision task which relies on shared feature information (Experiment 2). Both used large-scale regression designs with the same visual objects, and mixed-effects models incorporating participant, session, stimulus-related and feature statistic variables to model the performance. We found that concepts with relatively more distinctive and more highly correlated distinctive relative to shared features facilitated basic-level naming latencies, while concepts with relatively more shared and more highly correlated shared relative to distinctive features speeded domain decisions. These findings demonstrate that the feature statistics of distinctiveness (shared vs. distinctive) and correlational strength, as well as the task demands, determine how concept meaning is processed in the conceptual system.     

Features of graded category structure: Generalizing the family resemblance and polymorphous concept models

Many real-world categories contain graded structure: certain category members are rated as more typical or representative of the category than others. Research has shown that this graded structure can be well predicted by the degree of commonality across the feature sets of category members. We demonstrate that two prominent feature-based models of graded structure, the family resemblance (Rosch & Mervis, 1975) and polymorphous concept models (Hampton, 1979), can be generalized via the contrast model (Tversky, 1977) to include both common and distinctive feature information, and apply the models to the prediction of typicality in 11 semantic categories. The results indicate that both types of feature information play a role in the prediction of typicality, with common features weighted more heavily for within-category predictions, and distinctive features weighted more heavily for contrast-category predictions. The same pattern of results was found in additional analyses employing rated goodness and exemplar generation frequency. It is suggested that these findings provide insight into the processes underlying category formation and representation.     

Identification versus discrimination of distinctive features in speech perception

Dichotically presented CV syllables were presented to 24 normal subjects under two contrasting performance requirements: consonant identification and consonant discrimination. In the identification task, subjects made more errors identifying stimuli distinguished by two as opposed to one distinctive feature. Conversely, in the discrimination task, subjects made more errors for stimuli distinguished by one as opposed to two distinctive features. It was proposed that the results of the identification task reflect the degree of information load in short-term memory. The results of the discrimination task, on the other hand, reflect the degree of perceptual similarity between contrasting stimuli. Both results were accounted for within a distinctive feature framework. Analyses of the individual features which comprised one and two feature distinctions demonstrated the perceptual prominence of the feature [voice] in contrast to the two place features [compact] and [grave].     

Replicating distinctive facial features in lineups: identification performance in young versus older adults

Criminal suspects with distinctive facial features, such as tattoos or bruising, may stand out in a police lineup. To prevent suspects from being unfairly identified on the basis of their distinctive feature, the police often manipulate lineup images to ensure that all of the members appear similar. Recent research shows that replicating a distinctive feature across lineup members enhances eyewitness identification performance, relative to removing that feature on the target. In line with this finding, the present study demonstrated that with young adults (n = 60; mean age = 20), replication resulted in more target identifications than did removal in target-present lineups and that replication did not impair performance, relative to removal, in target-absent lineups. Older adults (n = 90; mean age = 74) performed significantly worse than young adults, identifying fewer targets and more foils; moreover, older adults showed a minimal benefit from replication over removal. This pattern is consistent with the associative deficit hypothesis of aging, such that older adults form weaker links between faces and their distinctive features. Although replication did not produce much benefit over removal for older adults, it was not detrimental to their performance. Therefore, the results suggest that replication may not be as beneficial to older adults as it is to young adults and demonstrate a new practical implication of age-related associative deficits in memory.     

A comparison of methods for measuring the interletter similarity between capital letters

Measures of interletter similarity are often required in perception experiments. The most reliable and valid of the available measures appears to be Townsend’s (1971) set of similarity parameters based on the Luce choice model. A simple mechanical measure offered a fairly strong prediction of the Luce choice-model similarity measure, as did a subjective rating measure based on the 10-point visual similarity ratings of eight subjects. By comparison, Gibson et al.’s (1963) matching-confusion matrix faired poorly, as did Gibson’s (1969) distinctive feature analysis based on a letter pair’s number of shared features. Distinctive feature analysis was significantly improved by substituting the feature set proposed by Geyer and DeWald (1973) or by weighting the features optimally via regression analysis. Such analyses suggested that figural curvature may be a particularly important perceptual feature, but in no case did these feature-analytic models predict the Luce measure as well as the mechanical or subjective rating measures.     

Global and Local Feature Distinctiveness Effects in Language Acquisition

Various aspects of semantic features drive early vocabulary development, but less is known about how the global and local structure of the overall semantic feature space influences language acquisition. A feature network of English words was constructed from a large database of adult feature production norms such that edges in the network represented feature distances between words (i.e., Manhattan distances of probability distributions of features elicited for each pair of words). A word's global feature distinctiveness is measured with respect to all other words in the network and a word's local feature distinctiveness is measured relative to words in sub-networks derived from clustering analyses. This paper investigates how feature distinctiveness of individual words at local and global scales of the network influences language acquisition. Regression analyses indicate that global feature distinctiveness was associated with earlier age of acquisition ratings, and was a stronger predictor of age of acquisition than local feature distinctiveness. These results suggest that the global structure of the semantic feature network could play an important role in language acquisition, whereby globally distinctive concepts help to structure vocabulary development over the lifespan.     

Common and distinctive features in stimulus similarity: A modified version of the contrast model

Featural representations of similarity data assume that people represent stimuli in terms of a set of discrete properties. In this article, we consider the differences in featural representations that arise from making four different assumptions about how similarity is measured. Three of these similarity models— the common features model, the distinctive features model, and Tversky’s seminal contrast model—have been considered previously. The other model is new and modifies the contrast model by assuming that each individual feature only ever acts as a common or distinctive feature. Each of the four models is tested on previously examined similarity data, relating to kinship terms, and on a new data set, relating to faces. In fitting the models, we have used the geometric complexity criterion to balance the competing demands of data-fit and model complexity. The results show that both common and distinctive features are important for stimulus representation, and we argue that the modified contrast model combines these two components in a more effective and interpretable way than Tversky’s original formulation.     

S-multiplicativity of a stochastic matrix and applications to visual identification

A stochastic matrix is called S-multiplicative if the conditional probabilities in a specific subset S of its cells can be factorized into the product of two functions depending on the rows and columns, respectively. The concept of S-multiplicativity is related to Goodman's concept of quasi-independence and generalizes Falmagne's concept of a multiplicative confusion matrix, where S consists of the cells lying outside the main diagonal ($\text{D}$-multiplicativity). In the present study, the notion of S-multiplicativity is applied to the analysis of visual identification performance. The confusion matrices obtained from three identification experiments are tested for predictions of S-multiplicativity derived from the multicomponent theory of perception. If the stimuli do not overlap, i.e., different stimuli have no features in common, then the theory predicts a $\text{D}$-multiplicative confusion matrix. Alternatively, in the case of overlapping stimuli the theory predicts S-multiplicativity with respect to a more severely restricted subset S. These predictions are confirmed by the data.     

An alternative for judging confusability of visual letters.

Numerical values for shared distinctive features were derived from Gibson's (1) feature analysis of the 26 uppercase alphabet letters. Due to the lack of agreement among the empirical matrices, it is proposed that this more general table is a useful and practical approach for judging confusability of visual letters for uppercase items. Reaction time data from a Posner-type of letter-matching task support the effectiveness of the table for judging visual confusability among uppercase letters.     

Eye-tracking the cancellation and focus model for preference judgments

By means of a relatively new eye-tracking method that allows for a test situation much closer to reality, we recorded and examined gaze time and fixation number within the cancellation and focus paradigm, a feature-matching model for preference judgments between two alternatives. In line with the cancellation and focus model we found that when subjects encountered the second option in each pair, shared features were canceled out and thus given less consideration whereas unique features were focused on more. We also investigated the role of feature attractiveness as a second important factor in preference judgments and found a U-shaped relationship between attractiveness and visual consideration intensity; that is, attractive and unattractive features received more attention than did those of intermediate attractiveness. Finally, we tested the ability of two models, Franklin’s rule and the WReSt (Weighted Recalled Stepwise Comparing) heuristic, to predict the preference ratings.     

Connectionist modelling of the separable processing of consonants and vowels

describe two aphasic patients, with impaired processing of vowels and consonants, respectively. The impairments could not be captured according to the sonority hierarchy or in terms of a feature level analysis. Caramazza et al. claim that this dissociation demonstrates separate representation of the categories of vowels and consonants in speech processing. We present two connectionist models of the management of phonological representations. The models spontaneously develop separable processing of vowels and consonants. The models have two hidden layers and are given as input vowels and consonants represented in terms of their phonological distinctive features. The first model is presented with feature bundles one at a time and the hidden layers have to combine their output to reproduce a unified copy of the feature bundle. In the second model a "fine-coded" layer receives information about feature bundles in isolation, and a "coarse-coded" layer receives information about each feature bundle in the context of the prior and subsequent feature bundle. Coarse-coding facilitated processing of vowels and fine-coding processing of consonants. These models show that separable processing of vowels and consonants is an emergent effect of modular processors operating on feature-based representations. We argue that it is not necessary to postulate an independent level of representation for the consonant/vowel distinction, separate from phonological distinctive features.     

Clarifying the nature of the distinctiveness by domain interaction in conceptual structure: comment on Cree, McNorgan, and McRae (2006)

The conceptual structure account of semantic memory (CSA; L. K. Tyler & H. E. Moss, 2001) claims that feature correlation (the degree to which features co-occur) and feature distinctiveness (the number of concepts in which a feature occurs) interact with domains of knowledge (e.g., living vs. nonliving) such that the distinctive features of nonliving things are more highly correlated than the distinctive features of living things. Evidence for (B. Randall, H. E. Moss, J. M. Rodd, M. Greer, & L. K. Tyler, 2004) and against this claim (G. S. Cree, C. McNorgan, & K. McRae, 2006) has been reported. This comment outlines the CSA, discusses Cree et al.'s (2006) critiques of the Randall et al. (2004) experiments and the CSA, and reports new analyses of property norm and behavioral data, which replicate the results reported by Randall et al. (2004).