Pooling of elementary motion, colour, and orientation signals into global perception
Date
2012-01-17
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Enamik senistest uurimustest on eksplitsiitselt või implitsiitselt eeldanud, et globaalse liikumismulje kujunemisse on kaasatud kõik stiimulis esindatud elementaarsignaalid. Väitekirjas esitatud tulemused näitavad, et see eeldus ei pea alati paika. Töös esitatakse konkreetne ja läbipaistev Bernoulli urnimudelitel põhinev meetod reaalse vaatleja efektiivsuse hindamiseks ideaalse vaatleja suhtes, mis ühtlasi võimaldab võrrelda sooritust erinevate ülesannete lõikes.
Olukordades, kus stiimulid koosnevad diskreetsetest ja kvantifitseeritavatest elementidest, on Bernoulli mudelil traditsioonilise Thurstone’i mudeli ees mitmed väga selged eelised. Klassikalise mudeli üheks eelduseks on sisemiste seisundite kontiinum, millele projitseeritakse väliste atribuutide stohhastilised representatsioonid. Erinevalt klassikalisest Thurstone’i eristusprotsesside mudelist asetub Bernoulli mudelite puhul juhuslikkuse komponent elementide alamhulga valikus kuvatud elementide koguhulgast. Empiirilise psühhomeetrilise funktsiooni kirjelduse tasandil on Bernoulli ja Thurstone’i mudelid formaalselt ekvivalentsed ja sellest lähtuvalt puuduvad esialgu argumendid nende adekvaatsuse ja bioloogilise tõepära võrdlevaks hindamiseks. Samas on Bernoulli mudel matemaatiliselt minimalistlikum ning lihtsamini falsifitseeritav.
Järeldati, et kui suhtelise arvukuse hindamise täpsust kajastav empiiriline psühhomeetriline funktsioon on paremini kirjeldatav binomiaalse kui hüpergeomeetrilise vastusmudeliga, viitab see üheselt, et teatud osa stiimulelementidest inspekteeritakse korduvalt, mis on võimalik ainult kahel või enamal ajahetkel (samas on välistatud olukord, kus paralleelne töötlusmudel imiteeriks seriaalset). Pakutud meetod võimaldab uurida mentaalse arhitektuuri üht seni vähest tähelepanu pälvinud aspekti – elementide korduvat loendamist – ning selle rakendamine on viinud kaalukate tõenditeni, mis viitavad, et teatud hulgal vaatluskordadest segistab inimene juba loendatud ning veel loendamata elemendid, võttes üht ja sama elementi arvesse korduvalt, mis saab sündida vaid seriaalselt.
Many previous studies have presumed, explicitly or tacitly, that in forming of the global motion percept, all elementary motion signals present in the stimulus are pooled together. As the presented results indicate this is not always the case. In the thesis, an approach of estimating observer’s efficiency in a straightforward fashion that is also highly comparable across different tasks and based on Bernoulli’s urn model is proposed. In all situations where stimuli consist of discrete quantifiable elements, the Bernoulli’s urn model has obvious advantages before the classical Thurstonian model which requires a continuum of internal states and a fuzzy projection of external attributes onto it. Alternatively to the classical Thurstonian model of discriminal processes, in Bernoullian models the randomness lies not in the internal representations of the stimuli, but in sampling of the elements out from the total number of elements in the display. It was shown that the Bernoullian model is formally equivalent to the Thurstonian discrimi¬nation model in terms of the description of empirically obtained psychometric function. Therefore, for the time being, it is impossible to decide which of the two models provides a biologically more adequate description. Nevertheless, the Bernoullian model is simpler and more easily falsi¬fiable. It was concluded that if an empirically determined psycho¬metric function for numerical discriminations between two sets of items is better described by binomial rather than hypergeometric response model, it would provide evidence (that would not suffer from potential model mimicking) that some of these elements are inspected twice or more times. This new method for identifying one neglected aspect of the mental architecture – avoiding repeated tagging – provided a strong proof that in a considerable number of trials human observer counted the same element twice or more times which can only be done at different time moments, that is serially.
Many previous studies have presumed, explicitly or tacitly, that in forming of the global motion percept, all elementary motion signals present in the stimulus are pooled together. As the presented results indicate this is not always the case. In the thesis, an approach of estimating observer’s efficiency in a straightforward fashion that is also highly comparable across different tasks and based on Bernoulli’s urn model is proposed. In all situations where stimuli consist of discrete quantifiable elements, the Bernoulli’s urn model has obvious advantages before the classical Thurstonian model which requires a continuum of internal states and a fuzzy projection of external attributes onto it. Alternatively to the classical Thurstonian model of discriminal processes, in Bernoullian models the randomness lies not in the internal representations of the stimuli, but in sampling of the elements out from the total number of elements in the display. It was shown that the Bernoullian model is formally equivalent to the Thurstonian discrimi¬nation model in terms of the description of empirically obtained psychometric function. Therefore, for the time being, it is impossible to decide which of the two models provides a biologically more adequate description. Nevertheless, the Bernoullian model is simpler and more easily falsi¬fiable. It was concluded that if an empirically determined psycho¬metric function for numerical discriminations between two sets of items is better described by binomial rather than hypergeometric response model, it would provide evidence (that would not suffer from potential model mimicking) that some of these elements are inspected twice or more times. This new method for identifying one neglected aspect of the mental architecture – avoiding repeated tagging – provided a strong proof that in a considerable number of trials human observer counted the same element twice or more times which can only be done at different time moments, that is serially.
Description
Väitekirja elektrooniline versioon ei sisalda publikatsioone.
Keywords
psühholoogia, taju, psühhomeetria, psychology, perception, psychometrics