Constructivist approach

  • View
    37

  • Download
    0

Embed Size (px)

DESCRIPTION

Constructivist approach. CSCTR Session 11 Dana Retov á. Constructivist approach. Start bottom-up Create cognition based on sensori -motor interaction Cohen et al. (1996) – Building a baby Cohen & Lederle (1998) - Dynamic maps as representations of verbs. Building a baby. - PowerPoint PPT Presentation

Text of Constructivist approach

Semiotics and Enactment

Constructivist approachCSCTR Session 11Dana Retov1Start bottom-upCreate cognition based on sensori-motor interaction

Cohen et al. (1996) Building a babyCohen & Lederle (1998) - Dynamic maps as representations of verbsConstructivist approachagent acquiring conceptual knowledge by sensorimotor interaction with its environmentUses notion of image-schemasGoalProduce conceptual structures from sensorimotor interactionsShow that very little prior structure is sufficient do to this (nature vs. nurture)

Building a babyPattern detectors or filters that map sensory streams onto redescriptions or partial representations.E.g .Image schemas

ANIMATE MOTIONANIMATE MOTIONWhen we see a cat walk across the room our ANIMATE MOTION image-schema produces a partial representation of the sceneBut when we see a car zoom past, our schema stays quiet4Simulated agentLives in simulated environment BabyWorldImplements Neos sensations, mental representations, mental and physical activities nad the behavior of objects.Neo implements everything that Neo doesLearning, moving, mouthing, looking, crying,StreamsWorld represents Neos environment, implements events that happen around Neo and in response to Neos actionsNeo senses its environment through a collection of streams in discrete time steps.

NEOStreamsAffect, pain, hunger, somatic, haptic, visual, auditory,Simple and probabilisticOnly random actionsE.g. If Neos eyes alight on a rattle then Neo will grasp the rattle with some probabilityAt the beginning experience has no apparent structureBabyWorldChanges in token values: Tokens in streams are augmented by noticing when they change value. Scopes: Neo finds pairs of correlated streams called scopes. Base fluents: Neo finds common token-value pairs within scopes. Context fluents: Neo finds base fluents that tend to follow each other in time. Chains: These temporal dependencies are combined into temporal chains, which represent activities. Chains are used for activity-based categorization.5 levels of redescription of Neos experienceEach level or redescription produces an intermediate representationRepresentations are interned in Neos memory when they have accrued sufficient evidence from Neos experience.Changes in token values : when the value of a stream changes (e.g. he looks elsewhere, the object moves, the sound changes)Scope: whenever there is red, there is also rattle-shapeFluent: Fluents represent things that dont change, or that change in highly regular, predictable ways. (e.g. sound made by a rattle, its shape, color etc.).The sensory experience is implemented as tokens for red and rattle-shaped in the appropriate streams, whereas the cognitive experience of a red rattle involves activating a fluent that represents the red rattle.Base fluents: coocuring tokens within scopes. Simultaneous start events and stop events are evidence that a single object or a single activity is making its presence felt in two streamsContext fluents: Neo compares the new values to the base fluents it has learned, and if it finds a match, it activates the asscociated base fluents. It activates a context fluent whenever one of its component fluents is activated. As soon as a fluent is activated, its level of activation begins to decline and after a perid of time it becomes inactive even if the sensory events that activated it are still present a form of habituationContext chains: Neo aggregates context fluents into chains7Agent was able to learn basic conceptsIt learned base fluents corresponding to the shape and color of most objects in its envirnomentIt learned activitiesGrasping, mouthing, movintgits arm, seeing its arm moveResults

Each level or redescription produces an intermediate representationRepresentations are interned in Neos memory when they have accrued sufficient evidence from Neos expreirence.Changes in token values : when the value of a stream changes (e.g. he looks elswhere, the object moves, the sound changes)Scope: whenever there is red, there is also rattle-shapeFluent: Fluents represent things that dont change, or that change in highly regular, predictable ways. (e.g. sound made by a rattle, its shape, color etc.).The sensory experience is implemented as tokens for red and rattle-shaped in the appropriate streams, whereas the cognitive experience of a red rattle involves activating a fluent that represents the red rattle.Base fluents: coocuring tokens within scopec. Simultaneous start events and stop events are evidence that a single object or a single activity is making its presence felt in two streamsContext fluents: Neo compares the new values to the base fluents it has learned, and if it finds a match, it activates the asscociated base fluents. It activates a context fluent whenever one of its component fluents is activated. As soon as a fluent is activated, its level of activation begins to decline and after a perid of time it becomes inactive even if the sensory events that activated it are still present a form of habituationContext chains: Neo aggregates context fluents into chains

9Context fluents

Context fluents: Neo compares the new values to the base fluents it has learned, and if it finds a match, it activates the asscociated base fluents. It activates a context fluent whenever one of its component fluents is activated. As soon as a fluent is activated, its level of activation begins to decline and after a period of time it becomes inactive even if the sensory events that activated it are still present a form of habituationContext chains: Neo aggregates context fluents into chains

10Context chains

Context fluents: Neo compares the new values to the base fluents it has learned, and if it finds a match, it activates the asscociated base fluents. It activates a context fluent whenever one of its component fluents is activated. As soon as a fluent is activated, its level of activation begins to decline and after a perid of time it becomes inactive even if the sensory events that activated it are still present a form of habituationContext chains: Neo aggregates context fluents into chains

The only difference between these chains is the object thatNeo grabs and mouths: in the first case it is wooden, in thesecond, plastic. We may form a class of things that Neo cangrab and mouth. The chains dont say exactly which objectsare in the class, but we know they are either wood or plastic,and they are graspable, and mouthable.GRASPABLE and MOUTHABLE are interactional properties(Johnson, 1987) that characterize Neos activities in its environment.Unlike TEXTUREwood or plasticthey are in asense subjective: Whats graspable by one agent isnt necessarilygraspable by another. Whereas TEXTURE is an inherentproperty of an object, GRASPABLE is a property of the objectand the agent who may try to grasp it. Interactional propertieslike GRASPABLE are the basis for categories in Lakoff andJohnsons theory of categorization (Johnson, 1987; Lakoff,1984; Lakoff and Johnson, 1980) and also in Mandlers theoryof conceptual development (Mandler, 1992). However,we believe categories are best defined in terms of activities,and the attractiveness of interactional features is due to themdescribing activities better than objective features such as texture(Cohen, Oates and Atkin, 1996).11Classes of objects are differentiated by how we interact with them,Concepts are abstractions over those classesMeaning of concepts are in large part predictive models of how interactions with objects will unfold.In the interactionist view, category distinctions are based on activityInteractionist account

For months, plastic frogs andspoons were functionally indistinguishable to Allegra: She would grasp either, put it in hermouth, and chew. The fact that we consider the frog a toy, and a spoon a utensil doesntmatter to her. These are adult categories, not infant categories. On the interactionistaccount, only when Allegra uses the spoon to eat food will she differentiate it from thefrog, and only then will she form a category that resembles in its membership those itemswe adults call utensils.So much for categories, but what about concepts and meaning? Here I want to point outthat except for formal, mathematical objects, many things perhaps most are defined interms of what we do with them, or how they were formed, or how they behave. One coulddefine spoons in volumetric terms, or in terms of the materials from which they arefabricated, but thats not how we think of spoons unless what were trying to do is designor fabricate spoons, so even in this case the definition is tied to activity. So the concept ofspoon is really a representation of the activities spoons are involved in, and the meaning ofthis concept is essentially predictive: What it means to be a spoon is just what happens tospoons in various activities.

12Thank you!!

13DiscussionSix lessons from babies (Smith & Gasser. 2005)It is claimed, that systems should be build incrementally. (not by selecting random x samples for training, and n-x for evaluation). Can you think of problems, where this does not work?[Martin]

It is said that it seems to be counter intuitive that children fail to locate an object in a room if the model that should help with this task is too similar to the room. Isnt this just because a very similar model has to be very complex and is therefore harder to understand?[Thomas]

Incremental learningHow could multimodal associations be established in an infant? How does a baby know to look in the direction of the radio when it hears a sound?(In case of a moving car, this would be easy, but if there are no obvious visual clues?) How does it know to look in the direction of its parents when they are talking (prior to it being able to focus well enough to perceive the mouth motions)?[Thomas]Multimodal associationsIt is mentioned that babies