Mandler, Jean;
The foundations of mind: Origins of conceptual thought
Oxford University Press, USA, 2004 [gbook]
ISBN 0195172000, 9780195172003
topics: | psychology | cognitive | developmental | language-acquisition
How to build a baby This was definitively one of my more influential books on cognitive psychology. It is developmental in its focus - the process she outlines is also called "how to build a baby" - but it touches upon many aspects that underlie adult cognition as well.
While the data she presents overwhelmingly rejects the Piagetian position that concepts are not available to the child before language, Mandler is not willing to completely break with Piagetian dogma, for her a concept is something that by definition is "declarative and conscious"; procedural schemas are "perceptual abstractions that do not constitute concepts".
So much of the categorization tasks that infants are repeatedly shown to be performing are a result of "perceptual schemas" - what we may call implicit cncepts. Clearly, a layer of subconscious, pre-declarative notions are at the base of all our concepts, but Mandler just stops shy of calling these concepts.
The emphasis on "propositional" knowledge - a legacy of several centuries of conscious modeling of the mind persists throughout this otherwise ground-breaking work.
Indeed it is possible that propositional representation does not exist in the human mind until language is learned. [Mandler 94] p.119
Excerpts : ch1: How to build a baby As I use the term, a concept refers to declarative knowledge about object kinds and events that is potentially accessible to conscious thought. The first time I said that there has been almost no research on the infant conceptual system, I was told that this was an exaggeration, because of research on the "object concept" and the "number concept". 4 Number concept: 5 month olds are sensitive to addition and substraction of small nums [Wynn 92 - see also Bloom] - but this is probably due to an implicit tracking mechanism rather than conceptual knowledge of cardinality (Simon 1997]. In this classic expt from 1992, Karen Wynn showed 3-mo babies either an impossible situation (right) or the usual outcome (left). Average looking time went up from abt 13 seconds (possible outcome) to 19 sec for the impossible. image source: Stephen Butterfill A system of object files governed by the perceptual system [Kahnemann and Treisman 84] may deliver information in implicit form that enables infants to keep track of small numbers of objects (Uller Carey Huntley-Fenner Klatt 99]. Also, implicit ability to estimate magnitudes that human infants share with many species (Whalen, Gallistel and Gelman 99]. These abilities are undoubtedly related to later mathematical achievements [Carey 01; Gelman 91] but there doesn't appear to be strong evidence [that this knowledge is conceptual] <-- How can it not be conceptual if it is converted to concepts Object concept: experiments mix implicit (unconscious) and explicit (conscious) processes. For example, the sensorimotor system delivers the implicit information that objects are 3D and solid and do not implode as they move behind barriers. On the other ahnd, some of the object permanence tests showing that 6 month olds represent specific information about hidden objects, such as that an object behind a screen is on top of the track rather than in front of or behind it (Baillargeon 86] seem difficult to account for in terms of implicit sensorimotor knowledge; the infants must remain aware of the information no longer in view. (Interestingly however, some of the demonstrations involving objects that disappear behind screens and reappear again [Baillargeon and DeVos 91] may be accomplished using the implicit object tracking mechanism.) The first concepts about objects are not necessarily "basic level" - term undefined - it is possible to know what a dog looks like and not have any idea of what it is and also to have a concept of animal without a commitment to any particular kind of shape or features. 5 [Challenges the view that] the foundation of the conceptual mind consists of associations accruing to perceptual schemas - pervades our thinking not only about infancy but also about the preschool years. 6 In my opinion the problem of relating perception and conception can be resolved only using a dual representational system ... brings new problems - e.g. how object files or implicit magnitude estimations are related to the explicit number system... ACCESSIBILITY, by a long standing definition [Tulving and Pearlstone 66] means the ability to come to conscious thought. But usage such as "accessible to consciousness" is necessary because the term is often used otherwise, and also defined differently (e.g. [Spelke and Hespos 02]). 7 During the behaviorist period, ... a single representational system was assumed. Of course, behaviorists would have rejected the notion of representation, so they talked instead about a common set of mechanisms that was assumed to apply to all processing. (We may be heading back to this view today, as connectionism has gained sway in the field.) 6 [Piaget] Babies don't start to develop a conceptual system until the end of the sensorimotor stage (at roughly a year and a half). How to study the transition from a sensorimotor creature to a conceptual one? Use the first words as the measure of the underlying conceptual system [p.8] ... Misuse of words is seductively easy to equate with conceptual misunderstanding, but it is just as likely to be due to trouble in mapping language onto existing concepts.Techniques in infant cognition
A. Nonverbal precursor to object sorting: relies on young children's spontaneous tendency, when given an array of objects, to touch sequentially those that are alike [Henry Ricciuti 65, Nelson 73, Sugarman 83]. B. Familariazation / Preferential-Looking: infants first shown stimuli fron one category, and then an exemplar from another category. In very small children, give them one object at a time and measure inspection time. Works better than sequential touching - upto 7 months - infants (<15 mos) tend to get overwlemed with "such a plethora of goodies" - tend to freeze up and interact with other objects. [Chapt 7] C. Deferred Imitation - observed by [Piaget 52] Babies as young as 9 months can act out (after a delay) what they remember having seen - events they have observed only a single occasion. [Chap 10] D. Inductive Generalization: model an event for infants using replicas; give a dog a drink from a cup. Substitute for the dog: bird, and a car. See which objects they use to imitate drinking = Q. "What sort of things drink?" --> observing events and making inductive generalizations. 9-10 Infants have an idea of container but are hazy about diff between "pan" and "cup"; know what an "animal" is but not diffs between animals. [NOT at the BASIC LEVEL]. Right from the beginning, infants form concepts in a way that looks remarkably like using defining FEATURES rather than overall physical appearance. We do not know, for the most part, what those "defining" features are, but we are beginning to discover them. Babies can recall objects and events from cues - mean that they are forming a declarative knowledge sysstem, they use this to give meaning to what they see - differs from what was described by Piaget. [12]Piaget's notions retained in Mandler thesis
a. Concepts are not innate but are constructed. b. Concepts based at least in part on perceptual knowledge - though he relied more heavily on motor learning c. Process for Conceptual schemas are similar to "perceptual meaning analysis" (earlier called "perceptual analysis" in Mandler 88/92) - e.g. Piaget's analysis of infant actions - e.g. blinking eyes or sticking-out tongue - but Piaget thought these depended more on action, so he thought it developed later than is being posited by Mandler. Piaget failed to relate (c) to the process of reflective abstraction Reflective abstraction (if I understand it correctly) refers to the workings of a thoughtful mind analyzing and systematizing aspects of the knowledge it has gained to date. 13 Misuse of words is seductively easy to equate with conceptual misunderstanding, but it is just as likely to be due to trouble in mapping language onto existing concepts.How early do infants start analyzing the world?
Infants are interpreters of the world around them from an early age. We don't know how early, but Werner and Kaplan's (1963) estimate of 3 months as the onset of contemplation of the world cannot be much more than 3 months off the mark! The very young infant who cannot act on objects is nonetheless construing the actions of others. Conceptualization is already on the march, perhaps even earlier than 3 months. [Chapts 3/4] I propose an organism that is born with the capacity to form two very different kinds of representation. One of these, largely sensorimotor in character, uses perceptual and kinesthetic information to form perceptual schemas of objects and motor schemas that control actions. This kind of learning is procedural; that is, it operates outside the bounds of consciousness, and the schemas it creates are not accessible to conscious thought. At the same time a mechanism of perceptual (meaning) analysis extracts and summarizes a subset of incoming perceptual information from which it creates a store of meanings or simple precepts ... arixe from attentive, conscious, analysis and are markedly different from perceptual and motor schemas. I have suggested that these meanings are represented in the form of image schemas, although there are other formats that could serve this purpose. 14ch2: Piaget's sensory motor infant According to Piaget, the sensorimotor child ... does not have a capacity for representation in the true sense, but only sensorimotor intelligence. Knowledge about the world consists only of perceptions and actions; objects are only understood through the child's own actions and perceptual schemas. It is a most unProustian life, not thought, only lived. Sensorimotor schemata... enable a child to walk a straight line but not to think about a line in its absence, to recognize his or her mother but not to think about her when she is gone. It is a world very difficult for us to conceive, accustomed as we are to spend much of our time ruminating about the past and anticipating the future. Nevertheless this is the state that Piaget posits for the child before 1 1/2. [Mandler 83, p. 424] 18 In Renee Baillargeon's experimentsl young infants must briefly remember that an object that has disappeared behind a screen is still there: 4.5-6.5-mo: do better if a copy of the object remains in view at the side of the screen. [Baillargeon:1991]Similarity is not enough for learning categories
To do inductive generalization without a conceptual system would mean that the infant would be dependent for such generalizations solely on its perceptual system; it could generalize from one stimulus to another only if they were perceptually similar. It would not be able to do inductive generalization on the basis of concepts. So, for example, it would not be able to generalize from a fish to a bird on the basis of their both being animals. 20 [BUT perceptual similarity also includes perception re: dynamics of events. [CONDITIONED expectations are not concepts] The infant might learn to anticipate the end of mealtime by raising its chin to have its bib removed. This does not mean the senorimotor baby could imagine the coming event, however, because imagery is another capacity that is dependent on a conceptual system. I believe that Piaget was correct in his assumption that imagery requires a conceptual system. We don't have a great deal of evidence, but what there is indicates that one cannot form an image on the basis of perception alone. e.g. Kosslyn[80] found it necessary to posit a propositional store of conceptual information in addition to a visual buffer in order to explain image creation. 21How we conceive images influences what we see
[Carmichael, Hogen, Walter 1932] (pdf at bryanburnham.net) showed people (nonsense[??]) line drawings that were given labels - e.g. a figure with two circles connected by a horizontal line - labelled either eyeglass or dumbbell - when asked later to draw from memory, it became obvious that the labels had influenced their images. 21 Stimuli line drawings were shown, along with labels - either left or the right. When subjects were asked to redraw the image shown, people who saw the two labels reproduced them differently. e.g. image produced for the top image above after eyeglasses label: , whereas after the dumbbell text: (figures from http://www.tuninst.net/LANG-MEAN/Human-learn/human-learn.htm) : Role of FREQUENCY data in conceptualization? ] imagery comes from imitation -- or more precisely, from detailed analysis of what one is observing. Piaget insisted that looking alone is insufficient to form visual image. He thought that to form an image you have to engage in some extra accommodatory effort ... 22bot But much of Piaget's description of imitation suggests that the very process of imitation requires a conceptual base. 22Perceptual meaning analysis
Perceptual meaning analysis : transduces perceptual information into conceptual form by creating image schemas. 22 [G. Mandler and Kuhlman 61] Subjects had to learn a randomly generated sequence of eight switches. Experiment was looking at "overlearning of motor patterns" - but resulted in learning an imagery - without the real panel in front, subjects could run through the pattern - had developed a body-feel - later, when performance became asymptotically fast, subjects reported visual imagery of the pattern - they could see the pattern as if it were lighted up, even though they had never seen the whole patern displayed at once. 23 Repetition alone is insufficient: [Nickerson and Adams 79] - can't image a penny very well, though handling it all the time. Mandler: Drawing of Old-Fashioned telephone dial --> most don't know where the letters are with the numbers, whether it says "operator" anywhere, etc. My tentative conclusion is that Piaget was correct: no amount of looking alone will result in imagery. One must analyze what one is looking at (or touching), and analysis requires a conceptual system --> imagery itself requires a conceptual base. 24 [SPEECH PATTERNS If actions are necessary, how do infants learn speech patterns, the first speech babbles already reflect this - need to look up these tests - around 9 months? [Calvin/Bickerton:00]: at six months, a Japanese infant can still hear the difference between the English /L/ and /R/ but, by age one, he or she no longer hears the difference, with a nearby Japanese phoneme capturing all nearby speech sounds as mere variants, standardizing them. "Rice" and "lice" would sound the same. ]Piaget's Sensorimotor infant
Sensorimotor Infant: conceptless creature who can't think independently of action, who cannot recall past or imagine the future, but can recog familiar objects and act appropriately toward them. Six Substages: 1-2: (1-4 mos) - SIGNALS - precursors to symbols - one perception indicates that another is to follow. e.g. sight of breast indicates that milk will enter the mouth, and baby salivates in response --> ASSOCIATIVE --> Conditioned Stimulus (CS) --> strictly procedural, no awareness. Such signals can't be reproduced by baby in their absence (how do we know?). Signals are EXTERIOR to the mind rather than in it. 25-6 3-5: (~ 6 mos) SIGNIFIERS - motor recognition - * Lucienne saw two parrots that used to be on her crib, now across the room - she shook her legs at them, using the gesture she had used when she played with them in her crib. Piaget suggested that the infant was expressing her recognition of a familiar toy and "naming it", so to speak, using an abbreviated version of her accustomed action on it. --> motor recognition, but not a symbol since it was executed only on seeing the object (but perhaps obsrvr wd not recog it if executed without it). 26 upto stage 6, mind is action-based 6: (18-24 mos) - Children move beyond trial-and-error problem solving. * Lucienne, tried to kneel on stool on casters - but it rolled away - immediately took it and wedged it against sofa and then climbed up. * Jaqueline arrived at a closed door with grass in each hand. She stretched out her hand toward the door, realized, put the grass on the floor and picked it up after opening door. 24 [Q. Can't these planning acts be done non-declaratively]The development of metaphor in signs
For conscious thought we must rely on IMAGES or WORDS [both are symbols] to present to awareness what is being conceptualized. Unclear in Piaget: whether first concepts are sensorimotor schemas made accessible as symbols, or does symbolization create a separate, less action-oriented layer... SIGN LEARNERS speak earlier (first signs earliest by 6mos, 8 mos on avg) [Mandler:1988] [Folven./Bonvillian:1991] [Meier/Newport:1990] [Goodwyn/Acredelo:1993]: This happens because babies gain control over their fingers before their vocal apparatuis, making gestural symbols easier. (see Lucienne parrot symbol p.26] Elissa Newport once gave me an evocative description of the contexts in which early signs are used. Daughter Susanna grew up bilingual in English and ASL. The first recognizable sign she produced was "finished", used by parents at end of meals; by 5-6-mos, [STAGE 2] she began using it herself at the end of a meal. Although clearly imitative in origin, it might not have had any conceptual import at that point and may have been no different from putting her face up to be wiped - like any other motor anticipation. However, by 7-mos Susanna began using this gesture when she didn't want any more to eat, turning her head away from the looming spoon while she executed it. Because it occurred before the end of the meal, it was unlikely to be merely anticipation; it seems to have become decontextualized to some extent to communicate "No more". 27Piaget's "indications"
Piaget called such abbreviated actions "indications" - a type of part-whole relationship, where the part primes the larger meanings. Later, some special aspects of imitation created images, which he considered to be the first true symbols. The characteristic of imitation that interested Piaget most was that it could not take place without active analysis of the model; that is, he believed that imitation is not a passive process but requires the imitator to analyze what the model is doing. 28-9 Piaget: Active analysis is required to imitate, at least to imitate anything complex. When anal has taken place and imitation carried out, it eventually becomes what he calls "interiorized" in the form of an image. The problem is that infants imitate from an early age - does this imply analysis? Why then does it take an year and a half to form "images"? Piaget's reconciliation of this paradox is that imitation in the early stages does not require active analysis as in more complex imitation. Nevertheless, as early as stage 2 (around 4-mos) infants can imitate clapping hands. That would seem to require analysis - must make the connection between its and the adult's hands - but, acc ro Piaget, this comes for free: "When the child... becomes capable of co-ordinating the movements of his hands with his vision, he acquires simultaneously the power of imitating certain movements of other hands, by assimilating them to his own." [Piaget 1951] [This seems amazingly precipient, given the modern motor neuron theory]Recognizing Imitation
When his 4-month old son waggled his thumb and Piaget imitated him, his son laughed "and compared our two hands several times" (P 1952, p.24) The comparisons made may be a more important measure of conceptual activity than novelty of the gesture. 30 Piaget began to try to get Jacqueline to blink her eyes at around 9-mos. He got nowhere, though the baby was interested and felt P's eyes with her fingers. Shortly before 1 year, she opened and closed her mouth in response to to his opening and closing his eyes. Shortly thereafter, when she was rubbing her eyes one day, P imitated her. She watched him rub his eyes, and then she rubbed her mouth. The next time she rubber her cheek and then her ear. These responses continued for several days until finally she rubbed her eyes in response to his doing so. That same day he blinked his eyes and she rubbed hers. There appears to be learning here throu8gh successive approximations ... similar observations with Lucienne. Around 11 months, she opened and closed her hand in response to Piaget's blinking his eyes, followed by opening and closing her mouth. At 1 year, she did both at the same time and also covered and uncovered her face with a pillow. 31 Piaget did not emphasize what seems to me as an expression of the concept of "opening" and "closing". They had the right idea but could not locate the right part. The concept was abstract - opening and closing per se (not of a particular obj). P recognized that there was a kind of analogical understanding going on, but I would add that analogical learning of this sort implies the presence of a working conceptual system. 32Similarity vs Association
two types of symbol formation similarity : symbol resembles referent (breast --> nourishment to follow) association: symbol is associated with referent 32 piaget: relies on association end of stage 1 - respond to signal of the breast stage 3 - motor recognition - movement stands for the object stage 4/5: anticipatory response to stimuli is more elaborate and less bound to immediate stimuli But none are true "symbols" for Piaget. Traditional objection to imagistic representation such as that proposed by Paivio (1978) - "conceptless" picture of a scene does not specify which aspects are to be thought about. For example, in an image of pouring milk from a pitcher into a mug, is the colour of the mug relevant to the meaning, or the fact that the pitcher is 10 inches high, or that it is held in the left hand and the mug in the right? 35 Baillargeon and Spelke's (1985) insight was that the technique (of familiarization / preferential-looking) could be used to study conceptual knowledge as well (orig designed for perceptual discriminations). This was done using impossible test events - i.e. events that could not happen in the real world - and the brilliant aspect of the idea was to make the impossible test event identical to the habituation event. e.g. by requiring the infant to remember something hidden behind a screen that made the habituation event no longer possible - rule out perception as the cause of the novelty leaving an explanation only in terms of conceptual knowledge. DRAWBRIDGE EXPT (Baillargeon 85): drawbridge rotates 180 deg - then a box placed behind it so it can't rotate more than 120 deg. In this condition rotating the drawbr by 180 deg (impossible event, but id to habituation) results in less preferential looking than 120 deg rotnch3: kinds of representation The assumption that infants live purely sensorimotor lives is deeply ingrained in the developmental community. We know that babies are learning to recog and manip objects but not rep them conc. [Meltzoff and Moore's 1977 claim for neonatal imitation] - "if their interpretn is true, we have to abandon [Piaget]; that's major trouble" - Mandler 1992b Levels of procedural knowledge - clenching ones fist, grasping a ball, tying a shoelace, driving a car.Learning to type
As we type a letter, the fingers that will type the subsequent letters are already beginning to rise... [Gentner 88] : Expertise in typewriting pdf A typical professional typist averages 50 words per minute (wpm) for 20 hours per week. Over the course of 10 years, that would amount to 150 million keystrokes... or 25 million words. A typing rate of 60 wpm corresponds to an average of five keystrokes per second. The fastest typists I have studied maintain an average of more than nine keystrokes per second over the period of an hour. [fastest: 112 wpm] Like other motor skills, typewriting, once acquired, is remarkably resilient. In a classic series of motor learning studies, Hill (1934, 1957; Hill, Rejall, & Thorndike, 1913) recorded data from three month-long efforts to learn typewriting that were separated by lapses of 25 years. Hill found significant saving of skill at the beginning of the second and third learning efforts, despite the intervening 25 years between efforts. Computer simulation of typing : [Rumelhart & Norman 82] (Tends to suggest that letters are typed in chunks and not as individual characters.] one-finger digraphs: "de" two-finger digraphs: (fingers on same hand), e.g. "dr" two-hand digraphs : e.g. "do". The possibility of overlapped movements for two-finger and two-hand digraphs was confirmed by analysis of videotape and high-speed film records of typists' finger movements (Gentner, Grudin, & Conway, 1980; Gentner, 1961). Numerous instances were found in the videotapes of expert typists when two, or occasionally three, keystrokes were in progress at one time.Verbal descriptions as "made up stories"
Although can describe shoelace-tying, actually this is a superficial construct. --> telephone example above 47 although one uses conceptual knowledge in learning how to tie shoelaces, these are just guiding the compilation, 50 [This concept level is LOST after the procedure is learned, because it is never repeated. We then "make up stories" to explain it declaratively] face recog - cannot articulate - can't even describe what it looks like - which side the hair is parted, whether the earlobes are droopy, and even whethter the face wears glasses 48 there are aspects of faces in general that no one knows how to describe. [Fagan and Singer 79] showed that 6-month olds can categorize faces as male or female, and do so before they are good at discriminating one male from another. Adults make up stories abt the delicacy of the female face or the strong jaw of the male, but the actual parameters are still unknown. [Abdi Edelman etal 95] 48 [Quinn Eimas+ 93] 3 mos old can look at pics of dogs + cats, just beginning to be able to maintain extended periods of alertness... [Colombo 01] but not able to conceptualize the classes in the few minutes of the expt.Implicit language
Language even though it is conceptual in nature, can on occasion be processed implicitly; it is registered but not attended to or elaborated. [Graf and G.Mandler 1984]: adults were asked either to study a list of words (semantic task) or to cross out the vowels in the words in the list (nonsemantic task). Then they were given a recall and recognition tasks, and a stem completion task, in which three-letter stems and asked to complete them with the first words that come to mind. The first words to come to mind are dependent to some extent on whether they have been activated recently [and] subjects given both set of instructions tended to produce the same number of stem completions from the list, showing the same activation of the material. However, on recall and recognition the vowel subjects were much poorer. The rate of decrement in performance over time was the same for the completion and recognition tests. The findings are discussed in terms of two processes: (a) activation of a mental representation, which also strengthens the relations among its components and increases its accessibility, and (b) elaboration, which establishes relations among different mental contents and increases retrievability. Implications for different memory tests and for understanding the amnesic syndrome are developed. REFERENCE: * Graf, P and G. Mandler 84: Activation makes words more accessible. but not necessarily more retrievable. J Verbal Learning and Verbal Behavior 23, 553-568. Three experiments: memory for words in semantic or nonsemantic conditions. semantic: on a scale of 1 to 5, mark whether you LIKE/DISLIKE it; or is the word; oo CONCRETE / ABSTRACT? non-semantic: report if the number of vowels is same for two words; report the number of "T" junctions in the written word, etc. When the tests required the completion of the initial letters (e.g., DEF—) of recently presented words with the first word that came to mind, semantic and nonsemantic processing conditions produced similar results. In contrast, free recall, cued recall, and word recognition showed substantially better performance under semantic processing conditions.Procedural information in inattentive listening
People not paying much attention when listening to a lecture do process the words they are reading or hearing and so show priming effect, but because of the shallowness of such processing (lack of semantic elaboration; GM 02), they have trouble with recall. The implicit priming effects stem from procedural info being used to perceive the visual or auditory structure of the info in q, such as the phonological or visual shapes of words. This kind of info is presemantic and preattentive and results in perceptual, not conceptual priming. These two kinds of priming appear to be indep of each other and to have diff neural underpinnings [Moscovitch, M. 00: Theories of memory and consc in E.Tulving/Craik: the Oxford handbook of memory. OUP] 52 [Moscovitch, M. 95: Models of Consc and memory, MS Gazzaniga, Cog NeuroSci 1995]. Thus, the procedural-declarative distinction - deals with representation, is not the same as the implicit-explicit distinction - which is a process. Procedural knowledge is always processed implicitly, we cannot process this kind of information in such a way as to bring it to awareness. With decl knowledge it is stored in a conceptual format and has the potential to be brought to awareness - whether that happens on a given occasion depends on the degree of attention (elaboration). 53 procedural knowledge is sensitive to frequency of occurrence - creates expectations of what will happen next 55-6 [*** QUOTE in context of linguistic priming. That this also occurs for language - is it indicative of the procedural nature of some aspects in lg? ]Class remembers the use of "she" for generic 3d person
Once Mandler thought she would use "he" and "she" more or less equally in her lectures. Towards the end of the course, she mentioned to the class that this is what she had been doing. At this there was a murmur in class, and the students said, “What do you mean you gave ‘he’ and ‘she’ equal time? You never said ‘he’ once during the entire course.” Further questioning showed that every student in the class thought she had used "she" 100% of the time. As it happens, she had taped the lectures. So, a couple of students were asked to count from two lectures at random... It turned out she had said "she" only 20% of the time. I was chagrined at how badly I had accomplished my mission, but the students were equally upset at having estimated 20% to be 100% and 80% to be zero -— an effect of the magnitude we would always like to find in the lab! What had happened, of course, was that [in those days]... the students were totally used to hearing "he" to refer to an individual. So each usage of "she" stuck vividly in their minds and when they examined their explicit memory (the only memory accessible to them), they found a great many she’s and no he’s. 56Illusory Correlation
[Hamilton and Gifford 1976] When distinctive or salient events (defined as statistically infrequent events) co-occur they are more likely to be associated with each other. being attended to are associated with socially unacceptable behaviours. Hamilton and Gifford : Created two groups of people labelled X and Y, |X| considerably larger than |Y|. The minority group are less frequently encountered --> more distinctive. Hence, can become associated with socially unacceptable behaviors, which are also less frequently and hence more distinctive. Made instances of positive behaviour ("contributed to a fund for the blind") and negative ("was rude to a shopkeeper"). +ve and -ve behavs were assigned in equal proportion to all indivs in X and Y. Group X Group Y Desirable bhv 18 (69%) 9 (69%) Undesirable 8 (30%) 4 (30%) Subjects : asked to read the behaviours, and who did it. Then estimate the num of undesirable behaviours in each group. --> People shown the lists of behaviours consistently rated the minority group Y as engaging in more unacceptable behaviour. In expt 2, the good behaviours were rarer and this time Y were judged to be better behaved than X. Conscious associations are built up more on the basis of what has been noticed than what has actually happened. 58 For example, we all know friends who relate an event like their going to the bathroom, with an Indian batsman getting out. Result: Hold the urge until the end of innings! see Illusory correlation (wikipedia)
ch 4: Perceptual meaning analysis how perceptual data is processed to form image schemas. Philosophers such as Locke and Hume assumed that we form images of perceptual displays and then associate them with one another... While accepting an associative account of meaning, psychologists want to know the exact mechanism, how exactly this transformn of percepts into concepts takes place. In my opinion Piaget came the closest, and in spite of my ultimate diwsappointment in his account of early conceptual development, I believe he was, in important ways, on the right track. He understood that we need something more than percepts and associations among them to create meanings. Some kind of analysis of perceptual sights and sounds must be carried out if meaning-bearing symbols are to be formed. 60 Lakoff 87: concept of containment is structured - there must be an inside and an outside and a boundary between. The components of inside and outside themselves have no meaning beyond the overall structure of the concept. 62Nativism vs Empiricism
[Piaget position = CONSTRUCTIVISM ] Nativists like Elizabeth Spelke 94, Alan Leslie 94 have been concerned w how infants come to learn abt the Piagetian "object concept" - how do infants learn that objects are solid and cannot pass through each other or be in two places at once, and esp that objects retain their properties when they are out of sight? Learn abt physical causality, that one obj makes another move, or that objects fall when unsupported. These are not specific concepts but ontological commitments. One can be a nativist on these issues while remaining an empiricsist on other aspects, like the concept of a dog or a cat. 63 Infants interpreting the world are ascribing meaning to what they perceive, and those meanings form concepts. Infants' concepts refer to the world, even though they do so less completely than do those of adults. 67 [WHAT IS THIS WORD MEANING being used as?] ---Chunking / Coding in working memory
Internal structures abstract the input from many different actions and entities, resulting in considerable efficiency in use (WM = 4 bits). The coding works by taking more frequent occurrences and giving them a smaller (more abstract) representation compared to infrequent entities which are not represented as compactly. Each of these may be called a "chunk" - each chunk is a primordial concept Internal Reference Conscious reference to concept (usually when it is not present), (in general thought, or when prompted by reading, or talking) mental attention --> ANALYSIS Subconscious reference: in dreams at night --> constitute structure for long term memory. --> MEMORY FORMATION Cognitively, the youngest babies can perceive spatio-temporal motion rather than appearance characteristics. Causality is understood in terms of contingent temporal sequencing, and the animate/inanimate distinction is based on contingent behaviour rather than shape / appearance. The visual acuity of the infant improves only after two months. Even as adults, we find animate properties in simple shapes if they are contingent; the shape etc are far less relevant...Fodor's mistake : Infants do not test hypotheses
Fodor 81: theory of concept formation - foundered on the rock of hypothesis testing - Infants analyzing perceptual displays do not do so with hypotheses in mind. 67 They merely apply an analytic mechanism to what they are looking at. This mechanism asks: How shall I interpret that? and not Could that be a ...? Nevertheless, there must be a core of possible meanings that human creatures can extract by means of this mechanism. Why are some meanings extracted while some are not? Why do infants analyze some kinds of displays [partly due to innate biases] 67 It is the conscious use of perception that teaches a child that apples and tomatoes are both red. It is possible -- perhaps even the rule -- to encounter tomatoes and learn what they taste like, and also to encounter apples and learn what they taste like, without making the observation that they have the same colour. When I first wrote about perceptual analysis I conceived of the mechanism as a comparator -- a device to compare one perceptual display with another, so that similarities and differences could be noted. [comparison seems too much a process like Vicarious Trial and Error, or VTEing (Muensinger 38) - in which an organism actually looks first at one thing then at another. Of course, VTEing does appear to be a kind of analysis - one has only to watch a rat hesitating in a choice point of a maze, its whiskers alert, looking back and forth between one arm of the maze and the other, to be convinced that the creature is comparing two alternatives. But comparison is only a part of PMA.Perceptual Meaning Analysis : "Apple"
For infants, comparison is not the most crucial aspect. For them the most crucial thing is that it is a concept-making engine, transforming perceptual info into another form. Unlike adult, the young baby doesn't have a concept of an apple in the first place... [?? Or at best, if it has seen apples, its concept is very impoverished] They baby only has the ability to attend to and analyze perceptual displays. These analyses result in the meanings that are used to create concepts such as an apple. Development of concept of "animate": For the first month or so, visual acuity is not great, so infants are not getting high-quality foveal information about what the objects moving around them look like. But thhey can parse the world into objects because of their coherence against the background when objects move [Kellman/Spelke 83] and they get good info abt at least three aspects of the movement: - by 3 months they discriminate biol motion from nonbiol, for both people (Bertenthal 93] and other mammals (Arterberry and Bornstein 01) - by 2 months they treat objects that act contingently as animates, as shown by smiling at them (Frye, Rawling, Moore and Myers 83; Legerstee 92; Watson 72). - between 4-6 months, infants are responsive to the diff between an object beginning to move without anything else coming in contact with it, and an object moving when touched by another (Leslie 82,84). 71 (all these ages are the earliest studied) [Arterberry and Bornstein 01] [Kellman/Spelke 83] Perception of partly occluded objects in infancy. Cog Psych 15 483-524. [Frye, Rawling, Moore and Myers 83] [Legerstee 92]; [Watson 72]. [Leslie 82,84]. None of these alone constitutes the concept of an animal - they are pieces of perceptual data that needs to be redescribed via PMAAction-Perception relation
Consider the baby who has not yet developed expectations about the world, whose foveal acuity is still poor, but who's eattention is attracted to moving objects (Arterberry/Craton/Yonas:93] [Kellman 93). What might this infant notice about events like the following: She sees an object nearby, she cries, the object begins to move, approaches, looms, and she is picked up. Leaving aside details such as whether she has a schema for "face" [Johnson/Morton 91] she might not be able to analyze much more than that an object began to move, along some irregular path, and did so contingently on her cries. 72 Even if the baby abstracts something abt the trajectory of the moving object, the way the trajectory begins and ends, and the contingency between the trajectory and her own behaviour, she will have gone a long way toward conceptualizing an object as animate. [Arterberry/Craton/Yonas:1993] [Kellman 93] in [Granrud 93] Kinematic foundations of infant visual perception, in CE Granrud ed. Visual perception and cognition in infancy, Hillsdale, NJ; Erlbaum. p. 483-524. [Book] [Johnson/Morton 91]Perceptual Meaning Analysis vs Representational Description (Karmiloff-S)
The PMA theory is related to the representational redescription (RR) [Karmiloff Smith 92]. Beyond Modularity (Fodor) suggests that initially in learning new domains procedures are implicit and encapsulated. 5 This may remain procedural for long periods of time, but at some point info is transformed into new representations indep of the procedures themselves - (KS calls this explicit, but JM would call it implicit because not accessible; in KM explicit knowledge is divided into several levels ). After still further redescription, some accessible concepts (sim to image schemas) emerge. 74 e.g. French children use un for both "one" and "a" but are not aware of this difference until after several years of use. JM: can still remember the first time she learned that breakfast was break + fast. Piaget assumes that the first concepts, in his view transforamtions of sensorimotor schemas into a symbolic form, constitute the basis on which lg develops. Does not discuss the representation of these first concepts. 76 conseq, concepts in infants have often been studied through the words newly verbal children use. But to access concepts only through the linguistic system, esp one that is in the process of being formed, will surely distort our picture of what the earliest concepts are like. Even more seriously, the first concepts are formed well before lg begins, so their origins must be nonlinguistic. It is a coincidence of linnguistic form that has led many philosophers (and psychologists) to assume that conceptual representation is propositional in nature. We appear to think in sentences, whose components are concepts couched in propositional format. This phenomenal experience of thought as equiv to lg is at least somewhat illusory. 77PMA vs Barsalou's "Perceptual Symbols"
Is it possible to represent a shape without an orientation, a path without a direction, or stripes without representing their number? [Barsalou 99] Different neural pathways are involved in handling each of these types of info. Both perception, as well as perceptual symbols (image schemas) are constructed out of bits and pieces which can be repr separately. ... For e.g. in Terry [Regier 1995] connectionist model, paths are repr in terms of starting and ending points, and a nonsequential static repr of the interim points. 81 Larry Barsalou sometimes uses the term image-schemas to his perceptual symbols [fig. 1, p.578). Many of the ideas in PMA could fit easily into Barsalou's system. The main diff is that Barsalou doesn't make, or at least doesn't emphasize, a distinction in format between perception and perceptual symbols, and these are not as clearly differentiated. ... Also, I emphasize the developmental aspects of the processes Barsalou proposes. E.g. he does not specify why attn focuses on some features rather than others, or how the cog sys olearns to categorize the world or to form the abstractions that enable inductive inferences. Nevertheless, there is a close affinity between our approaches. 83 In Regier, a source buffer contains a repr of a starting config of the trajector and the last time step in the buffer has the ending config. set of motion buffers builds up the static repr of the path in between. These constraints reflect innate biases to emphasize beginnings and endings of paths. It seems clear that some such biases must be built into the human system as well. But I suspect that the set of such innate specifications is not large. 84 --Not topological : IMAGE SCHEMAS are not TOPOLOGICAL - e.g. image schemas may repr whether the contents of a container take its shape or not, something that may get eliminated in a topological model, which approach symbolic representations likeor . [** There is no agreement on what constitutes a topol model. Topol models can jolly well encode surface contiguities] Image schemas may also contain directional indicators, e.g. in the ABOVE or BELOW schemas 82 The earliest meanings shd be spatial descriptions rather than, say, a figural descr of what something looks like. Given the limited perc capacity of young babies, spatial relations and movements may be available to them earlier. Although color is registered, shape is apt to be blurry. Infants in the first month often do not even scan the interior of objects and so get few of the details that identify particular kinds for adults. [Salapatek and Kessen 66] Second (probably more important) meaning of objects for humans dep on what they do or what is done to them, as Katherine Nelson pointed out [Nelson 74] Perhaps the reason why the infant attends more to motion than to appearance is determined by the needs of the species. What things do is the core of their meaning. 86 What about time? Infants may have no concept of time. Even as adults, we seem to unable to think of time independently of space. [Guyau:1890/1988]: We can easily imagine space: we have an inner eye for it, an intuition. Try, on the other hand, to represent time as such; you will only succeed by means of a represetnation of space. You will have to align successive events, placing one at some point along the line, the other at a second point. In other words, you must evoke a sequence of spatial images in order to represent time p.99 [mandler 87] Guyau suggested that we conceive of time in terms of a path from then to now, from one location in the past to the present. But [Boroditsky 2000] sometimes we conceive of time as a path that flows past us (e.g. when waiting). 87
ch 5 Some image-schemas and their functions In response to my movement of opening and closing my eyes, J... opened and closed her mouth..., L... opened and closed her hands. ... The child's mistake must ... be due entirely to analogy. When the child sees other people's eyes opening and closing, he assimilates what he sees, not to the visual schema related to other people's mouths, but to a general schea, partly visual but mainly tactilo-kinesthetic, of opening and closing something. [Piaget 1951] 93 (see p.31 for details) JM's model inspired by cog linguists - Johnson 87 Lakoff 87 WFDT Fauconnier 94 ANIMACY 9-month olds were distressed when a machinelike robot (i.e. an object that did not look like a person) began to move on its own. They were also distressed when the robot moved contingently upon verbal commands given by the mother, indicating they did not expect inanimate objects' movements to be influenced from a distance [Poulin-dubois, Lepage, Ferland 96] the -ve effect shown by the infants is a stronger reaction, of the sort one might expect if one's conceptual model of the world is being violated. [*** IDEA: have learnt the class of objects that behave contingently - and associated certain shape priors with this class] [Bennett Bertenthal 93] 3-mo: can distinguish between correct and incorrect human motion - technique orig invented by Gunnar Johansson with lights at joints. scenes shot in the dark - eliminating all figural info; infants discriminate between lights corresp to joints of a moving person vs others with equal displacement but not human joints. 95 [Arterberry/Bornstein 01] 3mos: distinguish between mammals and vehicles based on moving light displays alone - some parameters have been abstracted. [*** ABSTRACTION involves points of inflection] I originally thought that self-motion would be the aspect of animal motion most crucial in defining animals or infants, but I have come to suspect that contingency of motion between objects may be even more important. Responsivity to the contingency of events is present at least from birth and is one of the most powerful factors governing percptual learning and controlling attention. One of the first kinds of contingency that neonates learn is between two events in the environment, resulting in S-S conditioning [Sameroff/Cavanaugh 79] motor limitations of infants severely restrict their manipulation but they observe contingent interactions in others. 96 [John watson 72] seminal experimental work - showed at 2 months of age infants would learn to make a mobile hanging over their crib turn when the movement was contingent on their pressing their heads on a pillow. When the mobile did not turn or turned noncontingently, head presses did not increase. An interesting response of infants trained contingently was that after a few sessions of practice, they began to smile and coo at the mobile. [*** CONTRAST with [Poulin 96] - 9 months old are upset --> particularly telling evidence that shape class abstraction is a much later phenomenon] Watson hypothesized that the contingencies made the mobile become a social stimulus. I am not sure how to define the difference between a social stimulus and an animate one (see the section titled conceptualizing agents and goals), except that a social stimulus is one that reacts contingently to one's own movements, as opposed to reacting contingently to the movement of other objects. Infants learn to expect contingent interaction from their mothers very early. [Murray/Trevarthen;1985] see also [Muir and Nadel 98] found that 6 to 12 week old infants became upset if their mothers, shown on videotape, did not respond contingently to them. [Tronick/Als_etal:78] 2-month olds become upset if the mother presents only a still face and do not respond to them at all [Frye/Rawling_etal:83] 3 mos reacted similarly to their mother and to a toy when either of these was interacting contingently with the infant, and differently when either mother or toy did not act contingently. [Scaife/Bruner:75] [Butterworth:91] 6 months: infants will follow an adult's gaze if it moves away from the infant. Here of course, there are eyes to look at. [Johnson/Slaughter/Carey:1998] faceless animal-like object also elicits "gaze" following from 12-month olds if interaction is contingent. [Movellan/Watson:2002] completely mechanical looking robot that interacted contingently with 10-month olds elicited "gaze" following - the infants would laugh and converse with the robot if it acted contingently. 97 It is possible that infants analyze zelf-other interactions (e.g. give and take of parent-infant turn taking) precedes their analyzing contingent interactions that other objects do among themselves (dogs and cats or flocks of birds) [Murray/Trevarthen:1985] LINK schemas (1-way)--> associates e1 regularly followed by e2 - everytime the infant drops something over the side of her high chair her mother picks it up - conceptualizing an if-then relation - suggesting that this is one of the image schemas on which intuitive understanding of logic rests. 98 2-way: back and forth interaction - as in turn taking Link paths - objects follow joint trajectories - e.g. what hands do with objects --> may be vital to conceptualizing continuously contingent paths as goal directed, as in chase, even when an end point is not reached [*** --> CHASE writeup] [Pauen 2000b] - 7-mo-olds showed animal-like creature and a ball - then the two objects moved together in contingent fashion - then became motionless again. The infants looked longer at the animal in this third display, suggesting that the infants attributed the observed motion to the animal and not to the ball - and that they expected the animal and not the ball to move again. When animal and ball both moved together because of a hand, say, the infants showed no differential expectation that the animal would move in the 3d scene. difference between self-motion and caused motion is salient to infants at least by 6-mo and likely earlier. [Leslie 84] - data on 4-mo olds. Leslie 82 - similar to [Michotte 63] studied causal perception in adults.Nativism in Perception
[Leslie 94] speculates that infants come equipped with a domain-specific module that computes mechanical props of objects (Theory of Body Mechanism, ToBY) - has innate concept of force - used in analyzing motion and support. However, while such a case can be made, it may not be required. [ [Marsh:2002]: Leslie [Les88], confirming Michotte, showed that by 6 months infants can perceive causal relationships such as launching, suggesting that perception of causality is an innate perceptual ability. [KH01] D. Kerzel and J. Hecht. Visual causality. In M. May and U. Oestermeier, editors, Interdisciplinary perspectives on causality, pages 119-139. Books on Demand, 2001 ] [White 1988] Peter White suggested that because of the particulars of temporal integration function of the eye, we perceive the transfer of motion from one object to another. Michotte: Timing is crucial. If delay in 2nd ball starting to move is > 150 ms, perception of a causal reln between the two events disappears. 2nd ball also has to be present for at least 100 ms before being hit, else a single object exhibiting continuous motion is seen. - single object, pause less than 100 ms --> contiuty of motion - two balls with pause less than 100 ms --> causality White 88 --> spatial discontinuity tells us there are two objects --> causal. However, the continuous motion suggests to us a single object. This conflict is resolved by interpreting the event as involving two objects and the xfr of motion from one to the other (what Michotte 63 called ampliation). The impression of causality disappears if not within time frame of temporal integrn. 100 BASIS FOR CONCEPT OF FORCE: Important: because "Seeing causality" --> causal perception is an observational property given by the way the eye integrates info over time. Also, pushing and being pushed, as well as feeling and exerting pressure. [Profitt/Bertenthal:1990] no evidence that infrants are sensitive to dynamical, rather than kinematic or geometrical, constraints. even adults have dynamical (force) intuitions for only the simplest of object motions so it would be surprising if infants have more sophisticated notions. 101 No need to build innate knowledge for forces etc. [*** SISKIND discussion] Leslie shows infant's discriminate launching events from those with a spatial or temporal delay. I have emphasized spatial analyses because I consider these likely to be ontogenetically prior to analyses of experienced force, but the latter is also important. Johnson 87: force image schemas: compulsion, blockage, restraint, and removal of restraint intuitive understanding of necessity, possibility, and moral obligation rest on metaphorical extensions of the sensorimotor experience of force to the social and metaphysical worlds (see Sweetser 90) 102- [Leslie 84] : HANDS 4-mos: understand role of hands in making objects move - that hands pick up objects whereas blocks of wood do not; if obj held by hand, it is supported, whereas if the hand lets it go, it will fall. This also may be based on sensorimotor experience, and not innate. 102^ AGENCY From knowing what hands do to understanding agency seems like a small step and is taken up quite early. [Woodward:1998] Amanda Woodward 5-mo and 9-mo olds attended more to the goal of a reach than to its spatiotemporal properties [*** contrast cricket batsman's gaze]: two toys t1, t2 at A and B, reach shown for one or other. toys locations switched - Infants more sensitive to change in toy (target) rather than position. Did not work if the reaching hand was a mechanical arm and not a human hand. 103 [Woodward/Somerville:2000] 11.5 mos: expts looking at transparent box w toy, when hand is reaching for box. same toy in new transp box - no change, compared to new toy in old box. when toy outside box, no effect. So looking at hand touching box reflects idea that it may be going for toy, and thus new box is not imp, new toy is. [GOAL PLANS] INTENTIONALITY [Woodward:1999] perhaps as early as 5-mo, clearly by 9-mo: distinguish between someone grasping an obj with a hand unintentionally dropping onto the object; ie. between goal-directed activity vs similar but not purposeful motions [Baldwin/Baird_etal:2001] perceptual structures of intentional action - habituated infants to videos of everyday actions, then showed test videos where the motion was either suspended halfway or at the end - looked more when interrupted in the middle. 103 Gy\"orgy Gergely, Gergely Csibra and colleagues [Csibra/Gergely_etal:1990, Gergely/Nadasdy/Csibra/Biro:1995] 9-mo: Bar between two circles A and B. vid0 (habituation): A moves towards B, jumps over bar, stops touching B. vid1: bar is on other side of B. A's motion is identical, so the jump is not meaningful. vid2: A rolls in straight line, and ends up resting against B. infants looked longer at vid1 (the same motion they had been habituated to, than vid2 (novel motion). [Csibra etal 99] - is true for 9-mo but not 6-mo olds. GOAL-DIRECTED BEHAVIOUR, not SHAPE: Important in these experiments is the fact that the displays were circles w no figural info. Hence infants were interpreting (i.e. conceptualizing) goals purely only on interactive motion. [inferrring a causal agency to circles moving purposefully]. [Johnson/Sockaci:2000] 14-mo: purple blobs treated as agents if they engaged in goal-directed activity. While [Gergely/Nadasdy/Csibra/Biro:1995] conained animacy cues - A pulsated when it came next to B, and B pulsated in turn. In [Csibra et al 99] all indications of animacy were removed, including self-starting motion: A came from offscreen, sailing over the bar, and came to rest next to B. height of bar varied from trial to trial, each time A just clearing it. Same result as before. ==> by end of 1st year, infants have learned something abstract about the kind of route that an object traveling along a path to another object will take. Consistent w Woodward expts. 104^ [Lakoff 87] SOURCE-PATH-GOAL image schema -- closely related to AGENCY image schema [Csibra/Biro/Koos/Gergely:2003] large circle A "chasing" a smaller circle B. B went through a small hole in a horiz bar in front of it, too small to let A through. At that point, A veered around the end of the bar and continued after B until both went off screen. i.e. no end of path seen, only the kind of continuous interaction involved in a chase. test vid: hole large enough for A to pass through; A may go through or may go around bar. 12-mo, but not at 9-mo: infants dishabituated to the same motion when the hole was big enough for A to go through. Csibra claim: these dispalys more diff than Csibra etal 99 because end of path not shown. PATHs and LINK schemas: glass falls off the table and breaks. Telephone rings, and adult, wherever in the room they are, go to phone to pick it up. Objects taking direct paths to a goal (i.e. where something happens) is a very common occurrence. 106 [IDEA: expt: video of one day in baby's life: count how many path/goal associations] AGENCY = CONTINGENCY or GOAL-DIRECTED PATH? these two are not coterminous (coextensive, coterminous, conterminous: being of equal extent or scope or duration). Going from A to B repeatedly may be goal-seeking but not contingent. Turn taking may be contingent but not goal-seeking. [Rochat/Morgan/Carpenter:1997] 3-mo olds distinguish one object chasing another in continuously varying paths from objects that move indep of one another - but not known if this is done if they did so on the basis of goal-directedness or contingent interaction. [IDEA ***: CHASE: EXPT - contact mothers, have team of two visit, set up projector, take notes, and video of infant. ] The mechanism of PMA which conceptually represents objects in terms of the spatial paths they follow, including their beginnings, endings, and interactions, may be sufficient to account for the notion of a "goal", rather than having to build in this interpretation of the world as an innate proclivity. ANIMALS: NO evidence that infants' understanding of goal-based behaviour is restricted to animals. : opposed by [Carey/Spelke:1994] knowledge about animals is due to domain-specific learning and innate biases that apply solely to animals. 108 [Carey 2000],p.40 opposes view of early concept formation: "Where do the categories represented in the image schematic meanings themselves come from? If one cannot derive causality from spatiotemporal descriptions, or agency from from spatiotemporal descriptions (even those that provide the necessary input for attributions of each type of causality), then the problem of how these concepts arise has not been solved." I am suggesting however, that "categories" (ie. concepts) of both causality and agency can be derived from the spatiotemporal descriptions that perceptual meaning analysis produces. Tim Rogers and Jay McClelland : input layer --> representation layer that gradually becomes differentiated as learning proceeds. In addn to the representation layer, there is a relation layer that specifies whether a property (g. color), part (e.g. legs) or activity (e.g. walk) is being analyzed. This is a localist layer (something like PMA) in which one relation per trial is encoded. These two layers feed into a standard layer of hidden units. Finally there is an output layer that learns to reproduce the properties of the input objects. 109 An intriguing aspect of the R&M model is that it effectively differentiates visual input, which takes in a great deal of perceptual information in parallel, from a relatively narrow subset of perceptual information that I would call attended information. e.g. this input consists of what the object input on a given occasion is doing, such as walking or singing (although it also represents attributes such as legs or wings on other occasions). 110 [Landau/Jackendoff:1993] There are infinite degrees of variation in spatial relations, but the mind seems to mark out relatively few as important. Some of these are - containment (related concepts: opening and closing and going in and going out), - support (related - contact/attachment) - verticality (above / below/ up/down) - horizontality (left and right) [what of front and back] This is not meant to be an exhaustive list, but these are some of the basic spatial relations that we know infants are sensitive to, and many other relations such as between appear to be derivatives of these. 111 CONTAINMENT Baillargeon and colleagues: growth in understanding of containment between 2.5-12-mo: [Hespos/Baillargeon:2001a] 2.5-mo: If something is to go into a container, there must be an opening, and that something in a container will move when the container moves. [Spelke/Baillargeon:2002] 5-mo: distinguishing loose-fitting and tight-fitting containment [Aguiar/Baillargeon:1998] 6-mo: wide object will not go into a narrow container [Hespos/Baillargeon:2001b] 4-mo: tall object cannot disapp behind shorter screen [Casasolaetal:2003,not in Mandler: 6-mo: object-indep concept of containment, not dep on occlusion, ] 7.5-mo: taller object cannot disappear completely when lowered into a shorter container. 112 9-14-mo: abstract notions of containment - [Casasola/Cohen:2002,McDonough/Choi/Mandler:2003] 115 18-mo: comprehension of various terms for containment summary by Baillargeon: 2.5-mo - basically open/closed distinction, gradually add quantitative variables to it, such as the size relationships between container and contained. 112 [ IDEA:*** EMBARKATION - causal links? PATH-GOAL-PAUSE-PATH schema? combine topdown (causality) /bottom-up (image) causality - based on 150msec association - temporal integration in the eye - part of hardwire so nativist --> tempo: appears, stops, people "disappear inside" - tempo moves off and disappears. --> bus: appears, stops, people appear, others disapp, moves off. Stop's are regular - functions of object. "Bus stops to pick up people" --> causality in bus - how is it inferred? Hypothesis: Based on a PATH-GOAL-PAUSE-PATH schema. Learned when mother comes, stops, does things, and continues - and so on others as well. IDEA: Two levels of causality - a) SUBCONSCIOUS /unanalyzed - purely associative - multiple instances - lose support, fall. b) CONSCIOUS (theory/analyzed) - can provide theory + theoretical explanation in terms of chain of temporal contiguity. Subconscious/unanalyzed may arise from observation, and indeed provide the nucleus for the later theory. Refer to experiments mentioned in BLINK - [Antoine Bechara, Antonio Damasio et al 1997, Science: Deciding advantageously before knowing the advantageous strategy] - un-theorized subconscious model - behaviour modificn by tenth iteration, hunch around 50, but theory only by 80th iteration. ] For infants, hand - open/closed - as a notion of container. However, unlikely given motor skills around ages 0-6 months (grasping is reflexive, but not analyzed). Similarly, taking in food as a metaphor for containment (Johnson 87) - perceptual processes are more likely. 113 SUPPORT [Baillargeon/Kotovsky/Needham:1995] 3-mo: expect objects to be supported if in contact with surface 5-mo: expect to be supported if a part rests on the surface 6.5-mo: begin to distinguish partial support between adequate support (70%), and inadequate (15% overlap) As with containment, first notion of support is overly simple, and new considerations (gravity/weight distribn) added with experience over the first year. 113 Baillargeon suggests "noticing" (analysis) may be part of this abstraction process. [Lakoff 87] containment - basis for understanding logic; not only "in" or "not in" = P or not P, but also "if A in B and B in C then A in C". Combined with Link (if S1 then S2, p.98) --> roots of intuitions that ground logic.Infants distinguish Above/Below
[Quinn:2003] 3-mo: same object presented above sev times, and then below --> dishabituates above- PROPOSITION ("block-above-bar"=T/F; diff from "ball-above-bar") 6-mo: concept of "aboveness" - works for general objects - above (x,bar), where x = ball, box, etc. 114 infants create mental work spaces in roughly the same fashion as do adults [Fauconnier:1994, Fauconnier/Turner:2002] P expts in blinking --> elicits opening / closing of mouth/fist --> abstract image-schematic repr of opening and closing, though it does not tell us how the whole thought is assembled; maybe using what F&T02 call conceptual blending, which has been extensively documented in adults [Coulson 00]; 116 Janellen Huttenlocher 1974: 10-mo - has been playing peekaboo with diapers. One day, J removes all diapers and said: "let's play peekaboo" - picks up bowl. [? notion of covered-ness/hiding ? or purposefulness / affordance for objects ? ] [Veale/O'Donoghue:2000] shows in considerable detail how Fauconnier and Turner's conceptual blending can be implemented. For computational work on infant thought using a diff approach, see [Kuehne, Gentner and Forbus 2000] even in infancy concepts are represented in such a way that they are accessible for analogical reasoning, problem solving, and recall. 117 Imagery != images. Imagery formed by people depend not only on sensory data but construal see [Carmichael 32 above] If one is imaging a bottle on a table being picked up and milk being poured from it into a cup, various aspects of the conceptual understanding such as containment, above/below, and paths going out and in are already in image-schematic form. [Zwaan 2001][Zwaan/Stanfield/Yaxley:2000]: sentence mentioning a pencil in a cup speeded recognition of a picture of a vertical pencil, whereas a sentence mentioning a pencil in a drawer speeded recog of a picture of horiz pencilRelational aspects of language
How are these relations learned - particularly prepositions/modal verbs - may also be in terms of repre by image-schemas. [Brugman:1988][Sweetser:1990] Need interface from sensorimotor schema --> image schemas to which relational morphemes can be mapped. One way to do this is to "transform perceptual information into a still analog but more discrete form". Many people assume that this criteria refers to the Propositional lg of thought [Fodor:1975], but a propositional preverbal system is not necessary for - concept formation - image formation - preverbal recall - simple analogical reasoning - learning a natural language. Indeed it is possible that propositional repr does not exist in the human mind until lg is learned. [Mandler 94] [CHUNKING: In cognitive psychology and mnemonics, chunking refers to a strategy for making more efficient use of short-term memory by recoding information. - Magic Number Seven - Miller noted that according to this theory, it should be possible to effectively increase short-term memory for low-information-content items by mentally recoding them into a smaller number of high-information-content items. "A man just beginning to learn radio-telegraphic code hears each dit and dah as a separate chunk. Soon he is able to organize these sounds into letters and then he can deal with the letters as chunks. Then the letters organize themselves as words, which are still larger chunks, and he begins to hear whole phrases." Thus, a telegrapher can effectively "remember" several dozen dits and dahs as a single phrase. Naive subjects can only remember about nine binary items, but Miller reports a 1954 experiment in which people were trained to listen to a string of binary digits and (in one case) mentally group them into groups of five, recode each group into a name (e.g "twenty-one" for 10101), and remember the names. With sufficient drill, people found it possible to remember as many as forty binary digits. Miller wrote: "It is a little dramatic to watch a person get 40 binary digits in a row and then repeat them back without error. However, if you think of this merely as a mnemonic trick for extending the memory span, you will miss the more important point that is implicit in nearly all such mnemonic devices. The point is that recoding is an extremely powerful weapon for increasing the amount of information that we can deal with". Miller, G. A. (1956), The Magical Number Seven, Plus or Minus Two: Some Limits on our Capacity for Processing Information. Psychological Review, 63, 81-97. ] IMAGE SCHEMAS: repr events in a simple, abstract, spatial form ANALOGICAL REASONING: open/closed - as in hand --> apply to eyes - starts in infancy --> leads to metaphor [conceptual blending]
ch 6 Differences between percepts and concepts [Rosch:1973] work on categories as prototypes, based on Wittgenstein's famous notion of family resemblance, was an important corrective to the prevailing rigic views of categorization based on logical classes... One of the ideas though had a somewhat pernicious influence. --> notion that a particular level of categorization is primary or fundamental - a level that came to be known as the basic level. By the 1980s : ingrained that first concepts to be learned - basic level. ...not only do we not know what the term means but also, given any more or less understandable definition, the basic level is not the way to characterize the first concepts. 122 Wings, feathers, and beaks are not distributed randomly but instead form a cluster correlated with birds. Basic-level concptual cuts were assumed to be created at these discontinuities in the world, forming a unique level of abstraction. This level was said to carry the most information and to possess the highest "cue validity" - defined as the extent to which an attribute predicts a particular category - e.g. wings --> birds but not butterflies. 122 Failure of cue validity: if a cue predicts class X (dog), it also predicts the more inclusive (superordinate) class (mammal). Thus more inclusive categories always have a cue validity as great or greater than their subordinates. Rosch, 76 footnote notes this difficulty and says that concept cue validity refers to a psychological factor. [Murphy 82]: same problem if you use avg of features instead of sum [Medin 83]: instead of cue validity (cue => concept), use concept validity (concept => cue), but here too as many probs - lowest level subordinates are most predictive. [Tanaka/Taylor:1991][Mandler:1997] this level has never been unambiguously defined. 124-5 [Wendell Garner 1974] showed that verbal descriptions are highly dependent on the context of the surrounding items. E.g. a square drawn on paper with black ink is simply a "square". But in the context of red ink squares, it is apt to be described as a "black square", etc. 126 [Rosch etal 76] picture recognition is primed by words at the basic or subordinate level but not at the superordinate level. 127 concluded that the basic level is the highest level at which an image of an object can be formed - but not based on what details people may have imaged. For myself, when asked to image a achair, I image a straight backed chair with no arms - i.e. a subordinate category. Unless probed, I wouldn't say I was imaging an "armless dining chair". It is certainly true that terms like dog and chair are more frequent in early vocabularies than animal or furniture, which are often rather late. The q is how one should interpret this fact. Roger Brown 1958: How shall a Thing be Called? We use _dime for a particular coin, rather than "coin" or "1958 dime". 130 children's first words are taken from the vocab they hear - if these match their level of thought, so much the better - but we cannot use their first words to tell us about their concepts ... OVEREXTENSION [Clark 83, Rescorla 80] upto 40% of first 100 words are overextended - often "dog" refers to other animals - meaning more like "land animal" than "dog". Underextensions occur too, but they appear to be rarer [but may be because] it is easier to measure overextension than underextension in production. [AM: Child sees a new type of dog, never labels it - can't measure] If basic level terms match to their concepts, then one would not expect overextensions. [But] Children's early vocab much smaller than their comm needs - so one might well expect overextensions - this would tell little about their underlying concepts. 131 [McDonough 2002a] compared both comprehension and production and found that even at 2 years, 29% of both production and comprehension was overextended in the domains of animals, food, clothes, and vehicles. [See chap 11] Young children appear to be unclear about the boundaries of many basic-level terms - have no detailed specifications. So they have a double problems - must figure out the extension of the word being used in the framework of not being sure what the differenes are that make parents and others use different words -- not an easy task! 131 Rosch expts on child learning - that basic levels are learned first - confounds best prototypes of the superodinate class - mixing the prototype with levels - when separated [Bauer/Mandler] these expts don't hold up. 133 Rosch tended to emphasize phys similarity, particularly shape - so perhaps what was being measured was some level of perceptual categorization. 135 Following [Miller/Johnson-Laird 1981], [Smith/Medin:1981] proposed that concepts have a core meaning w nec and sufficient features. 135 In large part because of the work of Quinn/Eimas+ we have begun to learn a good eal about just how early young infants begin to form real-world perceptual categories. [Quinn/Eimas/Rosenkrantz:1993] 3-month olds learned to categorize real pictures of dogs and cats in as few as six 20-second trials. 136 Quinn etal use the preferential looking paradigm to show that infants are able to distinguish dogs and cats. {they look longer at a dog shown after a cat rather than after a series of dogs). Interestingly, the result is asymmetric - they are better at categorizing dogs after seeing cats than the other way around - possibly because dogs are more variable. See computational analysis of the phenomenon: Gupta et al 2011: A neuroplausible computational model of vision Asymmetry: Infants categorized better when familiarized (first) with cats than with dogs. [Quinn:93+] speculated that this was because dogs are more variable in appearance - but not so cats. A dog makes a bad cat. Confirmed further in work of [Mareschal/French/Quinn:2000] [Eimas/Quinn:1994] 3-mo olds distinguish horses as diff from zebra or giraffe, and cats as diff from tigers, though not lions (until 6 mos). [Behl-Chadha:1995] also show them being able to dist tables, chairs and beds. [Eimas/Quinn:1994] Eimas, P.D., and Quinn, P.C., Studies on the formation of perceptually based basic-level categories in young infants. Child Development, 65, 903-917. [Behl-Chadha:1996] Behl-Chadha, G. (1996). Basic-level and superordinate-like categorical representations in infancy. Cognition, 60, 105-141. The view of Ed Wasserman, who claims that pigeons can form abstract concepts, epitomizes the problem of not distinguishing perceptual and conceptual accomplishments. It has been known for some time that pigeons, which have a highly developed visual system, are capable of forming abstract categories such as trees. [Herrnstein:1979] [Wasserman:1995] Perceptual categorization, which pulls out the principal factors in patterns, is often abstract. 3-month old infants categorize biological motion, which is abstract, to say tht least. Wasserman: pigeons can differentiate displays with 16 pictures. A glance at the two kinds of array shows a striking perceptual difference. I'm not sure exactly how to describe it (because it is not well conceptualized!). 138 [IDEA:*** That we are not capable of describing it only says it is not linguistically conventionalized; but more likely what Mandler means is that the distinction is not conscious. This may then be a diff aspect from being conceptualized. We may also be able to describe distinctions linguistically - e.g. a man's walk vs a woman's, which we are not consciously able to analyze. So inability to describe is perhaps not a relevant criteria. ] [Smith/Carey/Wiser:1985] C. Smith, S. Carey, and M. Wiser On differentiation: A case study of the development of the concepts of size, weight, and density. Cognition. 21:177-137 --> scientists of the 18th c did not distinguish heat and temperature 142 To build a baby, we need more than one kind of categorization.
7 Some pre-verbal concepts Categorization is not governed by perceptual factors alone: 7-11-mo infants do not distinguish plastic models of dogs from fish, which, perceptually appear quite different. Figs. 7-2 154 later: Begin to dist by 1.5 years - possibly because fish were unknown - but birds and fish they distinguish. But 9-11-mos do distinguish birds and airplanes, which are perceptually rather similar Fig 7-3 155 So - although within category similarity (dogs, rabbits) is high, and so is between category dissimilarity, it is not enough. Sabina Pauen: 10-11-mo: systematically varied between-category similarity of little models of animals and furniture. In low-sim conditions, the items were realistic and varied greatly in shape, colour and texture. In high-similarity conditions, each had legs, curved as well as rectilinear parts, and b/w dots that could be interpreted as eyes on animals and knobs or decoration on furnitutre. Also furniture items were designed to resemble the overall shape of one of the animals, same colours as another animal, same texture as some other, etc. Even when there was high between-categ similarity, 10- and 11-mos distinguished the items as much as when between-categ sim was less. 155-56 [Pauen:2002] Sabina Pauen. Evidence for knowledge based categrory discrimination in infancy. Child Dev. 73, 1016-1033. CLAIM: first learn "broad, global" domains (similar to superordinate), before learning more detailed ones. 156 to some degree, "basic-level" distinctions are more likely to be discriminated on picture-looking tests than on object-examination tests, because, among other things, they grab the attention of the infant more actively, thus firing the attentional system, which need not be the case for learning purely perceptual schema or category (chapt 3 [53]). 162 [In this, it is quite likely that a greater degree of attention inheres to processes that are declaratively learned - but still not quite a complete description... ] Adults can attend to a location without looking [Posner:1988]... attentive looking is only a portion of looking time. 163 --> I guess what is meant is "consciously looking"
11 Language Acquisition I used to assume that learning names for objects might be possible on the basis of associative learning without having to call on the meaning of the objects in question. The assumption was that it is relatively easy to learn names for things because one can point to them, and so there is the opportunity for ostensive learning of a kind that does not exist for relations. One can point to a dog but can't point unambiguously to containment and can't point at all to the past. For nouns, the story goes, all one needs is something like [Markman:1991]'s whole object assumption, and the child should be in business. I have gradually come to understand that this enticingly simple procedure is insufficient. Words do double duty, having both sense and reference [Frege:1952] child words overextended, but do not cross global domain boundaries [Gelman, Croft, Fu etal 98] OVEREXTENSION: Studies on Comprehension [Mcdonough:2002b] Comprehension tests with detractor items drawn from the same superordinate class - shows that even at age 2 years, boundaries of animal, vehicle, food and clothing categs unclear. Overextensions upto 29% within domain, but rare across domain bdries, in both comprehension and production tests. E.g. asked to point to dog, first point to dog, then also include fox. Same error when asked to point to fox. It seems that 2-year-olds know what a typical dog and a typical fox looks like - and tend to pick a prototypical example first. Similarly planes and rockets; cakes and pies. [FN: little compunction abt using more than one basic-level label for an item. Markman:1991's hypothesis of mutual exclusivity of word meaning may be induced from experience w lg, rather than being an initial hypothesis ( [Nelson:1988], but see [Woodward/Markman:1998] for counterarguments) Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: MIT Press. Carey, S. (2000). The origin of concepts. Cognition and Development, 1, 37-42. Carey/Spelke:1994 Carey, S., & Spelke, E. (1994). Mapping the mind: Domain specificity in cognition and culture. In L. A. Hirschfeld & S. Gelman (Eds.), (p. 169-200). New York, NY: Cambridge University Press. Carey, S., & Spelke, E. (1996). Science and core knowledge. Journal of Philosophy of Science, 63, 515-533. Fauconnier /Turner Johnson, M. (1987). The body in the mind: The bodily basis of meaning, imagination, and reasoning. Chicago: University of Chicago Press. L. McDonough et al. / Cognitive Psychology 46 (2003) 229-259 257 Lakoff, G. (1987). Women, fire, and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press. Mandler, J. (2000). What global-before-basic trend? Commentary on perceptually based approaches to early categorization. Infancy, 1(1), 99-110. Mandler, J. (2002). On the foundations of the semantic system. In E. M. Forde & G. Humphreys (Eds.), Category specificity in mind and brain. East Sussex, U.K.: Psychology Press. (To appear in:) Mandler, J. M. (1988). How to build a baby: On the development of an accessible representational system. Cognitive Development, 3, 113-136. Mandler, J. M. (1990). From perception to conception. In P. van Geert & L. Mos (Eds.), Developmental psychology. New York: Plenum. Mandler, J. M. (1992). How to build a baby II: Conceptual primitives. Psychological Review, 99(4), 587-604. Mandler, J. M. (1997). Representation. In D. Kuhn & R. Siegler (Eds.), Cognition, perception, and language (Vol. 2). New York, NY: Wiley. Mandler, J. M. (2000). Perceptual and conceptual processes in infancy. Journal of Cognition and Development, 1, 3-36. Mandler, J. M., & Bauer, P. J. (1988). The cradle of categorization: Is the basic level basic? Cognitive Development, 3, 247-264. Mandler, J. M., Bauer, P. J., & McDonough, L. (1991). Separating the sheep from the goats: Differentiating global categories. Cognitive Psychology, 23, 263-298. Mandler, J. M., & McDonough, L. (1993). Concept formation in infancy. Cognitive Development, 8, 291-318. Mandler, J. M., & McDonough, L. (1996). Drinking and driving don't mix: Inductive generalization in infancy. Cognition, 59, 307-355. Mareschal, D. (2000). Infant object knowledge: Current trends and controversies. Trends in Cognitive Science, 4, 408-416. Mareschal, D., French, R. M., & Quinn, P. C. (2000). A connectionist account of asymmetric category learning in early infancy. Developmental Psychology, 36(5), 635-645. Mandler, J. M. (1992). How to build a baby: II. Conceptual primitives. Psychological Review, 99, 587-604. Mandler, J. M. (1996). Preverbal representation and language. In P. Bloom, M. Peterson, L. Nadel, & M. Garrett (Eds.), Language and space (pp. 365-384). Cambridge, MA: MIT Press. Mandler, J. M., & McDonough, L. (1996). Drinking and driving don't mix: Inductive generalization in infancy. Cognition, 59, 307-335. Mandler, J. M., & McDonough, L. (1998). On developing a knowledge base in infancy. Developmental Psychology, 34, 1274-1288. Mandler, J. M., & McDonough, L. (2000). Advancing downward to the basic level. Journal of Cognition and Development, 1, 379-403. McDonough, L., & Mandler, J. M. (1998). Inductive generalization in 9- and 11-month olds. Developmental Science, 1, 227-232. Nelson, K. (1974). Concept, word, and sentence: Interrelations in acquisition and development. Psychological Review, 81, 267-285. Quinn, P. C. (1999). Development of recognition and categorization of objects and their spatial relations in young infants. In L. Balter, & C. S. Tamis-LeMonde (Eds.), Child psychology: A handbook of contemporary issues (pp. 85-115). Philadelphia, PA: Psychology Press/Taylor and Francis. Quinn, P. C., Cummins, M., Kase, J., Maartin, E., & Weissman, S. (1996). Development of categorical representations for above and below spatial relations in 3- to 7-month-old infants. Developmental Psychology, 32, 942-950.
bookexcerptise is maintained by a small group of editors. get in touch with us! bookexcerptise [at] gmail [dot] .com. This review by Amit Mukerjee was last updated on : 2015 Aug 18