Summary Cognitive Psychology (c) 2001-06-07 F.M.Suchanek http://www.mpi-inf.mpg.de/~suchanek/personal/texts/summaries/cogpsy.txt This is a summary of the Cognitive Psychology lecture held by Prof. Franz Schmalhofer in the summer semester 2001 at the University of Osnabrueck. The course was based on "Foundations of Cognitive Science" (see FOC.TXT) and used the book "Cognitive Psychology -- A Student's Handbook" (revealing a strange idea of the size of a student's hand...). By reading the following text, you accept that the author does not accept any responsibility for the correctness or completeness of this text. If you have any corrections or remarks, please send me a mail. This is the only way to make the publication of this summary useful for me, too. My e-mail address is f.m.suchanek@zweb.de, but the letter 'z' has to be removed from the address. This text is divided into sections and each section contains small blocks. Each block has a headline stating the name of the block and the scientists involved. I have unified the blocks such that two types are predominant: theory ( ): Theory: Evidence: Counter-evidence: ... [self-] experiment ( ): Task: Result: Conclusion: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ INTRODUCTION ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 4 major approaches within CogPsy * Experimental CogPsy (does memory experiments) * Cognitive Science (develops computational models) * Cognitive Neuropsychology (works with brain damaged patients) * Cognitive Neuroscience (Neurobiology) Everyday-memory Research (Koriat & Goldsmith 1996): ... studies the performance of memory in everyday situations, in natural settings. Merely analizes the correspondance between reports and actual events. Content plays a role. Accuracy is not the main goal. Everyday memory is purposeful, has a personal quality about it and is influenced by situational demands. Traditional memory Research: ... studies memory in psychologist laboratories, measures the number of items recallable. Relies on the storehouse metaphor. Noetic awareness: ... is the faculty of thinking objectively about something one knows. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ FORGETTING ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Ebbinghaus (1850-1909): One of the first Cognitive Psychologists, used himself as a subject. Tested memory by memorizing meaningless material (CVC-triagrams, i.e. Consonant-Vowel-Consonant-Strings) and measuring the repetitions needed to reproduce it. Reduced-Savings Self-Experiment (Ebbinghaus ca. 1900): Task: Memorize meaningless material, reproduce it after a retention interval. Result: Performance becomes worse whith larger intervals Conclusion: Constant logarithmic function of forgetting. 2 theories of forgetting: * Trace decay theory: Memory fades away when time passes, not probable * Interference theory: Memory is negatively affected by other memory contents, verified Trace-Decay Experiment (Jenkins & Dallenbach 1924): Task: Memorize sth. and reproduce it after a retention interval. Subjects are either awake or asleep during the interval. Result: Subjects who were asleep performed better Conclusion: The experiment was to verify the trace-decay theory but rather supports the interference-theory. Interference: * proactive interference: previous learning negatively affects future learning (forward direction). * retroactive interference: learning negatively affects things learned previously (backward direction). Interference strength: * 2 responses associated with one stimulus --> strong interference * 2 similar responses associated with 1 stimulus --> weak interference * 2 stimuli associated with 1 response --> hardly any interference Two-process theory (Watkins 1979): Theory: Recall consists of search (retrieval) and decision (recognition). When asked to reproduce sth., people first generate items at random and then try to recognize the correct ones. Evidence: Probability-Experiment Counter-Evidence: Recall&Recognition Experiment Probability-Experiment (Bahrick 1970): Task: Memorize items, then either reproduce or recognize them Result: The probability of correct recall could be predicted by multiplying the probability of retrieval by the probability of recall, i.e. P(recall)=P(retrieval)*P(recognition). Conclusion: Recall consists of retrieval and recognition Recall&Recognition experiment (Tulving & Thomson 1973): Task: 1. Study a list of word pairs (A1,B2),(A2,B2),... 2. Freely associate words D1,D2... to given cues C1,C2,... 3. Try to remember whether the new words you generated (D1,D2,...) have been on the first list (B1,B2,...) 4. Try to recall B1,B2,... in the presence of A1,A2,... Result: Subjects perfom better on 4 (recalling list words) than on 3 (recognizing list words). Conclusion: Recall can be better than recognition, counter-evidence for the two-process theory. Encoding specificity theory (context effects theory) (Tulving 1982): Theory: The perfomance on recall tasks is better when the encoding conditions (during learning) are similar to the reproduction conditions. The to-be-remembered item and contextual factors are stored together in memory. Recognition is only affected by intrinsic context, i.e. context directly connected to the to-be-remembered item, wheras recall is affected by both extrinsic and intrinsic context. Evidence: Context effects experiments Context effects experiment (Tulving 1982): Task: Subjects study a list of cue-item pairs. One group gets a list with weak associative cues, the other group gets a list with strong associative cues. Then the items are to be reproduced in the presence of the cue. Result: Each group performs best with its specific cues. Conclusion: Support for the Encoding Specificity theory Context effects in recall & recognition exp. (Godden & Baddeley 1980): Task: Subjects have to learn words either in a wet environment or in a dry environment. Afterwards, they have to recall or recognize the words in a dry or wet environment. Result: Recall was better when the environment of study and test was the same. Recognition was not affected. Conclusion: Support for the Encoding Specificity theory "2 reasons for forgetting" theory (Tulving 1974): Theory: There is trace-dependent forgetting, where the information is no longer stored in memory and a cue-dependent forgetting, where the information is in memory but cannot be accessed. Evidence: Tulving's own experiments What can be memorized well: * Deeper levels of analysis produce more elaborate, longer lasting and stronger memory traces * Unique memory traces are more readily retrieved * continuos motor skills are only forgotten very slowly * recall on a list is better when _no_ starting-help is given, since this disrupts the participants' normal search processes (part-list cueing effect) * Students with high domain knowledge study better with incoherent texts which force them to work on their own while reading * Distributed practise produces better results than massed practise * Things that concern the subject's personality are remembered best (self-reference effect, Rogers et al. 1977) * Things should be encoded meaningful * Retrieval structures may help * peg systems (attaching items to well remembered items) help * SQ3R-technique (Survey, Question, Read, Recite, Review) helps ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ SUBDIVIDING MEMORY ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ A memory system (Masson & Graf 1993): ... is a collection of correlated functions served by different anatomical structures in the brain. Memory stores: * Sensory stores, modality-specific ultra-short-term memories * Short term memory with a limited capacity * Long term memory with an unlimited capacity Partial report experiment (Sperling 1965): Task: An array of 12 numbers is shown and is to be remembered. Then the numbers are recalled. Although participants claimed to have seen more numbers, the result was poor. Hence Sperling asked the participants to recall only parts of the array and the result was improved. Result: Visual information fades during report. Conclusion: The "iconic storage" (video buffer) can hold visual information for about 0.5 seconds. Shadow-message experiment (Treisman 1964): Task: Participants had to shadow (repeat back aloud) a message and ignore a second, identical message presented to the other ear. Result: When the second message preceded the shadowed one, the two messages were only recognized as being the same when they were within 2 seconds of each other. Conclusion: We have an "echoic storage" (audio buffer) of about 2 seconds. Working memory theory (Baddeley and Hitch 1974): Theory: The working memory consists of a central executive, a phonological loop (audio buffer) and a visual-spatial sketch-pad (video buffer). Two tasks can only be perfomed together when they use different components. Evidence: Word-length experiment The phonological loop: ... consists of a passive phonological store (speech perception) and an articulatory process (speech production). Visually presented words enter the phonological loop by subvocalization. All speech necessarily enters the loop. Similar phonological data can hardly be distinguished. The loop is of use for the learning of new words. Rehearsal uses the phonological loop. The visual-spatial sketch-pad: ... can be subdivided into a visual cache (stores form and color information) and the inner scribe (spatial and movement information, rehearses and transfers information). The maintenance of spatial information is not disrupted by a concurrent visual task. The central executive: ... switches retrieval plans, does time-sharing, controls selective attention and activates long-term memory. May consist of different systems. Random generation experiment (Baddeley 1996): Task: Participants had to generate random sequences of numbers and at the same time recite the alphabet. Result: The randomness decreased as the digit memory load increases. Randomness was also reduced by the alphabet task. Conclusion: The taskswitching overtaxes the central executive so that it does not produce good random numbers. However, random number generation does not only depend on the central executive but can also be supported by visibility of the alternatives (Towse 1998). Word-length experiment (Baddeley et al. 1675): Task: Reproduce a sequence of words. Result: The string reproduced was 2 seconds long. If the phonological loop was used for another task, participants had difficulties in reproducing the words. Conclusion: The capacity of the phonological loop is 2 seconds. Episodic and semantic memory (Tulving 1972): Tulving differentiates between memory of things happened (episodic memory) and general knowledge (semantic memory). Differences between the systems: Episodic memory Semantic memory ~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~ Acquisition one trial repeated trials Personal inv. yes no Contents event+context pure world knowledge Retrieval effortful effortless Orientation past present & future Stage in Dev. late early Implict and explicit memory (Graf & Schachter 1985): Implicit memory is memory gained in the abscence of conscious recollection (results in repetition priming effect, tested by fragment completion). Explicit memory is gained with the aim of reproducing it (tested by recognition & recall tasks). Explicit and implicit memory seem to have separate underlying brain systems. Differences between the systems: Explicit memory Implicit memory ~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~ Acquisition one-trial procedural learning Modalities multi-modal only one modality Reliability unreliable reliable Consciousness conscious unconscious Stage in Dev. late early Transfer appropriate processing theory (Roedinger 1990): Theory: There are two types of cognitive processes: * Data-driven, perceptually driven processes, which often underlie performance on tests of implicit memory * Conceptually driven processes, which often underlie performance on tests of explicit memory When processes used at test are those used at study, performance will be better. Evidence: Experiments by Jacoby (1983) Counter-evidence: Experiments by Mulligan (1998) Declarative & procedural memory (Cohen & Squire 1980 & Anderson 1976): Declarative memory is explicitly accessible and and verbalizeable while procedural memory is not. Inclusion and excl. exp. (process-dissociation proc.) (Jacoby 1993): Task: A list of words is studied. Then the subjects have to complete word stems in a first list with the words learned and in another list with avoiding the words learned. Result: Particpants with good explicit memory would fill the first list with nearly 100% of the original words and the second with nearly 0% of them. Conclusion: Contributions of implicit and explict memory can be measured. "Different implicit tests" theory (Tulving & Schacter 1990): Theory: There are two different implicit memory test strategies: * Perceptual implict tests, where the stimulus presented at study is similarly presented at test (e.g. fragment completion) * Conceptual implicit tests, where the tests provides information which is only conceptually related to the studied information Evidence: From experiments with Alzheimer's patients Toboggan experiment (Tulving 1982): Task: Subjects have to study a list of rare words (like "toboggan"). After a retention interval, they have to complete word stems which matched the words learned previously (test of implicit memory). Result: Implicit memory is hardly affected by the time while explict memory is. Divided Attention experiment (Gardiner & Parkin 1990): Task: Subjects have to memorize something either in a noisy environment or in a calm environment. Then the subjects have to do fragment completion to test what they remebered. Result: Implicit memory is not disturbed by divided attention while explicit memory is. Conclusion: Implicit memory exists independently form explicit memory. Amnestics implicit memory experiment (Graf 1984): Tasks: Amnestics and a normal control group have to remember sth. Result: Amnestics perform worse than the control group on explicit memory tasks but equally well on implicit memory tasks. Conclusion: Explicit memory is impaired separately. Division of memory by contents (Squire): * declarative memory (explicit memory) * episodic memory (events) * semantic memory (facts) * non-declarative memory (implicit memory) * procedural learning * priming * classical conditioning (Pawlow's Bello) * non-associative learning Hippocampus: Region in the telencephalon, manages transfer from short term memory to long term memory. It does not _contain_ long term memory. HERA-model: Encoding and retrieval stimulates assymetric structures in the human brain. The following parts of the prefrontal cortex are involved most: Episodic inf. Semantic inf. ~~~~~~~~~~~~~ ~~~~~~~~~~~~~ Encoding left side left side Retrieval RIGHT side left side ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ FALSE MEMORIES ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Memory distorting experiment (Loftus & Palmer 1974): Task: The subjects watch a film about a car accident. A week later, they are asked in different ways whether they saw broken glass. Result: The way the question is asked can trigger false answers. Conclusion: Eyewitness testimony is questionable. Misinformation experiment (Lindsay 1990): Task: Subjects had to watch a film and then listened to a false summary of it. Then they were told that the summary was false and had to remember the film. Result: Memory was nevertheless affected by the false summary. Conclusion: Subjects cannot differentiate where they heard what (source misattribution) and accept misinformation. Source misattribution is worse when the information is similar in both sources. Misinformation & individuality experiment (Tomes & Katz 1997): Task: Misinformation experiment, followed by an analysis of peoples' personality. Result: People accepting misinformation tend to have * poor general memory * good imaginery vividness * good empathy Verbal overshadowing experiment (Schooler 1990): Task: Subjects watch a film and either provide a detailed report on it or not. Then they are asked to recognize an actor's face. Result: Those who did not do the verbal report perfomed better Conclusion: Verbal reports interfered with the visual recollection ("verbal overshadowing of visual memories" phenomenon). True and false memories: It cannot yet be distinguished whether a memory for an event is real (i.e. has been repressed and now recoverd) or is false (i.e. has been created by suggestibility). Amnesia: * Amnestics often have widespread brain damage, which complicates making theories (and the patients' life, BTW) * Most common etiology: Hippocampus dysfunction (i.e. the transfer from short term memory to long term memory is blocked) * It is difficult to identify subtypes of amnesia * Retrieval probs are greater for memories acquired closer to the onset of amnesia than those acquired longer ago * Amnestics have an intact short term memory but an impaired long term memory * Amnestics have a normal learning rate at sensori-motor skills * Amnestics show repetition-priming effects * Amnestics have a context processing deficit * Amnestics have an impaired explicit/declarative memory * Amnestics have difficulties in storing integrated information * Alzheimer's patients have difficulties in task switching ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ AUTOBIOGRAPHICAL MEMORY ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Autobiographical memory: ... is the memory of one's own life, defines one's identity. It has three levels, each of which forms a part of the larger one: * Life-time periods (ca. >1 year), less vulnerable memory * General events (days and months) * event-specific knowledge (feelings &c, seconds to hours) Autobiographical memory is reconstructed rather than reproduced. Locations are remembered best, followed by actions and thoughts. Autobiographical memory is organized thematically. Recollections are rather broadly true than strictly accurate. Autobiographic memory may be euphemistic. Lifetime memory experiment (Rubin et al. 1986): Task: 70-year-olds are asked to think of personal lifetime memories. Result: Many events are remembered from age 5-30, less from 30-60 (reminiscence bump) and most from age 60-70. Conclusion: Novelty and stability (both present from age 5-30) produce best memories. Infantile amnesia explains the abscence of memory from age 0-5. Recency is the reason for good memory from age 60-70. Diary self-experiment (Wagenaar 1986): Task: Wagenaar recorded >2000 events of 6 years of his live. He then tried to remember them with cues. Result: The cues were of different usability: What-cues > Where-cues > Who-cues > When-cues. Furthermore, events with high levels of salience, emotional envolvement, pleasantness or rarity were recalled best. Diary self-experiment (Linton 1975): Task: Linton recorded two events each day and tried to recall some at random each month. Results: * If an event has already been tested, it was recalled more readily * similar events were forgotten * semantic memory/general knowledge increased whereas episodic memory/specific event memory decreased Dating events experiment (Brown et al. 1985): Task: People were asked to date events they remember Result: Events with low knowledge were dated as too remote Conclusion: When asked to date an event, we estimate the time by concerning the knowledge we have about the event. Furthermore, we use "landmarks" (e.g. lifetime periods) to determine the time of an event. Flashbulb memory theory (Brown & Kulik 1977): Theory: Memories of dramatic events (Flashbulb memories) are different from other memories in their longevity, accuracy and a special neural mechanism. Counter-Evidence: See evidence for Non-Flashbulb theory Non-Flashbulb memory theory (Finkenhauer et al. 1998): Theory: Flashbulb memories are the result of ordinary memory mechanisms, but the encoding was more efficient. Evidence: Flashbulb memories are forgotten like other memories and may be inaccurate. Prospective memory: ...means remembering to carry out intended actions. * Time-based prospective memory (remembering to do sth at a particular time) is usually difficult * Event-based prospective memory (remembering to do sth in the appropriate circumstances) is less thought about but better Prospective remembering is a "When"-question and thus less informative and more difficult than retrospective remembering ("What"-question). Both memory abilities are unrelated. Sth enjoyable is remembered better :-). Prospective memory is likely to be best for plans that are routine, high in priority and related to a plan network. Prospective memory is impaired by using the central executive since prospective memory involves conceptually driven and attentional processes. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ INTERNAL REPRESENTATION ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Different external represenations: * written/linguistic representation: symbols are arbitrarily realated to the objects they stand for. * graphical representation: has a closer relationship to reality For further differences see internal rep's Different internal representations: * Propositional representation: discrete, explicit, amodal (not modality dependent) and abstract. It consists of "propositions" * Analogical representation is continuous, implicit, vague and modality specific. It consists e.g. of a picture. * sub-symbolic representation in connectionist networks Concepts: Concepts are abstract representation of objects (object concepts) or actions/predicates/verbs (relational concepts). A combination of both results in a schematic representation of an event. Propositions: The definition of "proposition" in CogSci is a product of interdisciplinary research. A proposition... * is the meaning of a sentence * is the representation of an idea * represents conceptual relations between objects * is encoded in a fundamental (amodal) code * is written in technical notation, employing aspects of predicate logic Predicates: First-order predicates are like functions from objects to truth-values, e.g. hit(mary,franzerl). Second-order-predicates can get other functions as arguments, e.g. cause(hit(mary,franzerl),ishurt(franzerl)). Propositional representation theory (Pylyshyn 1973): Theory: Knowledge is represented by abstract, mental, conceptual and propositional structures. It is not sensory or pictorial. Evidence: Proposition test experiment, Reading time experiment, Description Experiment, Reinterpretation experiment Counter-evidence: Mental rotation experiment, Image scanning experiment. Proposition test experiment (Bransford & Franks 1971): Task: Subjects have to remember sentences. Result: Subjects remember very well the propositions studied but not their exact form. Conclusion: Sentences are stored independent of their form (i.e. as propositions) in memory. Reading time experiment (Kintsch & Keenan 1973): Task: Subjects had to read a sentence. The time needed is measured. Result: The reading time increased proportional to the number of propositions in that sentence. Conclusion: Sentences are divided up into propositions. Description experiment (Carmichael et al. 1932): Task: Schematic pictures with small descriptions are to be memorized. Two groups get the same pictures but different descriptions. Then the pictures are to be reproduced. Result: The reproduction depends on the previous description. Conclusion: The pictures were not stored as pure pictures in memory. Reinterpretation experiment (Chambers & Reisenberg 1985): Task: Subjects were shown a figure which could be interpretated in different ways. Then the figure was taken away and the subjects were asked to find out another interpretation of the picture. Result: People were not able to do this. Conclusion: The image was not stored as an image but as information about what it depicted. Semantic decomposition theory (Schank 1972): Theory: All actions can be explained by a combination of 15 basic verbs. Counter-Evidence: Studies failed to prove Schank's theory Script theory (Schank & Abelson 1977): Theory: Standard action sequences (e.g. "eating at a restaurant") are stored as "scripts" ("schemata") in memory. Evidence: Theory makes good predictions about implicit inferences. Counter-evidence: Scripts are not flexible enough, schemata may interfere. Psychological counter-evidence. Connectionist theory (McClelland & Rumelhart 1986): Theory: Content is stored subsymbolically in memory by distributed representations (neural networks). Evidence: Explains the other theories and lacks contradictions, explains internal representation, is less complicated than exact symbolic descriptions, explains why any part of the memory can reinstate all of the original memory, explains automatic generalisation. Distributed versus local representations: In a distributed representation, the units represent small feature- like entities and the pattern as a whole is the meaningful unit. In a local representation, there is a one-unit-one-concept implementation. Slot Interdependency theory (Rumelhard & Ortony 1977): Theory: There is an interdependency among possible slot-fillers, i.e. as soon as one variable is set, the default values for the others are changed as well. Mental rotation experiment (Cooper & Shepard 1973): Task: Participants had to decide whether a letter (char) is correct or mirror-reversed. The figures were shown in a rotated position. Result: The time needed to make the decision increased with the angel of the figure (180 degrees = maximal time). Conclusions: Participants did mental rotation, counter-evidence for the propositional representation theory Image scanning experiment (Kosslyn 1983): Task: Participants studied a map of an island. They then had to imagine a flight from one location on the island to another. Result: the time needed was proportional to the distance. Conclusion: Participants stored the map as a picture in memory. Problem: In these experiments, subjects are instructed to behave as if they were seeing something real in the outside world. Neuropsychology on imagery: The imagery generation component seems to be a left hemisphere function whereas the right hemisphere has all other imagery components. No-answer theory (Anderson 1976): Theory: The question about internal representation cannot be answered. Evidence: See current research Mud theory (Newell 1973): Theory: Clarity is never achieved, matters simply become muddier and muddier as we go down through time. Evidence: See current research Internal representation functions: Encoding inputs to internal structures and decoding internal structures to outputs can be seen as functions: InternalStructure = Encoding(Input) Output = Decoding(InternalStructure) Problem: Once we have a theory of Encoding(x) and Decoding(x) and and Encoding2(x), we can compose an Decoding2(x) such that the new pair of functions mimicks exactly the first theory. Imagery & perception theory: Theory: Imagery processes have similar properties to perceptual processes. Dual coding theory (Paivio 1971): Theory: There are two separate systems in our brain. The verbal system deals with and stores linguistic information. The non- verbal system carries out image-based processing and representation. Both systems are connected to distinct input and output systems. Both systems can be subdivided into modality-specific subsystems (although smell and taste have no corresponding units in the verbal system). "Imagens" represent images and "logogens" represent words. Imagens and logogens are referentially connected. Spatial processes are shared by perceptual and image-based processing in the non-verbal system. Evidence: Picture remembering experiment, concrete word remembering experiment, neurophysiological evidence (left and right hemisphere of the brain) Counter-evidence: Imagery instruction experiment (1970) Picture remembering experiment (Paivio 1971): Task: Participants either have to remember pictures or words and afterwards recognize or recall them. Result: Both recall and recognition are better with picutures. Conclusion: Pictures are named spontaneously while memorizing them. They were thus stored in both the verbal and the non-verbal system. Concrete word remembering experiment (Paivio 1968): Task: Participants have to remember words and then reproduce them. Result: Recall was much better on concrete words than on abstract words. Conclusion: Concrete words are more likely to produce images and thus get encoded in both the verbal and non-verbal system. Imagery instruction experiment (Bower 1970): Task: Participants have to remember pairs of concrete words. Different groups are given different instructions: interactive-imagery, separation-imagery (i.e. construct an image of two objects separated in space) or rote-memorization. Result: Participants with interactive-imaginery instructions performed better than those with separation-imagery instructions. Conclusion: Both instructions use the non-verbal system but perform unequally. Relational organization plays a role. Map experiment (Schmalhofer 2001 (?)): Task: A map of two countries with cities is studied. Then questions concerning the relative position of two cities are answered. Result: With incongruent borders between the countries, the error rate was much higher, i.e. when country A in general lies west to country B but the border is concave such that one city of A lies east to one city of B. Conclusion: We deduce knowledge about relational positions of small objects from knowledge about relational positions of large areas. Computational imagery theory (Kosslyn 1994): Theory: Visual images are represented in a spatial medium, which ressembles a TV screen. Its highest resolution is at the center. Once the image is generated, the medium begins to fade. Long term memory contains image-files and propositional files, which are often linked together. Images are constructed by imaging the outline and then adding components according to the propositional knowledge about the object. Imagery abilities are described by algorithms (like PICTURE, ROTATE, SCAN, TRANSFORM). Evidence: Image-tracing experiment, Mental walk experiment Image-tracing experiment (Kosslyn 1975): Task: Participants have to imagine a large and a small animal and are asked whether they can "see" the details of the small animal. Result: Participants zoomed in the picture to see the details. Conclusion: The imagery store works like a TV screen. Mental walk experiment (Kosslyn 1978): Task: Participants have to imagine a large object and then mentally walk towards it until it fills the whole "screen". Result: The distance to the point of overflow increases linearly with the object's size. Conclusion: The imagery store works like a TV screen. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ CATEGORIES ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Categorizing the world ... means dividing the world into concepts (classes). Each object is an instance (member) of a concept. Categorization obeys 3 principles: * cognitive economy: The goal is to reduce the amount of data to be stored * Informativeness: The structure should keep as much information as possible * Natural coherence: The categorization should accord to natural categories People classify things to make predictions about their attributes (Murphy & Ross 1994). Theories about concepts: * Defining attribute concept theory * Prototype concept theory * Instance-based concept theory * explanation-based concept theory Defining attribute concept theory (Frege 1952): Theory: * Concepts are defined by lists of necessary attributes. * Each concept has an intension (list of attributes) and an extension (set of members of this concept) * Concepts are not fuzzy. * All members are equally representative for the concept * Concepts are organized hierarchically and each subconcept inherits the attributes of the superconcept Counter-Evidence: Typicality experiment, Fuzzy category experiment, Superordinate experiment, Irrational typicality experiment, some concepts do not have defining attributes (like "game") Typicality experiment (Rosch 1973): Task: Answer questions concerning the attributes of certain objects. Result: The less typical an instance is for a concept, the longer it takes to answer the question. Conclusion: Counter-evidence for the Defining attribute concept theory Irrational typicality experiment (Armstrong & al. 1983): Task: Rate the typicality of an instance of a clearly defined class (e.g. the typicality of 18 being an odd number). Result: In spite of the clearly defined class, people assigned different typicalities to equally representative instances. Conclusion: This cannot be explained by the Defining attribute view (but who can explain this anyways??) Fuzzy category experiment (McCloskey & Glucksberg 1978): Task: Subjects had to tell the categories of objects. Result: Different people had different categories. Sometimes people had different views at different times. Conclusion: Concepts are fuzzy. Superordinate experiment (Smith, Shoben & Rips 1974): Task: people had to answer several questions of the form "Is a a ?". Result: Some questions with more distant superordinates were answered faster than others with closer superordinates. Conclusion: This cannot be explained by the Defining attribute theory. Prototype concept theory: Theory: * Concepts are fuzzy * Concepts are built around a certain instance (prototype) * Membership is determined by similarity to prototype * The typicality gradient of a member is a function of its similarity to the prototype Evidence: * Color experiment * The typicality gradient is a good predictor of the categorization time * Typical members are mentioned first in brainstorming tasks * Typical members serve as reference points in phrases like "An ellipse is almost a circle." Counter-evidence: * Abstract concepts have no prototype * People know about the relations between the attributes * No explanation for ad-hoc categories without a prototype Color experiment (Berlin & Kay 1969): Task: From a bag of colored chips, pick out the one which you think is most typical for the color "blue" (or any other of the 11 "basic" colors). Result: People all over the world selected the same chip. Conclusion: This color serves as a prototype for the color "blue". Instance-based/exemplar-based concept theory: Theory: * Categories are just a collection of instances * Categorization occurs by finding all instances matching a certain criterion * Instances are grouped by similarity metric Evidence: The theory lacks the disadvantages of the other theories, Variability experiment Counter-evidence: Irrational typicality experiment, no explanation for general knowledge Variability experiment (Rips & Collins 1993): Task: Judge whether an object of 19 inches in size is a pizza or a ruler. Result: People knew that rulers have a normal length of 12 inches but pizzas are variable in size. Hence they picked pizza. (Or maybe they just preferred the idea of having a 19 inch pizza to having a 19 inch ruler :-) Conclusion: The result can be explained by the Instance-based concept theory. Explanation-based concept theory: Theory: * Concepts can have attributes and explanatory relations between these attributes * Concepts can be dynamically created using background knowledge * Naturality emerges from the theoretical knowledge about concepts Evidence: Dissociation between similarity and categorization can be explained, it is known that background knowledge can influence the application and learning of categories, Cat&Sim experiment, Disjunctive concepts experiment Counter-evidence: none (?) Cat&Sim experiment (Rips 1989): Task: Subjects in the "similarity group" have to judge the similarity of a round object to either a coin or a pizza while subjects in the "categorization group" had to categorize the object into either the coin or pizza class. Result: Although the similarity group decided that the object was more similar to a coin, the categorization group categorized it as a pizza. Conclusion: Similarity and categorization differ. Disjunctive concepts experiment (Pazzani 1991): Task: People had to decide whether the balloon would inflate after having seen a picture with the balloon being involved in some action (being stretched e.g.). Result: People learned the disjunctive concept of "The balloon will inflate if it was either stretched OR an adult is involved." more easily than a conjunctive concept (which is normally learned more easily). Conclusion: Background knowledge about causal relations influences the learning of concepts. Three-level concept theory (Berlin 1972): Theory: Concepts are subdivided into three levels: * The superordinate level: very general categories * The basic level: main level at which categorization is carried out * The subordinate level: specifc types of objects Evidence: Three-level concept experiments Three-level concept experiments (Rosch 1976): Task: List all the attributes to given concepts of one of the three levels. (And other tasks) Result: People listed most attributes for the basic and subordinate level. The basic level is the most common level. It is the one at which people spontaneously name objects. It can be characterized by the motor movements normally associated to objects of this category (i.e. same movements with all instances of this class, e.g. "chair" -- "sit on"). A mental image can capture a basic level. Basic level objects are recognized fastest. It seems that a basic level category is the balance between informativeness and economy. Different personal experience entails different basic level categories. Conclusion: Support for the Three-level concept theory. Instable concept theory (Barsalou 1987): Theory: The way people represent concepts changes according to the context. Ad-hoc concept theory (Barsalou 1983): Theory: Some categories are not well established in memory but can be constructed by people to achive certain (categorization) goals. The associations between the objects and the concept are not fixed but can be created if required. Evidence: Barsalou's own experiments Conceptual combination: New concepts can be created by combining existing concepts (like "fake gun"). This can correspond to "multiple inheritance" in C++ (i.e. a class being the subclass of multiple superclasses) or the interface concept in Java. However, the above theories have difficulties in explaining these assembled concepts (no prototype etc.). Hampton (1983) suggested a "composite prototype", but many combinations are not intersective ("horse races" are not both "horses" and "races"). Complex, background-knowledge based mechanisms enable people to understand and produce combined concepts. Contrast theory of similarity (Tversky 1977): Theory: s(a,b) = theta*f(A&B)-alpha*f(A-B)-beta*f(B-A) * s(a,b) is the similarity of two objects a and b * theta, alpha and beta weight the importance of the factors * A and B are the sets of attributes of the objects a and b * A&B is the intersection of A and B, ie. the common attributes * A-B is the difference, i.e. the attributes distinctive to a * f is a function morphing a set to a value In slightly less obscure terms, the above formula says: When the number of common features increases and the number of distinct features decreases, the two objects become more similar. Evidence: Several experiments Similarity experiment (Tversky 1978): Task: One group had to judge the similarity of country A to country B. Another group had to judge the similarity of country B to country A. Result: Similarity is not commutative (s(a,b)<>s(b,a)). The small country was always more similar to the big country than vice versa. Conclusion: This can be taken into account by the weighting factors alpha and beta in Tversky's similarity formula. Relation similarity experiment (Medin & al. 1990): Task: Judge the similarity of a group of geometric objects to two other groups of geometric objects. Result: People considered relational attributes (e.g. all objects of one group are of the same color) more important than absolute attributes (e.g. two groups contain the same object). Conclusion: Similarity is sensitive to relations. Connectionist concept learning: Neural networks make good concept learners because they can learn from specific instances and implement similarity mechanisms automatically. IAC net (McClelland 1981): McClelland constructed a neural network and gave it attribute descriptions of the individuals in "West Side Story". Related attributes are grouped into "pools". Attributes and the corresponding person are linked excitatorily while the attributes among themselves are linked inhibitorily. To find out which of the persons satisfy a specific attribute, you "clamp" this attribute, i.e. you fix the activation of this node to "on". The system now runs thru activation cycles, performing the "excite" and "inhibit" effects throughout the net. In the end, it reaches a stable state and all nodes with the persons satisfying the specific attribute are "on". The net has answered a categorization question with just instance-based knowledge about the individuals. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ LANGUAGE ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Grammar (according to Altmann 1997): Both the speaker and the hearer share some common knowledge regarding the significance of one combination of words or another. This shared knowledge is grammar. Pragmatics: ... is the study of the intended meaning of a sentence. Parsing: ...means running through a sentence and assembling its grammatical structure. Serial parsing theory: Theory: One syntactic analysis of the sentence is chosen initially and replaced by another as soon as the first proves unworkable. Parallel parsing theory: Theory: Multiple syntactic analyses of a sentence are all considered together while parsing the sentence. Modular parsing theory: Theory: Lexical access strictly precedes parsing which in turn strictly precedes semantic processing. Interactive parsing theory: Theory: A single process combines lexical, semantical and world- knowledge contraints while reading a sentence. Garden-path parsing theory (Frazier & Rayner 1982): Theory: The simplest syntactical structure is chosen, making use of * Minimal attachment: Prefering the easiest grammatical structure * Late closure: New words are attached to the existing structure as long as possible (=gramatically permissible) Evidence: The well-known phrase with the horse, so far the only sentence with which this problem occurs at all. Counter-evidence: Context affects initial parsing decisions (bye, bye, horse!). Also, punctuation affects parsing decisions. Constraint-based parsing theory (MacDonald & al. 1994): Theory: All relevant sources of information or constraints are available immediately to the parser. 3-force theory (Austin 1976): Theory: There are three aspects of a sentence: * The locutionary force (literal meaning) * The illocutionary force (intended meaning) * The perlocutionary force (effect on the listener) Literal meaning theory (Clark & Lucy 1975): Theory: The literal meaning is constructed before the intended meaning. Evidence: Clark and Lucy's own experiments with rather artificial sentences. Counter-evidence: People can work out the intended meaning without processing the literal meaning (as shown by Gibbs 1983). Inner speech: ... means subvocalizing a text while reading. It can be measured by recording the activities of some muscles involved via electromyographics (EMG). Suppressing the subvocalization has effects on the understanding of difficult texts, but not on easy texts. Experiments suggest that inner speech is useful for retaining information about the word order in a sentence. It has also been shown that phonetically similar words make it harder to judge the syntactic correctness of a sentence. The difference between oral and silent reading speed is because a time-taking motor response for pronouncing each word is not needed (Rayner & Pollatsek 1989). Inner speech might provide the prosodic structure (i.e. rythm, intonation) for a written text, which is normally present in spoken text. Capacity theory (Just and Carpenter): Theory: The working memory is used for both storage and processing during comprehension. The working memory has a limited capacity. (This idea is very similar to a von-Neumann-Computer :-) This capacity can be measured by the reading span experiment. Evidence: Reading span experiment, Inanimacy cue experiment, Read-and-maintain experiment, Operation-span experiment, Theory also works for aphasia people (impaired language abilities). Reading span experiment (Danemann & Carpenter 1980): Task: Read a text, then try to recall the final word of each sentence. The largest number of sentences for which a participant can do this 50% correctly is defined as the "reading span". Result: The reading span correlates with the ability to answer questions about the text and also with verbal intelligence. Conclusion: The following working memory load is assumed: Intelligent folks: |-final-words-----------|--text-comprehension-| Stupid folks: |-final-words-|------------text-comprehension-| The less memory is needed to understand the text, the more space can be used to store final words => larger reading span. Inanimacy cue experiment (Just & Carpenter 1992): Task: Read a sentence like "The evidence examined by the lawyer shocked the jury." and another where "The evidence" is replaced by "The defendant". Result: People with a high reading span needed less time to understand the sentence with "the evidence". Conclusion: They made use of the inanimateness of "the evidence", preventing it from being the subject of the sentence "The evidence examined". With "The defendant", there was no such cue. People with low reading span took more time for both sentences. Read-and-maintain experiment (Just, Carpenter & Keller 1996): Task: Read sentences for comprehension and either maintain the last word or not. Result: As measured by fMRI, there was much more activity in the Wernicke's area when subjects had to maintain the last words. Conclusion: Processes in maintenance and comprehension overlap. Operation span experiment (Turner & Engle): Task: Similar to the reading span task, except for the fact that people had to solve arithmetic tasks instead of understanding a text. Words following each formula were to be maintained. Result: Similar to the reading span task. Conclusion: The reading span does not only determine the efficiency of understanding but also the efficiency of any cognitive task. Inferences: Two types of inferences are drawn while reading a text: * Bridging inferences, establishing coherence between the current part of the text and the preceding text. One bridging inference is the anaphora inference, connecting a pronoun ("she" or "he") to the corresponding subject. It has been shown by Clifton & Ferreira (1987) that the distance between the pronoun and the subject has no influence. * Elaborative inferences, adding details to the text. As investigated by O'Brien & al (1988), some obvious inferences are drawn without any loss of reading time. However, Singer (1979) found out that only strong and obvious inferences are drawn immediately. Johnson-Laird (1980) suggested that readers construct a complete mental model of the situation while reading. Bransford & al (1972) argued that inferences are stored in the same way as information presented directly in the text. Local inferences are drawn automatically while global ones are not. However, experimentors failed to distinguish between inferences drawn while reading and those drawn while being asked. The following two theories on inferences are both half-true, the last one is fully true. Minimalist inference theory (McKoon & Ratcliff 1992): Theory: Only very few inferences are drawn immediately by the reader. Strategic inferences (in contrast to automatic inferences) are only drawn if they are consistent with the reader's goal in reading. Constructionist inference theory: Theory: Numerous automatic inferences are drawn while reading. Search-after-meaning inference theory (Graesser,Singer&Trabasso 1994): Theory: Readers draw only those inferences which are necessary to satisfy the following goals: * Constructing a meaning of the text that addresses the reader's goals * Establishing local and global coherence in the text * Explaining the action, events and states mentioned in the text. Theories concerning the relation between thought & language: * Thought is dependent on and reflects language (Bruner 1983) * Language determines our concepts and we can only think thru concepts (Sapir & Whorf 1956) * Thought can occur without language (Heider) Theories concerning the origin of human laguage: * Language is learned like any other faculty by e.g. conditioning (Skinner, behaviorist) * There is an innate grammar, which biologically determines the basic structure of our language abilities (Chomsky, linguist) Criteria for "true" languages (Schmalhofer (?)): * Semanticity: Use of symbols that refer to objects & relations * Displacement: symbols refer to things in another time or place * Creativity: Symbols can be newly combined * Structure dependence: Syntax matters Projects with apes and language: * "Washoe" (Gardeners 1960) * "Sarah" (Premacks 1972) learned 100 symbols and complex sentences * "Kanzi" (Savage-Rumbaugh 1990) learned 256 geometric symbols and understood complex instructions All projects proved semanticity and displacement in the ape's "language", but there was only little evidence for creativity and structure dependence. Interactive activation reading theory (McClelland & Rummelhart 1981): Theory: Reading is based on three processes, namely * The feature level (recognizing e.g. horizontal and vertical bars in letters (chars)) * The letter level (recognizing a letter) * The word level (assembling letters to words) These processes are connected excitatorily, but also inhibitorily. So the word being assembled influences e.g. the letter recognition. Evidence: Word superiority experiment Word superiority experiment (Reicher 1969): Task: While reading a text, the eye fixations on letters (chars) are measured. Result: letter recognition is faster, when the letter is part of a word than of a non-word. Conclusion: Letter recognition is influenced by other than visual processes. Story memory theory (Gomulicki 1956): Theory: Story memory resembles a summary. Evidence: Written summary and mental summarization cannot be distiguished. Story grammar theory (Thordyke 1977): Theory: Stories obey a grammar. Thordyke proposed a hierarchical grammar with setting, theme, plot and resolution at its top. Evidence: Stories obeying this grammar are remembered best. Counter-evidence: All grammar theorists have different grammars. Schema theory (Bartlett 1932): Theory: Schemas play an important role in determining what we remember from a story. Schema-driven reconstructive processes occur at time of retrieval. Evidence: Culture experiment, Adolf experiment, Schema retrieval experiment Counter-evidence: Schemas also influence comprehension process (Bransford & Johnson 1972), there are not always reconstructive processes (Wynn & Logie 1998). Culture experiment (Bartlett 1932): Task: Read a story taken from a different culture. Recall it. Result: People tended to adjust the story to their own culture. Conclusion: Can be explained by the schema theory. Adolf experiment (Sulin & Doolin 1674): Task: Read a story about a dictator. In one story, the dictator was a fictional character, in the other story, it was Adolf Hitler. Afterwards, say whether the sentence "He hated the Jews and persecuted them" has occured in the text (while it did not). Result: People who read the story with Adolf Hitler thought the sentence occurred. Conclusion: Their schematic knowledge about Adolf Hitler distorted their recollection. Schema retrieval experiment (Anderson & Pichert 1978): Task: Read a text about a a house from either burglar's or a housebuyer's point of view. Afterwards, recall the story, then switch the point of view and recall it again. Result: On the second recall, people recalled other information than in the first recall. Conclusion: Retrieval is schemata-driven. Script-pointer + tag theory (Schank & Abelson 1977). Theory: * Information from the story is combined with schema information * Atypical (= non-schema-matching) information is tagged separately to the schema and remembered better than the typical information after a short retention interval but worse after a long retention interval. Evidence: General experimental evidence for good memories on atypical events. Counter-evidence: The theory does not specify how different types of atypical events are recalled. MicroMacro Theory (Kintsch & van Dijk 1978): Theory: Understanding bases on propositions and their arguments. At the micro-structure level, propositions are formed to a connected structure. At the macro-structure level, the gist of the story is assembled. In processing cycles, the propositions read to the short term memory are linked and all irrelevant propositions are garbage-collected (hello Java :-). With the help of simplification rules, the macro-structure of the story is created from the micro- structure. Evidence: Reading time experiment (s.a.), MicroMacro experiment Counter-evidence: The coherence of a text does not depend on the same argument being repeated several times, as suggested by Kintsch & van Dijk. MicroMacro experiment (Kintsch 1974): Task: Read a text, afterwards judge the accuracy of implicit and explicit inferences. Result: Explicit inferences (those present in the text) were judged faster after a short retention interval, but after 15 minutes, there was no time difference between the verification of implicit & explicit inferences. Conclusion: Implicit inferences are present in the micro- and macro- structure directly after reading, and thus verified faster. After a retention interval, the micro-structure faded and both inferences are equally well representated. Construction-integration theory (Kintsch 1988): Theory: 1. Sentences are turned to propositions (-> propositional net) 2. Inferences and related propositions are added from long term memory (-> elaborated propositional net) 3. A spreading activation process ("integration") stresses the highly connected propositions and sorts out the irrelevant ones (-> text representation) 4. Information about the relationship of two propositions is included if they were together in the short term buffer (-> episodic text memory) Three levels of representation are constructed: 1. surface level (the text itself) 2. propositional level (database of propositions) 3. situational level (mental model, gist of the story) Evidence: 3-level-forgetting experiment, Literary text experiment Counter-evidence: This is a bottom-up approach, the propositions are generated without considering the global context. But experiments have shown that the global context affects proposition building (as pointed out by Schmalhofer and some irrelevant other researchers). 3-level-forgetting experiment (Kintsch 1990): Task: Read a text. After a retention interval, recall situational, propositional or surface information. Result: Surface and propositional information faded away while situational information did not. It even increased, as pointed out by Ganseforth (2001). Conclusion: Support for the experimentor's theory. Literary text experiment (Zwaan 1994): Task: Read a text which is either labelled "newspaper report" or "literary text". Recall surface and situational information. Result: "Literary text" readers cared more for surface information (stylistic devices etc.) than "newspaper report" readers. Conclusion: The reader's goals influence the construction of the three levels of information. Support for the Construction-integration theory. I-want-two-more-levels theory (Graesser & al 1997): Theory: In addition to Kintsch's three levels, there are two more levels of text information: The text genre level (identifying the text as a joke, exposition or narration) and the communication level (storing the point of view of the text). Event indexing theory (Zwaan & al 1995): Theory: Readers monitor five key aspects while reading a story: 1. The protagonist 2. Time relationships 3. Causality 4. Spacial relationships 5. Relationship between the character's goals and the present event Discontinuity of any of these aspects makes reading more difficult. Evidence: Zwaan's own experiments show that discontinuity (incoherence) in each of these dimensions polonges reading time. Counter-evidence: The model treats the the 5 aspects as independent dimensions, which seems unplausible.