Transcript Learning
Implicit Learning I Dr Magda Osman Room 2.25 Office hours Mondays & Tuesday What is learning? Learning Learning is the process by which relatively permanent changes occur in behavioural potential as a result of experience As opposed to memory Memory is the relatively permanent record of the experience that underlies learning. What is implicit learning? Implicit learning e.g. of hidden „grammar rules” Learning new skills in patients with amnesia Automatic processing (fast, inflexible) Implicit memory: memory without any sensation of remembering Changes in performance (e.g. Word-stem completion task, anagram solution) Performance of people with amnesia on indirect tests of memory The case for implicit learning I We sense that we can learn things intuitively – without effort Language, is a good example in which we are able to acquire a lot of information very quickly – does this suggest implicit learning? We can train animals to learn to do many things, they don’t seem to be conscious of what they know – they must be using a form of implicit learning, right? Therefore, by extension if there are basic learning mechanism that animals have, then perhaps we have a basic unconscious systems as well? The case for implicit learning II We can apply knowledge without thinking We are able to acquire complex information quickly – so, doesn’t this mean that we can learn implicitly? There seem to be different types of learning, some which are harder to implement than others Implicit and explicit learning refer to different states of consciousness, if these states exist, then surely there are different learning mechanism that generate them? What does implicit learning imply? Implicit is unconscious learning, that is, processes that operate independently of consciousness (Reber & Reber, 2001) Knowledge that is stored in memory that cannot later be recalled or recognized, but can exert influence over behaviour Implicit learning relies on knowledge and representations that are encoded and stored differently from explicit knowledge. (Claparede (1911) – amnesic was pricked by concealed needle in their hand, days later they refused to shake the physician’s hand “You never know what people carry around”) Newell & Shanks (in press) Examples of implicit learning Priming/subliminal perception Perceptual motor learning/Rule learning Clinical Dissociations How can we know when a process is unconscious? Necessary to demonstrate a dissociation between a behavioural measure of performance and measures of awareness. Two stages of information processing: encoding & retrieval: Amnesics show difficulty in retrieving the details (declarative knowledge), but that doesn’t mean to say they weren’t conscious at the time of learning (encode the experience – as procedural knowledge) To demonstrate the encoding process as unconscious, you need to demonstrate that at the moment the stimulus was presented, the individual’s awareness of the stimulus was absent How can we know when a process is unconscious? Problems with definition of consciousness and detection threshold (at encoding) Cheesman & Merikle (1984) Subjective threshold – discriminative responding for which individuals CLAIM they are guessing, but are NOT responding at chance level Objective threshold – discriminative responding for which individuals are ACTUALLY responding at chance level Subliminal priming - what does it mean to learn without attention? Presentation of stimuli subliminally – without conscious detection Demonstrations of unconscious detection rests on two different points: Accessibility of information at encoding (retrieval) Availability of information stored after it was presented Some say that unconscious detection = the inaccessibility of information at encoding (Brody, 1989) – but can you know? Others say that unconscious detection = availability to consciousness is greater than accessibility (Erdelyi, 1986, 2004) Subliminal priming Market researcher James Vicary (1957) reported a study involving 700 cinema goers in the states. The two messages shown during the film were “Eat Popcorn” and “Drink Coca-Cola”. A message was flashed for 3msc of a second every five seconds – in lab tests this was claimed to be too fast for conscious attention. People Claimed this was too fast for conscious awareness. Companies claimed that over a six-month period sales of popcorn rose 57% and sales of Coca-cola rose 18.1%. This was the beginning of the concept of subliminal advertising. However, these impressive results were never replicated, and in 1962 Vicary admitted the study was a hoax. Subliminal priming/Masked Priming Subliminal priming: stimuli too short (20-110ms with masking)/ or too low intensity to reach conscious level But in both case processing of incoming stimuli is said to occur Classic demonstrations Semantic priming (Marcel’83) Affective priming (Zajonc’80) 13 Simple Lexical Decision with subliminal priming BREAD TABLE kurtka PHOUD BUTTER Word Nonword 14 Subliminal Priming Prime Mask Target TABLE XXXXXX PHOUD BREAD XXXXXX BUTTER HORSE XXXXXX BALLOON 20ms 20-200ms 1000ms 15 Masked Priming Find that the masked stimulus (prime) influences the speed of judgment of word/non-word discrimination TABLE PHOUD Incongruent trials – slow responses BREAD XXXXXX XXXXXX BUTTER Congruent trials (semantically related trials) – speed responses HORSE XXXXXX BALLOON Congruent trials (semantically unrelated trials) – baseline Opstal et al (2010) Opstal et al (2011) 16 Priming Bargh, Chen, & Burrows (1996) study Priming Old Age and Walking to the Elevator Primes were presented at a supraliminal level (conscious level) Repetition Priming Jacoby et al (1989) Judgments of fame effect Based on the effect reported by (Claparede (1911) - Mere exposure to names of people will make them seem famous (people can’t figure out why the names are familiar so judge them as famous) Jacoby et al’s (1989) Exposure phase –non-famous names: Sebastian Weisdorf, Valerie Marsh, Adrian Marr either 1 or 4 times Test phase –non-famous names & famous names – judgement of fame Manipulation – test administered directly after exposure, or a day after. Repetition Priming Jacoby et al’s predictions were: Famous names – judged accurately, New names – judged innacurately, Old non-famous names [1x] more inaccuarte than [4x] 0.7 0.6 0.5 Famous New 1x 4x 0.4 0.3 0.2 0.1 Famous names – judged accurately, New names – judged more accurately compared to Old non-famous names [1x] Old names [4x] judged inaccurately 0 Immediately One day Is diverted attention, still a lack of attention? Attention and consciousness are highly correlated phenomena – so lack of attention towards the to-be-learnt object is still classed as implicit learning (e.g. dual task paradigm) E.g. Mulligan (1997, 1998) Asked people to memorize word lists, while repeating back stings of numbers varying length Later recall for words learnt while repeating long lists was impaired, but on indirect test (stem completion, e.g. Fisherma_?) show that people did learn the words. Perceptual-motor/rule learning Logic of Experiments 1. Present people with simple/complex stimuli that follow a rule 2. Include a training period 3. Include a test period 1. 2. Explicit (or direct) measures of knowledge Implicit (or indirect) measures of knowledge Explicit measures: Implicit measures: 1. Ask questions that directly relate to information that was presented during training 1. Ask people to report their confidence or how much they like the stimulus they are presented 2. Ask people what they think they learnt 2. Produce stimuli that don’t follow the pattern/rule that you were exposed to in training Artificial Grammar Learning Reber (1967) Present people with stimuli that follow a complex rule Training period: MTTV Try to remember as many of these strings of letters as possible Test period: Which is grammatical? VXM or VMR? BPPQ or BQF? How did you decided which is grammatical What is the grammar? General findings People show above chance accuracy (54%-69%) in discriminating between grammatical and non-grammatical letter strings. But are unable to report how they make their discriminations (Reber, 1967– replicated (Brooks, 1978; Dienes, Broadbent & Berry, 1991; Saffran, 2001), People can transfer their ability to discriminate to letter strings they haven’t seen before, but without knowing. (Dienes & Altmann, 1997) Cross modal transfer (from one sensory form to another)– people learnt letter string but at test were able to discriminate between sequences of musical tones (Altman, Dienes, & Goode, 1995) If during training people are made aware that there is a grammar, their ability to later discriminate declines. (Reber, 1967; Reber, Kassin, Lewis, & Cantor, 1980) But…. E.g. People may not learn rules, but just pairs of letters (bigrams) that the grammar allows. People trained on grammatical pairs were just as good as people trained with complete strings (Dulany, Carlson, & Dewey, 1984; Perruchet & Pacteau, 1990; Shanks, 2004) Grammaticality is conflated with familiarity, people only tend to rate grammatical strings if they feel that they are familiar to them. When looking at unfamiliar letter strings only, discrimination for grammaticality was at chance (Kinder & Assmann, 2000; Johnstone & Shanks, 2001) Familiarity – is a fluency effect (the easier something is to process, the more familiar it seems), when told to deliberately ignore the fluency of processing the letter strings, discrimination for grammaticality was at chance (Kinder, Shanks, Cock, &Tunney, 2003) Measures of explicit knowledge not exhaustive, and not inclusive of what the individual may have actually learnt. Serial Reaction Time Task Nissen & Bullemer (1987) Present people with stimuli that follow a rule. E.g. deterministic sequence: 13423413241 Training period: Block of trials in which people perform the task Test period: Transfer task Indirect Measures Direct Measures General findings People show dissociation between the ability to perform the sequence and accuracy of their knowledge of the sequence (Hartman et al, 1989; Nissen & Bullemer, 1987; Willingham et al, 1989) Learning through observation impairs performing the sequence, but not knowledge of the sequence (Kelly & Burton, 2001; Kelly et al, 2003) Process dissociation in free generation tasks show that when people are told to explicitly avoid generating the sequence, they still generate the sequence (Cleeremans & Jiménez, 1998) Dual Task studies show that you can learn the sequence even when having to complete a secondary task at the same time – like tone counting (Cohen et al, 1990; Frensch, Lin, & Buchner, 1998; Hsiao & Reber, 2001) But… Learning of the sequence is affected if you have to discriminate between high and low tones and indicate the result using foot-tapping, or a highly visual demanding task. (Heuer & Schmidtke, 1996; Rowland & Shanks, 2004) Poor discrimination between old and new sequences was improved from guessing to well above chance, by introducing performance related pay to motivate individuals to respond accurately. (Johnstone & Shanks, 1999) Usually, only partial elements (6-items) of the sequence are presented for later recognition tests, when trained for a long time, and then presented whole (12items) old or new strings people were faster to respond to old than new, and were able to discriminate old items from new (Shanks & Johnstone, 1999; Shanks & Perruchet, 2003) Theoretical Question Given that there are so many different forms of knowledge: 1. Are there multiple types of learning processes/mechanisms 2. Is there one single learning process/mechanism Implicit learning = rule abstraction 1. 2. 3. 4. Learning is a sophisticated process that tracks and exacts patterns and forms them into rule without any conscious effort As a result, what becomes processed are the deep underlying structures (e.g. language learning, intuitive reasoning) From this we can make accurate predictions and generalize and transfer knowledge (e.g. cross modal transfer) We do not need to consciously attend to information for it to be learnt. Implicit learning = exemplar based 1. 2. 3. 4. 5. Given that we don’t always know what to expect in our environment, we often begin by learning by example, when we have enough examples that are stored in memory, we can begin to make predictions by comparing how similar new examples are to the ones we have in memory. We do not need to attend to information to learn it implicitly There is no implicit learning TAP – transfer appropriate processing 1. 2. 3. (Morris et al, 1977) – learning the properties of the words (semantic or phonemic), when tested based on either properties, when training & test matched, performance was better. Our cognitive system is designed to find short cuts. We often begin by learning chunks of information, which makes it easier to store information in memory From this simple rules can be formed which enable us to make predictions. Attention is absolutely necessary for learning (Shanks & St John, 1994; Shanks, 2004) Describe the different methodological approaches used to investigate implicit learning