How are sentences stored in LTS? Explicit tasks; episodic memory Sachs (1967) STUDY subjects hear a story “…A wealthy manufacturer, Matthew Boulton, sought out the young inventor…” TEST yes-no recognition (1) identical sentence (2) change form, but not meaning (formal) (3) active/passive change (4) semantic change correct response yes no ONLY (4) CHANGES MEANING no no Sachs Results Form Change 50% Active/ Passive False Alarms Semantic Change 10% | immediate | 80 syllables later | 160 syllables later Test is You forget the form of the sentence, but remember the meaning In STS, you remember the actual words of the sentence. In LTS, you remember the meaning, but forget the wording. False Alarm Method Try to get subjects to make a false alarm on a yes-no recognition test It shows what part of the episode they remembered and what part they forgot e.g., in Sachs (1967), false alarms to items with same meaning, but different wording Sentences in LTS are stored as propositions Ratcliff & McKoon (1978) Study list of sentences . . . “The geese crossed the horizon as the wind shuffled the clouds.” . . . 1. CROSS (GEESE, HORIZON) 2. SHUFFLE (WIND, CLOUDS) 3. AS (1,2) Predict horizon “closer” to geese than to wind Item recognition priming test “clouds” “yes” “chair” “no” . . . “geese” OR “wind” “yes” PRIME “horizon” “yes” TARGET RT to horizon when geese is prime 600 msec when wind is prime 630 msec Why? Either because GEESE is closer to HORIZON in sentence (surface structure) Or because GEESE and HORIZON are in the same proposition (propositional hypothesis) STUDY “The kitten that the girl was carrying scratched the lawyer” SCRATCH (KITTEN1, LAWYER1) CARRY (GIRL1, KITTEN1) TEST (item recognition priming) PRIME “kitten” “girl” TARGET “lawyer” “lawyer” Integration Hypothesis Propositions that contain the same concepts connect together in memory Example: “A car hit a tree. The tree fell on a wire, and the wire touched another car.” HIT (CAR1, TREE1) FALL-ON (TREE1, WIRE1) TOUCH (WIRE1, CAR2) Concepts already in semantic memory Car Tree Wire Integration Hypothesis Propositions that contain the same concepts connect together in memory Example: “A car hit a tree. The tree fell on a wire, and the wire touched another car.” HIT (CAR1, TREE1) FALL-ON (TREE1, WIRE1) TOUCH (WIRE1, CAR2) Concepts already in semantic memory Tree Car is Car1 is hit Tree1 Wire Integration Hypothesis Propositions that contain the same concepts connect together in memory Example: “A car hit a tree. The tree fell on a wire, and the wire touched another car.” HIT (CAR1, TREE1) FALL-ON (TREE1, WIRE1) TOUCH (WIRE1, CAR2) Concepts already in semantic memory Tree Car is Car1 is hit Tree1 Wire is fall on Wire1 Integration Hypothesis Propositions that contain the same concepts connect together in memory Example: “A car hit a tree. The tree fell on a wire, and the wire touched another car.” HIT (CAR1, TREE1) FALL-ON (TREE1, WIRE1) TOUCH (WIRE1, CAR2) Concepts already in semantic memory Tree Car is Car1 is hit Tree1 is Car2 Wire is fall on Wire1 Based on Bransford & Franks (1971) tree in shaded man front yard tall smoked pipe Actually presented “The tree shaded the man who was smoking a pipe” Never presented (but consistent) “The tree in the front yard shaded the man” Never presented (inconsistent) “The tree broke the window” Support for integration hypothesis Bransford & Franks (1971) McKoon & Ratcliff (1980) “The lawyer gestured to a waiter.” The waiter brought coffee. The coffee stained the napkins. The lawyer flourished documents. The documents explained a contract. The contract satisfied a client. Propositions GESTURE TO (LAWYER1, WAITER1) . . . etc. is is is Waiter 1 Coffee 1 Napkins 1 Document 1 is Contract 1 is Client 1 is Lawyer 1 is Waiter is Coffee Napkins is is is Waiter 1 Coffee 1 Napkins 1 Document 1 is Contract 1 is Client 1 is Lawyer 1 Lawyer Document Contract Client Nodes for Concepts already in semantic memory is is is Waiter 1 Coffee 1 Napkins 1 Document 1 is Contract 1 is Client 1 is Lawyer 1 is Document to Waiter = 2 links Document to Napkins = 4 links Prediction “waiter” primes document more than “napkins” does Recognition Test client pen lawyer sofa waiter OR napkins documents PRIME TARGET RT to say “yes” to documents 665 msec with waiter as prime 704 msec with napkins as prime Supports integration hypothesis “Sally likes pets. She has a black cat.” LIKES (SALLY, PETS) HAS (SALLY, CAT1) IS (CAT1, BLACK) humans keep pets cats chase mice dogs like bones “Sally likes pets. She has a black cat.” LIKES (SALLY, PETS) HAS (SALLY, CAT1) IS (CAT1, BLACK) humans keep pets is Sally cats chase mice dogs like bones “Sally likes pets. She has a black cat.” LIKES (SALLY, PETS) HAS (SALLY, CAT1) IS (CAT1, BLACK) humans keep pets is Sally cats chase cat1 mice dogs like bones “Sally likes pets. She has a black cat.” LIKES (SALLY, PETS) HAS (SALLY, CAT1) IS (CAT1, BLACK) humans keep pets is Sally cats chase cat1 mice black dogs like bones Directly stated propositions and inferences “Sally likes pets. She has a black cat.” Directly-stated SALLY HAS A CAT Inferences Propositions that follow from the directly-stated propositions or from other inferences SALLY LIKES CATS (1) Logical inferences necessarily follow from directlystated propositions e.g., Sally has a pet Sally’s cat is a mammal (2) Pragmatic inferences are probably true, but not necessarily true e.g., Sally takes good care of her cat Sally buys cat food Inferences in Real Life Commercials “Four out of five doctors recommend the ingredients in Anacin” Court Room Harris, Teske & Ginns (1975) Witness: “I went up to the burglar alarm” Memory test “Did the witness say that they rang the burglar alarm?” Nearly everyone said “yes” even when instructed not to draw inferences Turtles Experiment Bransford, Barclay & Franks (1972) Group 1 STUDY “Three turtles rested beside a floating log and a fish swam beneath them.” TEST “Three turtles rested beside a floating log and a fish swam beneath it.” Result: very few false alarms Group 2 STUDY “Three turtles rested on a floating log and a fish swam beneath them.” TEST “Three turtles rested on a floating log and a fish swam beneath it.” Result: many false alarms! Group 2 propositions rest on floating log swim beneath 3 turtles fish Inference added to memory: causes false alarms to test sentence with fish swimming beneath log Conclusion: Memory contains inferences as well as directly stated propositions Schemas An organized set of propositions that describes the general characteristics of some thing or activity. (Stored in semantic memory) Restaurant Schema roles: CUSTOMER (human) SERVER (human) 1. ENTER (CUSTOMER, RESTAURANT) 2. SIT-AT (CUSTOMER, TABLE) 3. GREET (SERVER, CUSTOMER) 4. . . . BRING (BILL, TO CUSTOMER, BY SERVER) PAY (CUSTOMER, BILL) LEAVE (CUSTOMER, RESTAURANT) Other examples: Quiz show, coins Restaurant Schema Human is Restaurant is in Customer at Table Menu Waiter takes Order Schemas and Inferences Schema Instantiation Inferences propositions that link concepts in a text to concepts in a schema TEXT: “John went to the White Horse” “He sat down . . “ John is Restaurant Schema go to customer White Horse go to restaurant is Schema instantiation inferences IS (JOHN, CUSTOMER) IS (WHITE-HORSE, RESTAURANT) How memory and comprehension fail when you do not make schema instantiation inferences. Bransford and Johnson (1972) “Clothes washing” experiment No title given Title given before reading Recall (number of “ideas”) 2.8 5.3 Rated comprehension (1-7 pt. scale) 2.3 4.5 If you don’t know it’s about clothes washing, you can’t connect the text to your clothes washing schema. Bridging Inferences Inferences that fill gaps in the text by using information from a schema. “John went to the White Horse. He ate and left.” Inference JOHN PAID HIS BILL is John ate go to White Horse Restaurant Schema left is customer something pays bill go to restaurant McKoon & Keenan (1974) Investigation of Memory for Bridging Inferences Read “A camper carelessly threw a match. A great forest was destroyed.” Camper 1 Forest 1 Great threw was Match 1 Destroyed McKoon & Keenan (1974) Investigation of Memory for Bridging Inferences Read “A camper carelessly threw a match. A great forest was destroyed.” Camper 1 Bridging Inference made by consulting “forest fire” schema Great threw start Match 1 Fire 1 burn Forest 1 was Destroyed True-false Test Directly-stated A great forest was destroyed. T or F ? Inferences The match started a fire. T or F ? Conclude: Inference stored in LTS RT to say “true” immediate test 20 minutes later ‘Nancy’ Experiment Owens, Bower & Black (1979) p.355 in text Nancy woke up feeling sick again and wondered if she was pregnant. How would she tell the professor she was seeing?… Nancy arrived at the cocktail party. She looked around the room to see who was there. She went to talk with her professor. She felt she had to talk to him but was nervous. … Nancy went over to the refreshments. The hors d’oeuvres were good but she wasn’t interested in talking to the rest of the people. …. Nancy Experiment Recall Results Directly-stated Propositions Inferences (Nancy got sick at the party) with theme 29.2 without theme 20.2 15.2 3.7 •Schema improves memory for directly-stated proposition •It also promotes recall of inferences which were never mentioned Childhood Amnesia Recall birth now serial position for your whole life Repression? (Freud’s theory) Language acquisition? Difference between infant and adult schemas? Waldvogel (1948) number of memories 14 Recall your childhood memories 12 - 10 86- 4- females 2 | 0 | 1 | 2 males | 3 | 4 | 5 | 6 | 7 Sheingold & Tenney (1979) Asked questions about birth of younger sibling. Answers confirmed by parents. accuracy score | 1-2 | 3-4 | 5-6 Age | 7-8 No memories before 3 | 9+ Do animals show “childhood” amnesia frogs yes guinea pigs no Rats yes May depend on how “advanced” animal is at birth Stoloff & Spear (1976) 15-day old rats (still infants) 36-day old rats (young, but not babies) shock Trained until perfect in T-maze Tested 1 day later and 21 days later 36-day olds 100% correct turn chance training 1-day later 21-days later Older rats forget what they knew as babies Conclude Little evidence for repression explanation Language may play a role (but only in humans!) Schema Explanation Schemas for early memories are different than those for adults can’t consciously access them Early sensory-motor schemas Later proposition schemas The effects of schemas on judgment Hindsight Bias The outcome of an uncertain situation is judged to be more likely if you already know what happened. Rochester Nuke Plant Case Hindsight bias in assigning blame in accident steam tube tools reactor core Experimental Demonstrations of Hindsight Bias Fischoff (1975) •subjects read about war between British and Gurkas •then they judge the likelihood of the war outcome Estimated Probability British Gurkas Nobody win win win Told nothing (no hindsight) .40 .20 .40 Told “British actually won” .55 .12 .33 Told “Gurkas actually won” .32 .35 .33 Arkes et al. (1981) Doctors given a case history to read Is it disease A or disease B? Group 1 - told nothing about real diagnosis Group 2 - told “correct” diagnosis was disease A Group 3 - told “correct” diagnosis was disease B Result: probability estimates were higher for “correct” diagnosis So: Hindsight Bias •Hindsight bias happens when you warn people to avoid it •It does not happen when subjects don’t believe the outcome information Explaining Hindsight Biases by Schemas •Outcome info activates schema •Schemas guide retrieval of facts •Retrieved facts bias judgments of probability Medical example: heart attack or indigestion Symptoms smokes felt pain after dinner skipped heart beats burps a lot Explaining Hindsight Biases by Schemas •Outcome info activates schema •Schemas guide retrieval of facts •Retrieved facts bias judgments of probability Medical example: heart attack or indigestion Symptoms smokes felt pain after dinner skipped heart beats burps a lot heart attack schema indigestion schema Mnemonic Devices External Aids -- notes, string . . . Internal Aids used at encoding used at retrieval Internal Aids used at Encoding Basic Strategy make each item distinctive (reduce interference) make a collection of items meaningfully related (instantiate a schema) use a retrieval strategy to make sure you don’t miss anything Method of loci used for serial recall makes items distinctive and allows for a good retrieval strategy Peg words serial recall distinctive good retrieval strategy one - bun two - shoe three - tree four - door five - hive six - sticks seven - heaven eight - gate nine - wine ten - zen Keyword method for foreign language vocabulary Paling (Dutch) = Eel (English make image of Emphasizes distinctiveness and useful retrieval strategy Chunking Strategies (for lists) BROWN EEL sentence story image NAIL PAPER Emphasize distinctiveness and meaningfulness of collection People with Good Memories Exceptional ability or just good strategies? How specific are the abilities S. (studied by Luria) had very vivid mental imagery serial recall of 70-word lists retained for years used method of loci An exceptional memory with no imagery V.P. (studied by Hunt & Love) Continuous Paired Associate Task JUK - 23 ROQ - 29 CUH - 13 JUK - ? CUH - 97 ROQ - ? CUH - ? VP 100% | 2 | 4 College Students Recall | | 6 8 Lag | 10 | 12 etc. VP’s digit span = 25 Story recall - nearly perfect after 1 year •Used rapidly generated semantic associations No mnemonic devices at all Elizabeth (Stromeyer Eidetic Memory “Photographic” 10,000 dots Results are questionable No mnemonic devices at all Elizabeth (Stromeyer Eidetic Memory “Photographic” 10,000 dots Results are questionable No mnemonic devices at all Elizabeth (Stromeyer Eidetic Memory “Photographic” 10,000 dots Results are questionable Conclusions •People don’t just have good or bad memory as a whole -- they have good or bad memoryrelated skills. (e.g., good imagery, good ability to form semantic associations) •For verbal memory, mnemonic devices are needed for exceptional performance, but these don’t need to involve imagery. THE CAR CLIMBED THE HILL. THE CAR CLIMBED THE HILL.