How analogical processing (and spatial language) fosters spatial learning Dedre Gentner Northwestern University Benjamin Jee, Northwestern University David Uttal, Northwestern University Cathy Manduca, Carleton College Thomas Shipley, Temple University Brad Sageman, Northwestern University Carol Ormand, Carleton College 1 Basil Tikoff, UW-Madison Two lines of research • Analogical processes support spatial learning and cognition • Spatial language supports spatial learning and cognition Overarching question: How do we come to be fluent relational thinkers 2 http://www.spatialintelligence.org Analogy as Structure-mapping base target An analogy conveys that partly identical relational structures hold between objects in different domains Corresponding objects need not resemble each other (though it’s easier if they do) Gentner (1983, 1989); Falkenhainer, Forbus & Gentner (1989); Gentner & Markman (1997) 3 http://www.spatialintelligence.org Analogy highlights common relational structure and fosters relational abstractions e.g., A small force at a long distance from the fulcrum can Balance a large force at a small distance from the fulcrum 4 Gentner, 1983; Gentner & Markman, 1997; Gentner & Namy, 199; Gick & Holyoak, 1983 http://www.spatialintelligence.org Progressive alignment from easily aligned to hard-to-align is a good way to start early learning Once understood, the relational abstraction can often be extended to more dissimilar pairs via Progressive alignment (Kotovsky & Gentner, 1986; Gentner, Anggoro & Klibanoff, in 5 press; Jee et al, in prep) http://www.spatialintelligence.org How analogy leads to spatial learning: Detecting differences Alignable difference Alignable difference : different elements that each occupy the same role in the aligned structure: e.g., “Motorcycles have two wheels, cars have four” 6 Gentner, 1983, 1989, 2003; Gentner & Markman,http://www.spatialintelligence.org 1997; Gick & Holyoak, 1983 Task: Name a difference between these two images 7 Gentner & Sagi, 2006 http://www.spatialintelligence.org Task: Name a difference between these two images Nonalignable Difference Alignable Difference 8 Gentner & Sagi, 2006 http://www.spatialintelligence.org Finding Fault: Using alignment to supporting spatial learning Alignable difference 9 http://www.spatialintelligence.org Progressive alignment of pairs • 64 Psych 110 students • 10 similar & 10 dissimilar pairs (No feedback on responses) Similar first Dissimilar first 10 http://www.spatialintelligence.org Comparison and learning • Structural abstraction: Receiving multiple structurally-alignable comparisons helps generalize the concept • Progressive alignment: Transitioning from close (easy to align) pairs to less similar pairs helps novice learners abstract a common relation (Gentner, Loewenstein, & Hung, 2007) Teaching students to recognize faults • Alignable pairs showing fault/no-fault > same examples separately • Order of pairs matters: Progressive alignment High-similar (easy to align) pairs low-similarity (hard to align) pairs is better than the reverse order 11 http://www.spatialintelligence.org Spatial Language helps children perform a spatial mapping task (Loewenstein & Gentner, 2005) “the winner” Hiding Box Finding Box For 6 trials (twice per location): Baseline: “I’m putting it here” (Noninformative Language) Spatial Language: “I’m putting it on/in/under the box” (Informative Language) 12 Children are shown the correct answer on each trial http://www.spatialintelligence.org Loewenstein & Gentner, 2005 Spatial Mapping Task – Neutral Version Search Trials 1.0 • Young (3 ½) children benefit strongly from relational language • Older (4 year-olds) perform well with or without spatial language Proportion Correct Results * On, In, Under 0.8 0.6 0.4 Baseline 0.2 0 3;8 4;1 N=20 N=20 13 http://www.spatialintelligence.org Cross-mapped Task Neutral 1.0 Results • 4 ½ & 5-year-olds benefit strongly from overt spatial language Spatial Language * 0.8 Proportion Correct Pattern repeats at older age: • 4-year-olds fail entirely Cross-mapped * 0.6 Baseline 0.4 0.2 0 3;8 yrs 14 http://www.spatialintelligence.org 4;1 yrs 4;7 yrs 5;2 yrs N=20 N=20 N=20 Thank you 15 http://www.spatialintelligence.org Ongoing Study in Chicago Children’s Museum with Susan Levine, Sonica Dhillon, Jessica Saunders, Ashley Poltermann & Tsivia Cohen Does alignment foster acquisition of principles of stable construction? with Susan Levine, Sonica Dhillon, Jessica Saunders, & Tsivia Cohen Participants: Children 3-8 with families in Build-a-skyscraper activity Key principle in engineering and architecture: Diagonal braces confer stability Vertical Brace Diagonal Brace Horizontal Brace 16 http://www.spatialintelligence.org High Alignability Low Alignability “Which one do you think is stronger?” “OK, now see if you can wiggle them” The building with the vertical piece bends. The building with the diagonal brace does not bend. 17 “So which one was stronger? Right! The one with the brace was stronger” http://www.spatialintelligence.org Ongoing Study in Chicago Children’s Museum with Susan Levine, Sonica Dhillon, Jessica Saunders, Ashley Poltermann & Tsivia Cohen Post-test : Show me where you would put this to make this building strong Results: High-alignment > Low-alignment 18 http://www.spatialintelligence.org Recognizing Faults Fault = a fracture in a rock along which movement has occurred 19 http://www.spatialintelligence.org Learning through examples • Most introductory geology courses include instruction about faults & other geological structures • Students see multiple examples of faults • Students vary in rate of learning Sources: Wikipedia.org, USGS.gov 20 http://www.spatialintelligence.org Method: Transfer task & post-tests 14 single images, 7 fault and 7 no-fault Is there a fault? If so, place 2 Xs on its location X X X X Spatial ability tests: Mental rotation and water levels Background experience questionnaire (10 questions) • Key question: How many previous course in geology? 21 http://www.spatialintelligence.org Results: Effects of condition Does similar-first sequence lead to higher performance in the initial phase? No geo courses At least 1 geo course n=44 n=20 2 ns 1 0 Dissimilar-first Similar-first Initial Phase d-prime Initial Phase d-prime May depend on background experience 2 * 1 0 Dissimilar-first 22 http://www.spatialintelligence.org Similar-first Results: Effects of condition No geo courses At least 1 geo course n=44 n=20 Transfer task d-prime Transfer task d-prime Main effect of geo course experience in the transfer task 2 1 0 Dissimilar-first Similar-first 2 1 0 Dissimilar-first 23 http://www.spatialintelligence.org Similar-first Results: Geoscience experience and spatial ability Are geoscience experience effects due to spatial ability? • No significant difference on spatial ability measures between those with geo experience and those without (no experience actually higher on each measure) What knowledge leads to higher performance? • 17 participants asked to explain the information they used to find faults in images, and how they know when they have found a fault • 2 main types of information present in responses: use of cracks/fractures, and use of movement/displacement of layers Crack/Fracture mentions Movement/Displacement mentions Accuracy above mean 6 16 Accuracy below mean 14 7 24 http://www.spatialintelligence.org Further directions • More initial instruction that explicitly distinguishes faults from fractures may be especially beneficial for novices • e.g., diagrams, photos, or simulations of faults Fracture, No fault Fault • Provide feedback during initial phase • May enhance novice’s ability to profit from comparison • Examine performance on new set of transfer items • Examine eye movements to observe processing (search and comparison) in more detail 25 http://www.spatialintelligence.org How analogy leads to learning: Detection of differences Analogical comparisons make alignable differences more salient (Gentner & Markman, 1994; Markman & Gentner, 1993, 1996) Example task: Find the wrongly placed bone (Kurtz & Gentner, in prep) Target 26 http://www.spatialintelligence.org Standard Ongoing Study in Chicago Children’s Museum with Susan Levine, Sonica Dhillon, Jessica Saunders, Ashley Poltermann & Tsivia Cohen Does alignment foster acquisition of principles of stable construction? with Susan Levine, Sonica Dhillon, Jessica Saunders, & Tsivia Cohen Participants: Children 3-8 with families in Build-a-skyscraper activity Key principle in engineering and architecture: Diagonal braces confer stability Vertical Brace Diagonal Brace Horizontal Brace 27 http://www.spatialintelligence.org High Alignability Low Alignability “Which one do you think is stronger?” “OK, now see if you can wiggle them” The building with the vertical piece bends. The building with the diagonal brace does not bend. 28 “So which one was stronger? Right! The one with the brace was stronger” http://www.spatialintelligence.org Method Vertical Piece Diagonal Brace Horizontal Piece -- Pre-training • Comparison: Diagonal brace vs. Horizontal/Vertical piece • Conditions: High-alignable Low-alignable Control (no pre-training) -- Task: Child builds skyscraper w/ family -- Post-task: Questions & assessment tasks 29 http://www.spatialintelligence.org High Alignability “Which one do you think is stronger?” “OK, now see if you can wiggle them” The building with the vertical piece bends. The building with the diagonal brace does not bend. 30 “So which one was stronger? Right! The one with the brace was stronger” http://www.spatialintelligence.org Results – Reaction time RT (ms) RT (ms) “Different” Responses Same-Diff (n=24) Name-a-diff (n=20) 1. Same-Diff: High Sim slower than Low Sim (p<.01) 31 2. Name-a-Diff: High Sim faster than Low Sim (p<.01) 3. Same-Diff faster than Name-a-Diff http://www.spatialintelligence.org (p<.01) Structure-Mapping Principles Gentner, 1983, 1989 1. Structural Consistency • 1-1 Correspondences • Parallel Connectivity a b c a’ b’ c’ BASE TARGET 2. Systematicity • Common Connected Systems of Relations • Higher-Order Constraining Relations 3. Relational Focus • Common Relations > Common Object Descriptions Embodied in SME (the Structure-mapping Engine) (Falkenhainer, Forbus & Gentner, 1989; Forbus, Gentner & Law, 1995) Many of these principles now used in other models: ACME, IAM, AMBR, SIAM, LISA, EMMA 32 http://www.spatialintelligence.org Structural alignment supports spatial learning Structural abstraction Candidate inference • Highlights common spatial relational structure • Supports abstraction of common spatial structure • Invites new spatial inferences 33 Gentner, 1983, 1989, 2003; Gentner & Markman,http://www.spatialintelligence.org 1997; Gick & Holyoak, 1983