Analogy


Lecture notes

Phil/Psych 256
Feb. 4, 1997

Analogy preview:

1. 	Representation - proportional, geometric, predicate, 
    domains, levels

	Constraints on mapping - similarity, structure, purpose

2.	Computation - analogy formation

	Problem-solving - decision, explanation, learning

	Psychological plausibility - similarity, metaphor

	Applications - Case Based Reasoning

Q: What is an analogy?

	A1: "proportional," (Hofstadter) e.g.,

		abc : ijk :: abd : ?
		 (ijl, ijd, ijj)

		"abc" is to "ijk" as "abd" is to __?

		"What education is to the individual human being, 
		that revelation is to the whole human race." (Lessing)

	A2: "geometric" (Evans)

	A3: recognition, e.g.,

		"He looks just like Elvis!"

	A4: categorization, e.g.,

		"He's the Michael Jordan of bowling"
		"Way to go, Einstein!"

	A5: "me-too," e.g.,

		"I'm going to pay for my beer now"
		"Me too"

	A6: remindings

	A7: explanations, e.g.,

		Bohr model of the atom

Q: What are the components of an analogy?

	A1: a source (base) domain - a concept, usually well 
	understood, e.g.,

		planets, sun, orbits (trajectory)

	A2: a target domain - a concept, often not so well 
	understood, e.g.,

		electrons, nucleus, orbits (quantum)

Q: How are the domains organized (multiconstraint theory - 
Holyoak & Thagard)?

	A1: attributes - basic objects and their properties, e.g.,

		planet, sun, gravity

	A2: relations - relations that hold among attributes, e.g.,

		orbits(planet,sun) 

	A3: systems - systemic facts that hold among other predicates, 
	e.g.,

		because(orbits,gravity) 

	So, the Bohr model analogy could be expressed by the mappings:

		Source                   Target

		planet                   electron
		sun                      nucleus
		gravity                  charge

		orbits(planet,sun)       *orbits(electron,nucleus)

		because(orbits,gravity)  because(orbits,charge) 

	The *orbits predicate is imported into the target domain, 
	to explain the structure of the atom.

Q: What constitutes a good analogy (MT)?

	A1: similarity - corresponding predicates are similar in 
	meaning, e.g., near neighbors in a semantic network, visual 
	similarity

	A2: structural isomorphism (systematicity) - each predicate
	corresponds with one other predicate, and their slot-fillers 
	correspond similarly

	A3: purpose - mappings contain predicates that address the 
	problem to be solved

Example:

   Armor, n.  The kind of clothing worn by a man whose taylor 
   is a blacksmith.

   Source                     Target

   clothing                   armor
   taylor                     blacksmith
   man                        man
   needle, thread, cloth      hammer, rivet, metal

   makes(taylor,clothing)     makes(blacksmith,armor)
   uses(taylor,n-t-c)         uses(blacksmith,h-r-m)
   buys(man,clothing,taylor)  buys(man,armor,blacksmith)
   wears(man,clothing)        wears(man,armor)

   in-order-to(uses,makes)    in-order-to(uses,makes)
   because(makes,buys)        because(makes,buys)
   in-order-to(buys,wears)    in-order-to(buys,wears) 

Q: What about analogies in arguments?

	A1: The truth of the source predicates needs to be examined, 
	e.g., guns are like cricket bats


Phil/Psych 256
Feb. 6, 1997

Q: What is the process of analogy formation?

	1. Characterize the target 

	2. Pick out a source that matches target:
		- given, or
		- search memory and retrieve, or
		- construct one
	(fishing expedition)

	3. Map source to target; find correspondences

	4. Evaluate solution 

	5. Learn schema from target for next time

Q: What factors affect the different stages?

	A1: retrieval (2) depends on similarity

		- mere-appearance matches, e.g., "A 
		  cloud is like a sponge" --> "both are  
		  round and fluffy" (5-year old) or 
		  "both hold water" (adult) (Gentner)

		- tumor problem (Holyoak & Thagard)

	A2: mapping (3) depends on structure and representation 
	(Gentner)

		- competing mappings

		- the "frame problem," e.g., 
		  hotter-than(a,b) vs. colder-than(b,a), 
		  cause(a,b) vs. because(b,a)

	A3: solving (4) depends on purpose (Holyoak & Thagard)

		- are electrons like planets?

		- are guns like cricket bats?

Example: Analogy and explanation

	Darwin used several analogies to describe the concept 
	of natural selection:

	1. Artificial selection, e.g., dogs, pigeons

	2. Laissez-faire economics (Smith)

	3. Exponential population growth vs. linear resource 
	growth (Malthus)

	4. The "argument from design" (Paley)

Q: How does analogy lead to learning?

	A1: a useful analogical source may be generalized to 
	form a schema 

		- irrelevant detail is ignored

		- specifics are made defaults or stripped away

	A2: multiple sources may suggest a schema (convergence), 
	e.g., armies & firemen --> tumors

Example: Analogy and decision

	Operation Desert Storm was conceived in response to several 
	previous analogues:

	1. WWII - Hussein as Hitler, massive invasion

	2. Vietnam - Hussein as Ho Chi Min, incremental involvement

	3. Vicksburg - Civil War battle in which Grant outflanked 
	the Confederate Army (Schwarzkopf as Grant)

Q: How are metaphor and analogy related?

	A1: good metaphors share the properties of good analogies, 
	e.g., "my job is a jail" 

		- my job is like a place of confinement 
		  (categorization), but also,

		- my job is unpleasant

		- my job is not easy to leave

		- ignore some details of jail

	A2: the source domain is re-conceived or extended 
	(unlike analogy); consider

		- "That acrobat is a hippopotamus"
		   vs.
		- "That hippopotamus is an acrobat"

	A3: literal and metaphorical interpretations interfere, 
	e.g., (Glucksberg)

		- some desks are junkyards  [slow]
		- some desks are roads      [faster]

Example: Case Based Reasoning

	- Case Based Reasoning systems solve problems by identifying 
	and adapting previous "cases" similar to the given (current)
	one.

	- previous cases are stored in a case library (often static)

	- good systems can apply several library cases (perhaps 
	piecemeal) to a given problem

Review of analogy: 

	1. Database - semantic network of concepts, e.g., case library

	2. Knowledgebase - characterization, retrieval, mapping, 
	evaluation, learning

	3. Goals - categorization, decision, explanation, learning, ...

	4. Learning strategies - generalization, convergence

	5. Accords with basic psychological data, but still in early 
	stages of research 

Next week:

	- sample midterm
	- essay 2
	- visual imagery
	- Kosslyn

Further materials


Return to Phil/Psych 256 home page