L14 - Similarity & Analogy Flashcards
What are the similarity theories of Mental Representations?
Spatial Models: How close in “map” in your mind.
Feature Models: Each concept has features, and degree they’re shared is similarity.
Structured Models: Analogy
Explain Spatial Models of similarity
- Claim: represent stuff in mental space, and distance function of similarity.
- Latent-Semantic Analysis: huge matrix based on encyclopaedia, similarity of words is the correlation of 0’s and 1’s; the more two words appear in the same entries = more similar. Does well on ESL synonym test.
- > used in google translate
Explain Feature Models of similarity
Similarity of A and B is the sum of features common to A and B, minus the features A has that B doesn’t, and minus the features B has and A doesn’t.
-> S(A,B)=qf(A similarity B) - af(A-B) = bf(B-A)
Features are a list of comparisons.
Each set can be weighed as more or less important according to context.
Similarity judgements highlight common features and difference highlight distinct.
E.g. East and West Germany would have both more common AND distinct features than Sri Lanka and Nepal, because you know more.
Explain why Tversky rejected the Spatial Model of similarity before proposing the Feature Model?
- Tversky believed spatial models are violated.
Because: - Similarity is asymmetrical (Canada is like USA and vice versa)
- Metric spaces show “triangle inequality”.
Explain the Structured Models of similarity
- Representations are structured, objects are bound by RELATIONS among them.
- We find similarity in roles of objects AND features.
e. g. easier to list differences in hotel and motel than motel and coconut because perceptual dimensions are the same.
What is an analogy?
When two conceptual domains share relational similarity, not feature or object based similarity.
Beneficial because relational structures are computationally expensive.
What is the Structure-Mapping theory of analogy?
Comparisons involve an alignment of relational structures.
Constraints:
- One-to-one mapping: each element can align with only one element in other domain.
- Parallel connectivity, e.g. sun -> nucleus -> electrons, not sun -> electrons.
- Systematicity: deeply rooted relational structures make better analogies