Matches in SemOpenAlex for { <https://semopenalex.org/work/W6168977> ?p ?o ?g. }
- W6168977 abstract "Heuristics for Choosing Features to Represent Stimuli Matthew D. Zeigenfuse (mzeigenf@uci.edu) Michael D. Lee (mdlee@uci.edu) Department of Cognitive Sciences, University of California, Irvine Irvine, CA 92697 USA Abstract another. We use these heuristics to begin answering the ques- tion of specifying what properties of a feature cause people to represent it. In this paper, we compare three heuristic methods for choosing which of a set of features to use to represent a domain of stim- uli when we know the categories to which those stimuli belong. Our methods are based on three measures of category differen- tiation: cue validity, category validity, and their product, collo- cation. In a comparison of their ability to predict human simi- larity ratings in the Leuven Natural Concept Database, we find collocation to have the best performance, suggesting people use both cue and category validities in choosing which features to represent. Keywords: Feature representation; basic-level categorization; similarity judgment. Representation and Basic-Level Categories Introduction Of all the aspects of their world that could be represented, which do people actually choose? Imagine you are standing in front of a black dog named “Rover” with a small white patch of hair under its left eye. Which of its features do you choose to represent: its tail and four paws, its name, “Rover”, and the spot under its eye? The last two of these may be useful for a representation of this particular dog, but are probably less useful to representing dogs as whole. Conversely, the first two may be useful for representing dogs, but are probably less useful for distinguishing Rover. One method of learning about which aspects of a particular set of concepts people represent is the feature generation task (Rosch & Mervis, 1975). Often in this task people are asked generate a fixed number of features for each exemplar in a domain. In some cases, additional participants are asked to rate whether an exemplar has a feature for each combination of features and exemplars in a domain (Deyne et al., 2008). This leads to a large number of features describing each ex- emplar; however, not all of these features will be important to a person’s representation. Zeigenfuse and Lee (2008, 2010) provide a computational- level (Marr, 1982) approach to the problem. Similar to the theory of second-order isomorphism in perception (e.g. Shep- ard & Chipman, 1970), they argue that people represent those features that determine the similarity between objects and de- velop a model to infer which features are important using sim- ilarity judgments. Unfortunately, their method does not of- fer a psychological rationale for why one feature is important vis-`a-vis an unimportant one, since it is more of a statistical solution than an account of feature importance. This paper expands upon the computational approach of Zeigenfuse and Lee (2008, 2010) by exploring psychologi- cal theories of what makes a feature important. To this end, we propose heuristic methods for choosing important features based on how well a feature distinguishes categories from one Our heuristics are based on measures of category differenti- ation that have been proposed to explain basic-level catego- rization. Basic-level phenomenology refers to people’s pref- erence to categorize objects at a particular level in a cate- gory hierarchy, known as the basic level. Key finds are ob- jects are categorized into basic-level categories more quickly than sub- or super-ordinate categories, basic level objects are named faster, objects are described preferentially with ba- sic level names, more features are listed at the basic level than at the superordinate level, basic level names are learned before names at other levels, and basic level names tend to be shorter (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976). These results suggest an intimate relationship between an object’s basic-level category and its mental representation. Category-Based Measures Category Differentiation Given a feature representation, many theories of basic-level categorization score potential categorizations of the concepts in a domain through the infor- mation its categories give about the features of category mem- bers and vice-versa. Examples include, cue validity (Rosch et al., 1976), category validity, collocation (Jones, 1983), fea- ture predictability (Corter & Gluck, 1992), category statis- tical density (Kloos & Sloutsky, 2006), and strategy length and internal practicability (SLIP: Gosselin & Schyns, 2001). Inverting this logic, given a set of categories, we can score features on their usefulness in providing information about which of the set of categories a concept belongs to, the infor- mation knowing a concepts category provides about whether it has the feature, or a mixture of the two. Usefulness Measures The heuristics described here for choosing feature representations are based on three measures of feature usefulness. Suppose we have a domain of cat- egories {c 1 , . . . , c M }. Let f be an arbitrary feature. The first heuristic is maximum cue validity, which we define as max 1≤ j≤M p(c j | f ). The quantity p(c j | f ) is known in the lit- erature as the cue validity of feature f (implicitly, with re- spect to category c j ). Psychologically, it expresses how well having a feature predicts whether a stimulus belongs to a par- ticular category. We also look at maximum category validity, defined as max 1≤ j≤M p( f |c j ). Here p( f |c j ) is often referred to as the category validity f (again, implicitly, with respect to category" @default.
- W6168977 created "2016-06-24" @default.
- W6168977 creator A5030538261 @default.
- W6168977 creator A5069667659 @default.
- W6168977 date "2010-01-01" @default.
- W6168977 modified "2023-09-26" @default.
- W6168977 title "Heuristics for choosing features to represent stimuli" @default.
- W6168977 cites W191277985 @default.
- W6168977 cites W1969949803 @default.
- W6168977 cites W1974275080 @default.
- W6168977 cites W1976790980 @default.
- W6168977 cites W1980797923 @default.
- W6168977 cites W1999100266 @default.
- W6168977 cites W2000255081 @default.
- W6168977 cites W2018475972 @default.
- W6168977 cites W2027796863 @default.
- W6168977 cites W2059799772 @default.
- W6168977 cites W2059975159 @default.
- W6168977 cites W2063910430 @default.
- W6168977 cites W2068502752 @default.
- W6168977 cites W2072588367 @default.
- W6168977 cites W2091489452 @default.
- W6168977 cites W2109166628 @default.
- W6168977 hasPublicationYear "2010" @default.
- W6168977 type Work @default.
- W6168977 sameAs 6168977 @default.
- W6168977 citedByCount "1" @default.
- W6168977 crossrefType "journal-article" @default.
- W6168977 hasAuthorship W6168977A5030538261 @default.
- W6168977 hasAuthorship W6168977A5069667659 @default.
- W6168977 hasConcept C103278499 @default.
- W6168977 hasConcept C111919701 @default.
- W6168977 hasConcept C115961682 @default.
- W6168977 hasConcept C119857082 @default.
- W6168977 hasConcept C127705205 @default.
- W6168977 hasConcept C138885662 @default.
- W6168977 hasConcept C153180895 @default.
- W6168977 hasConcept C154945302 @default.
- W6168977 hasConcept C173801870 @default.
- W6168977 hasConcept C177264268 @default.
- W6168977 hasConcept C17744445 @default.
- W6168977 hasConcept C199360897 @default.
- W6168977 hasConcept C199539241 @default.
- W6168977 hasConcept C204321447 @default.
- W6168977 hasConcept C2776359362 @default.
- W6168977 hasConcept C2776401178 @default.
- W6168977 hasConcept C41008148 @default.
- W6168977 hasConcept C41895202 @default.
- W6168977 hasConcept C80023036 @default.
- W6168977 hasConcept C94124525 @default.
- W6168977 hasConcept C94625758 @default.
- W6168977 hasConceptScore W6168977C103278499 @default.
- W6168977 hasConceptScore W6168977C111919701 @default.
- W6168977 hasConceptScore W6168977C115961682 @default.
- W6168977 hasConceptScore W6168977C119857082 @default.
- W6168977 hasConceptScore W6168977C127705205 @default.
- W6168977 hasConceptScore W6168977C138885662 @default.
- W6168977 hasConceptScore W6168977C153180895 @default.
- W6168977 hasConceptScore W6168977C154945302 @default.
- W6168977 hasConceptScore W6168977C173801870 @default.
- W6168977 hasConceptScore W6168977C177264268 @default.
- W6168977 hasConceptScore W6168977C17744445 @default.
- W6168977 hasConceptScore W6168977C199360897 @default.
- W6168977 hasConceptScore W6168977C199539241 @default.
- W6168977 hasConceptScore W6168977C204321447 @default.
- W6168977 hasConceptScore W6168977C2776359362 @default.
- W6168977 hasConceptScore W6168977C2776401178 @default.
- W6168977 hasConceptScore W6168977C41008148 @default.
- W6168977 hasConceptScore W6168977C41895202 @default.
- W6168977 hasConceptScore W6168977C80023036 @default.
- W6168977 hasConceptScore W6168977C94124525 @default.
- W6168977 hasConceptScore W6168977C94625758 @default.
- W6168977 hasIssue "32" @default.
- W6168977 hasLocation W61689771 @default.
- W6168977 hasOpenAccess W6168977 @default.
- W6168977 hasPrimaryLocation W61689771 @default.
- W6168977 hasRelatedWork W138582014 @default.
- W6168977 hasRelatedWork W1528714290 @default.
- W6168977 hasRelatedWork W2003663929 @default.
- W6168977 hasRelatedWork W2131012547 @default.
- W6168977 hasRelatedWork W2379245051 @default.
- W6168977 hasRelatedWork W2523571579 @default.
- W6168977 hasRelatedWork W2563490378 @default.
- W6168977 hasRelatedWork W2564072317 @default.
- W6168977 hasRelatedWork W2577043533 @default.
- W6168977 hasRelatedWork W2580972828 @default.
- W6168977 hasRelatedWork W2587028620 @default.
- W6168977 hasRelatedWork W2590484290 @default.
- W6168977 hasRelatedWork W2766153660 @default.
- W6168977 hasRelatedWork W2767467192 @default.
- W6168977 hasRelatedWork W2767512750 @default.
- W6168977 hasRelatedWork W2767556146 @default.
- W6168977 hasRelatedWork W2767825521 @default.
- W6168977 hasRelatedWork W2768653494 @default.
- W6168977 hasRelatedWork W2770036471 @default.
- W6168977 hasRelatedWork W2788832751 @default.
- W6168977 hasVolume "32" @default.
- W6168977 isParatext "false" @default.
- W6168977 isRetracted "false" @default.
- W6168977 magId "6168977" @default.