- Slides: 30
Classical Categories versus Experientially Based Categories Prototypes and the Basic Level Linguistics 5430 Spring 2007
The Mysteries of Categorization § § § Using common nouns for things requires us to categorize. But what does it mean to know the name of something—knowing a description, a context, an image? How do we use names appropriately? How do our purposes affect the way we categorize? How do we learn names (Plato’s problem)? Are linguistic and nonlinguistic categories the same (Jackendoff’s problem)?
Answers from Philosophy § Putnam (1975: 227) on semantic externalism: “’Meanings’ just ain’t in the head’. § This view follows from Kripke’s (1980) causal theory of names: a name’s reference is fixed by an original act of dubbing, after which the name becomes a rigid designator of that object. § Later uses of the name succeed in referring to the referent by being linked to that original act via a causal chain. § This view is opposed to Russell’s descriptive theory of names: proper names are abbreviated definite descriptions.
Answers from Philosophy § The descriptive theory was devised to account § § § for the coherence of sentences like Santa Claus does not exist. But Kripke argues that to use a name appropriately, you need only be part of a long historical chain reaching back to the original naming event. Further, we can make counterfactual statements about the referents of proper names. Putnam extends the causal theory to natural kind terms.
The Division of Linguistic Labor § Putnam says: for me, both elms and beeches are ‘deciduous trees growing in North America’. § Yet if Putnam claims, ‘The elm is the most popular ornamental tree in North America’ this statement can be evaluated as true or false. § Nothing in Putnam’s head fixes his elm reference; rather, the linguistic community contains people who do know the difference between the two trees. § The experts ensure that when he says elm he is talking about elms. This is 'the division of linguistic labor'.
Answers from Machine Learning Merely using the right word in the right place is in itself an adaptive ability. A child can usefully learn that the place she lives is Colorado, […] a college student that operant conditioning is related to learning, a businessperson that TQM is the rage, before needing any clear idea of what these terms stand for. Many well read adults know that Buddha sat long under a Banyan Tree (whatever that is) and Tahitian natives lived idyllically (whatever that means) on breadfruit and poi (whatever those are). More-or-less correct usage often precedes referential knowledge, which itself can remain vague but connotatively useful. Thus the frequent arguments over the meaning of words and the livelihood of lexicographers and language columnists who educate us about words we already partially know. Moreover, knowing in what contexts to use a word can function to amplify learning more about it by a bootstrapping operation in which what happens in response provides new context if not explicit verbal correction. (Landauder & Dumais 1996)
What’s Important about Categorization? § § Most words refer to classes rather than to specific entities. Perception, motor activity, linguistic behavior all involve categories. Inference is based on categorization (e. g. , Linnaean taxonomy, Euler’s formula: VE+F=2, medical diagnosis. Categories and category boundaries matter in the search for universals of human cognition.
The Classical Model of Categorization § § The folk model based on common properties (conditions) is also the classical model. Categories have no internal structure; all members are equal. Categories are based on inherent properties. Humans have access to category structure, but may make naïve errors.
The Classical Model § Reason is the mechanical manipulation of § § § abstract symbols. Symbols get their meaning from their ability to refer to things in the world or a possible world. The meaning of a sentence is the set of condition(s) under which it is true or false. Questioning the classical model of categorization is questioning the classical view of reasoning.
The Classical Model My proposals will also not conform to the expectations of those who, in analyzing meaning, turn immediately to the psychology and sociology of language users: to intentions, sense-experience, and mental ideas, or to social rules, conventions and regularities. I distinguish two topics: first, the description of possible languages or grammars as abstract semantic systems whereby symbols are associated with aspects of the world; and second, the description of the psychological and sociological facts whereby a particular one of these abstract semantic systems is the one used by a person or population. Only confusion comes of mixing these two topics. —David Lewis, “General Semantics” (1972)
Questioning the Classical Model § Human categorization is essentially a matter of human experience and imagination—of perception, motor activity and culture (on the one hand) and of metaphor, metonymy and mental imagery on the other (on the other). § To change our view of categories is to change our view of the world, since categories are categories of things: species, substances, artifacts, colors, kinsmen, emotions, etc.
Stereotypes Associated with the Classical Model § Meaning is based on truth and § § reference. The mind is independent of the body. Reason transcends human concerns. Only reason has conceptual content. Language is a poor window into cognitive processing, because it tends to be sloppy.
The Alternative Model § Every category has a prototypical member and belongs to a taxonomy. § Prototypes give rise to: § membership gradience § markedness in language § polysemy in language § contested categories § generativity § reference-point (metonymic) reasoning
The History of Prototype Theory § Prototype theory can trace its origins back to late work of Ludwig Wittgenstein. § Philosophical Investigations (1953): Words are not defined by reference to objects in the external world or by mental representations, but rather by how they are used in effective, ordinary communication. § His famous example is the category GAME. § How can GAME be defined in such a way as to include basketball, solitaire, tag, musical chairs and chess?
Application to Categories § There may be no property which characterizes § § § all members of a category, whether necessary or sufficient. This doesn’t entail that the category has unclear boundaries: GOE ≠ DOM (C&C, p. 79) It does mean that the category label may be extensible. The SALAD example
Radial Categories § Lakoff proposes a model of category structure to explain (most) prototype effects: the best exemplar is defined by a cluster of converging properties. § Peripheral members lack one or more of the properties that define the central member. § Occasionally, there is also chaining within a category (e. g. , Jello salad is chained to lettuce salad via fruit salad).
Polysemy § J. L. Austin (1961) asked, “Why do we call different kinds of things by the same name? ” § He argued that neither similarity or equivocation were involved. § The word healthy: healthy body vs. healthy activity vs. healthy complexion § Fillmore on climb: The airplane climbed to 30, 000 feet vs. She climbed out onto the balcony.
Category Boundaries are Extensible § Lounsbury: Fox skewing rules (e. g. , merging rule: X’s same sex sibling can be referred to by X’s title; mother’s sister= mother). § Berlin & Kay (1967): languages don’t carve up color spectrum randomly. Hierarchy of basic color terms via agreement on best exemplars. • • • black, white red yellow, blue, green brown purple, pink, orange, gray
Rosch on Prototypes § What Rosch describes are asymmetries that she calls prototype effects: subjects judged members of some categories as more representative than others. § Paradigms included: direct rating, reaction time, production of examples, asymmetry in similarity ratings, directionality of generalizations.
Rosch on Prototypes § Rosch initially attributed the effects as caused by perceptual salience, then as a theory of category structure, then as having many underlying causes. § Barsalou on ad hoc categories. § Linguistic prototype effects: are subjects’ hesitations to apply a given label due to gradience or goodness of fit? (e. g. , lie) § How does metonymy interact with a theory of prototypes?
The Basic Level § Roger Brown: “While a dime can be called a coin or money or a 1952 dime, we somehow feel that dime is its real name”. § Categorization “begins at the level of distinctive action”; superordinate and subordinate names are “achievements of the imagination”.
The Basic Level § It is the level of distinctive actions. § It is the level which is learned earliest and at which things are first named. § It is the level at which names are shortest, most frequent, and least likely to be borrowed. § It is the highest level at which a single image can stand for the whole.
The Basic Level § It is the level at which folk categories correspond most closely to scientific categories. § It is the level with the greatest cultural significance. § It is the level at which distinctions are made most easily. § It is the level at which most the knowledge is concentrated.
Prototypes and the Basic Level § How do prototype-based categories interact with basiclevel categories? § Are they the same thing? § Could there be prototype effects are all levels in a taxonomy? § What does it really mean for a single image to stand for the whole category?
Linguistic Prototype Effects § We often refer to linguistic categories as § § § marked or unmarked. What does this mean? Are there numerous conceptions of markedness? Is markedness a useful concept in linguistics?
Shortcomings of Prototype Theory (Croft & Cruse 4) § The features used are not context sensitive. § Features may not vary independently. § Features may need to be weighted to distinguish basic-level categories from one another. § Features need to be ‘grounded’ in order to be meaningful. § But frames can behave like features, as in Lakoff’s account of the category MOTHER.
A Dynamic Model of Prototypes § Smith & Samuleson (1997) are critical of the ‘fixed categories’ approach. § They argue that categories are created based on life experience, priming and current plans and goals. § All three phenomena come into play in the determination of boundaries, framing and level of categorization.
Boundaries § Context determines the placement of a category boundaries. § The following don’t appear to qualify as games, but in some contexts may be considered as such: § A jigsaw puzzle (no active agent against whom one competes) § Roulette (goals external to the activity) § A race (player cannot interfere with opponent)
Frames § Peripheral members of a category may be central depending on the frame invoked. § Pizza is a prototypical breakfast if the ‘timed meal’ frame is invoked. § A genetic mother is prototypical if the hereditary disease frame is invoked. § A guide dog or a horse is a prototypical pet if not ‘functional’, thereby promoting the ‘companion’ frame.
Level of Categorization § What constitutes the basic level vs. the superordinate level is not merely a function of expertise. § One’s purpose in categorizing may influence the ‘granularity’ of perspective that one takes. § Framing influences level of categorization; these superordinates act like basic-level terms: § beverage (in a restaurant frame) § produce (in a grocery store) § liquid (at airport security )