Abstract Sassoon

Acquisition order helps bootstrapping category structure

The more typical instances of a concept are acquired earlier by children and adults, remembered best, retrieved faster etc. (Mervis & Rosch 1981). These effects are robust, but scholars hardly agree about the mental models underlying them, and their relation to predicate meaning. The need for an account is strengthened by accumulating evidence for typicality effects in the categories of grammar (Langacker 1987, Lakoff 1987).

Rosch states that evidence about a prototype (an optimal instance or feature set) poses constraints, but does not in itself constitute a mental model (ibid).

Formal accounts based on membership degrees have been proven highly problematic: Fuzzy Logic (Zadeh 1965) predicts that no entity is more typical of a brown apple than of an apple. However, typical brown-apples are counter-examples (atypical apples). Partee & Kamp’s 1995 solution to this problem fails to account for cases with no syntactic modification: They predict that no entity is more typical of an ostrich than of a bird. However, typical ostriches are counter-examples (atypical birds); Worse yet, a prototype cannot be associated with tall, because there is no maximum tallness (ibid), so the intuition that there are typical tall people remains unaccounted for. Similarly for composed concepts: For ex. what is the prototype of non-bird: a dog, day, number?

I propose different conditions for a feature / entity to be considered typical.

(i) Features like nesting are typical of a bird if they correlate with ordering (not membership!) in bird. The correlation is ceteris paribus:

Any entity more typical in nesting than other entities (equally typical in other features) is more typical of a bird.

(ii) This typicality ordering (scale) is no more than a side effect of the order in which we learn that entities fall under bird, which remains encoded in memory. While disambiguating predicate meaning within a particular context, the status of entities deviating from given examples (given birds) is uncertain. Thus:

d1 is more typical of P than d2 in a context iff: Either the P-hood of d1 is established earlier than the P-hood of d2 Or the non-P-hood of d2 is established earlier than the non-P-hood of d1.

This correctly predicts that: The earlier we learn that d is a bird the more typical we consider d to be; Atypical birds (lately acquired birds) may be typical ostriches (even the earliest ostriches known!) Similarly, even the latest apples acquired might be the earliest brown apples acquired; The scale of non-bird is, by this definition, inverse to the scale of bird in each context (see evidence in Smith et al 1988); Context independent prototypes (maximal tallness) are not needed: In each context some entities are the best: the earliest given (among the available entities); Finally, concepts like tall have fixed ordering features, but typical associates with them more contextual features. Thus, the NP typical tall person, like typical bird, associates with a more context dependent scale.

This is a small part of the facts covered by my proposal.