Page 91

		5 Strategies for Learning

		The paradigms TxtEx and Ex allow scientists to be any computable device whatsoever. This conception seems excessively general if only because human scientists (and the computers they build) operate under resource constraints involving time and memory. In addition, real scientists are apt to impose various methodological constraints upon their theorizing, for example, requiring that present hypotheses account for all available data. In the present chapter we formalize and investigate such constraints by considering the inductive competence of alternative subsets of the class of computable scientists. Any such subset is called a strategy, and the principal question it invites is whether every identifiable collection of languages or functions is identified by a member of the strategy. We shall see that the answer is often negative, even for strategies that appear to embody a rational canon of hypothesis selection. In this sense, many of the strategies to be defined below are restrictive constraints on scientists.

		In terms of paradigms, the present chapter may be conceived as exploring alternative conceptions of scientist, while leaving untouched the interpretations of theoretically possible reality, data etc. that we introduced in Chapters 3 and 4. Let us also note the connection of strategies to language development in children. To the extent that human children are essentially identically endowed for language acquisition (which seems manifestly to be the case), their learning methods are drawn from an extremely narrow strategy in the sense defined above. So it is suggestive to examine the inductive capacity of diverse constraints on scientists in hopes of finding one that allows identification of just the languages children can assimilate. (See Section 3.1.2 for more discussion.)

		In what follows, we first consider strategies for the identification of languages, and then of functions.

		§5.1 Strategies for Language Identification: Introduction

		We divide strategies for language learning into four groups:

		1. constraints on potential conjectures,

		2. constraints on the information available to a scientist,

		3. constraints on convergence, and