Page 121

		The question of order-independence was first considered by Blum and Blum [18]. They showed that order-independence is not a restriction on TxtEx. The proof presented in this chapter is due to Fulk [69, 71]. In the proof he showed that not only can a scientist be converted into an order-independent scientist without any loss in learning ability, but the resulting order-independent scientist can also be made to satisfy several additional properties such as rearrangement-independence. The reader is referred to Lange and Zeugmann [121] for a treatment of order-independence and set-drivenness in the context of uniformly decidable families of recursive languages. A related result is also due to Jain [84].

		The notion of prudence was introduced by Wexler and Culicover. Fulk [69, 71] showed that prudence is not a restriction on identification. The reader is referred to Kurtz and Royer [115] and to Jain and Sharma [91] for additional results on prudence.

		The question of identification by reliable scientists was considered by Minicozzi [132]. The reader is directed to papers by Case, Jain, and Ngo-Manguelle [28] and by Kinber and Zeugmann [110, 111] for additional material on reliability.1

		Angluin [6] began the investigation of conservative scientists. The study of monotonicity was initiated by Jantke [99] who also defined the notion of strong-monotonic strategies. Wiehagen [200] introduced monotonic strategies and Lange and Zeugmann [119] defined weak-monotonic strategies. The duals of these strategies were proposed by Kapur [103]. Lange, Kapur, and Zeugmann have obtained numerous results on the relative learning abilities of these strategies in the context of identifying uniformly decidable families of recursive languages, e.g., see [122], [209], and [120]. A survey of this work can be found in [208]. The study of monotonicity for r.e. languages is due to Jain and Sharma [88, 96], where the distinction between global and class versions of a strategy is formalized. Kinber and Stephan [109] also independently investigated monotonicity in the context of r.e. languages.

		1In our discussion of strategies, we have required that the underlying realities a scientists investigates are to be modeled by recursively enumerable languages or computable functions. In the case of reliability, it is mathematically interesting to modify this requirement and consider the behavior of scientists on non-r.e. "languages" and non-recursive functions. For example, in the language case one can require, not unreasonably, that a reliable scientist should fail to converge on any text that has a non-r.e. set as its content. In the language case, this strengthened reliability requirement turns out to be equivalent to ordinary reliability. In the function case, matters are quite different and rather involved. For this reason, we have called the strategy discussed in Definition 5.66 as recursive-reliable to distinguish it from the notion of reliable scientists in the literature reserved for this stricter notion of reliability. Case, Jain, and Ngo-Manguelle [28] discuss these subtle issues. Fortunately, for other strategies discussed in this chapter, both for languages and functions, the analogous stricter version of global strategies do not place any additional constraint on identifiability.