Much writing in linguistic theory appears to be driven by a certain common wisdom, which is that the simplest theory either is the most aesthetically elegant or has the fewest components, or that it is the theory that eschews extra conceptual resources. This common wisdom is reflected in a 1972 paper by Paul Postal entitled “The Best Theory,” which appeals to simplicity criteria for support of a particular linguistic proposal. A lot of linguists would wholeheartedly endorse Postal’s remark (pp. 137–138) that, “[w]ith everything held constant, one must always pick as the preferable theory that proposal which is most restricted conceptually and most constrained in the theoretical machinery it offers.”

This claim may seem pretty intuitive, but it stands in need of clarification, and once clarified, the claim is much less intuitive, if not obviously false. As an alternative, I will propose that genuine simplicity criteria should not involve appeal to theoretical machinery, but rather a notion of simplicity in the sense of “simplicity of use”. That is, simplicity is not a genuine property of the object of investigation (whether construed as the human language faculty or something else), but is rather a property that is entirely relative to the investigator, and turns on the kinds of elements that the investigator finds perspicuous and “user friendly.”

Let’s begin by considering Postal’s thesis that the simplest (and other things being equal the best) theory is the one that utilizes less theoretical machinery. It may seem natural to talk about “theoretical machinery,” but what exactly is theoretical machinery? Consider the following questions that arise in cross-theoretical evaluation of linguistic theories of the sort discussed in Chapter 1. Is a level of linguistic representation part of the machinery? How about a transformation? A constraint on movement? A principle of binding theory? A feature? How about an algorithm that maps from level to level, or that allows us to dispense with levels of representation altogether? These questions are not trivial, nor are they easy to answer. Worse, there may be no theory neutral way of answering them.

The problem is that ‘machinery’ can be defined any way we choose. The machinery might include levels of representation, but then again it might not (one might hold that the machinery delivers the level of representation, but that the level of representation itself is not part of the machinery). Alternatively, one might argue that levels of representation are part of the machinery (as they are supported by data structures of some sort), but that the mapping algorithms which generate the levels of representation are not (as they never have concrete realization). Likewise one might argue that constraints on movement are part of the machinery (since they constrain other portions of the machinery), or one might argue that they are not (since they never have concrete realizations).

Even if we could agree on what counts as part of the machinery, we immediately encounter the question of how one measures whether one element or another represents more machinery. Within a particular well-defined theory it makes perfect sense to offer objective criteria for measuring the simplicity of the theoretical machinery, but measurement across theories is quite another matter. (Ludlow 2011: 153)

References

Ludlow, P. 2011. The Philosophy of Generative Grammar. Oxford University Press.
Postal, P. 1972. The best theory. In S. Peters (ed.), Goals of Linguistic Theory., pages 131-179. Prentice-Hall.