chapter  9
18 Pages

Kinds of Evidence for Linguistic Theory

A well-known crucial struggle for theoretical linguists is the search for the appropriate kind of evidence that should guide them in the choice of the optimal formal analyses for empirical linguistic data of various sorts. Data do not come with a label of what the formal mechanisms involved in their computations are, nor do they come with a comment of what their overall signicance is for the general functioning of the speakers’ internal grammar. Furthermore, linguists do not have any direct access to the functioning of the internal computational system at work in (different) human language(s), a well-known aspect of linguistic research within the domain of cognitive sciences. At the same time, then, data are not clean nor does there exist an easy procedure available to the linguist to clean them up and understand what they mean, what they reveal of the formal properties of the I-language. A very general practice in the generative tradition since its beginnings has been that of taking so-called ‘grammaticality judgments’ given by native speakers as a privileged source of evidence to support or disconrm general hypotheses and particular formal analyses. Various reasons explain the fortune of this quite widely adopted practice, mostly good ones. One such reason is that speakers’ grammaticality judgments are relatively easy to obtain: The linguist must be smart enough to construct the appropriate examples to ask about and careful enough to clean them up in such a way that interfering factors do not obscure the relevant aspects of the hypothesis to be tested. In this sense, the theoretical linguist builds up experiments, and interfering factors and variables have to be controlled for exactly as in all experimental research. The examples to be presented to native speakers in order to get their grammaticality judgments, constitute, in a sense, the purest and simplest kind of experimental design that the linguist can build up: Neither machinery nor any special lab is needed for that. The easiness ends here, though: In order to construct the relevant examples to be judged, a very precise theoretical hypothesis is always presupposed. One further, and rather obvious, reason for the wide use of speakers’ grammaticality judgments in linguistic research can be identied in the somewhat

trivial fact that native speakers belong to a category of experimental subjects that is very easy to nd: They are numerous and since the experiments to be undertaken do not involve any special setting nor do they involve any potential danger for the speakers’ health and they usually take relatively little time to be performed, native speakers are generally ready to take part in the experiments by providing their grammaticality judgments.1 There is no deep reason, however, why grammaticality judgments should constitute the only source of evidence for linguistic theory. To make a parallel, consider, for instance, the important progresses that have become possible in theoretical syntactic studies within the P&P model once the comparative perspective has been seriously adopted as a general practice.2 Until then, even abstracting away from the widespread prevalence of English as the language considered, different languages were typically analyzed as autonomous grammatical systems, and there was no systematic attempt to compare their properties with those of other languages, in particular of closely related languages. The comparative perspective, sometimes referred to as the ‘new comparative syntax,’ has allowed for substantial advancements in the understanding of UG and its possible parameters of variation. Thus, not only was there at the time no principled reason to limit the analysis to one single language, there turned out to be important reasons to extend it and adopt the comparative perspective. The careful study of the subtle variations instantiated by the different northern Italian dialects is an especially revealing case in point: As Kayne rst put it, the numerous dialects provide an ideal laboratory for comparisons; small variations can often be neatly singled out, as other related properties are kept constant, since invariant (Kayne 2000, 2005c). Hence, the overall signicance of the variation and its possible limits can be best appreciated. The parametric perspective to the study of language variation has provided a greatly innovative approach to this domain, and it has contributed to bringing dialectal studies to the fore of the cognitive approach to the formal study of language.3