I first came to MIT in the Fall 1979, under Noam’s urge after his visit to Pisa in the Spring of that year, and then I was a visiting Scholar several times (for about 6 terms between 1979 and 1989). Since the beginning of my student career in the early 1970s I had been fascinated with the issue of how aspects of grammatical diversity cluster across languages and can be scientifically described.
Thus, the most intriguing problem for me could be formulated as follows:
1) Which (and how abstract) syntactic properties can crosslinguistically vary independently of each other?
In other words, which variable properties are ultimately the real entities of grammatical diversity (regarded as one of the most central features of human culture and cognition)?
In those years at MIT, I found the development of the Principles&Parameters framework an absolutely illuminating way of addressing this problem, as well as the most interesting strategy to carry out comparative practices in linguistics since the classical historical method.
Now, over thirty years later, parametric theories have become a standard form of successfully expressing contrastive generalizations and typological clustering of variable grammatical properties.
In this sense, we can claim that the parametric format has attained some high degree of “crosslinguistic” descriptive adequacy. What is more dubious in my view is whether parametric models have achieved further levels of scientific success, first of all whether they are able to address concerns of classical explanatory adequacy, as represented in the following question:
2) Do P&P theories represent realistic models of language acquisition?
This conjecture has gone largely untested, mostly owing to the lack of a reliable and sufficiently wide sample of plausible parameters and to the difficulty of defining a set of triggers for each of them.
In fact, it is very difficult to imagine a viable alternative to a parametric model (broadly understood as any finite set of discrete predetermined choices) of grammar acquisition. In particular, empiricist views, undermined by poverty-of-stimulus considerations, but also previous nativist theories relying on evaluation measures, seem to represent no such alternative. However, empirically, parametric theories are not yet sufficiently corroborated, since nobody has so far indisputably assessed their effectiveness as acquisition models by implementing a parameter setting system over a large and realistic collection of parameters (Fodor 2001, Yang 2003; cf. Chomsky 1995, 7: “The P&P model is in part a bold speculation rather than a specific hypothesis. Nevertheless, its basic assumptions seem reasonable…. and they do suggest a natural way to resolve the tension between descriptive and explanatory adequacy”)
A plausible strategy in order to find evidence for P&P or its variants is that of collecting relatively many hypothetical parameters, set in relatively many languages, though all contained within a single submodule of grammar (in order to downsize the complexity of the task and the risk of missing some of the close interactions between contiguous parameters).
A substantial though still manageable database of this type can be subjected to various tests which are not possible with isolated parameters, e.g. a study of its abstract learnability properties. Thus, this practical approach makes it possible to meaningfully raise questions like 3):
3) Are (fragments of) parametrized grammars mathematically learnable?
However, I think similar databases may even more immediately allow for an original empirical way of testing of parametric approaches.
In the same year the P&P model was proposed, David Lightfoot happened to publish his Principles of Diachronic Syntax, now regarded as the forerunner of all the foundational work in historical generative syntax which has boomed for the past 20 years (think of Lightfoot’s notion of “local causes”, Clark and Roberts’ and Berwick and Niyogi’s concept of “logical problem of language change”, and Keenan’s idea of “inertia”). The glimpses of understanding of syntactic “change” achieved so far permit us, in my view, to empirically evaluate P&P with respect to their “historical adequacy”, i.e. their ability to provide correct insights on the actual history of languages and populations through space and time. Precisely this kind of success established linguistics as a well respected discipline in the 19th century. Therefore, I believe that a great deal of insight and respect among neighboring sciences can arise for generative linguistics if questions like 4) are successfully addressed:
4) Do P&P theories represent realistic models of language transmission through time and space?
Can parametric syntax, e.g., provide us with insights about the (pre-)history of human diversity parallel to and better than those achieved by lexical comparative linguistics? After thirty years of P&P, I hope I can still participate in an effort in that direction.
In sum, pursuing problems of the type of 3) and 4) represents in my opinion much more structured and updated ways to address the concerns which intrigued me and drew me to MIT in the 1980s. Research programs revolving around such questions seem now feasible and promising. The very concrete possibility of raising and answering them appears as the best witness of the progress of cognitive science since those exciting years.