Séminaire de Recherche en Linguistique

This talk takes as its starting point the three factors in language design proposed in Chomsky (2005): Universal Grammar (UG, F1), Primary Linguistic Data (PLD, F2) and  non-domain-specific principles of computational efficiency. I first briefly sketch in more detail what each factor consists in: in particular that F2 may involve a version of Fasanella & Fortuny (2013)’s Chunking Procedure, and that the relevant aspects of F3 are Feature Economy (FE) and Input Generalisation (IG) in the sense of Roberts (2007, 2011). Following earlier work, I sketch how the interaction of the three factors allows us to view parametric variation as an emergent property of the interaction of the three factors.

It is clear on the one hand that UG is indefeasible, in that it consists of “hard constraints” (e.g. Merge cannot create anything other than binary-branching structures). The F3 constraints, on the other hand, are clearly defeasible: FE is “outranked” by PLD cuing the presence of a given feature (if not, parametric options favouring relatively more features than others could not be acquired); IG is “outranked” by sub-regularities and irregularities in the PLD (otherwise microparameters and irregularities – nanoparameters in the sense of Biberauer & Roberts (in progress) -- could not be acquired). But is F2 defeasible? Certainly, it is “outranked” by UG, in that for example linear strings are internalised as binary-branching structures, unbounded dependencies are internalised as locality-compliant, etc. Does F2 always “outrank” F3? I suggest that this is not always the case, and that certain types of diachronic change show this.

In the diachronic domain, where FE “outranks” F2, we have loss of some property; where IG “outranks” F2 we have analogy. Here, we see that the “outranking” really concerns the relation between the acquired (innovative) grammar and the target (conservative) grammar (assuming, following Lightfoot (1979), that change is driven by reanalysis of various kinds in acquisition). In the child’s synchronic grammar, F2 may always either outrank or perhaps, following Clark & Roberts (1993), rank equally with F3.

I will then give examples from each of the five parameter hierarchies (H1-H5) proposed in Roberts (2011) of cases where F3 outranks F2 (the opposite case will also be illustrated, although this is more expected and hence less interesting). In relation to H1 (word order/linearisation), the case of FE >> F2 is change from head-final to head-initial order: here the data must be such at some point in the timecourse of the change that acquirers are exposed to some head-final orders but nonetheless posit head-initial orders. If head-initial orders involve the lack of a linearization trigger on the relevant head (Biberauer, Holmberg & Roberts 2013), then we see that here FE >> F2. IG >> F2 generalises harmonic orders (of either kind). For H2 (null arguments), FE >> F2 involves loss of φ-features and concomitant loss of licensing capacity by the relevant head(s); IG >> F2 involves generalisation of argument-licensing capacity across heads.  For H3 (word-structure/incorporation/head-movement), FE >> F2 involves loss of head-movement, e.g. V-to-T in the history of English; IG >> F2 is a less clear case, perhaps involving generalisation of head-movement, as seems to have happened in the development of general V2 in the history of Germanic. For H4 (A’-movement) FE >> F2 involves loss of wh-movement, as in recent French; IG >> F2 involves generalisation of e.g. resumption strategies in relatives in the history of Greek.

Finally, I will briefly discuss the question of the relative “weighting” of F2 and F3, again developing one or two remarks in Clark & Roberts (1993).