The Significance of What Hasn’t Happened
Naturally enough, the focus of diachronic syntax – and, indeed of historical linguistics more generally – has been on documenting and analyzing recorded instances of change. In a parametric model, this means trying to observe, describe and explain cases of parametric change. However, if change is viewed as abductive reanalysis of Primary Linguistic Data (PLD) in language acquisition, which, in part, also involves resetting parameter values of the underlying grammar (Lightfoot 1979, 1991, 1999), we expect acquisition mostly to be convergent and, thus, that little will change. This is the Inertia Principle of Keenan (2002) and Longobardi (1994), which we can phrase in parametric terms as:
(1) Most of the time, most parameter values don’t change.
(1) is almost certainly true, perhaps a truism. But in order to seriously understand both change and the nature of parameters, we need to qualify both occurrences of most. In other words, which parameters change and when? Are certain parameters more amenable to change than others? If so, what can we learn about parameters more generally from these changes? These are the questions this paper investigates. As we shall see, the cases where a given parameter does not change can be as revealing as those where it does.
In this connection, consider the following cases of long-term historical conservation of known parametrically variant properties:
(2) a. (Multiple) Incorporation in the Algonquian languages (Branigan 2012)
b. Harmonic head-final order in Dravidian (Seever 1998:31) and Japanese/Korean
c. “Radical pro-drop” in Chinese and Japanese
According to Goddard (1994) and Branigan (2012), Proto-Algonquian was spoken 2000-3000 years ago. In that time numerous structural, lexical and phonological features have changed, but incorporation has remained as a “signature” property of the family. Assuming for concreteness that a new generation of native speakers emerges every 25 years, in 3000 years we have 120 iterations of the learning cycle. Proto-Dravidian is dated by Seever (1998) to 4000BC, i.e. 6000 years ago, so this parameter has remained constant over roughly 240 iterations of the learning cycle. Similarly, the oldest texts in Japanese date from around 700-800AD, and so are over 1000 years old, again showing conservation of head-finality and radical pro-drop over 40 iterations. We observe then three cases, each independently thought to be macroparameters, which are conserved for millennia. Macroparameters affect all relevant categories in a uniform way.
On the other hand, it is easy to observe examples of relatively short-lived parameter settings. Assuming that the class of English modals emerged through grammaticalisation in approximately the 16th century, we can see in contemporary English, less than 500 years later, that many of the modals are moribund: this is true in most varieties for need and dare, and in US English for must and may. Moreover, individual modals differ in the naturalness/possibility of inversion: in contemporary UK English for all uses of may and deontic might and in US English for all uses of might. Here, then, the relevant parameters concerning attraction of T by interrogative C are relativised to individual lexical items (the restrictions on “conditional inversion” in contemporary English show that irrealis C interacts with a different set of lexical items). This is a clear case of microparametric change, a change affecting a small set of lexical items, possibly just one, in relation to a specific feature property of a functional head. The class of modals seems to have started to change in this way in the 18th century, two hundred years, a mere 8 iterations of the learning cycle, after its creation through grammaticalisation. Another example of the same kind in a different domain concerns the subject-clitic systems of North-Western Romance (including “advanced” varieties of French – Zribi-Hertz 1994): here we see synchronically a range of systems featuring extreme microparametric variation concerning which clitics have reanalysed from their earlier pronominal status as functional heads in T/Agr and C systems (on Northern Italian dialects, see Poletto 2000, Manzini & Savoia 2005). Again, these systems appear to have emerged quite recently: Poletto (1995) observes that 16th-century Veneto did not have subject clitics, and conservative varieties of contemporary French also do not. “Jespersen’s Cycle” provides a further case: the bipartite negation of Stage II, in particular, can be short-lived (cf. Kiparsky & Condoravdi 2006; the fact that it has survived several centuries in Standard French is plausibly due to normative pressure). Further, the fate of the earlier preverbal negator in Stage III varies: in West Flemish, it functions as a polarity-emphasis marker (Breitbarth & Haegeman 2010); in French it is a “minifier” (i.e. an operator selecting the smallest possible value in a set of alternatives; see Rooryck 2008), and so on.
To summarise, we observe values of macroparameters affecting large classes of categories being conserved over millennia, in opposition to values of microparameters, affecting very small classes of lexical items, undergoing rather frequent change. Note that the same formal operations are involved in our examples: head-movement (incorporation, T-to-C) and licensing null arguments (radical pro-drop, subject clitics).
Finally, there are “intermediate” cases which we dub mesoparametric change. Mesoparameters concern entire syntactic categories and, as such, are “smaller” than macroparameters (which concern all possible categories), but “larger” than microparameters (which affect subclasses of lexical items). An example is the null-subject parameter in Latin and Romance. This parameter involves T licensing null arguments, and has been stable from Latin through most of the recorded histories of Italian, Spanish and European Portuguese. It has, however, changed in French and Northern Italo-Romance, presumably under contact influence from Germanic. Another likely case is (root) V2 in Germanic; although its diachrony is obscure and the evidence from Gothic, Old High German and Old English suggests it was not present in Proto-Germanic, it has remained remarkably stable across nearly all North and West Germanic varieties. English is of course the exception here, and again contact may explain why this language diverges (cf. Kroch & Taylor 1997). In the domain of word order, the West Germanic pattern whereby all categories in the extended projection of V (except C) are head-final is an example. This pattern is stable across West Germanic, and has been for at least a millennium; again, it changed in English, arguably under contact with VO North Germanic (Trips 2000) and also with Norman French. It has also changed in Yiddish at the T- and arguably v-levels, although VP remains variable (see Wallenberg 2009); note that this “downward propagation” of word-order change is dictated by the Final over Final Constraint (FOFC; see Biberauer, Holmberg & Roberts 2007, 2011).
We conclude that it is possible to isolate three classes of parameter: macro, meso and micro. Macroparameters concern whole classes of heads, and are diachronically very stable. Mesoparameters concern individual syntactic categories (T, V, etc) and are diachronically stable, but subject to change through contact. Finally, microparameters concern small classes of lexical items and are relatively prone to change (unless they are particularly high-frequency elements). Grammaticalisation, since it affects small classes of lexical items, is microparametric in nature. To the extent that grammaticalisation can be endogenous, microparametric change can be.
In line with the general view of parametric change as involving abductive reanalysis of PLD through language acquisition, macroparameters must be “easily” set; hence they resist reanalysis and are therefore strongly conserved. Meso- and microparameters are correspondingly less salient in the PLD. This view is consistent with the view of parametric hierarchies put forward in Roberts (2011): macroparameters represent the higher parts of a hierarchy, microparameters the lowest and mesoparameters an intermediate position. Importantly, this view does not imply that UG prespecifies the parameter types: the hierarchies emerge thanks to third-factor motivated acquisition strategies, possibly acting on minimal UG-specified content, possibly along the lines of the schema-based model suggested by Gianollo, Guardiano & Longobardi (2008). Macroparameters may be set at a stage of acquisition at which categorial distinctions have yet to be acquired, and hence their nature may be due to the “ignorance” of the learner (Branigan 2012). As categorial distinctions emerge, mesoparameters become available, refining the early acategorial system. As functional categories emerge, microparameters become possible. This view then explains how “superset” parameters can be set early without a “superset trap” arising; hence it is consistent with the Subset Principle (cf. Berwick 1985, Biberauer & Roberts 2009).
Finally, it is important to note that we are not proposing that macroparameters cannot change at all (this view would be incompatible with the principle of connectivity). Presumably, sufficiently intensive contact can lead to change in these parameters too: the evidence of head-initial to head-final change in the Southern Semitic languages under intensive contact with Cushitic may be an example (cf. Leslau 1945).