Sections 3.2 and 3.3 have shown that the persistence characteristics of TD are quite different from the persistence characteristics of ING. For ING, persistence is independent of lexical repetition, while for TD, persistence arises only in cases where the trigger and target are the same word. What might account for this difference? Two fairly obvious possibilities based on previous analyses of these variables are their relative degree of socio-stylistic evaluation and their loci within the grammatical system. ING is often held up as a socially meaningful, stylistically dynamic variable par excellence, correlating well with social class, gender, social networks, social evaluation, stylistic context, and more. TD, on the other hand, is most often discussed with respect to its linguistic conditioning, except when it has been
argued to be relatively insensitive to social and stylistic context (Labov, 2001b; Hazen, 2011). It could be, then, that we find what appears to be more robust persistence with more stylistically-dynamic variables, potentially as a result of stylistic clustering rather than true sequential dependence across tokens. Making such an account fit with the patterns of lexical repetition seen in this chapter, however, would not be straightforward.
A second possibility, opened up by my argument in favor of a morphological locus of variation for verbal ING, is that the difference in persistence behavior stems from the distinction between phonological versus morphological variation. My suggestion here is that morphological variation induces generalized persistence while phonological variation gives rise to only lexically-specific persistence. In chapter 4 I will expand on this sugges- tion with respect to the temporal decay of the different variables, and consider how such a distinction might be explained with respect to the operation of different memory systems interacting with the structure of the lexicon and the grammatical architecture. Here, how- ever, I will note only that the investigation of the non-verbal categories of ING will be of particular importance for this suggestion. There is both the empirical question of whether nominal and root-attached ING behave like TD in showing only lexically-specific persis- tence, and the more theoretical question of whether making a distinction between nominal and root-attached as categories is justified at all. Unfortunately, answering these questions more deeply will require a substantially larger dataset or perhaps experimental work; once the prime and target are constrained to be of the same nominal category and then further sliced up by lexical repetition and prime variant, the number of tokens is too low to draw even reliable descriptive statistics from.
When DH-stop and DH-null, as discussed in section 3.4, are considered in tandem with ING and TD, we can see that they line up in somewhat unexpected ways. Although on the face of it one might expect to see alignment between DH-null, being by hypothesis morphological, with ING and DH-stop, being by hypothesis phonological, with TD, in
fact the opposite pattern obtains. DH-stop shows generalized persistence with a lexical boost, while DH-null is lexically restricted. The behavior of DH-null might be understood as a quite different variable with a word-level locus, which would then look more like a case where each lexical item constitutes its own variable for the purposes of persistence. In other words, if the choice a speaker makes is between ‘them’ and ‘’em’ rather than between overt and null versions of the /dh/, then the use of ‘’em’ would not have any reason to impact a subsequent choice between ‘that’ and ‘’at’ because they do not share a structural relationship. This would then be consistent with the persistence pattern we see, and despite the surface similarity the lexically-specific TD persistence would have its origins in a different framework. Indeed, I suspect that nominal ING may show a similar pattern, with some lexical items having two forms, one velar-final and one apical-final, and others having only the velar-final version. This suspicion is partly driven by the observation that there are five nominal ING lexical items (wedding, beginning, clothing, evening, and
housing) that occur at least five times in the dataset without occurring in the apical form. Although standard variationist practice might exclude these, I object to that practice on the grounds that it does not distinguish between lexical items that cannot alternate (for example, the place namesReading, PA andFlushing, NY are not supposed to be variable (Labov, 2001c)) and ones that simply happen not to because they occur rarely and are expected to have low rates of /in/ anyhow. This is a subject for future work, making the prediction that nominal ING persistence should be lexically restricted as with DH-null. The behavior of DH-stop, in the framework I have been developing here, may seem somewhat more puzzling. Its resemblance to the verbal ING and past tense TD patterns suggests an analysis in which the variation in DH-stop is actually morphological. I will revisit this possibility in the next chapter.
All of the variables that I consider here sit in some way at the interface of phonology and morphology. Note that one of the most prominent variables in the corpus persistence
literature, /s/-deletion in Spanish and Brazilian Portuguese, could also be characterized, in at least some contexts, as morphological rather than phonological variation (that is, the variation could be described as a suffix alternation). It would be instructive to revisit this variable with an eye to teasing apart its persistence behavior in monomorphemes and mor- phologically complex words. However, while this interface position makes such variables of great interest for the complexity of their behavior and the possibility that the surface variation we see may be decomposable into multiple processes, it does tend to muddy the waters when considering the interpretation of the results.
If the distinction I am beginning to outline in this chapter is to stand, it will be of critical importance to anchor these results to comparable behavior from less ambiguously-located variables. A crucial testing ground for the hypothesis that phonological variation gives rise to only lexically-specific persistence is low-level phonological processes that are not understood to have any plausible morphological locus of variation, such as TH-fronting in British English. Conversely, it will be important to ask whether variables that target higher levels of the grammar or more abstract forms show the generalized persistence that I hypothesize characterizes morphological variation, particularly given my situation of the morphological structure of ING within a framework that treats morphology and syntax as a single system (Embick, 2008). The behavior of morphological variation under such a framework should bear a striking resemblance to at least some kinds of syntactic variation.