The primary goal of this chapter has been to demonstrate a connection between the phe- nomenon of persistence and questions of grammatical structure, both categorical and vari- able. There is evidence for the mediation of observed persistence effects by grammatical structure: clear evidence for the circumscription of persistence by grammatical contexts
for surface-similar instances of variation, and evidence that is at least consistent with the possibility that variation at different levels of the grammar gives rise to different types of persistence behavior. The sensitivity of persistence to linguistic structure, insofar as my proposal can be shown to hold across new cases, makes it a useful source of quantitative evidence for distinguishing morphological from phonological variation and for identifying the underlying unity or disunity of surface-similar strings. Specifically, I have suggested that persistence appears in a generalized form across instances of a variable only when the variation is morphological. In contrast, it does not appear in variation that is the outcome of a probabilistic phonological process, except when the prime and target are the same word. Persistence can thus be used as a source of evidence on two points. One is the distinction between variation at different levels of the grammar, particularly morphology and phonol- ogy; the other is the diagnosis of multiple underlying sources for variation that appears unified on the surface. Moreover, the very observation that persistence reflects underlying, rather than surface, structures gives us reason to see a connection between persistence and the psychological factors at play in the production of language. The next chapter extends the analysis of ING, TD, and DH to make explicit and strengthen the connection between the persistence patterns observed here and their cognitive bases.
Chapter 4
Persistence in cognitive perspective
In chapter 3 I argued that persistence reflects the underlying grammatical origins of ob- served variation and demonstrated that different variables have different quantitative per- sistence patterns, particularly with respect to the interaction of persistence and lexical repe- tition. Although those analyses did include a lag term to capture the distance between prime and target, plus the interaction of the lag term with the prime variant and lexical repetition effects, I did not present or discuss the lag-related model coefficients in that chapter. Here I take up that analysis with the same set of variables (that is, verbal ING, monomorphemic and past tense TD, DH-stop, and DH-null), aiming to characterize the basic decay patterns of the persistence effects that were demonstrated in chapter 3. This analysis is motivated by the interest in the experimental literature in using decay profiles to disentangle priming effects with different underlying mechanisms.
4.1
Decay in the priming literature
The degree to which priming effects last over time has been of substantial interest recently in the structural priming literature. There is an apparent split between results where the
effect is no longer detectable with even a single intervening trial (Wheeldon and Smith, 2003; Branigan et al., 1999), and results where the effect is detectable at some distance (Bock and Griffin, 2000; Branigan et al., 2000), possibly even up to a week (Kaschak et al., 2011). To some extent these differences may be attributable to the methods used. For example, the week-long ditransitive priming effect that Kaschak et al. (2011) observe in their first experiment is after an exposure phase where the participants are exposed to 20 instances of one type of prime and 0 instances of the other (i.e., 20 double object sentences and 0 prepositional object sentences, or vice versa). This dramatic bias may induce a stronger effect than in the case where the design is to present a single prime and have the participant respond to a related target some moments later in the experiment; although the time delay is much shorter, the emphasis on the prime is also substantially weaker.
Another direction that has been pursued in an attempt to reconcile the temporal differ- ences across studies is the development of multi-factorial analyses of structural priming. The lexical boost, where a syntactic structure makes a more effective prime if its verb is re- peated in the target, has played a key role in the development of such analyses. Hartsuiker et al. (2008), for example, suggest that the seeming disparity between Branigan et al. (1999) and Branigan et al. (2000) is due to the latter study using primes and targets with lexical overlap while the former uses lexically distinct primes and targets. What Hartsuiker et al. show is that the basic effect of structural priming is relatively long-lasting (persisting at a stable magnitude through up to 6 trials), while the enhancement of the effect provided by lexical repetition is present only when the prime and target trials are adjacent.
There have been many similar efforts to explain differential decays across different fea- tures of repetition priming (“cat” primes “cat”) and morphological priming (“cats” primes “cat”) of words by localizing different effects to different systems. The most common proposal is that word recognition implicates a short-lived activation effect and a long-term episodic effect (see Tenpenny (1995) and Versace and Nevers (2003) for more comprehen-
Estimate Std. Error z value p-value
/in/ prime -1.59 0.19 -8.326 <0.001
Same word 0.72 0.33 2.154 0.031
Lag (log2 sec.) -0.08 0.03 -2.623 0.009
/in/ prime x same word -1.28 0.47 -2.731 0.006
/in/ prime x lag 0.19 0.04 4.952 <0.001
Same word x lag -0.06 0.09 -0.681 0.496
/in/ prime x same word x lag 0.08 0.14 0.575 0.565
Table 4.1: Estimated coefficients for ING parameters relevant to persistence, with verbal primes and targets (N = 4238)
sive overviews, as well as more recent work from Kouider and Dupoux (2009)). I will discuss the exact temporal findings of such results in section 4.5, in addition to suggest- ing that there are not two but three sources for priming effects to unify the set of results across the repetition priming and structural priming literatures and make sense of the com- plex set of persistence decay profiles of the morphophonological variables at stake in this dissertation.