The third and fourth morphophonological variables that I turn to are both found in the function words beginning with voiced alveolar fricatives. I will refer to them as DH-stop and DH-null, and I will treat them as two separate variables although it would be possible to unify them as a single variable. DH-stop is the use of a non-continuant segment (a stop or flap) in the place of initial eth, giving rise to the well-known but stigmatized forms written ‘dis’ and ‘dat’ for this andthat. Previous research indicates that, unlike in other nearby
speech communities such as New York, only word-initial and voiced alveolar fricatives are variable in Philadelphia (so thatthinkdoes not become ‘tink’ andfatherdoes not become ‘fadder’ (Labov, 2001c). DH-stop has often been compared to ING in terms of its stylistic sensitivity in Philadelphia; see for example Labov (2001a). One advantage of DH-stop over ING and TD, as a variable to study, is that it occurs many times in any stretch of speech, since all of the contexts for its use are high frequency function words. Of course, this also means that coding DH-stop is much more time-consuming (from a time-per-interview perspective) than coding ING or TD; accordingly, I coded only 42 interviews for DH-stop rather than the full set of 122 used for ING and TD.
As a methodological note, I count any degree of perceptible continuancy in a DH-stop token to be indicative of the relatively standard instance of the variable, following Labov (2001c), who notes that affricate-like tokens do not arouse the same sort of stigmatization as a fully non-continuant (stop) variant does. DH-null occurs in the same set of function words as DH-stop does, but instead of the variant being a non-continuant segment it is null, so that for example themis pronounced ‘’em’. When examining the influence of lexical repetition in both DH variables, I ignore contraction so that, for example, ‘they’ll’ is treated as an instance of ‘they’. I also exclude the lexical item the because it is of such high frequency that it swamps the rest of the data set, and is also quite difficult to code because many instances of it are so fleeting that they are reduced to near-imperceptibility. I do, however, treat the occurrence of the as an exclusion context that interrupts the formation of prime–target pairs, so that the potential impact of DH variability in the on other DH words is mitigated. Finally, I implicitly assume the ordering where DH-null comes before DH-stop, so that the proportions for DH-null represent the rate of null tokens out of all DH tokens in the relevant categories whereas the proportions for DH-stop represent the rate of null tokens out of only the tokens that remain after DH-null tokens are excluded. This relationship is represented schematically in figure 3.14.
Figure 3.14: Assumed relationship between DH-null and DH-stop
The previous two variables have shown interesting patterns in the relationship between persistence and lexical repetition. Comparing this interaction in DH-null and DH-stop reveals that there is a difference between the two DH variables. Figures 3.15 and 3.16 show the raw proportions of how persistence is affected by lexical repetition. From these figures, it appears that while there is a persistence effect in both cases, for DH-stop it is a generalized persistence effect reminiscent of the one seen with verbal ING, whereas for DH-null it is an effect that appears only when the prime and target are the same word, just as we saw for monomorphemic TD.
Tables 3.8 and 3.9 present the estimated parameters from the multivariate modeling in the same vein as the previous variables. For both DH-stop and DH-null, the control predictors are speaker sex, speaker age (centered), preceding segment (vowel, consonant, pause), and speech rate (same measure as for ING and TD). These models also include the lag term and its interaction, which will be discussed in the next chapter, as well as random intercepts by speaker and random slopes by prime variant. Again, the full model results are available in the appendix. In table 3.8, the main effect of a stop shows that there is a significant persistence effect for DH-stop even when the prime and target are different words. The positive main effect of lexical repetition and the negative interaction of the stop prime and same word indicate that there is a lexical boost, for the effects of both stop
0.00 0.25 0.50 0.75 1.00
Diff word Same word
Probability of contin
uant
Previous_DH
other stop
Figure 3.15: Persistence in stop/continuant variability in DH-stop, error bars are 95% Clopper-Pearson CIs 0.00 0.25 0.50 0.75 1.00
Diff word Same word
Probability of retention
Previous_DH
null other
Figure 3.16: Persistence in /dh/ presence/absence in DH-null, error bars are 95% Clopper- Pearson CIs
Estimate Std. Error z value p-value
Stop prime -0.45 0.12 -3.82 <0.001
Same word 0.25 0.11 2.37 0.018
Stop prime x same word -0.68 0.19 -3.61 <0.001
Table 3.8: Estimated coefficients for DH-stop parameters relevant to persistence (N = 7462)
Estimate Std. Error z value p-value
Null prime -0.09 0.17 -0.53 0.598
Same word 0.44 0.15 2.98 0.003
Null prime x same word -2.08 0.29 -7.18 <0.001
Table 3.9: Estimated coefficients for DH-null parameters relevant to persistence (N = 8409)
and continuant primes. For DH-null, in contrast, table 3.9 shows that the main effect of a null prime variant is not significant, but its interaction with lexical repetition is: just as in monomorphemic TD, the persistence effect in DH-null is restricted to cases where the prime and target are the same word.