56 Jen Hay and Margaret Maclagan
the F3 than social class suggests the latter may be more likely. We will return
to speculation about how such a link might arise in the discussion.
3.3 Is the variation in F3 perceivable?
Given the results outlined before, we were interested in whether it was pos-
sible to “hear” degrees of intrusive /r/. For variation to be socially mean-
ingful, it presumably has to be accessible to the listener. Yet it is not clear
whether individuals are really able to hear subtle differences in the degree
of constriction of an /r/.
We asked a trained linguist to listen to each /r/ in the data-set and dis-
tinguish between whether it was a “strong” /r/ which could plausibly form
a syllable onset, or whether it was a weaker form of /r/ which seemed to
contain a reduced degree of constriction. She coded 165 of the 192 intrusive
/r/s analyzed in section 3.2 as strong /r/s, and 27 as lesser /r/s. Obviously
what we were asking the analyst to do here was to impose an arti¿ cial
boundary on what is likely to be a continuum. Nonetheless, we were inter-
ested in whether degrees of “/r/-ness” (even if just two) could be perceived
by a listener.
A logistic regression of this analysis reveals that the single best pre-
dictor is the F3 of the consonant—the lower the F3 was, the more likely
the listener was to rate the /r/ as a “strong” /r/ (p<.01). The variation in
F3, then, seems to have been at least one criterion that this listener used,
suggesting that the variability in F3 was to some degree audible. Inter-
estingly the rate of /r/-insertion of the speaker also reaches signi¿ cance
(p<.05). That is, speakers are more likely to produce an audibly more /r/-
like /r/ if they are also speakers who have high levels of /r/-insertion over-
all. The model is shown in Tables 2.6 and 2.7, and the predicted analysis
based on the percent /r/ produced by the speaker and the F3 of the token is
shown in Figure 2.10. This binary analysis echoes the result shown before,
which demonstrated that the F3 of a consonant was well predicted by the
degree to which the speaker producing the consonant tended to produce
intrusive /r/, over all.
This suggests that it is at least plausible that this F3 var iance is behaving
as a sociolinguistic variable which carries social meaning. It is possible for
a trained linguist, who is permitted to listen to the signal as many times as
she wants to, to distinguish between degrees of /r/-ness. Of course, whether
this is also true of the general listener, on a single À eeting occasion, is not so
clear. This suggests interesting lines of enquiry for future perceptual work.