The Frequency and Degree of “intrusive /r/” in New Zealand English 63question of whether this link between frequency and degree may be echoed in
other variables. That such a link could exist follows well from exemplar models
of speech production and perception, in which lexical representations consist
of distributions of remembered exemplars (see, e.g., Johnson 1997; Pierrehum-
bert 2001, 2002). In such models, speech perception proceeds by matching the
acoustic signal to the distribution which it most resembles. Speech production
proceeds by activating a subpart of the stored distribution. Pierrehumbert (2002)
describes the process associated with producing a particular vowel: “to produce
an /i/, for example, we activate the exemplars in some area of the /i/ region in
the vowel space. This group of /i/s serves as a goal for the current production,
much as a perceived object can serve as a goal for a reaching motion.”
Hay and Sudbury (2005) have argued that the historical evolution of intru-
sive /r/ in NZE is best modeled by an exemplar theoretic approach in which
words and frequent phrases are stored, and in which “the alternation exists as
a set of correspondences between /r/-ful and /r/-less exemplars in the lexicon”
(819). Hay and Gibson (2005) have also argued for an exemplar account of
intrusive /r/. They conducted an /r/ phoneme-monitoring experiment involving
linking, intrusive and “real” /r/s. New Zealanders “heard” the intrusive /r/ much
less than other /r/s, and were more accurate across word-internal morpheme
boundaries than across word boundaries. Hay and Gibson argue that their
results can be well accounted for if one assumes that the more the /r/ is present
in the representation of a word, the more participants “hear” it in this task.
If the production target constitutes an averaging over a subpart of the
exemplar space, it follows that the production target could potentially gradi-
ently vary depending on the nature of that exemplar space. Certainly if all
previously encountered exemplars of the appropriate type were produced with
/r/, this should lead to an /r/-ful production. It should also lead to a robustly
/r/-ful production, given the overwhelming force of evidence in favour of the
/r/. However if the exemplar space is variable, it is possible that averaging over
/r/-ful and /r/-less exemplars could lead to an /r/-ful target with a relatively
weak constriction. That is, variability in the probability of /r/-production, may
lead to variability in the /r/-fulness of /r/s that are produced.
In this way, speakers of lower social classes will have encountered more
/r/s, and so will be both more likely to produce an /r/, and the /r/ that they pro-
duce will be relatively /r/-ful. For speakers of higher social classes, a smaller
proportion of their stored exemplars will be /r/-ful.
This account may also help explain some reports in the literature that intru-
sive /r/ may involve a lesser constriction than linking /r/, or than “real” onset /r/
(Mullooly 2004; McCarthy 1993). An onset /r/ is lexically present, and will be
produced for all exemplars of a particular word. Linking /r/ is also categorically