The Frequency and Degree of “intrusive /r/” in New Zealand English 51about where that constriction may be. Regardless of the articulatory means, it
seems fair to say that, if a speaker increases the magnitude of the articulatory
gestures associated with /r/ (whatever they may be for that speaker) so that the
realization becomes more strongly /r/ like, the result will be a lowered F3. Our
general hypothesis was that there may be structured variation in how /r/-like
our recorded /r/s were.
In this analysis, we considered only tokens which had been analyzed as
containing an /r/. We excluded 14 examples of sofa words which contained an
intrusive /r/, but were produced without the base-¿ nal vowel (as in, e.g., “sof/r/
ish” rather than “sofa/r/ish”). These were excluded from the analysis, in order
to maintain comparability—i.e., all of the analyzed /r/s are clearly intervo-
calic. We also excluded the Oprah bases from this analysis. As seen in Figure
2.3, tokens of Oprah were highly unlikely to be produced with intrusive /r/—
this is because they were most often produced without the second syllable of
Oprah as in, e.g., Oprese. Only 7 Oprah tokens were actually produced with
an intrusive /r/. Because we expected the identity of the base to be a contrib-
uter to the F3 of /r/, we excluded these Oprah tokens from this analysis. All
other bases were produced with reasonable frequency (between 20 tokens (for
plough) and 54 (for claw)). The total data set analysed in this section contains
192 obser vations.
Obviously, comparing the raw, non-normalized F3 measurement is prob-
lematic, because of the inherent differences in this value due to vocal tract
length. Rather than measure the entire vowel space in order to normalize
the formant values, we chose to measure, for each speaker, their F3 during
“regular” non-intrusive /r/s. The read materials contained four instances of
the name Sarah (see Appendix 2.1). For each speaker we used Praat to mea-
sure the lowest point of F3 during the /r/ for all Sarah instances where the
spectrogram was clear enough for us to be con¿ dent of the analysis. The main
analysis problems occurred, as indicated previously, when the female speak-
ers used particularly high fundamental frequencies, so that it was dif¿ cult to
distinguish F1 clearly.
We then included this mean value as a predictor in an ordinary least
squares linear regression analysis. If the degree of constriction of intrusive /r/
is not sociolinguistically variable, then the value of F3 during Sarah should
theoretically account for all of the variation in F3 in intrusive /r/. This mea-
surement, then, effectively acts as a vocal tract length normalizer. We might
also expect some differences across speakers depending on which bases they
produced intrusive /r/ with, since the identity of the preceding vowel could
affect the value of F3. However by including the identity of the base in the
model, this variation should be accounted for.