360 Renée van Bezooijen and Vincent J. van Heuven
analysis, and their speech should be spontaneously produced rather than read
out from paper. To aggravate matters, the type of speaker we were targeting is
not easily accessible. These are typically well-known public ¿ gures, celebri-
ties who will not be easily persuaded to participate in a scienti¿ c study. As a
feasible alternative we decided to record a televised series of weekly talk shows
featuring precisely the type of speakers that we were looking for. The particular
talk show, Het Blauwe Licht (The Blue Light), was produced by the “high-
brow” VPRO television network in the Netherlands. In each show two guests
discussed recent television programs, press photos and newspaper articles.
From the winter season of 1998–1999 onwards, the ¿ r st 16 male a nd 16
female Dutch-speaking guests who appeared in the television talk show were
recorded. The mean ages of the men and women were the same (47 years of
age). Per speaker some 6 minutes of speech were recorded. For each speaker
10 tokens of the target diphthong /͑i/ were selected from the recordings, along
with 5 tokens of /i/ and 5 tokens of /a/. Tokens to be selected into the data-
base preferably occurred before obstruents in stressed syllables of content
words. Multiple tokens of the same word by the same speaker were avoided
(for details see Edelman 2002).
2.2 Procedure
Both authors independently judged the vowel height of the onsets of the 320
/͑i/ tokens along a scale from 0 (maximally close onset) to 10 (maximally open
onset). Tokens were made audible with a minimal acoustic context of 500 ms
both preceding and following the target diphthong token. The scores for per-
ceived onset height were averaged per speaker, for each rater separately, so
that 32 pairs of onset height scores were obtained. The scoring was done with
substantial between-rater consistency, as is evidenced by the correlation coef-
¿ cient that was found between the 32 pairs of scores, r = 0.80 (p < 0.001).^1
2.3 Results
Figure 15.2 presents the perceptual ratings collapsed over both raters (after
Z-normalization per rater) for the 16 male and 16 female speakers, ordered
pair-wise from left to right in ascending order of conservatism, i.e., in descend-
ing order of perceived openness of the /͑i/ onset. It is quite clear from Figure
15.2 that the female speakers lead the male counter par ts in the lowering of the
/͑i/ onsets; in the ordering, every female member of the pair has a more open
onset than the male counterpart, t(15) = 6.61 (p < 0.001, one-tail).