1 Statistical analysis of the data inFigures 2–4 is precluded because we do not have access to all the numerical data on which they are based. We are extremely grateful to Paul Warren for providing us with the right-hand graph inFigure 4.

2 We are aware that vowel length varies with phonetic context. Because our data were drawn from connected speech, it was not possible to control the phonetic context. Because 1,437 tokens ofnurse were analyzed across the speakers, we consider that there are sufficient data so that contextual differences will not affect the vowel length results.

3 Even though thenurse andgoose vowels are very close in F1–F2 space, their formant frequencies remain significantly different. There is a significant difference for both genders between the F1 frequencies ofnurse andgoose (female:F(1,24) = 86.4,p < 0.001, male:F(1,22) = 217.60,p < 0.001) and their F2 frequencies (female:F(1,24) = 22.4,p < 0.001, male:F(1,22) = 26.01,p < 0.001). In addition there is a significant vowel group interaction for both the male and female speakers, for both F1 (male:F(2,22) = 30.0,p < 0.001, female:F(2,24) = 26.13,p < 0.001) and F2 (male:F(2,22) = 6.17,p < 0.01, female:F(2,24) = 9.04,p < 0.01). For the male speakers there was a significant difference between the F1 values ofnurse andgoose for all three groups (HE:t(4) = 10.29,p < 0.01, PE:t(9) = 11.40,p < 0.001, Y:t(9) = 4.63,p <0 .05) and between the F2 values ofnurse ofgoose for the PE group (t(9) =5 .87,p < 0.01). For the female speakers there was a significant difference between the F1 values ofnurse andgoose for both the historical and present-day elders (HE:t(5) = 10.00,p < 0.01, PE:t(9) = 5.98,p < 0.01)) and between the F2 values ofnurse andgoose for the historical elders (t(5) = 6.78,p < 0.01).

4 We thank an anonymous reviewer for drawing this possibility to our attention.