Dear Sarah, dear all, One of the misconceptions about the Watt and Fabricius method that tends to float around (and you are by no means alone on this, Sarah!) is the idea that the point corner vowels to be used HAVE to be beet, bat and the u' construct. This is not what we wrote in the original article. What we wrote was that each speaker's vowel space should be normalised using a) the vowel which has the highest F2 and lowest F1, i.e. at the top left corner of the vowel space; b) the vowel that has the highest F1, wherever in the back/front dimension it is - the modified W&F method (which is proving to be the better performing one of the two in some studies) leaves out the F2 value of this vowel, so that it does in effect find the lower centre point of the vowel space, and c) the u' value which is constructed using the F1 of the top left corner, and F1=F2. For any individual spaker, then, it has to be determined which measurable vowels in the high front vowel space and the low vowel space best correspond to the descriptions in a) and b9 above. It will not necessarily be the same vowel categories that will fulfill these conditions, *especially if you are dealing with regionally-varied origins. You don't specify where your data come from, but I would suggest that looking at each speaker from this perspective would be a place to start the detective work. Another important issue to consider is that the choice of normalisation algorithm method should be based on the method NOT removing the data details that we hear and want to see documented in measurable data. Testing explicitly just exactly what any normalisation method does to the data we give it is an important methodological step, and we need to document this and understand how we made that decision. In other words, if, after you have followed the steps I suggest above, the problem remains, then the next step is to go to a closely-related method such as Lobanov or the formant-intrinsic, vowel-extrinsic version of Nearey, and see how the three methods perform with your data. On that basis you will have a solid foundation to be able to argue "I chose X, because I tested X, Y and Z and X did the optimum job according to the criteria I had in the study, ie that I wanted to be able to see features a b c in the data after normalisation". Hope this helps! Anne _____________________________________________________ Dr. Anne H. Fabricius Associate Professor of English Language Director of Studies, English Programme, Department of Culture and Identity Member of CALPIU Research Centre www.calpiu.dk Bldg 3.2.5 Roskilde University DK-4000 Denmark __________________________________________________ -----Original Message----- From: VAR-L automatic digest system Sent: Friday, April 29, 2011 1:03 AM To: [log in to unmask] Subject: VAR-L Digest - 20 Apr 2011 to 28 Apr 2011 (#2011-41) There is 1 message totaling 47 lines in this issue. Topics of the day: 1. Normalisation of vowels ---------------------------------------------------------------------- Date: Thu, 28 Apr 2011 23:41:32 +0100 From: Sarah Haigh <[log in to unmask]> Subject: Normalisation of vowels Hi everyone, I was wondering if someone with more experience of vowel normalisation could possibly share it with me as I'm facing something of a dilemma. To try to summarise, I'm working on three vowels (GOAT, NORTH and PRICE) and have been analysing them individually. I have 16 male speakers, which I am analysing in four regionally-based groups using the Watt & Fabricius method on the NORM suite website. When considering monophthong variants in particular, GOAT and PRICE seemed to give me results that were broadly in line with what I was expecting, but the results for NORTH did not correspond at all with what I could hear through plain auditory analysis. I subsequently realised I'd made a mistake and hadn't included the "corners" of the vowel spectrum on the NORM form - ie, the 'bat', 'beet' and 'school' values. However, having gone back and included these, the results look very much the same as they did without! I'm not discounting the possibility that I'm making some other error, or even that normalisation isn't appropriate for the data I'm trying to put into it. So if anyone could give me any thoughts I'd be very grateful! And if I *am* doing things right (or even if I'm not), why should GOAT and PRICE give me the results I predicted, while NORTH does not seem to at all? Any thoughts appreciated! Many thanks, Sarah ######################################################################## The Variationist List - discussion of everything related to variationist sociolinguistics. To send messages to the VAR-L list (subscribers only), write to: [log in to unmask] To unsubscribe from the VAR-L list, click the following link: http://jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=VAR-L&A=1 ------------------------------ End of VAR-L Digest - 20 Apr 2011 to 28 Apr 2011 (#2011-41) *********************************************************** ######################################################################## The Variationist List - discussion of everything related to variationist sociolinguistics. To send messages to the VAR-L list (subscribers only), write to: [log in to unmask] To unsubscribe from the VAR-L list, click the following link: http://jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=VAR-L&A=1