Take-home lab exercise 3 on "Method of Adjustment"

Ling 600.01 Phonetic Theory

We will do this take-home exercise during the fourth and fifth weeks of the quarter and the final report is by the beginning of class on Tuesday, November 1. The final report on in-class lab #4 (on "Measuring formants in English vowels") should be included as an appendix to the report on this take-home lab exercise.

Part 1 -- Production data.

The data for the first part of the exercise are measurements of the first and second formants in the vowels that you and your language consultant produced for the English words that you recorded for in-class lab 4 -- i.e., the measurements that that lab was about. Here are the steps you should take to make those measurements.

1. Decide where you are going to measure the formants. For example, will you take the measurements at the midpoint of the voiced part of the vowel? Or will you try to identify a "target" or "steady state" point for the measure?
2. Make the measurements.
   • If you decide to measure at each vowel's midpoint, look at a few of the words and make sure that you can identify the beginning and the end of the vowel in a consistent way across all of the words, both for your productions and for the language consultant's productions. Then make a TextGrid object that has an interval tier, in which you place a boundary at the beginning and at the end of each vowel, and a label identifying each relevant interval by a tag for the vowel or the word. If you decide to measure at the vowel midpoint and want to automate this a bit, ask us and we can help you write a script in class that calculates where the midpoint is and puts a candidate tagged point in a points tier and then extracts the formant values there for you.
   • If you decide to measure at the "target" of "steady state" point for the vowel, look through all of the words, and define some criterion (or set of criteria) for identifying this point in a reasonably consistent way across the different vowels. Then make a TextGrid object that includes a points tier, where you can place a marker for each steady state and a label identifying the vowel or word that each marker points to. If you decide to measure at the steady state and want to automate this a bit, ask us and we can help you write a script in class that opens the file and zooms in on the word and puts your cursor at the tagged point for the steady state in the word and then extracts the formant values there for you.
   Whichever way you measure, if you decide to write a script to extract the formant values, be sure to make it an interactive one, designed to "pause" at each labeled interval or point so that you can adjust the LPC model as appropriate if the formant values that you see in the red speckles do not match the formant values that you see in the spectrogram on which the red speckles are overlaid. As you extract the formant values -- either by copying and pasting values from the info window or by running a formants extraction script that you write, note the LPC model that you are using. That is, note how many formants you are asking Praat to return and what frequency band you are telling Praat to return that many formants for. Keep track of each word token for which you have to alter the LPC settings to tailor the red speckles specifically for that word.
3. Whichever method you choose, you should write a short "Methods" section specifying what method you used, what the criteria were for placing the interval boundaries or the steady state points, and what LPC model you used. (By "short" we mean a picture possibly, plus four or five sentences of text at most.) If there were any words for which you had to alter the LPC setting, you might note these in your methods section (and in that case, you can go over the four-sentence limit to accommodate).

Part 2 -- Run the Method of Adjustment experiment.

The data for the second part of this exercise are formant values that you will use to estimate two different "perceptual vowel spaces" for these English words. To get these formant values, you need to run two experiments that probe your vowel space and your consultant's vowel space using the "method of adjustment" paradigm described in:

Johnson, Fleming, and Wright (1993). The hyperspace effect: Phonetic targets are hyperarticulated. Language 69, 505-528.

Here's what to do in order to be able to run these experiments and get the two sets of formant values.

1. Point your browswer to the ChooseVowels folder that is linked in at this URL and download the MoA.zip file. Put this zip file into the directory where you want the stimuli and results to be stored.
2. Extract the Sounds directory and the 330 vowel stimulus files that it contains from the MoA.zip file. You might check to see that this worked by opening one of the files in Praat and playing it.
3. Extract the MoA.txt file from the MoA.zip file, putting it in the parent directory to the Sounds directory that you extracted.
4. Rename the MoA.txt file to be IDcode.txt, where IDcode is the ID code that you want to give to the person who will be the subject. For example, when Mary Beckman ran this on herself, she renamed the file MaryBeckman.txt
5. Open Praat and use the Read ... command to read the IDcode.txt file into the Praat Objects: window. It should be identified as an "ExperimentMFC" object. Select the object and run it, having you or your consultant serve as the subject.
6. After the subject has finished choosing and rating vowel stimuli for each of the 11 words that will be presented as the stimuli, use the "Extract results" command to get a ResultsMFC object. Select that object and use the "Collect to Table" command to make a Table object. Use the Rename... command below the Praat Objects: window to assign the IDcode as the name of the Table. Use the "Write to table file..." command under the Write pulldown menu to write the results to a file that will be called IDcode.Table (or the like).
7. Repeat steps 4-6 with your language consultant as the subject. If you are having any trouble at all doing the above steps -- for example, if you are using a long-distance consultant -- ask us right away for help, so that we can help you figure out what to do this part of the exercise.
8. Once you have the results, either download the addFormantValues.R code for the ChooseVowels URL and adapt it and run it on your results files to get the formant values corresponding to the vowels that you and your language consultant used, or upload the files to the drop box by the beginning of class on Tuesday, Ocotber 25, and ask us to do this for you.

Part 3 -- Your report.

For each of the participants in the experiments (i.e., yourself and the language consultant), make a plot of the production vowel space in ERB space, with a plot of the perceptual vowel space overlaid. We will write a script in class together called h_dVowels.R that you can adapt to do this.

Your report should include your methods section from Part 1, a results section which is the two figures, and a short (no more than two short paragraphs) discussion section.

In the first paragraph of the discussion section, you should note the patterns that you see in each plot, discussing the relationship between the production space and the perceptual space of each speaker. Speculate about any gross discrepancies. Can these be attributed to the difference in vocal tract size between the subject and the perceived speaker of the stimuli that you listened to in the MoA experiment?

In the second paragraph of the discussion section, you should compare across the two MoA figures. What differences do you see? Choosing the most salient one of the differences, discuss it in terms of the different first languages that you and the language consultant speak.