Articulatory Phonetics Final Project Part 3

Assignment overview

Now that you've spent lots of time thinking about your questions, hypotheses, and predictions, you need to work out the details of how you'll test your hypotheses to see which of your predictions are born out. The description of those details is called a "method", because it's the method by which you test your hypotheses.

As in the previous steps, you might discover during this phase that something you settled on earlier might not work so well anymore. That's part of the process! Be flexible and open to change that makes your work better. We're doing this in lots of parts so you have time for exactly this kind of rediscovery.

You'll do this phase in three installments: a first draft assignment, a peer review assignment, and a revise-and-resubmit assignment. The revise-and-resubmit assignment should be about 250-350 words. In the rest of this document I describe the things you'll need to do for this assignment ("Steps to success"), then offer some advice for getting the most out of it ("Recommendations from your teacher").

Steps to success

Step 1: Plan your method

Participants

First, you have to describe who you're going to collect data from. How many people? Are they experts at the behavior, and if so how do you evaluate expertise? Are you comparing multiple groups? What languages do they speak and how proficient are they? How old are your participants? How were they compensated for their participation? Did they report any relevant disabilities; and if so, how did you accommodate these?

You don't have time to answer all of these questions in this little assignment, but you should at least explain details about your participants that you think are important for your study. If someone were to repeat your experiment, they should be able to draw participants from the same groups you did to try to get the same results with the fewest confounds possible.

Data collection

In order to test your hypotheses and quantitative predictions, you need something to measure. That "something" you measure is data. In articulatory research, data are usually parts of spoken utterances, controlled to have a particular format that gives you a good chance to have enough measurements to test your hypotheses. The measurements themselves come from turning the data into a number. In a study of vowel duration, the spoken vowels would be data, and the actual duration of the vowel (acoustic duration from Praat or articulatory duration from ultrasound or another articulatory measurement tool) would be the measurement.

If you just ask your participant to talk (or whatever the behavior you chose is), they'll probably just give you one example and hope that's what you wanted. But one example is never enough, because articulation can be highly variable. What if the one example you got happened to be a weird one? Or worse, what if you can't measure the data because it doesn't include the feature you're studying? You want to plan what you'll ask your participants to do for you in advance. The more you control about the data you elicit, and the more similar your different elicitations are to each other, the better your measurements will be.

You also want a whole bunch of examples to measure so you can draw conclusions from general patterns. Eliciting 20-100 data points from your participants is common for a single measurement in many areas of science; if you're comparing measurements of different conditions (e.g., a test condition and a control condition), you need 20-100 data points for each condition. Your study might be like that, or you might need less; some articulatory studies are done with only a few data points. Talk to your consultant about how many repetitions of the vocal behavior would be reasonable to ask from a participant, and try to pick a number that gets you the most data without tiring your participant out.

Measurement tool

By now, you've read in Articulatory Phonetics about several different tools that are used in articulatory research. (There are a couple more in the first couple of chapters we skipped that deal with brain and muscle measurements, and another couple in chapters that have yet to be assigned.) Pick one of them, then describe what it measures and why it's a good measurement tool for testing your particular hypotheses.

Remember that each tool has advantages and disadvantages—that's why there are so many of them! You don't have to comment on your tool's disadvantages right now, but you should be prepared to defend your choice if your reviewer thinks there's a better way.

Measurements

What, specifically, are you going to measure? There are often many ways to measure the same thing, even with the same tool. Someone replicating your experiment should be able to use the same measurement you did. I give an example later in this document.

Common measurements in articulatory phonetics include the duration of an articulator movement, the difference in time between when two different articulators reach their targets, the location of a constriction, or the degree of a constriction how narrow the constriction is). There are plenty of options, and it all depends on your hypotheses and their predictions.

Step 2: Write

Write 250-350 words that include the following:

  1. Briefly summarize your hypotheses and predictions.
  2. Describe your participant(s).
  3. Describe the data you will ask your participants to produce.
  4. Identify which articulatory measurement tool(s) you will use and explain why the tool is a good fit for assessing your predictions.
  5. Describe what you plan to measure with your tool.
  6. Written in a scientific style.

A person who wants to replicate your experiment (to further support or reject your hypotheses, or to use as part of a related study) should be able to do exactly what you did. If you were reporting on a real experiment, you probably need more than the 250-350 words you have here to include all the details. But some journals require this kind of brief method description for abstracts or similar summaries, so it's good practice to explain as much as you can about the essentials of your work.

Submit what you write to the course site. This part of the project has three stages: the first draft assignment, peer review, and the revise-and-resubmit assignment. The details for these assignments will be posted closer to their due dates.

The next pages include some advice for finding citations for this project.

Notes and examples

Participants

A note on style: we say "participant", not "subject", to describe a person who gave us behavioral data. This is true in phonetics research in general, and especially important in this project which is all about doing science in collaboration with the population being studied.

Experiment design

For example, imagine you were trying to study the difference in tongue position for nasalized and non-nasalized vowels, predicting that vowel height would be lower . You might have each participant say two utterances like this pair:

  • "I asked for a pan again."
  • "I asked for a pad again."

The only difference between these two sentences is the minimal pair "pan" and "pad", which only differ by the condition we're testing: nasality. "Pan" and "pad" each have the same vowel (or at least, we assume they do in this study), so if I compare the tongue position of the vowel in "pan" against the tongue position of the vowel in "pad", I should be able to tell whether the tongue positions are the same or not.

I'd probably come up with another 4-5 pairs of sentences to make sure that I'm covered in case there's something weird about the "pan"/"pad" pair (5-6 pairs total = 10-12 sentences). Those sentences should have the same structure ("I asked for a _ again."—this is called a frame sentence), and I should use more nasal/non-nasal minimal pairs with the same /æ/ vowel so I can pool all my measurements together. When it's time to collect data, I'd ask my participant to say each sentence 5 times (10-12 sentences x 5 repetitions = 50-60 data points, 25-30 per nasal/non-nasal condition).

Of course, if my question and hypotheses aren't specifically about the vowel /æ/, I would probably want to collect data for some other vowel pairs to make sure the conclusions I draw are as general as my question. That means adding at least two more vowels like /i/ and /u/, making sure that I have just as many minimal pairs for those vowels as I did for /æ/, and those minimal pairs should be as close to the /æ/ minimal pairs as possible in terms of phoneme structure (e.g., /pæd/ —> /pud/). In the end, that's: 2 conditions (nasal + non-nasal) x 3 vowels x 5-6 pairs x 5 repetitions = 150-180 data points.

150-180 repetitive sentences is a lot of boring things for a participant to say, no matter how much they're getting paid. So, ideally, your experiment has a way to keep your participant's attention. This might include periodic scheduled breaks, gamification, or attention tests. It also helps if you can find a way to reduce the number of data points you need. (The best way to do this is with a question that has really specific hypotheses!)

This is the classic way to plan data collection in phonetics, but there are other things you can think about too. Do your participants need to provide data at multiple times throughout their lives (e.g., as they age, or before and after a medical procedure)? Do you need a confederate (which here means an actor or friend who knows about the experiment but isn't taking measurements) to have a conversation with your participant to guide them toward certain topics or make the speech more natural?

Measurements

In my example of tongue height for nasalized and non-nasalized vowels, it's not enough to say "I'm going to measure tongue height" because there are several ways of measuring tongue height, and how you take your measurement depends on the place of articulation. (It might have been better to measure constriction degree rather than height, but again, it depends on the question you ask.) For example, you might try to measure an /u/ vowel by finding the distance between the highest point of the tongue and the lowest point of the velum, since /u/s are often made with constrictions near the velum.

Putting aside that this might not be a good measurement for people who make very fronted /u/s that are still high but farther forward on the palate, there's a much bigger problem: this experiment tests nasality, and the velum changes height during nasals. With a velum-to-tongue distance measurement, I could accidentally conclude that the short distance during nasalized vowels comes from a very high tongue position instead of the low velum position. There are plenty of ways to fix this, but I might next try measuring tongue height by finding the distance between the edge of the tongue and the highest position the velum ever reaches for this participant in my data set. That way, my measurements have a fixed landmark.

When determining the measurements you'll do in your study, try to think about what exactly you need to measure to test your hypotheses. The tongue height measurements above can boil down to numbers in millimeters (the distance between the edge of the tongue and something else). What units will your measurements be in, and what tool lets you measure in those units?