Linguistics 203: Handout
Week 9: November 20,2002

Phonetic Experiments


  1. Types of phonetic studies:
    • Descriptive studies – e.g. measure vowel formants, duration of long and short vowels, etc.
    • Experiments – hypothesis testing. E.g. ‘is semantic focus marked by increased duration?’
  2. Experiments allow us to test for causal relationships.
    Correlations vs. causality.
    • In children, shoe size correlates with reading ability.
  3. Basics of experimental method
    • Manipulate some aspects of the experimental situation (independent variables).
    • Measure other aspects of the experimental situation (dependent variables).
    • If a dependent variable changes in response to a manipulation of an independent variable, then we can conclude that the change in the independent variable caused the change in the dependent variable IF no other factor is systematically influencing the dependent variable.
  4. Controlling ‘nuisance variables’
    • The ideal is often held to be an experiment in which the independent variables change, but everything else remains constant.
    • In the best of circumstances, this is not practical, and in linguistic experiments the best circumstances rarely obtain (see below).
    • More practical:
      • Eliminate alternative explanations for the hypothesized effect.
      • Keep conditions the same where practical, to be on the safe side.
    • Nuisance variables can also be controlled by randomization, e.g. subject selection.
  5. Many aspects of language co-vary, so it is often impossible to vary only the desired independent variable. E.g.:
      • Intensity and duration generally vary with vowel height.
      • Can’t find languages that are identical in all respects except vowel inventory.
    • Many properties of subjects in a given population also co-vary (e.g. age and shoe size of children).
    • Often alternative explanations are best eliminated by a series of experiments, rather than attempting to control for all of them within a single experiment.
  6. Some independent variables cannot be manipulated directly.
    • E.g. Lindblom (1963) studied the effect of vowel duration on target undershoot, but actually manipulated the stress and position of vowels to elicit a range of vowel durations.
  7. Controlling variability – random variation can obscure an effect.
    • Within-subject designs help to control for variation between subjects.
      • Between subjects: Each subject is in only one experimental condition.
      • Within subjects: Each subject is in multiple experimental conditions.
      • Classic example: measure subjects before and after training.
      • Phonetic example: measure Voice Onset Time of voiced and voiceless stops for each subject.
    • Include independent variables which aren’t of experimental interest,
      • e.g. following vowel in a study of stop VOT. This models the effect of vowel quality on VOT rather than treating it as random variation.
    • Try to control other sources of variability through the design of the experiment, e.g. speech rate.
  8. Statistical analysis - Test whether observed differences could have arisen by chance.
    • Given random variation within each experimental group, is an observed difference between groups due to sampling, or due to a real effect?
  9. Take copious notes through the course of any experiment.
    Keep materials (recordings etc) organized, labeled, dated, etc.
  10. Perception experiments.
  11. Some resources:
    • The phonetics lab.
    • Portable recording equipment (check out from Coco).
    • The lab RA (currently Rob Podesva, podesva@stanford.edu).
    • L205 Phonetics

Making a recording


  1. Microphone placement:
    • close to the speaker, to maximize intensity of signal.
    • not directly in front of the speaker, to avoid overloading from stop bursts ('pops')
    • at a constant distance from the speaker throughout the recording.
  2. Recording levels should be set as high as possible without overloading. This is to maximize the signal-to-noise ratio - the stronger the signal is, the less significant tape noise etc will be by comparison.
    • check levels frequently. Speakers may change volume.
  3. Begin recording with a statement of date, place, speaker names and material recorded.
  4. Play back recording (check).
  5. Keep good notes and records of contents of recording.