Grundfragen der Experimentellen Phonetik

1
Grundfragen der Experimentellen Phonetik
3. Sitzung
Wintersemester 2001/2002Jacques Koreman
2
The experiment
A theory should not be based on introspection,
but on objective (or at least intersubjective)
data. To this end, an experiment must be carried
out. An example of intersubjective data naive
judges grammaticality judgments of sentences. An
example of objective data duration measurements
of naive speakers natural productions of
utterances.
Session 3
3
Hypothesis
The hypothesis is the starting point for the
experiment. It describes the expect-ations of
the experimenter w.r.t. the outcome of the
experiment and is formu-lated so that it can be
falsified.
Session 3
4
Dependent and independent variables
Once the hypothesis has been formulated, it must
be operationalised by choosing the independent
and dependent variables. The dependent
variable(s) is/are the measurements that are made
to objectively evaluate the hypothesis. The
independent variables are the factors whose
influence on the measured variable(s) are under
scrutiny.
Session 3
5
Stimuli
On the basis of the dependent and independent
variable the stimuli are chosen. These can be
nonsense syllables or words (possibly in a
carrier phrase), sentences, free conversation
(map task, story telling), etc. Utmost care
should be taken to control other possible
influences on the dependent variable. The
possibilities for control are often limited by
the demands on naturalness, which is often a
prerequisite for the generalisation of the
hypothesis to everyday behaviour (which is often
the goal of the experiment).
Session 3
6
Speakers
If possible, more than one speaker should be used
as subjects, in order to make sure that no
speaker-specific properties are measured. Also,
the speaker group should be controlled for
factors which can possibly influence the results
(sex, dialect, age, language proficiency, etc.)
and put restrictions on the possibility to
generalise them. The speakers sex may affect the
measurability of the dependent variables in the
case of formant values, for instance, female
speakers often present the experimenter with
measurement problems. The subjects are chosen

• randomly
• layered reflecting all possible influences (see
  above)
• proportionally layered reflecting the
  distribution of these influences in the population

Session 3
7
Experiment preparation
First, the experiment groups present their
hypotheses, independent and dependent variables
and the stimuli for their experiment. If it is
important to hide the aim of the experiment (e.g.
because what is investigated can be consciously
controlled by the speakers in case of a
production experiment) fillers are needed to
detract from the aim of the experiment. It is
important to note that the artificiality of
controlled experiments, which were often carried
out on the basis of read speech, has led to a
huge phonetic knowledge base of which it is not
immediately clear whether the results can be
generalised to natural situations (e.g.
spontaneous speech), nor whether other effects in
natural situations (other independent variables
which operate at the same time, but were
controlled in the experiments) overrule the
effects found.
Session 3
8
Stimulus presentation
In production experiments with read sentences,
speakers may contrastively accent the stimulus
word in a carrier phrase, or they may produce the
sentences with a list effect. Contrastive
accenting can be avoided by embedding the
stimulus words in different phrases instead of
using the same carrier phrase for all stimuli. Of
course, one has to be careful not to introduce
error by using different sentences.List effects
can be reduced by presenting the stimulus
sentences on separate cards (or on a computer
screen sentence by sentence). This will cause the
subjects to breathe in after each sentence and
produce each sentence as a separate prosodic unit.
Session 3
9
Stimulus presentation
The order in which the stimuli are presented can
be important in perception experiments, the
subjects response to a stimulus may not be
independent of the preceding stimulus, or there
may be a learning effect or a fatigue/boredom
effect. We should either present the stimuli in
random order or counter-balance them (blocks of
stimuli presented in one order to one subject
group and in reverse order to another subject
group). Of course, counter-balancing does not get
rid of the order effect, it just makes sure it is
not mistaken for an experimental effect (i.e. the
effect an independent variable has on the
dependent one) - it makes the variance in the
data greater and therefore reduces the likelihood
of finding significant effects to support the
experimental hypothesis.
Session 3
10
Random and constant errors
Random errors can be caused for instance by
measurement errors. If measurement errors depend
on the independent variable, a constant error is
introduced. An example would be if we compare
vowel duration after a semi-vowel versus after a
voiced plosive. Since the end of a voiced plosive
is traditionally segmented where the burst ends
in the speech signal, and semi-vowels at the
midpoint of the transition into the vowel, there
may be differences in vowel duration merely due
to segmentation criteria and not so much by
actual differences in vowel duration. This could
be considered as a constant error, leading to
misinterpretation of the results w.r.t. vowel
duration (or maybe we should say that the results
may give reason to reconsider the definition of
what is a vowel in the signal and to the
segmentation criteria).
Session 3
11
Speakers
The subjects must be chosen so that they form a
representative group for the population to which
the experimental results shall be generalised. If
we want to generalise to all speakers of German,
the speakers sex, age and dialectal background,
for instance, may be important factors. The
speakers sex may affect the measurability of the
dependent variables in the case of formant
values, for instance, female speakers often
present the experimenter with measurement
problems.
Session 3
12
Exercise
Decide what subjects you choose for your
experiment. Write an instruction for the subjects
in your experiment. The instruction should state
clearly what the task is which the subject has to
carry out. Care should be taken not to influence
the subjects behaviour w.r.t. the aim of the
experiment.
Sitzung 3
13
Literature list
Day, R.A. (1998). How to write publish a
scientific paper. Phoenix Oryx Press.
Session 3