RELG 456 Lecture 11: trarwr
15 Jun 2018
School
Department
Course
Professor
Intonational Bestiary
Goodhue, D., Harrison, L., Su, Y. T. C., and Wagner, M. (2015).
Toward a bestiary of English intonational tunes.
In
Poster presented at the 46th Conference of the North Eastern Linguistic Society, Concordia University.
Our Goal
Our goal is to test which intonational contours speakers use to convey certain intentions, when they are not aware
that the choice of contour is what the experiment is about.
In this particular experiment, there were three types of dialogues, which each tried to elicit a certain type of
response, either a contradiction, an incomplete response, or an incredulous response.
Intended Contradiction:
Emma:
So yesterday Sarah asked me if I was going to John’s Birthday party and I said no, I don’t even like him.
[Stage directions: Your friend Emma spent the whole day with John yesterday and you know for a fact that she likes
him. So you’re very surprised by what Emma says, and your answer should reflect that.]
Participant’s Response: You like John
Intended Incomplete Answer:
Emma:
I don’t feel like going to this party tonight, I have the feeling I might not like any of the people there
[Stage directions: You know your friend John is attending the party, and you know Emma knows and likes him, but
you’re not sure whether she’ll like anyone else, and your answer should reflect that.]
Participant’s Response
: You like John
Intended Incredulous Response:
Emma:
Yesterday Sarah kept saying mean things about John and I was really uncomfortable because John’s a nice
guy, I really like him.
[Stage directions: Just the other day your friend Emma was bad talking John, so you know for a fact that she doesn’t
like him. So you’re very surprised by what Emma says, and your answer should reflect that.]
Participant’s Response
: You like John
Results: The contours speakers used by context
(Note that these plots are based on one annotator’s decisions, and as you’ll see, some of them are annotation
errors…)
From <http://prosodylab.org/data/bestiary/contour/>
Ethics module included getting
consent to release data as corpus
online
-
your participants may
refuse this
Bestiary
-
type of books collecting
bizarre animals, very interesting
as half of them used to be
fictitious
-
for linguistics, looking
at the intonational tunes they
have in their inventory to use for
diff purposes
These intentions were not chosen
at random
-
it had already been
shown that they exist in English
Target utterance at the exact
same words
Three conditions
-
three
intentions
One target set
-
one item set
Ideally, we want a lot of item set
-
in total they had nine
One choice pt for every given item
set, does the participant see all
three conditions or only one? 3*9
= 27 pieces of dialogue
The motivation for this
being that the first
utterance of "You like John"
might affect the way they
pronounce it the second or
third time ('between
subjects' design)
-
But this might not be ideal
as people might start
repeating the intonational
tune, and we can't really
see what they do
systematically (within
participants)
-
But at least you can get
more data from each
participant
-
This is called a Latin
square design
○
Three of item sets for
each condition
○
Then, we might want
to have 18 item sets -
then we would have 6
pieces of dialogue per
each condition from
each participant
○
But can't we just randomize
it?
-
Isn't there a less artificial
way to get this data?
Spontaneous production?
-
Another possibility would be to
run this so you only see one
condition from each item set (per
participant)
10 diff participants per
dialogue
-
Latin square
-
make three
playlists
-
randomize the order and
play the same no of participants
for each playlist
You can also make three
playlists, but each participant
in this design is exposed to
all three playlists, and within
each playlist, the order of
stimuli within each playlist
will also be randomized
-
We can retroactively make
this a Latin square design by
only considering the first 9
results from each participant
-
If you want to show them all the
dialogue (27 trials)
Verum focus
-
stress on 'like' / 'do
like'
Shouldn't we also have a
perception component,
because even though they
have a variable distribution
of resulting data, we don't
know if it would be
perceived the same way?
-
Results
Falling contradiction tune conveys
the sense that information being
conveyed is already known to the
people in the conversation
'Other' - weren't able to classify
these
Intended Incomplete / Insinuation
answer = Rise Fall Rise (RFR)
The intended incredulity
tone is used as the question
form in British English
-
People use yes/no (question)
contour - people rarely used the
intended incredulity tone when
that was the target form
All three conditions (other than a
verum focus) were meant to
produce rising intonations
People in the 70s had claimed that
women use more rising intonations
In this study, little more than
twice the females than
males
-
gender was not
controlled for
-
The problem with this claim is that
they never specify which rising
intonation to use (question
contour is rising, but so is CC)
As annotations based on only one
annotator, some errors (if you click
on CC, some will not sound like CC
Perception ratings based on 10
people - perception experiment -
each utterance in three conditions
till got 10 ratings per condition
-
item combination
People actually rated CC as good
for the intended incomplete
answer condition
However, not all tunes
Have similar data in French
For contradiction condition, actually used verum focus
Sometimes, second language speakers will try and transfer intonational tunes to the L2, although they
might not be part of the L2's inventory
LING 455
30 January 2018 11:46 AM
Jan 30^J 2018 Page 1
