Can Prosody Disambiguate Mathematical Structure? It Depends

Can Prosody Disambiguate
Mathematical Structure?
It Depends Who's Listening.
Michael Phelan
@phelanguist
SVALP
April 1, 2016
1
2
Three times five minus two times two is eleven, but
three times five minus two times two is eighteen.
-L. Bloomfield (1939)
3
Three times five minus two times two is eleven, but
three times five minus two times two is eighteen.
-L. Bloomfield (1939)
(3 · 5) – (2 · 2) = 11
3 · ((5 – 2) · 2) = 18
4
Do talkers disambiguate mathematical
expressions? Do listeners understand?
Are there limits to disambiguation?
5
Do talkers disambiguate mathematical
expressions? Do listeners understand?
Are there limits to disambiguation?
Do the answers to these questions
depend on math ability?
6
Prosody often signals sentence structure
7
Prosody often signals sentence structure
- Complex NP (Lehiste, 1973)
- PP attachment (Cooper & Paccia-Cooper, 1980)
- Prep/verb particle (Price et al, 1991)
8
Prosody often signals sentence structure
- Complex NP (Lehiste, 1973)
- PP attachment (Cooper & Paccia-Cooper, 1980)
- Prep/verb particle (Price et al, 1991)
Listeners make use of these distinctions
- With explicit attention to meanings
(Lehiste, Olive, & Streeter, 1976)
- Without explicit attention
(Speer, Schafer, & Warren, 2011)
9
A few studies have used mathematical
stimuli:
A + (E ∙ O)
(A + E) ∙ O
(Streeter, 1978)
10
A few studies have used mathematical
stimuli:
A + (E ∙ O)
(A + E) ∙ O
(Streeter, 1978)
A, + E, ∙ O
(Wagner & Crivellaro, 2010)
11
A few studies have used mathematical
stimuli:
A + (E ∙ O)
(A + E) ∙ O
(Streeter, 1978)
A, + E, ∙ O
(Wagner & Crivellaro, 2010)
12
Two studies with more complex stimuli
O'Malley, Kloker, and
Dara-Abrams (1973)
Holm, Bailey, & Laborde (1999)
13
Role of mathematical ability very unclear
- Aware/unaware of ambiguity
- “Used to listening to equations”
14
Why not a math test?
(Phelan, 2012)
15
Many speakers
used felicitous
prosody to
disambiguate
Prosodic grouping
mostly matched
syntactic grouping
(Phelan, 2012)
16
Felicitous prosody
positively
correlated with
accuracy (p < 0.05)
Infelicitous prosody
negatively
correlated with
item score (p < 0.01)
and overall test
score (p < 0.05)
(Phelan, 2012)
17
Speakers are producing disambiguating
prosody.
Can listeners use it?
18
Speakers are producing disambiguating
prosody.
Can listeners use it?
(Phelan, 2014)
19
Stimuli:
- Clips from previous production exp.
- 44 with two answer choices (Easy)
- 22 with three (Medium) or five (Hard)
(Phelan, 2014)
20
Stimuli:
- Clips from previous production exp.
- 44 with two answer choices (Easy)
- 22 with three (Medium) or five (Hard)
- 8 non-math English items
- PP, RC, NP ambiguities
- Pick appropriate paraphrase
(Phelan, 2014)
21
Subjects:
15 Children, ages 7 – 17
14 Adults, ages 30 – 59
Recruited at science museum
22
Results:
Trial Type
Accuracy (Chance)
Easy
60.7% ***
(50%)
Medium
39.4% **
(33%)
Hard
19.9%
(20%)
English
77.1% ***
(50%)
(Phelan, 2014)
23
English performance correlated with
math performance for both groups, but
more strongly for adults.
Kids
Eng ~ Math
Correlation
r = .588
Adults
r = .712
(Phelan, 2014)
24
Results by age:
25
Results by age:
26
Results by age:
?
27
Follow up:
More subjects, demographic info:
28
Follow up:
More subjects, demographic info:
- 117 subjects, ages 9 – 73
29
Follow up:
More subjects, demographic info:
- 117 subjects, ages 9 – 73
- Questionnaire asking about
- Math use in job/major
(y/n)
- Self rating of math skill (1-7)
- Nervous/uncomfortable (1-7)
30
Only “Easy” math trials
31
Only “Easy” math trials
Mixed with picture matching for English
32
Results
Math Accuracy by Age Group
100
90
% Chose Intended Expression
80
70
60
67.6
68.4
67.8
YA – No
YA – Yes
OA – No
65.2
55.5
50
40
30
20
10
0
Kids
OA – Yes
Age Group
33
34
35
36
Across age group, neither self-rating
nor nervous rating do correlates with
math accuracy
37
Within age group, self-rating and
nervous rating do correlates with math
accuracy
Group
Self Rating
Nervous
Kids
r = 0.182
-.187
YA – No Math
0.396
0.290
YA – Math
X
0.383
OA – No Math
0.226
X
OA - Math
0.228
-0.239
38
Accounting for differences?
39
Accounting for differences?
Blocked vs. Unblocked design
40
Accounting for differences?
Blocked vs. Unblocked design
Pictures too distracting?
41
Accounting for differences?
Blocked vs. Unblocked design
Pictures too distracting?
Individual differences in prosodic
processing swamp group effects
42
Conclusions There are interesting variations in how
people use prosody to disambiguate speech.
Some people are terrible at this in math, and
in non-math. What accounts for this
variability?
Everyone has important limits on prosodic
disambiguation.
43
Acknowledgments
Many thanks to Shari Speer, Cynthia Clopper,
Mary Beckman, Laura Wagner, and the
Speerlab and Phonies groups at OSU
Important feedback from Speech Prosody
2012 and 2014 attendees
44