The study evaluated the effectiveness of a computer assisted pronunciation training (CAPT) system called PARLING for teaching English pronunciation to Italian children. 28 children participated and were split into a control group that received teacher-led training and an experimental group that used PARLING. Both groups showed significant improvement in pronunciation quality from pre-test to post-test, with no significant differences between the groups, indicating that PARLING was as effective as teacher-led instruction. Difficult and unknown words showed greater improvement than easy words known by the children.
1. The Effectiveness of Computer Assisted
Pronunciation Training for Foreign
Language Learning by Children
Presenter: Sze-Chu Liu
Instructor: Dr. Pi-Ying Teresa Hsu
2102/10/15
1
2. Citation
Neri, A., Mich, O., Gerosa, M., & Giuliani, D.
(2008). The effectiveness of computer
assisted pronunciation training for foreign
language learning by children. Computer
Assisted Language Learning, 21(5), 393-
408.
2
4. Some Abbreviations
CAPT = Computer Assisted Pronunciation
Training
ASR = Automatic Speech Recognition
ITC-irst = Istituto Trentino di Cultura –
Istituto per la Ricerca Scientifica
e Tecnologica
4
5. Introduction
Accurate
perception
and
production
Advantage L1
acquisition
over adults hampers L2
Pronunciation
training for
young
learners
5
6. Introduction
Typical FL The use of
learning setting computer
Abundant
Limited oral
exposure spoken
examples
Rare
Self-paced
interaction with
practice
native speakers
6
7. Introduction
Interactive
Role-plays with
speech-based
the computer
games
CAPT
with ASR
Automatic Fun learning
feedback experience
7
8. Introduction
Research questions
Is the effectiveness of CAPT systems for
children better than that of teach-led class?
Is CAPT able to help learners to learn
pronunciation of difficult words?
8
9. The CAPT system considered:
PARLING
Developed by ITC-irst
• Pronunciation quality
• Isolated word level
Providing automatic feedback
• Presentation of oscillograms
• Animated characters
Meeting the requirements
• Match traditional training
• Highlight pronouncing isolated words
9
10. The CAPT system considered:
PARLING
The user interface of PARLING: 4 modules
10
11. The CAPT system considered:
PARLING
The ASR component
Training Process
• Native British English speakers + Italian
learners of English
Recognization process
•Forced time-alignment likelihood (A)
•Phone recognition likelihood (B)
Decision making
• If A>B, respond “accept”; Otherwise,
“reject” 11
12. Method
Participants
• Total: 28
Number of
• Control group: 15
samples • Experimental group: 13
• 11-year-old Italian native speakers
Profile • Same public school
• Same curriculum
• All had 4 years of English FL
Background classes
12
13. Method
Training Procedure for Control Group
Teacher • 4 British teachers
Schedule • 4 sessions, 60 minutes for each
session
• Hansel and Gretel (Englsh version)
Material • Printed handout
•Teacher-led
•read
Teaching •explained
activities •Provided the correct
pronunciation
•Prompted to repeat aloud
•Played printed word game
13
14. Method
Training Procedure for Experimental Group
Teacher • Work with PARLING
Schedule • 4 sessions, 30 minutes /session
• Hansel and Gretel (Englsih version)
Material • Story excerpt shown on screen
• Students-driven
Teaching • Listened and repeated
activities • Repeated a word until
permitted
• Played a word game
14
15. Method
Testing Procedure
Training
Pre-test Post-test
Procedure
Children read and record 28 isolated words
Recordings scored by 3 experts
The 28 words were classified as
easy (n = 21)/difficult (n = 7)
known (n = 21)/unknown (n = 7)
15
16. Method
Rating Procedure
The pronunciation quality of each utterance is scored on a 10-point scale.
S1 S2 S28
Word #1 Audio file #1
(e.g. away)
Word #2 Audio file #2
(e.g. birds)
Word #28 Audio file #28
Speaker
Speaker Speaker
#2 #28
#1
In total, each rater assigned 1656 scores.
16
17. Method
Table 1. Audio files scored by each rater
Single-word Rater Speaker scores
scores reliability
17
18. Results
Reliability of ratings
Table 2. Reliability coefficients (Cronbach’s alpha)
These coefficients indicate high inter- and
intra- raters reliability.
18
19. Results
Single-word scores vs. Speaker scores
A strong, positive
correlation between the two
scores (r = 0.884, p<0.01)
The speaker scores
were for the rest of
the analysis.
Figure 3. Correlation between single-word and speaker scores
19
20. Results
Table 3. Pre-test and Post-test speaker scores for the two
groups
T-test results: t = .321, p = .754
Pronunciation quality is NOT significantly
different in the pre-test.
20
21. Results
Table 3. Pre-test and Post-test speaker scores
for the two groups
ANOVA #1: F(1,26) = 78.818, p <0.05
A significant effect for test time
indicated! 21
22. Results
Table 3. Pre-test and Post-test speaker scores
for the two groups
ANOVA #2: F(1,26) = 0.610, p = 0.442
Training group has NO significant
effect. 22
23. Results
Table 3. Pre-test and Post-test speaker scores
for the two groups
ANOVA #3: F(1,26)=0.548, p = .446
No significant test × training
interaction is found. 23
24. Results
Pronunciation quality of specific types of words
Difficult/Unknown Easy/Known
Mean SD Mean SD
Pre-test 3.06 1.10 5.11 1.08
Post-test 5.59 0.93 5.74 0.79
Over all 4.32 1.01 5.44 0.93
ANOVA test #4: F(1,26) = 144.729, p < 0.01
A significant effect is revealed for the
test.
24
25. Results
Pronunciation quality of specific types of words
Difficult/Unknown Easy/Known
Mean SD Mean SD
Pre-test 3.06 1.10 5.11 1.08
Post-test 5.59 0.93 5.74 0.79
Over all 4.32 1.01 5.44 0.93
ANOVA test #5: F(1,26) = 57.531, p < 0.01
A significant effect is shown for word
type.
25
26. Results
Pronunciation quality of specific types of words
Difficult/Unknown Easy/Known
Mean SD Mean SD
Pre-test 3.06 1.10 5.11 1.08
Post-test 5.59 0.93 5.74 0.79
Over all 4.32 1.01 5.44 0.93
ANOVA test #6: F(1,26 ) = 60.080, p < 0.01
A significant effect for word type is
indicated in the Pre-test.
26
27. Conclusions
The improvement in pronunciation quality
of isolated words in the two groups are
comparable.
The improvements in pronunciation quality
of difficult/unknown in the two groups are
comparable.
27
28. Reflection
If the participants change to college
students, the material should be carefully
selected.
The improvement of pronunciation quality
at sentence level can be one direction for
future research.
The sample seems to be small.
28