Evaluation of a Reading Tutor with Synthesized Speech Feedback for Dutch Speaking Elementary School Children with Reading Difficulties Leen Cleuren Lukas Latacz Yuk On Kong Maaike Vandermosten Werner Verhelst Hugo Van hamme Pol Ghesqui

1
Evaluation of a Reading Tutor with
Synthesized Speech Feedback for Dutch Speaking
Elementary School Children with Reading
DifficultiesLeen Cleuren Lukas Latacz Yuk On
Kong Maaike Vandermosten Werner
Verhelst Hugo Van hamme Pol Ghesquière
2
Introduction

• Children with reading difficulties (RD) in Dutch
• Dutch is an orthographic transparent language
  grapheme-to-phoneme-rules are much more
  consistent than in English
• RD are characterized by speed problems, rather
  than accuracy problems
• Early, regular and adequate intervention needed
• To support reading development
• To support appropriate functioning in all school
  domains

3
Introduction

• Intervention to speed up reading?
• Not focused on phonological skills (e.g. knowing
  how to split up a word in its constituent
  phonemes)
• ? improves accuracy but not speed
• Focused on
• READING READING READING!!
• (To overcome the Matthew effect)
• What kind of reading?
• ? still subject of discussion

4
Introduction

• A general intervention approach
• Let children read aloud under supervision
• Give feedback on what they read
• Repetitive versus non-repetitive strategies
• Repeated reading read the same passage over and
  over again until a preset criterion is reached
• Wide reading no rereadings, always new reading
  material
• Different feedback techniques
• General feedback indicates whether performance
  is right or wrong
• Corrective feedback idem provides help to
  correct the error

5
Introduction

• Daily reading practice necessary
• Not always possible due to classroom issues and
  governmental health insurance constraints
• A computerized reading intervention program could
  offer a welcome solution
• ? error detection and child tracking speech
  recognition
• ? feedback and decoding assistance
  synthesized/digitized speech ability to
  highlight words or word parts
• SPACE project Speech Algorithms for Clinical
  and Educational applications

6
Remark

• Synthesized versus digitized speech
• Digitized speech sounds like recorded speech it
  is speech that is converted into a digitized
  format and then reconverted back into speech.
• Synthesized speech is computer generated speech
  that is created by concatenating pieces of
  recorded speech that are stored in a database,
  based on various acoustic/phonetic algorithms.
• Programs can be open so that its user is not
  restricted to the learning material chosen and
  implemented by the authors of the program.

7
Present Study

• Evaluation of the Dutch Reading Tutor with
  synthesized speech feedback developed within the
  SPACE project
• Can substantial improvements in reading accuracy
  and speed be obtained in reading disabled
  elementary school children that used the SPACE
  Reading Tutor?
• ? intervention study
• Instructional-level non-repetitive text reading
• Corrective phoneme-by-phoneme feedback
• e.g., speelgoed

8
Participants

• 10 pairs of 2nd 6th grade elementary school
  children
• Normal IQ, Dutch speaking, normal vision and
  hearing
• 8 girls 12 boys
• Age range 6.75-10.75 years
• Individual matching on chronological age, IQ,
  mastery text reading level (AVI)
• ? no significant group differences (ps gt .10)

9
Pre- and Posttest Materials

• Paper-and-pencil tests
• One-Minute-Test 1 list of words
• Klepel 1 list of pseudowords
• AVI test stories
• Computerized tests
• Real Word Reading Test 3 lists of words
• PseudoWord Reading Test 3 lists of pseudowords
• Story Reading Test stories
• accuracy and speed independently registered
• 1-syl, 2-syl, 34-syl (pseudo)words separately
  assessed
• 3 stories at mastery, instructional and
  frustrational level

10
Intervention

• Instructional-level stories (commercial reading
  materials)
• Presented on a touch screen, paragraph-by-paragrap
  h
• No re-readings, always new reading material
• Wizard-of-Oz-method no computer speech
  recognition
• Help for a word child touches word and receives
  phoneme-by-phoneme feedback
• Errors erroneous word highlighted
  phoneme-by-phoneme feedback, child asked to blend
  phonemes together

11
/k/ /w/ /a/ /t/ Herhaal jij het woord ook eens?
En dan mag je verder gaan.
12

• Example Boy, 2nd grade, Instructional level
  AVI1
• poes kijkt naar de boom.
• daar zit nog een poes.
• op een tak, heel hoog.
• jan ziet lien.
• lien woont hier pas.
• ze zoekt haar poes.
• dag, zegt lien.
• hoe heet jij?
• ik ben jan, zegt jan.
• lien kijkt naar de poes op het hek.

13
Design

• Randomized pretest-posttest matched-group design
• Matched pairs randomly assigned to intervention
  or control group
• 20-minute reading sessions 4.5 hours of training
  in total

14
Motivation

• Improvement graph after each session
• Verbal encouragement by the Reading Tutor
• Stickers that could be exchanged for a present at
  the end of the study

15
Analysis Strategy

• Effects of intervention
• Mixed model analyses of covariance (ANCOVA)
• Pretest of each variable covariate
• Matched pair random factor
• Intervention-Control between-subjects factor
• Measure of effect size R²
• Influence of performance level on intervention
  effects
• Why? Because not all children obtained a
  clinical pretest OMT and Klepel score
• Therefore Clinical Average
• How?
• Intervention-Control effect allowed to vary at
  pair level
• groupperformance added to the fixed part of
  the original model

16
Results

• Feedback use of synthesized speech
• ? speech synthesis was good sounded naturally,
  was clearly audible and well understandable
• ? children got quickly used to the reading
  tutors voice and were very understanding w.r.t.
  occasional mistakes

17
Real Word Reading Results
Average children p .02
Average children p .06
p .15 R² .10
p .04 R² .19
p .003 R² .36
p .09 R² .17
p .15 R² .20
18
Real Word Reading Results
p .11 R² .29
p .16 R² .43
p .15 R² .20
p .32 R² .58
19
PseudoWord Reading Results
p .05 R² .40
p .16 R² .23
p .07 R² .37
20
PseudoWord Reading Results
p .03 R² .46
Clinical children p .01
p .01 R² .55
p .01 R² .57
21
Story Reading Results
Average children p .03
p .11 R² .19
p .30 R² .14
p .01 R² .61
22
Story Reading Results
p .23 R² .20
Average children p .01
23
Summary

• Aim of intervention study
• investigate whether improvements in reading
  accuracy and reading speed could be obtained in
  reading disabled Dutch speaking elementary school
  children that used the SPACE Reading Tutor with
  synthesized speech feedback
• Design
• randomized pretest-posttest (-posttest)
  matched-group design in which each child of a
  matched pair was randomly assigned to an
  intervention or control group
• Intervention
• individualized intervention sessions focused on
  non-repetitive instructional-level text reading
  with phoneme-by-phoneme feedback when an error
  was made or help was asked for

24
Summary

• Reading accuracy
• Posttest
• intervention gt control 1syl 2syl real words,
  instructional level stories
• average children
• intervention gt control 34syl real words
• Follow-up
• intervention gt control 1syl 2syl real words,
  1syl 2syl 34syl pseudowords, OMT
  SS
• average children
• intervention gt control 34syl real words,
  frustrational level stories
• 1 counter-intuitive result at both posttest and
  follow-up!
• intervention lt control mastery level stories ?
  suggestions?

25
Summary

• Reading speed
• Posttest
• intervention gt control 1syl 2syl real words
• average children
• intervention gt control frustrational level
  stories
• Follow-up
• intervention gt control 1syl 2syl real words
• 1 counter-intuitive result at follow-up!
• intervention lt control 1syl 2syl 34syl
  pseudowords
• ? explanation found

26
Conclusion

• Despite a very limited amount of training (M
  4.5 hours), beneficial influence of
  extra-curriculum practice with the SPACE Reading
  tutor on the reading performance of elementary
  school children, mainly w.r.t word and pseudoword
  reading!.
• Both clinical and average performing children
  benefit from exercising with the Reading Tutor.
• Gains obtained for both reading accuracy and
  speed.

27
Thank You

• For listening!
• To Lukas Latacz (VUB), Yuk On Kong (VUB), Werner
  Verhelst (VUB), Hugo Van hamme (KUL, ESAT), Pol
  Ghesquière (KUL, ORTHO)
• Questions?