ERIC ED341026: Talking to the Computer: A Prototype Speech Recognition System for Early Reading Instruction. Report No. 91-3.

14 Pages·1991·0.81 MB·English

by  ERIC

DOCUMENT RESUME
CS 010 801
ED 341 026
Kantrov, Ilene
AUTHOR
Talking to the Computer: A Prototype Speech
TITLE
Recognition System for Early Reading Instruction.
Report No. 91-3.
Education Development Center, Inc., Newton, Mass.
INSTITUTION
91
PUB DATE
NOTE
14p.
Reports - r2scriptive (141) -- Reports -
PUB TYPE
Research/Technical (143)
MF01/PC01 Plus Postage.
EDRS PRICE
*Beginning Reading; *Computer Assisted Instruction;
DESCRIPTORS
Computer Software; Grade 1; Kindergarten;
Kindergarten Children; Man Machine Systems; Primary
Education; Reading Instruction; Young Children
Prototypes; *Speech Recognition
IDENTIFIERS
ABSTRACT
While computer software that talks has become
A
commonplace, software that listens still remains quite rare.
early reading
demonstration spebch recognition program designed for
market due
instruction developed in the mid-1980s never reached the
volatility of the
to the immaturity of the technology and the
used an
microcomputer hardware and software business. The program
(providing meaningful and
interactive speech technology system
18-word
appealing contexts in which young children explored an
vocabulary list) that did not make many errors in recognition,
of the
compared to other low-cost systems. The four components
software were: a training program; a story program; three games
that
designed to help children learn vocabulary; and a program
words.
allowed parents or teachers to enter additional vocabulary
Three field tests of the system over a period of 6 months wsre
2
conducted with 42 kindergarten and first-grade children in
(1) despite
Boston-area public schools. Results indicated that:
(2) the
technical problems, children enjoyed using the proqram;
the screen by
opportunity for children to control what happened on
and
reading words aloud was a powerful motivation to learn to read;
just
(3) the program was most successful with children who were
beginning to recognize words and with those having difficulty
learning to read. As the prices of speech recognition systems come
demonstration
down and quality goes up, the experiences of this
beginning to
speech recognition program may prove relevant to others
instruction and
explore speech recognition's potential for reading
the
remediation. (Two figures representing video images produced by
included.) (RS)
program and three tables of data are
********************************************************0.**************
Reproductions supplied by EDRS are the best that can be made
from the original document.
***********************************************************************
MMNIOFMTERUMNWMX
FacIlity has assivwd
The ERIC
thIs vosurnent for procespeng
to:
In our lodgment. this document
la all* of 4nterest 10 MI OW
ocelot/sae noted to the Nen
Indestng shoutd (elect thee
special points of vote.
Liking to the Computer
A Protptype Speech Recognition System
for Early Reading Instruction
Ilene Kantrov
Report No. 91-3
farming
cc
inw
.
,
,
_,
s
z
EACHiNG
w
&
A
A.
0
A 0
OF EDUCATION
S DEPARTMENT
neeee,cfl end /mP,Oveur1
"PERMISSION TO REPRODUCE THIS
Off,c e it Educef.ona
ORMATION
RESOURCES INF
MATERIAL H
BEEN 'RANTED BY
EOUCATIONAL
CENTf R tERtOp
teotocluted as
rlIfi men
o+Qandatil
j1.4.1Za document
pereet
e.ect from the
OF vnIting tt
.rniwOve
010
mare l'Ater, M10510
CI Mostar changes
----
oerrOduCtron autlftv
914 CICK u
prnment *tate ,n
rea
0
rainy/ken! uthrtsi
Poetts
r,elte&sertiv
To THE EDUCATIONAL RESOURCES
merd 110 001
pohey
(N RI p08,liOn of
INFORMATION CENTER (ERIC,
EDC
EDUCATION DEVELOPMENT CENTER
55 CHAPEL STREET NEWrON, MA 02160 .
IFS? COPY AVAll ARI F
Thlking to the Computer
A Prototype Speech Recognition System
for Early Reading Instruction
Ilene Kantrov
Report No. 91-3
3
Publications in the series Reports and Papers in
Progress grow out of
the work of the over twenty projects of
the Center for Learning,
Teaching, and Technology (M) of Education
Development Center.
LIT was established in 1982 as the Center
for Learning Technology to
study and develop applications for information
technologies. Early
work explored new and productive
uses in education for
microcomputers, video, CD-ROM, and multimedia. Building
on this
work with technology, a talented and skilled
staff of fifty currently
engage in research and policy, product design and development,
professional development
programs, and dissemination. LTT now
represents a wide-ranging capacity that
spans all levels from early
childhood to college and in-service,
a variety of content domains, and
learning and teaching in and out of schools.
The Center for Learning, Teaching, and
Technology is part of
Education Development Center, Inc. (EDC),
an international,
nonprofit, research and development
organization with offices in
Newton, Massachusetts, and Washington,
D.C. For over three decades,
EDC has been recognized
as a leading center for curriculum reform,
training, research, and institutional
development. EDC works
throughout the United States and around the world
to solve a wide
range of educational, health, and social problems.
To receive a list of current publications,
please write to
Publications
Center for Learning, Teaching, and Technology
Education Development Center
55 Chapel Street
Newton, MA 02160
Talking "iv the Computer
A Prototype Speech Recognition System for Early Reading instruction
Ilene Kantrov
This paper reports on the field tests of one speech recognition system
for early reading instruction. The software expanded the use of
speech recognition technolory beyond simple drill and practice by
providing meaningful and appealing contexts in which young chi!.
dren (here, kindergarteners and first-graders) explored an eighteen-
word vocabulary list. The results of these field tests reveal both the
particular challenges involved In designing educational programs
using speech recognition technology, and promising directions for
future work.
omputer software that talksthrough speech synthesis or digitized speech
has become commonplace (Parham 1988). We now have a variety of talking
111
cad processors and programs that issue spoken instructions, as well as programs
that read stories and speak the dialogue of plays.
Computer software that listens, on the other hand, still remains quite rare. The
technological and economic barriers to good quality speech recognition at an
affordable price remain formidable. A number of products have appeared on the
market, but none has met with great commercial success, and nearly all have been
aimed at the business rather than the school market .1 The few speech recognition
products targeted at schools,2such as those produced by Chatterbox Voice Learning
Systems, focus on drill in reading and math. The Chatterbox Voice Reading Ability
Texas Instruments chip. The system ran on an
Drill software, for example, drills students on
lists of words (from a total vocabulary of 800
Apple II+ or lie computer with 64K and one disk
words), providing feedback for correct and in-
drive.
correct responses.3
The Dragon system could recognize up to thirty-
It was speaker
two words at any given time.
Yet the educational possibilities of speech rec-
dependentmeaning that each speaker had to
ognition technology go beyond drill and prac-
Hopes and expectations for significant
°train" the system to recognize his or her pro-
tice.
applications of speech recognition have been
nunciation of the words in its vocabulary. And
especially high for reading instruction
it recognized only isolated, not connected,
speechmeaning that each word had to be
(Blanchard et al. 1987; Strickland et al. 1987).
preceded and followed by a brief silence.
This report describes a set of software programs
for early reading instruction developed in the
The great advantage of the Dragon system was
mid-1980s by me and my colleagues at Educa-
that, compared to other low-cost speech
tion Development Center in Newton, Massa-
recognizers (and even some not so low-cost
chusetts. The software was designed to demon-
systems), it had great reliability. That is, it did
strate some of the capabilitiesbeyond simple
not make many errors in recognition. Using a
drill and practiceof speech recognition,
standard test of isolated word recognition, the
combined with speech output, for young chil-
system performed with 99.3 percent accuracy.
dren learning to read. We were particularly
Since the standard test was performed under
interested In using the technology to help
optimum conditions, however, one object of
children learn to read words in meaningful and
our project was to determine whether the sys-
appealing contexts, as many reading specialists
tem was sufficiently rel fable when used by young
advocate.
children in typical school settings to permit its
Building on the original set of demonstration
use in early reading instruction.
programs, we had hcped to develop a series of
The Early Reading Software
speech-based reading software that would take
full advantage of the possibilities of the technol-
We developed and tested four software compo-
ogy. Because of the immaturity of the technol-
nents:
ogy, however, as well as the volatility in the
A training program, which both introduced
microcomputer hardware and software business
the child to the recognizer's initial eighteen-
at the time (1984-85), our demonstration pro-
word vocabulary (divided into two chunks
grams never reached the market and we never
of ten and eight words) and allowed the
got to develop additional software.
In the
recognizer to collect the speech samples
1990s, as prices of speech recognition systems
needed to run the other programs.5 Clever
come down and quality goes up, our experience
graphics and music made the training pro-
with the technology may prove relevant to
Jam fun for children to use.
others beginning to explore speech recognition's
potential for reading instruction and
A story program, entitled "Kody's Jungle
Adventure," which helped the child to read
remediation.4
the initial vocabulary in a meaningful and
enjoyable context.
The Technology
A set of three games designed to help the
Our product used an interactive speech technol-
child master the reading vocabulary.
ogy system developed by Dragon Systems, Inc.
A program entitled "Your Words," which
ia
The system included speech recognition soft-
enabled a teacher or parent to enter addi-
ware (which was integrated with our applica-
tional vocabularies (of up to sixteen words
tion software) and a printed circuit board with
each) which children could then train and
speech output as well as speech input capabili-
use to play the games.
ties. The high-quality speech output, achieved
All components provided instructions and
using custom-encoded LPC (linear predictive
prompts usingspeech output; even the program
coding) speech synthesis, was produced with a
TALKING TO THE COMPUTER
For instance, the message °Here comes the el-
menus used speech as well as readily recogniz-
ephant.* might appear, with "elephant" high-
able icons. Professional artistJim Carson created
lighted. The program would say, °Here comes
the programs' visually appealing graphics, and
the.' If the child said °elephant,* the elephant
original music was composed by a talented
would appear on the screen (see figure I), ac-
young musician, Paul Tegels.
companied by a musical theme.
The following sections describe the story pro-
Next, the message 'sKody wants to lift it." might
gram and the games in some more detail.
appear, with "lift" highlighted. The program
lure
Kody's Jungle Adve
would say, "Kody wants to." When the child
The story program, which we developed with
said "lift,' Kody would lift the elephant, and the
music would play.
the guidance of reading sped alist Carol Chom sky
of the Harvard Graduate School of Education,
If the child said something other than the word
was designed to give children practice In read-
called for, the program would suggest that she
ing words in a meaningful context. The program
"Try again." If the child still could not read the
featured a friendly koala named Kody. Speech
word, the graphic of the obstacle or action
output set the sceneKedy was lost in the
would appear, and she would get another chance
jungleand provided simple instructions: The
to read the word. The graphics therefore served
child could help Kody find his way home by
as prompts as well as rewards for successful
reading the words highlighted on the screen. A
If the third attempt failed, the
performance.
series of two-part episodes then presented Kody
program would read the entire sentence and
with a variety of obstacles (such as an elephant,
then proceed.
a zebra, a tree, or a flower) and options for action
At the end of each episode (obstacle plus action),
(such as riding, tickling, climbing, or jumping).
a graphic showing Kodyss progress through the
By reading the words highlighted on the screen,
jungle appeared at the top of the screen, and he
the child got the obstacles to appear and then
was shown running toward home. After four or
got Kody to perform the actions.
.
Tt°:fr-44
Kajltt
4..
r4k
4,
Figure 1.
V
3
7
BEST COPY AVAILABLE
MNTROV
Figure 2.
five episodes (depending on the number of
for a range of difficulty and for words that would
words trained), he would reach home. Each
make the program interesting to children.
time the child read the story, it would consist of
The Comes
a different set of episodes; a total of twenty-
eight combinations of obstacles and actions
The software included three reading games:
were available.
Elephant Game (Rearrange the Words): Words
The story program was modeled on the
appeared in a 3 x 3 grid, with one blank square.
way a
parent or teacher may read a story to a child.
The child could move any word adjacent to the
The adult will often stop at various points and
blank square into that square by reading It. The
ask the child to read the next word. Of course,
object was to move the word in the bottom
we had no intention of replacing such adult-
lefthand corner to the top rightthereby let-
child interaction; rather, we wanted to give
ting the elephant out a door at the top of the
grid. This game provided a lot of practice in
children additional practice in recognizing words
in meaningful contexts. We also expected the
reading the words, as most words had to be
program to require some adult supervision, es-
moved more than once.
pecially when first used. We felt that adultsor
Zebra Game (Tic Tac TOO: The child read the
older childrenshould interact with young-
word in the box he wanted to mark with
an X;
sters learning to read, at the computer as well as
the program played 0 (see figure 2). When the
away from it.
child won, the zebra danced.
As for the story vocabulary, all the words were in
Monkey Game (Concentration):
Words ap-
the kindergarten to second grade range. Because
peared in a 3 x 3 grid. The child could reveal the
our goal was to demonstrate the speech
colored rectangles "behind" a word by reading
recognizer's capabilities, rather than to supple-
it, and then matched pairs of rectangles to
ment any particular classroom reading program,
reveal a piece of a picture of a monkey. When
we made no attempt to correlate the vocabulary
all the pairs were matched, the picture of the
with standard reading texts. Instead, we aimed
monkey was complete.
4
8
TALKING TO THE COMPUTER
A
average, and high ability as rated by their
The games were independent of the story. Chil-
dren could play the games with the story vo-
teachers.
cabulary or with any other set of words that they
The children who tried out the system in these
trained the program to recognize. (These addi-
field tests enjoyed using it, despite some techni-
tional kts could consist of anywhere from seven
cal glitchesespedally in the first two trials.6
to sixteen words.) A number of additional
Our field testing showed that the opportunity
vocabularies were provided, or teachers and
the program provided for the child to control
parents could use their own word lists. Even
what happened on the screen by reading we:Cs
children who had mastered the story vocabu-
aloud was a powtrful motivation to learn ti
lary enjoyed practicing their reading skills by
read. The program proved especially successful
playing the games, and they could be further
with younger children (ages four to five) of
challenged by more advanced vocabularies.
average to high ability who were just beginning
to recognize words, and with slightly older
Field Testing and Revising the System
children (ages six to seven) who were having
We conducted three field tests of the system
difficulty learning to read. Tables 1-3 (below
over a period of about six months with a total of
and on page 6) show the mean numbers of
forty-two kindergarten and first-grade children
words learned by children at each of the three
in two Boston-area public schools. In the first
trials.
two field tests, which took place in Watertown,
Ironically, the technical problems with recogni-
Massachusetts, children were brought in pairs
tion, which were most severe in our first trial,
to the school gymnasium, where we had set up
seemed to account for the children learning to
the system. An adult introduced the program
read more of the words in that trial. That is,
and helped the children as needed. An observer
because children had to repeat the correct words
and video camera operator were also present.
more timesboth to train the system and to
The third field test, in Newton, Massachusetts,
read the story and play the gamesthey had
took place in the school's computer laboratory,
more practice with them, which probably ex-
a setting more typical of actual school computer
plains why, according to the pretests and
use. Students again worked in pairs with an
posttests we conducted, they learned more
adult, and the session was likewise observed and
words.
videotaped. Before each session, the adult ad-
ministered a pretest to see if the children could
After each trial, we worked to improve both the
read the words in the story vocabulary before
reading software and the recognition software.
In revising the reading software, we focused
encountering the program. Children were tested
on these words again after the session. Partici-
particularly on the nature of the child-machine
pants in all three tests included students of low
interactionthe dialogue that occurred between
Table 1. Number of words
learned in Trial *1
(out of ten)
Kindergarteners - 10th month of the school year
Teacher
Mean Age
Mean
Mean Words
Mean
Rating
Pretest Posttest
Learned
Yrs-Mos
of Ability
Score
(Range)
Score
Low (n=5)
0.0
3.6
5-7
3.5 (2-6)
.
-
.
Average (n=5)
6.4
6.4
6-0
6.4 (5-8)
-
.
d
WO (n=7)
5-8
7.7
7.7
6.6 (4-8)
5
9
Table 2. Number of words (out of ten) learned in Trial 02
Kindergarteners - 3rd month of the school year
Mean Words
Mean
Teacher
Mean
Mean Age
Learned
Pretest
Posttest
Rating
Yrs-Mos
Score
of Ability
Score
(Range)
-
,
0.0
0.0 0.0
Low (n=1)
5-2
..
.
0.0
3.0
3.0 (2-4)
Average (n=2)
4-2
,
3.7 (2-8)
High (n=3)
5.0
5-3
1.3
Unrated (n 1
0.0
2.0
2.0
5-4
First Graders - 3rd month of the school year
-
-
0.5
7.0 (6-8)
6-6
7.5
Low (n=2)
2.0
8.5 6.5 (6-7)
6-8
Average (n=2)
1.5 (0-4)
6-6
High (n=4)
10.0
8.5
Table 3. Number of words (out of ten) learned in Trial #3
Kindergarteners
- 4th month of the school year
Teacher Mean Words
Mean
Mean Age
Mean
Pretest
Rating Posttest Learned
Yrs-Mos
(Range)
of Abiiity
Score
Score
.
.
.._
0.0
Low (n=3)
2.7
2.3
5-4
.
.
,
,
0.5
Average (n=2)
2.5
2.0 (1-3)
5-0
.
,
High (n=5)
3.4
4.0 (2-8)
7.4
5-3
.
_