THE OBJECTIVE MEASUREMENT OF NONSPEAKING  CHILDRENS' INTERACTIONS
  	WITH A COMPUTER-CONTROLLED PROGRAM FOR THE
          STIMULATION OF LANGUAGE DEVELOPMENT


            KENNETH MARK COLBY
		  AND
	    HELENA C. KRAEMER

	One  difficulty  in  evaluating  treatment  improvements   in
disorders  whose  defining  characteristics  are purely behavioral is
that we lack objective measurements. An objective measurement is  one
which is intersubjectively confirmable and impartially independent of
individual opinions, intuitions, and judgemments. When humans try  to
be   both   participants   in   and   observer-recorders   of,  their
interactions, objective measurement becomes  difficult.     But  when
one participant in an interaction is a machine such as a computer, an
opportunity arises for the machine itself to record observations  and
to  collect data.  This capacity of a computer allows us to develop a
measurement standard , a basis of comparison  in  which  interactions
can  be  considered  similar  or  different  according to objectively
defined measurement criteria.
	While  developing  a  computer-aided  treatment  method   for
stimulating language behavior in nonspeaking autistic children (Colby
and  Smith,1971,Colby,1973),  we  became  interested   in   how   the
interactions  between  these  children  and the machine differed from
those of (1) normal children  and  (2)  other  types  of  nonspeaking
children.   To  evaluate  the treatment method we attempted to follow
changes in interactions over time and to assess whether these changes
could  be  considered as an improvement. If no changes occurred or if
the changes were judged as retrogressive, then the  treatment  should
be  discontinued.    We  chose  to  define improvement as a change in
interactions  towards  those  characteristic   of   normal   speaking
children.
	The   treatment   involves   letting  a  child  play  with  a
computer-controlled   audio-visual    device    consisting    of    a
typewriter-like  keyboard and a television-like video display screen.
When a child presses a key, a symbol,  letter,  word,  expression  or
drawing  appears  on  the screen accompanied by a sound, mainly human
voice sounds and some animal or machine sounds. The  overall  program
is divided into "games" of varying complexity. A sitter who sits with
the child changes the games in accordance with the child's  interests
and  abilities.  The simplest game is Game #1 in which pressing a key
produces that key's symbol on the video screen accompanied by a voice
pronouncing  the  corresponding  letter or number. We shall limit our
discussion of objective measurement to the data collected in this one
game.
	To keep track of the child's interactions with the machine, a
program  was  written  (by  Earl Sacerdoti, a graduate student in the
Department of Computer Science, Stanford University)  which  recorded
the  game  the child was playing, which key was pressed at what exact
time and whether the sound for this key was  played  over  the  audio
device.  The  computer-controlled  system  is designed so   that if a
child presses a single key or several keys in bursts of less  than  a
second's duration, the sound for the first pressing is played but the
rest are not, in order to avoid confusing the child. As soon  as  the
child  pauses  at the end of such a burst, a buffer is cleared of all
the symbols accumulated during the burst, and when the  next  key  is
pressed, its sound is played. Striking the keys at an extremely rapid
rate indicates a child is ignoring the  sound  and  paying  attention
only to his visual and/or tactile experience.
	In  Fig.  1 the first column on the left indicates the number
of the game, the second column indicates the key  struck,  the  third
column  shows  the time of striking in hours-minutes-seconds, and the
fourth shows whether the sound for that key was played ("played")  or
not  (blank space). The data of Fig. 1 illustrates the interaction of
a normal speaking 17 month old boy. Fig.2 shows the  interactions  of
an  eight  year  old nonspeaking boy clinically diagnosed as autistic
and with a score of +26 on Rimland's E2 scale.     (A  score  greater
than  +20  is  considered by Rimland to indicate a true case of early
infantile autism (Rimland,1971)).   It is noteworthy how,  on  simple
visual  inspection,  the  data  of  the  8  year  old  autistic child
resembles that of a 17 month  old  normal  child.   In  this  way  an
objective  measurement  allows  us  to establish equivalences between
different children.  Fig.3 demonstrates the interactions of a  normal
speaking 4 year old boy which are strikingly different from the other
two children.
	To analyze data from several children  playing  Game  #1,  we
examined  for  each  child  (1)  the  total  number  of  runs  (a run
consisting of the same key being pressed),(2) the distribution of the
lengths  of  runs,  (4)  the  number of gaps (blank spaces indicating
non-listening to sound),and (5) the distribution of the length of the
gaps.   From these observed values we computed statistics descriptive
of the performance of the child and  relatively  independent  of  the
length  of  the  game. These were (1) average run and gap length, (2)
relative frequency distributions of runs and gaps, and (3) numbers of
gaps per length of  test.    Discarded  from  this  list  were  those
measures  which for normal children were insensitve to the age of the
child.  The two  most  sensitive  measures  were  found  to  be:  (1)
proportion  of runs of length 1 (%R1) and (2) relative number of gaps
(total number of gaps divided by total length of runs). The data  for
these  measures  from  fifteen  normal and three types of nonspeaking
children (autistic,  aphasic,  organic  brain  syndrome)  appears  in
Tables 1. and 2.
	From  the  normal children data we can construct a prediction
line as shown in the graph of Fig. 4. Normal  children  from  age  17
months   to   10   years   appear  to  progress  in  the  performance
characteristics of their  interactions  along  this  line.   One  can
project  the  observed  point describing a normal child's performance
onto the prediction line, and, as can be seen, the projection  points
with but one exception, line up according to the age of the child. If
one graphs the abscissa of the projection  against  the  age  of  the
child,   one  can  produce  an  age-prediction  curve  based  on  the
performance characteristics. (See Fig. 5).
	Using curve fitting procedures on these data we found that
the numerical  formula for prediction was:
	A(i) = 0.960 - 1.91 log (1 - % R1 + 0.809 TNG/TLR)
	For each normal and each nonspeaking child, the age-level  of
performance  was computed by this formula and appears in Tables 1 and
2.  In the age  range  of  particular  interest  (1  -4  years),  the
age-level of the performance of normal children is closely comparable
to  their  actual  chronological  age.  On  the   other   hand,   the
performances  of  the  nonspeaking  children are comparable to normal
children less than 4 years of age.  Over  time  it  can  be  assessed
whether  a  nonspeaking  child  is  progressing  towards  more normal
interactions (i.e.  achieving higher age-level performances), whether
he has reached a plateau, or is retrogressing.
	As yet we do not have complete data from start to  finish  of
treatment  on  an  improved  case  of  a  nonspeaking autistic child.
However, Fig.  6 shows some current (Aug.l973) interactions of D.,  a
nonspeaking  autistic  10  year old boy whom we treated for two years
three years ago and whose language development gained markedly.   (We
were  not  collecting  this  type  of data when he was in treatment.)
Stretches of D's  interactions  are  quite  normal  looking  and  his
location  on  the  age-equivalent  curve  of  Fig.  5 shows him to be
performing slightly better than a normal three year old.  (It  should
be  noted  that  between ages 3-4 a normal child achieves grammatical
complexity roughly equivalent to adult  colloquial  language,  making
occasional   mistakes.   Between  ages  4-5  language  becomes  fully
established with all  adult  grammatical  forms,  deviating  only  in
style).
	We have several cases of failure in which the final  sessions
of  treatment  show  interactions indistinguishable from those of the
early sessions.  The data of the child in Fig.2  over  the  past  two
years   reveal   short   periods  of  retrogression  and  no  overall
improvement.  We will try for another year and if no  change  occurs,
we  will  discontinue. We must also be prepared for the possibilities
that (1) a child's comprehension and speech  improves  as  judged  by
clinical  and parental observation but his interactions do not change
or (2) a child's interactions change towards the normal but he  still
does  not use speech for social communication.    As yet, we have not
observed either of these paradoxical outcomes.
	In  summary,  we  have  presented an objective measurement of
children's  interactions  in  playing  with   a   computer-controlled
audio-visual  device  programmed  to  stimulate language development.
This measurement is useful in 3 ways: (1) it reveals  where  a  child
stands  on  an  interaction  curve  relative to normal and other non-
speaking children; thus treatment can be planned to suit the  child's
position on an  age-equivalent curve, (2) changes in the interactions
over time can be evaluated to see if a child is improving or not, and
(3) if no change takes place or a child reaches a plateau and remains
there for a long  time,  discontinuation  of  the  treatment  can  be
justified.    Thus an objective measurement of interactions serves as
a useful instrument in planning and terminating treatment of language
deficiencies in nonspeaking children.

                          References

Colby, K. M. (1973). The rationale for computer-based treatment of
     language difficulties in nonspeaking autistic children.
     Journal of Autism and Childhood Schizophrenia, 3, 254-260.

Colby, K. M. and Smith, D. C. (1971). Computers in the treatment
     of nonspeaking autistic children.  In J. H. Masserman
     (Ed.), Current Psychiatric Therapies, Grune  & Stratton,
     N.Y.

Rimland, B. (1971). The differentiation of childhood psychoses: an
     analysis of checklists for 2,218 children.  Journal of
     Autism and Childhood Schizophrenia, 1, 175-189.