refinements were petformed wlth X-PLOR (27). computed from the refned single-conformermodel. A X-ray Ciysta!!

el. A X-ray Ciysta!!ography and NMR (Yale Univ Press,

20. W. I. Weis. K. Drickamer. W. A. Hendrckson. Nature conventional (model-phased)difference electron den- New Haven. CT. 1992): develorsmenta verslon of
360, 127 (1992);W. I. Weis and K. Drckamer, Struc- sity map consisting of a Fourier synthesis of F
(lI, - X-PLOR, avQiabk on request. '
ture 2, 1227 (1994). F,I) e~p(i+,,~,) is an approximation of the vector 28. D. T. Cromer, J. App!. Ciysta!!ogr. 16, 437 (1983);
21. Multiple conformers [J. Kuriyan et a!., Proteins 10, difference map [J. Drenth, Principies of Protein X-ray a n d D. L. Liberman,J. Chem. Phys. 53,1891
340 (1991)] (9) of the sub-MBP-A protein molecules Ciystallography (Springer-Verlag,New York, 1994), (1970).
were generated stariing with the model that was pp. 150-1521, As a consequence, vector dlfference 29. W. A. Hendrickson and W. I. Weis, unpublshed data.
obtained after single-conformer refinement (79). The maps of sub-MBP-A were observed to have less 30. L. M. Gregoret, S. D. Rader, R. J. Fletterick, F. E.
water molecules and Yb3+ were restrained to their noise than conventional difference maps. Cohen, Proteins Struct. Funct Genet. 9, 99 (1991).
inita postons. Conformers of the proten were ren- 23. M. M. Teeter, Proc. Nat!. Acad. Sci. U.S.A. 81, 6014 31. We thank P. Gros for assistance in designing the
dered invisible to each other in terms of chemical (1984). X-PLOR crystallographic language for phasing; V.
interactions. Each conformer in the ensemble con- 24. G. M. Core eta!., Structure 2, 89 (1994). Agrawal, S. Shah, and M. Wiis for help with data
tributed equally to the calculated structure factor 25. M. Levitt and R. Sharon, Proc. Nat!. Acad. Sci. collection; C. Ogata for beamine support; and P.
w t h the occupancy set to the reciprocal of the num- U.S.A. 85,7557 (1988);Y. Komeiji et a!., Proteins 16, Adams, M. Gersten, M. Levitt, and L. Rice for crltical
ber of conformers. Evaluation of the free R value as a 268 (1993); M. Karplus and G. A. Petsko, Nature readng of the manuscript. The coordinates and the
functon of the number of conformers produced a 347, 631 (1990); P. J. Steinbach and B. R. Brooks, diffraction data have been deposited with the
shallow minmuqYor elght conformers, which were Proc. Nat!. Acad. Sci. U.S.A. 90, 9135 (1993); V. Brookhaven Data Bank (accession number 1YTT).
used for subsequent anaysis. Lounnas eta!., Biophys. J. 66, 601 (1994). Supported by the Howard Hughes Medical Institute
22. Vector dlfference electron density maps were com- 26. D. C. Rees and M. Lewis, Acta Ciysta!!ogr. A39, 94 (A.T.B.), NSF (DIR9021975, A.T.B.), and NIH
puted by taking the vector difference between com- (1983); E. Arnold and M. G. Rossmann, ibid. A44, (GM50565, W.I.W.). W.I.W. IS a Pew Scholar In the
plex structure factors F I[, exp(i+,,) - IF,,, 1, 270 (1988); L. M. Rice and A. T. Brunger, Proteins Biomedca Scences.
exp(i+,,,J], the experimental structure factor wlth Struct. Funct. Genet. 19, 277 (1994).
centrod phases obtained from Phq(+)(7 7) mnus that 27. A. T. Brunger, X-PLOR. Version 3.1. A System for 8 September 1995; accepted 1 November 1995

Tempor'al Processing Deficits of childhood conditions (for example, in the

presence of middle ear disease) under which
Language-Learning Impaired Children acoustic i n ~ u t sare consistentlv muffled.
Ameliorated by Training Cortical pla'sticity studies indicati that tem-
poral processing deficits like those that
emerge in LLI children would be expected
Michael M. Merzenich," William M. Jenkins, Paul Johnston, from these learning scenarios, if thev were
Christoph Schreiner, Steven L. Miller, Paula Tallal to be undertaken in a training regime ap-
plied in a monkey model (3, 4).
Children with language-based learning impairments (LLls) have major deficits in their Visual psychophysical studies have al-
recognition of some rapidly successive phonetic elements and nonspeech sound stimuli. ready shown that very great improvements
In the current study, LLI children were engaged in adaptive training exercises mounted in the recognition of brief, successively pre-
as computer "games" designed to drive improvements in their "temporal processing" sented stimuli can be achieved with prac-
skills. With 8 to 16 hours of training during a20-day period, LLI children improved markedly tice in adult humans (1). . , In this studv,
, . we
in their abilities to recognize brief and fast sequences of nonspeech and speech stimuli. asked: C a n we substantially alter the defi-
cient temporal processing capacities of
young, school-age LLI children by similar
acoustic ~ r a c t i c e ?
Experiments conducted in human ( 1 ) and els of temporal sequence learning, earlier For training tools, we produced two au-
monkey (2) neurological models of percep- studies of the receotive deficits of LLI chil- diovisual (AV) "games" designed around a
tual learning have demonstrated that the dren (7) had show; that they have a "tem- circus theme to engage 5- to 10-year-old
capacity for segmentation of successive poral processing deficit" expressed by limit- children at high levels of attention and
events in sensory input streams can be ed abilities at identifying some brief pho- enthusiasm within highly repetitive learn-
sharpened, apparently throughout life, by netic elements presented in specific speech ing tasks. T h e first A V game (Circus Se-
practice. Electrophysiological studies of contexts and by poor performances at iden- quence) was a perceptual identification task
learning-induced plasticity conducted in tifying or sequencing short-duration acous- in which a correct response in the exercise
the neocortices of monkeys have provided a tic stimuli presented in rapid succession (8). was a faithful re~roductionof the order of
growing body of evidence about the neural Consistent with a temporal processing def- two-stimuli sound sequences by touch-
processes that underlie practice-based im- icit hypothesis, LLI children can distinguish screen button-oress seauences (9). . . The
provements in both temporal segmentation these brief speech features and can correctly nonverbal stimuli applied In training were
and spectral (spatial) discrimination perfor- reconstruct stimulus inout seauences if 16 octave-per-second upward- or down-
mances (3-5). These studies have shown stimuli are presented in slower forms or at ward-gliding (U and D, respectively) fre-
that the abilitv of an adult animal to make slower event rates (8).
. . quency-modulated (FM) tonal pairs (U-U,
fine distinctidns about the temporal or Taken together, these experimental U-D, D-U, or D-D). Stimuli in each FM
spectral features of complex inputs can be findings led us to hypothesize that the def- pair swept across the same frequency range.
sharply improved, or degraded, by a period icits underlying the phonetic reception lim- These stimuli were in the range of sweep
of intensive behavioral training (2-4, 6). itations of a LLI child might arise in early frequencies and speeds that apply for the
In parallel with animal and human mod- life as a consequence of abnormal percep- formant transitions of English consonants
tual learning that then contributes to ab- that LLI children have difficulty recogniz-
M. M. Merzench, W. M. Jenkins, P. Johnston, C. Schrei- normal language learning (3, 4). For exam- ing (7, 8). T h e interstimulus intervals
ner, W. M. Keck Center for Integrative Neurosciences ple, a child might simply make more-limit- (ISIs) and FM frequencies were adaptive
and Coleman Laboratory, Unlverslty of California, San ed-than-normal use of t e m ~ o r a linforma- oarameters.
Francisco, CA 941 43-0732, USA.
tion as he or she learns to make distinctions T h e second game was a phonetic ele-
S. L. Miller and P. Tallal, Center for Molecular and Behav-
ioral Neuroscience, Rutgers University, 197 University
Avenue, Newark, NJ 071 02, USA.
about speech inputs as a learning akerna-
tive, or a child might generate a represen-
-
ment recognition exercise imolemented as a
two-alternative forced-choice task in which
*To whom corres~ondenceshould be addressed tation of phonetic information under early the child was presented two consonant-

SCIENCE * VOL.271 * 5 JANUARY 1996

vowel (CV) stimuli with contrasting con- trial performances by an audio and visual ments in the temporal event recognition
sonants (for example, [be] versus [de] pre- signaling of a correct response and by a and sequencing abilities for tonal stimuli
sented in rapid sequence) (10). The child's point accumulator that advanced for "hits" centered at 1.2 kHz, for all children (study
task was to identifv the seauence Dosition of but not "misses." Entertaining reward ani- 1 in Fig. 2). Specifically, there were signif-
a target CV (in this example, of either a mations were presented at frequently icant performance improvements in these
precued [ b ] or [d~]).The main variables in achieved performance benchmarks. Chil- children's abilities to sequence both stimuli
this exercise were (i) the durations of syn- dren were also rewarded for their progress of shorter durations and stimuli separated
thetically produced consonants (and recip- by working in a token economy in which by shorter ISIs, although all training for the
rocally, of the following vowels; the total earned points could be traded for prizes. latter was conducted with stimuli of a fixed
CV duration was constant), (ii) the magni- The first trial of these temporal process- (60-ms) duration.
tudes of a 0- to +20-dB amplification of ing games was conducted with seven 5.9- to For the Phoneme Identification game,
consonant elements versus vowel intensity, 9.1-year-old LLI children (1I). Training at all children in study 1 performed at near-
and (iii) the ISIs between presented CV the nonverbal Circus Sequence temporal chance levels when they attempted to iden-
pairs. training exercise was applied for 19 to 28 tify stop consonants presented with brief
These two games were begun with stim- training sessions of 20 min each conducted formant transitions in the initial sessions.
uli that LLI children could easily di_stin- over a 4-week training period. Five of these All reliablv identified the same brief conso-
guish and recognize: They had long nonver- LLI children benefited from the reinforce- nant transitions with a high performance
bal stimulus (60 ms) or consonant transi- ment contingencies used in the game and reliability in some later sessions. A bench-
tion (65 to 70 ms) durations, were present- progressed regularly for extended periods of mark test of phonemic reception (13) con-
ed with long ISIs (500 ms), and for CV training across the daily sessions of adaptive ducted before and after training showed
stimuli, were presented with a maximum training. For the four children who were that six of seven children made significant
differential amplification of the consonant under the best psychophysical control, per- improvements at phonetic element identi-
transition (+20 dB). These variables were formances over successive daily practice ses- fication. Their average gain translated to
then altered adaptively in training, trial by sions asymptoted at progressively higher -1.5 years in language development age.
trial, to drive each child in the direction of performance levels (Fig. 1). Two of these Because stimulus delivery in this training
normal performance levels. four children reached or exceeded normal exercise was relatively slow and because the
Correct performances at both of these ~erformancelevels at this task: that is. after stimulus categories were not changed in a
exercises were rewarded in several ways. idaptive training they could' consistently fixed pattern, performance improvements
The children received feedback about their order stimuli that were presented in imme- at this game were somewhat erratic and
diate or nearly immediate succession. Note were not easily compared between children.
that normal children can reliably identify A second study was conducted to deter-
75-ms-duration stimuli that are separated mine if these results could be replicated in a
by -10 ms (12). Three other children made larger, independent sample of LLI children
severalfold im~rovements in task ~erfor- (14). The games were modified in an at-
Early
mance with practice, operating with ISIs tempt to increase performance consistency
300 at or below 100 ms for some or all FM and-to more strongly maintain the atten-
400 stimulus categories. One child never dis- tion of LLI children on these tasks (15).
(D FM 0.5 kHz
g.500- 9 played a consistent response pattern dur- Children progressed at the task of identi&-
ing this exercise and never progressed to ing rapidly successive stimuli (the Circus
3 100 Late ISIs shorter than -300 ms. Another child Sequence game) more reliably and under
showed a clear progression in performance, better response control than in the initial
but that performance achievement was lim- game version, with 10 of 11 children sys-
ited to stimuli separated by -200 ms for all tematically improving their performances

- 7
stimulus categories.
A benchmark test of temporal process-
over the majority of time spent at this ex-
ercise. With extended stimulus sets that

01 ill=
0 Late ing ability, the Tallal Repetition Test (I2), included stimulus duration as a Darameter
1 4 was conducted before and after training. It and with more reliable on-task behavior,
200 Early revealed statistically significant improve- LLI children were also driven to higher

400 FM 2.0 kHz

Fig. 2. Performance of A B
500 1
1 5 10
1
15 20
study 1 and study 2 children 2
foreandimmediatelyafter , , ,
Trial block
the conclusion of training, as g 400 g loo
Fig. 1. Game performance functions at the se- evaluated with a benchmark .E 300 $ 80
quence recognition game for three of the four measure of temporal se- 2 TJ 60
stimulus sets (9)for one LLI child, recorded near quence recognition ability, 2 *0° =I 40
the beginning (open dots) and in the final week the Tallal Repetition Test. 5 100 2 20
(filled dots) of 4 weeks of testing (study 1). All Bars are standard errors. 0
children in both studies who worked under re- The Tallal Repetition Test = Study 1 Study 2 Study 1 Study 2
sponse control at these games showed substan- measures (A) threshold lSls Pretralntng Posttrain~ng
tial, progressive gains in their abilitiesto sequence at which the sequence or-
these fast, brief stimuli. The symbols represent ders of rapidly sequenced tonal stimuli pairs of 0.8 and 1.2kHz are correctly identified with 0.8 probability,
changes in performance level; for positive chang- and it also measures (B)the minimum durations of stimuli that can be sequenced [see (1211.The effects
es in level (decreasing ISls) a dot may represent of training were significant for all pre- and posttraining comparisons. For study 1, pretraining versus
from one to three correct responses. After eight posttraining threshold ISls, ANOVA (repeated measures, two-tailed)F(1,6)= 36.7, P < 0.001; for study
reversals in diiculty level, the child progressed to 2 ISls, F(1,6)= 12.8,P < 0.05. For study 1 durations,F(1,6)= 7.3, P < 0.05;for study 2 durations,F(1,6)
training with another FM stimulus set. = 28.9, P < 0.01.

SCIENCE VOL. 271 5 JANUARY 1996

performance levels in this revised game. ms. Three children also progressed to a guage training and equivalent periods of
Despite a highly variable number of total perfect (IS1 = 0) performance level when nonadaptive video game playing per day
trials for different children (640 to 3739 in the duration of the FM stimuli was lowered over the 20-day training period (7) did not
17 to 22,20-min training sessions), 10 of 11 to 40 ms, and one child mastered the task significantly improve in their sequenced-
children showed progressive gains and were for 20-ms-duration FM stimuli in this cate- stimulus recognition abilities, as measured
able to complete a substantial program of gory. Differences between early and late by this benchmark test (17).
learning within at least two FM stimulus training performance levels were statistical- The phoneme recognition game was also
categories; 9 of 11 completed a substantial ly significant [ANOVA (analysis of vari- revised for testing in the 11 study 2 LLI
program of learning and showed progressive ance) repeated measures, two-tailed: children. Six of the 11 children progressed
gains in all four categories (Fig. 3). F(1,18) = 54, P 5 0.0001; see Fig. 31. to stimulus categories for several CV con-
For example, for distinguishing between Although some individual performance dif- trasts at which tKey could identify the brief-
FM stimuli of 60-ms duration with a start- ferences by category were recorded, overall, est consonants (35 ms) at the fastest ISIs
ing or ending frequency of 1 kHz, the aver- performance gains did not significantly vary (10 ms) (see examples in Fig. 4). Two other
age performance improved about fourfold by FM stimulus category (16). children also learned to reliably identify the
(see Fig. 3A), with four children achieving Again, the performance gains measured fastest forms of natural (nonamplified) con-
a perfect (IS1 = 0) performance level at this in these training experiments were mirrored sonants. Two others showed significant im-
task. The initial reversal-defined threshold by results obtained before and after training provements but, with this limited training,
(9) averaged 268 ms for these children; with on the benchmark Tallal Repetition Test did not achieve normal recognition abilities
training, it dropped to an average IS1 of 77 (12). All 11 children showed gains by this for CV stimuli in which consonants were
performance measure (see study 2, Fig. 2). presented in very fast and nonenhanced
l+kHzFM
The pre- and posttraining group differences forms. The 11th child, again, never consis-
were again significant. O n the average, tently performed above chance at this
12+ kHz FM minimal ISIs dropped to -1115 of their game. Notably, performance gains were
pretraining values. The mean IS1 threshold again mirrored by improvements in a pho-
of about 20 ms achieved after training ap- netic element recognition benchmark test
proaches normal performance abilities (8). (13).
. . Analvses of results from these tests
Similarly, trained children could identify indicated that with training at these exer-
much briefer stimulus tones presented in cises supplemented (i) by two other games
almost immediate succession (Fig. 2). After designed to promote speech perception gen-
training, children could distinguish the se- eralization (18) and (ii) by special language
quence order of almost immediately succes- training with acoustically modified speech
sive tonal stimuli that were, on the average, (7), temporal processing improvements ap-
18 ms in duration, which was about one-
-f loo fifth the pretraining value. This perfor-
plied generally to these children's language
comprehension abilities [see (7)].
S mance ability again approaches that of nor- It is likely that performance at these
E
z * mal children (8). These training effects on
the ability of study 2 LLI children to iden-
adaptive training tasks could be further im-
proved by additional practice in almost ev-
fi tifv the seauence order of briefer sound ery tested LLI child (19). The five children
20 stimuli was significantly greater than those who had the lowest performance results
0
for study 1 children, possibly because they played the game for the least number of
were trained with stimuli that varied in sessions and trials. Moreover, the number of
Training session duration. In contrast, 11 matched LLI chil- training sessions and training trials com-
Fig. 3. (A) Average performance differences on dren who were undergoing classical lan- pleted by different children were directly
the Circus Sequence game for the 11 children in
study 2, recorded early and late in a 20-day (20
minlday) training period. "Early" scores were Fig. 4. Performance levels pat&l

from the first trials in which there was a clear dem- achieved early and late in
onstration that the children understood the "rules training in five LLI children
of the game" (that is, played the game on-task)
and reached a stable response level revealed by
working at the Phoneme
Identification game. The
12 20
w n g
stimulus reversals within a training session. These bottom of each colored bar a
initial measures overestimate the early perfor- marks a stable performance lo
mance abilities of these children in some cases level achieved during the 0
because children often practiced these games for first five to six training days;
a session or two before any clear stimulus rever- the top of each colored bar 0

-
sals indicating stable performance limits were re- marks the performance level 1 2 3 4 5
corded. (B) Progression in performance at the 1- achieved during the final Subject
or 2-kHz FM stimulustasks for 6 of the 11 children days of training. Note that in [bal vs. Ida1 rn [ a h ] vs. [cda]
in the second experimental series performing the this exercise, training began [bdVS. [&I rn[bail vs. [dd]
stimulus sequence recognition (CircusSequence with differentially prolonged r [fa] vs. [va]
game) exercise (9). The highest difficulty levels and amplified consonants
reached on every trial day on which children delivered initially in CV stimuli presented at slow repetition rates. Every child could initially identify these
worked at these particular stimulus sets are synthetically disambiguated CVstimuli. As the child's performance at the task improved with training, task
shown. Similar progressive learning functions difficulty was progressively increased (see right margin of the graph). The IS1 was first decreased in
were recorded for nine other children in these two 100-mssteps to initially stabilize at 200 ms; then the consonant duration was decreased progressively to
studies (that is, in 15 of 18 trained LLI children). natural fast transition rates (35-to 40-ms durations) in seven steps; then the differential consonant versus
More difficult stimuli had progressively shorter lSls vowel amplification (consonant "emphasis") was faded in seven steps; then the IS1 between CVs were
and durations. further incrementally decreased in eight nonlinear steps to an IS1 of 10 ms.
SCIENCE VOL. 271 5 JANUARY 1996
correlated with a benchmark measure of 55, 873 (1990); N. N. Byl, M. M. Merzenich, W. M. dren demonstrated a severe delay n recept~veand
Jenk~ns,J. Neurol., In press. expressive language development (mean language
language outcome (20) for both study 1 [n 7. P. Tallal et al., Science 271, 81 (1996). age = 4.8 years) as well as marked temporal process-
= 7; r (correlation coefficient) = 0.85; P 5 8. P. Tallal and M. P~ercy,Nature 241, 468 (1973); Neu- ing def~c~ts. These school-age children also had read-
0.011 and study 2 LLI children (n = 11; r = ropsychologia 13, 69 (1975); P. Tallal, Brain Lang. 9, ing defc~ts.All were w~thoutother prmary deficits
182 (1980); and R. E. Stark, Ann. Dyslexia 12, The Tallal Repetton Test determ~nesthe threshold
0.73; P 5 0.01) (21).
32, 163 (19 8 2 ) ; , E. D. Mell~ts,Brain Lang. 25, IS1 at which sequences of two tonal stmull (in th~s
These experiments once again confirm 314 (1985); P. Tallal, S. Miller, R. H. F~tch,In (3),pp. case, of 1000 and 1400 or 800 and 1200 Hz) that are
that LLI children have major temporal pro- 27-47. It has been argued that the fast-element 150, 75, 40, or 17 ms n durat~onare perceved with
cessing, fast-speech-element recognition (usually consonant) phonet~crecognit~ond~fficulty their dellvery sequence reproduced with a 75% ac-
arlses because of an napproprlate "~ntegrat~on" curacy. lSls vary n the test from 500 to 0 ms. See P.
deficits (8). Those deficits have presumably w ~ t hor "masking" by follow~ngor precedng speech Tallal, in Non-Speech Language and Communica-
been in olace for most of the orior vears of elements (usually vowels). Although parametr~cpsy- tion, R. Sch~efelbusch,Ed. (Un~versityPark Press,
the speeih reception histories 'of these chil- chophysical stud~esare lack~ng,th~smay apply only Baltimore, MD, 1980), pp. 449-467.
w ~ t h nfrequency channels as these children appar- 13. The Goldman-Fstoe-Woodcock Dagnostc Aud~to-
. , With a total of 5 to 10 hours of
dren (22). ently have less diff~cultydentitying consonant-vowel ry Discrm~natonTest (Amercan Gu~danceSewce,
intensive practice at each of these two tasks pairs w ~ t hbref consonants n whch consonant tran- Circle P~nes,MN) was used as a standard bench-
in brief daily sessions extended over only 20 s~t~ons and vowel formants are w~thinlargely non- mark. It was desgned to defne an ndv~dual'sability
overlappng channels. to ident~fyphonic elements w ~ t h ~words.
n
days, a substantial remediation of these def-
9. Cort~calplastclty and learnng studes conducted In 14, In study 2, 22 chldren (8 females, 14 males) ranglng
icits was achieved in nearly all the LLI prlmate and human models had shown that such In age from 5.2 to 10.0 years (mean age = 7.4 years,
children studied. These studies strongly in- tranng (I)had to be appled w ~ t ha heavy schedule of standard dev~at~on = 1.4 years) who had a mean
dicate that the fundamental temporal pro- practice tr~als,(i~)would be ideally conducted on a nonverbal ntelligence score of 96.4 (standard devl-
series of successive traning days, ( ~ iwould
) require at~on= 9.7) part~c~pated. The group included 18
cessing deficits of LLI children can be over- relat~velyintense practlce schedules des~gnedto whites, two H~spancs,one Asian, and one Afr~can
come by training. dr~vecont~nuousperformance Improvements, and Amercan. All ch~ldrenwere from m~ddleSES fam-
These studies also stronelv (IV)would have to be conducted under conditions of Ihes. All children demonstrated a severe delay n re-
" , indicate to us
high mot~vationaldr~ve.We chose to accomplish ceptive and expressive language development
that there may be no fundamental defect in these training objectves in LLI chldren by uslng en- (mean language age = 4.9 years; standard dev~at~on
the learning machinerv in most of these tertaining and highly rewardng CD-ROM-mounted = 1.4 years) as well as marked temporal processing

children, bekuse they so rapidly learn the exerclses dlsgused as games. The Crcus Sequence deficts. These school-age ch~ldrenalso had reading
game was developed w ~ t hAuthorware Profess~onal deficts. All were wthout other prlmary deficits.
same skills at which t h ~ yhave been defined and Drector (Macromedia) software. The FM stimul~ 15. After inital testing w ~ t hthese training exerclses in
to be deficient. That finding suggests in were generated with a 22.05-kHz sampl~ngrate and these seven children, both games were rev~sed,then
turn that the physical differences and dis- 16-bit processor in Audio Interchange File Format retested in study 2 in 11 LLI children (74). In the
(AIFF) with Matlab (Mathworks) software. Stimuli second version of the Circus Sequence game, the
tributed functional response differences re-
were ramped on and off to reduce spectral spatter. number of stimulus variations in each set was ex-
vealed in evoked potential and imaging The compact disks were produced with Sony Hybrid tended to 135 by including FM stimuli with durations
studies of the brains of LLI individuals (23)
~, software and a Sony 900 CDR. Games were played of 60, 40, and 20 ms. An animated performance
may substantially be effects of the learning on Macintosh computers with CD-ROM drives or barometer was added to the game to further signal
with these exercises copied and played off of Macin- performance progress in yet another compelling vi-
histories of these soecial children. Further- tosh hard disks for convenience. Children received sual manner, and points were subtracted for re-
more, it may also imply that inherited fac- sound stimuli through Sony model MDR-V600 head- sponse "misses." To further ensure that children
tors contributing to LLI origin (24) may sets. Four stimulus sets were mounted in this game were kept on task in these games, five misses in a
in version 1: 60-ms-duration FM sweeps with start- row resulted in a suspension of a change in difficulty
relate to the initiation of a scenario that ing or ending frequencies at 0.5, 1.0,2.0, or 4.0 kHz. level; task difficulty was then maintained constant
embeds, through learning, a defective rep- Three successive correct responses resulted in a until the child again recorded four "hits" in a row. For
resentation of speech phonetics-and does shortening of either the IS1 or the stimulus duration. the Phoneme Identification game, an animated per-
An error resulted in a one-step lengthening of the IS1 formance barometer was also added to the AV dis-
not necessarily mean that these children or stimulus duration. This learning schedule assured plays, and points were subtracted for incorrect re-
have irreversible defects in the molecular that at least 79.3% of the child's responses were sponses. Most importantly, the game was recon-
and cellular elements of the learning ma- correct. When the child achieved a predetermined structed as a progressively adaptive exercise in
number of positive and negative task difficulty rever- which task difficulty was (i) increased progressively
chinery of their brains. sals in any single training session, he or she was when any three-trial block was completed without
advanced to a new FM stimulus set. Training was error, (ii) decreased by one step in difficulty for any
REFERENCES AND NOTES initiated at a new stimulus set at a performance level error, and with the drop in game difficulty, (iii) halted
established by a prior session's performance whenever a child made five errors in succession.
1. A. Karni and D. Sagi, Proc. Natl. Acad. Sci. U.S.A. achievements. Task difficulty was increased by first reducing the
88,4966 (1991); Nature 365,250 (1993); M. Ahissar 10. The Phoneme ldentification game was created with duration of the consonant stimulus elements, then a
and S. Hochstein, Proc. Natl. Acad. Sci. U.S.A. 90, Authorware Professional (Macromedia) software. differential amplification of fast consonant elements
5718 (1993). Synthetic speech stimuli were produced with 16-bit progressively faded, then interstimulus lSls for suc-
2. G. H. Recanzone, M. M. Merzenich, C. E. Schreiner, processing and with a 22.05-kHz sampling rate by cessive CVs were progressively reduced.
J, Neurophysiol. 67, 1071 (1992); X. Wang, M. M. using SenSyn (Sensimetrics) software, based on a 16. For the main effect of stimulus category, F(1,18) =
Merzench, K. Sameshima, W. M. Jenkins, Nature Klatt cascade-parallel formant synthesizer [D. Klatt, 1.7, not significant (n.s.);for the interaction of time
378, 71 (1995); G. Bertini, A. Karni, P. De Weerd, R. J. Acoust. Soc. Am. 67, 971 (1980)l. Children in and stimulus category, F(1,18) = 2.5, n.s.
Desimone, L. G. Ungerleider, Neurol. Abstr. 21, 276 study 1 were trained with [be], [de], and [gel target 17. Group B children, who received equivalent language
(1995); R. Beitel et a/., ibid., p. 1180. and foil stimuli. Children in study 2 were trained with training but with natural speech materials and who
3. M. M. Merzenich, C. S. Schreiner, W. M. Jenkins, X. five CV pairs: [ba] versus [da], [be] versus [de], [fa] played video games rather than these adaptive au-
Wang, in Temporal Information Processing in the versus [va], [aba] versus [ada], and [bail versus [dai]. ditory-speech training games (71, did not improve
Nenrous System: Special Reference to Dyslexia and In the initially tested version of this game, chldren significantly with respect to either IS1 or duration
Dysphasia, P. Tallal, A. M. Galaburda, R. R. Llinas, C. performed trials under given parametric conditions in thresholds measured by the Tallal Repetition Test.
von Euler, Eds. (New York Academy of Sciences, training blocks of 10. When performance criteria For ISI, ANOVA (repeated measures, two-tail) F(1,7)
New York, 1993), pp. 1-22. were met, consonant durations were shortened or = 2.8, n,s.;forduration,F(1,7)= 3.9, n.s. Incontrast
4. M. M. Merzenich and W. M. Jenkins, in Maturational differential amplification of fast consonant elements to the experimental treatment group A, in which ev-
Windows and Adult Cortical Plasticity, B, Julesz and was reduced (or both) for another set of CVs through ery child improved at this benchmark, the majorlty of
I. Kovacs, Eds. (Addison-Wesley, New York, 1995), the next trial block. In study 2, an adaptive staircase group B children had equal or poorer performances
pp. 247-272. training procedure identical to that used in the Circus on this test after their 1-month-long training period.
5. For reviews of ths complex subject, see M. M. Mer- Sequence game (9) was used. Although the mean performance of group B LLI chil-
zenich and W. M, Jenkins, in Memov Concepts, P. 11. Seven children (four females, three males) ranging in dren was modestly better than was that for group A
Anderson, 0. Hvalby, 0. Paulsen, B. Hokfelt, Eds. age from 5.8 to 9.1 years (mean age = 7.3 years, chldren, children in these groups did not differ in
(Elsevier, Amsterdam, 19941, pp. 437-451 ; M. M. standard deviation = 1.5 years) who had a mean their pretreatment Tallal Repetition Test measures of
Merzenich and C. DeCharms, in The Mind Brain nonverbal intelligence score of 106 (standard devia- threshold lSls or durations.
Continuum, R. Llinas and P. Churchland, Eds. (MIT tion = 18.25) participated in the study. The group 18. Children in study 2 were also trained at two addition-
Press, Cambridge, in press). comprised four whites, two Hispanics, and one Afri- al games, both designed to facilitate the generaliza-
6. M. M. Merzenich, G. M. Recanzone, W. M. Jenkins, can American. The children were from lower and tion of training gains from these frst two described
K. A. Grajski, Cold Spring Harbor Symp. Quant. Biol. middle socioeconomic status (SES)families. All chil- games to the wider range of temporal sequence

SCIENCE VOL. 271 5 JANUARY 1996

 events and phonetic element contexts and contrasts
that occur in natural running speech. The third game Language Comprehension in Language-Learning
(Old McDonald's Flying Farm),produced with Direc-
tor (Macromedia) software, was a limited hold reac- Impaired Children Improved with Acoustically
tion time task in which the child maintained a touch-
screen "button" press while repeated stimuli were Modified Speech
delivered in regular sequence. The child's task was
to release the button when there was a change in
phonetic element identity. The durations of a wider
Paula Tallal," Steve L. Miller, Gail Bedi, Gary Byma,
array of synthetic consonant elements and the inter- Xiaoqin Wang, Srikantan S. Nagarajan, Christoph Schreiner,
stimulus times between repeated stimuli were the
main exercise variables. The fourth game (Phonic William M. Jenkins, Michael M. Merzenich
Match), also developed with Director (Macromedia)
software, was a sound-matching exercise in which A speech processing algorithm was developed to create more salient versions of the
button presses resulted in soundings that the child
had to locate a match for, on a 2-by-2 to 5-by-5 rapidly changing elements in the acoustic waveform of speech that have been shown to
touch-screen button array. The button array size and be deficiently processed by language-learning impaired (LLI) children. LLI children re-
the temporal structare of elements and of element ceived extensive daily training, over a 4-week period, with listening exercises in which all
sequences in individual consonant-vowel-consonant
stimuli were game variables. Stimuli applied in this speech was translated into this synthetic form. They also received daily training with
exercise were synthetically processed to prolong and computer "games" designed to adaptively drive improvements in temporal processing
differentially amplify brief phonetic elements [see ( I ) ] . thresholds. Significant improvements in speech discrimination and language compre-
Children also played both of these games for approx-
imately 20 midday throughout the 20-day training hension abilities were demonstrated in two independent groups of LLI children.
period. In general, children's performances at these
two games paralleledtheir progressiveachievements
at the time order judgment and phonetic element
recognitiontasks described in this report. All LLI chil-
dren who were trained at these games also under- Exposure to a specific language alters an rapid frequency changes (formant transi-
went training with acoustically modified speech stim- infant's phonetic perceptions within the first tions) that occur during the initial few tens
uli, as described by Tallal eta/. (7).
months of life, leading to the setting of pro- of milliseconds. Interestingly, LLI children
19. Children were still improving at their game perfor-
mances when these exercises were arbitrarily ter- totypic phonetic representations, the building are able to identifv these same svllables when
minated at the end of the +-week training period. block on which a child's native language de- the rates of chaige of the criiical formant
Their ultimately achievable performance limits are velops (1 ). Although this occurs normally transitions are simply synthetically extended
unknown.
20. The Token Test for Children (Teaching Resources
without explicit instruction for the majority of in time by about twofold (8). A strong pre-
Corporation, Boston, MA, copyrighted 1978) is de- children, epidemiological studies estimate diction is suggested by these findings: If the
signed to test the ability to follow auditory commands that nearly 20% of children fail to develop critical acoustic cues within the context of
of increasing length and grammatical complexity.
normal speech and language when exposed to fluent, ongoing speech could be altered to be
21. The intensity of practice at three FM stimulus catego-
ries were all significantly correlated with Token Test speech in their native environment (2). Even emvhasized and extended in time. then the
(languageoutcome) results. For the 1+ kHz catego- after all other primary sensory and cognitive phonological discrimination and the on-line
ry, trial numbers versus language outcome, r = 0.75, deficits are accounted for, approximately 3 to language comprehension abilities of LLI
P 5 0.01; for 2+ kHz FM stimulus, trial numbers
versus language outcome, r = 0.73, P 5 0.01 ;for 4+ 6% of children still fail to develop normal children should significantly improve.
kHz FM stimulus, trial numbers versus language out- speech and language abilities (3). Longitudi- T o test this prediction, we have conduct-
come, r = 0.84, P 5 0.01. The 0.5+ kHz category nal studies have demonstrated a striking con- ed two studies with LLI children who have
practice trial numbers were not significantly correlat-
ed with language outcomes (r = 0.48). vergence between preschool language delay been trained with the application of tempo-
22. A. A. Benescish and P. Tallal, in (31,pp. 312-314; and subsequent reading disabilities (such as rally modified speech. These same children
infant Behav. Dev., in press. dyslexia). A broad body of research now sug- also received training at making distinctions
23. A. M. Galaburda, G. F. Sherman, G. D. Rosen, F. gests that phonological processing deficits about fast and rapidly sequenced acoustic
Aboitiz, N. Geschwind, Ann. Neurol. 18, 222 (1985);
J. P. Larson, T. Hoien, I. Lundberg, H. Odegaard, may be at the heart of these language-learning inputs in exercises mounted in the format of
Brain Lang. 39, 289 (1990); T. L. Jernigan, J. R. impairments (LLIs) (4, 5 ) . computer "games" (9). Modification of flu-
Hesselink, E. Sowell, P. A. Tallal, Arch. Neuroi. 48, Tallal's earlier research has shown that ent speech was achieved by application of a
539 (1991);J. M. Flynn, W. Deering, M. Goldstein, M.
H. Rahbar, J. Learn. Disabii. 25, 133 (1992); R. Du- rather than deriving from a primarily linguis- two-stage processing algorithm (10). In the
ara et al., Arch. Neurol. 48, 410 (1991); J. 0. Hag- tic or cognitive impairment, the phonologi- first stage, the duration of the speech signal
man, F. Wood, M. S. Buchsbaum, L. Flowers, W. cal and language difficulties of LLI children was prolonged by 50% while preserving its
Katz, ibid. 49,734 (1992); A. Kusch eta/., Neuropsy-
chologia 31, 81 1 (1993); A. M. Galaburda, Neuroi.
may result from a more basic deficit in pro- spectral content and natural quality. In the
Clin. 11, 161 (1993); and M. Livingstone, cessing rapidly changing sensory inputs (6). second processing stage, fast (3 to 30 Hz)
Ann. N.Y Acad. Sci. 682, 70 (1993); R. Naas, Curr. Specifically, LLI children commonly cannot transitional elements of speech were differ-
Opin. Neuroi. 7, 179 (1994).
identify fast elements embedded in ongoing entiallv enhanced bv as much as 20 dB. This
24. P. Tallal, J. Townsend, S. Curtiss, B. Wulfeck, Brain
Lang. 41, 81 (1991); B. A. Lewis, J. Learn. Disabil. speech that have durations in the range of a two-step acoustic modification process was
25, 486 (1992); , N. J. Cox, P. J. Bayard, few tens of milliseconds, a critical time frame applied to speech and language listening
Behav. Genet. 23, 291 (1993); B. F. Pennington, J. over which many phonetic contrasts are sig- exercises that were recorded on audiotapes,
Child Neuroi. 10, S69 (1995).
25. WethankT. Jacobson, B. Wright, X. Wang, G. Bedi, naled (7). For example, LLI children have as well as to the speech tracks of children's
and G. Byma for their technical assistance, and C. particular difficulty in discriminating be- stories recorded on tapes and on educational
Checko, N. Reid, and A. Lipski for assistance in tween many speech syllables, such as [ba] CD-ROMs. T h e differential emphasis of fast
programming the animation reward sequences for
these AVexercises. T. Realpe, I. Shell, C. Kapelyn,A.
and [da], which are characterized by very elements also resulted in a speech envelope
Katz-Nelson, L. Brzustowicz, C. Brown, A. Khoury, that was more sharply segmented. This pro-
J. Reitzel, K. Masters, B. Glazewski, A. Rubenstein, cessed speech had a staccato quality in
P. Tallal, S. L. Miller, G. Bedi, G. Byma, Center for Mo-
and S. Shapeck assisted in the training of these
lecular and Behavioral Neuroscience, Rutaers University, which the fast (primarily consonant) ele-
children at Rutgers University. This research was
Newark, NJ 071 02, USA. ments were exaggerated relative to more
funded by the Charles A. Dana Foundation with sup-
X. Wang, S. S. Nagarajan, C. Schreiner, W. M. Jenkins,
portive assistance by Hearing Research, Incorporat-
M. M. Merzenich, W. M. Keck Center for Integrative Neu-
(primarily 'Ow-
ed. For further ~nformationabout this and related els) in the ongoing speech stream. W e rea-
rosclences and Coleman Laboratory, University of Cali-
subjects, contact: http://www.ld.ucsf,edu/
fornia, San Francisco, CA 94143-0732, USA. soned that amplifying the fast elements
6 October 1995; accepted 30 November 1995 'To whom correspondence should be addressed. should render them more salient, and thus

SCIENCE VOL. 271 5 JANUARY 1996 81