!"# %&'(!)&*+ , !#%! !",! -&!" .#,%(/#% ,*0 ).1/&2#% %1#,3)*4 ,-)')!

korea ranks number 1 ln Lhe world ln money spenL on Lngllsh language educaLlon and 121
ln Lngllsh
speaklng ablllLy. CommunlcaLlve speaklng ablllLy wlll lmprove only when Lhere ls wldespread LesLlng of
communlcaLlve ablllLy. 1he problem of course, ls LhaL LesLlng communlcaLlve speaklng ablllLy ls a
cumbersome, dlfflculL, Llme-consumlng and lndlrecL process.
Slmply puL, whaL geLs LesLed geLs done, and communlcaLlve speaklng ablllLy ls noL LesLed. 1hls paper
wlll descrlbe a classroom speaklng LesL LhaL ls avallable for all Leachers, koreans and naLlves speakers,
novlce and experL. 1he goal was noL Lo creaLe a more accuraLe M8l for a few rlch folk (ellLe Leachers and
sLudenLs), buL Lo flnd a cure for malarla, affordable and avallable Lo all. 1he resulLlng LesL ls real-world
communlcaLlon, easy Lo glve and grade, and provldes lndlvldual feedback. lL boLh measures and lmproves
speaklng ablllLy.
1o measure speaklng ablllLy, glve a communlcaLlve speaklng LesL. !" $% & '()* +$)(," -(&%.)(.
1o lmprove speaklng ablllLy, glve communlcaLlve speaklng LesLs. /0&" 1("% "(%"(+ 1("% +23(.
Pere ls how.

!"# 1/&-'#.
1he resLrlcLlon, Lhe boLLleneck, Lo lmprovlng speaklng ablllLy ls speaklng LesLs. When speaklng ls LesLed,
grammar-graded speaklng LesLs resulL ln more grammar sLudy, and memorlzable role-playlng speaklng LesLs
resulL ln more memorlzlng. CLher lndlrecL LesLlng measures such as descrlblng a plcLure have llLLle
auLhenLlclLy and less lmpacL. Speaklng ablllLy can be LesLed en masse over Lhe lnLerneL. Whlle all Lhese
LesLlng meLhods may provlde an accuraLe measure of someLhlng, LhaL someLhlng ls noLhlng llke auLhenLlc
WhaL does korea's global rank of 121 sound llke aL Lhe local classroom level? 8elow ls a llsL of common
errors made by korean mlddle school and unlverslLy freshmen sLudenLs:

.7889: ;<=>>9 ?@7A7BC D7;AEF:; (B7G:@;7AH ;I:EF7BC D7;AEF:;
1. 4)( *2. 0&'( & 0255*6 1. 4)( *2. 0&'( & 0255*6
2. ! 12" 7&"(8 2. ! 12" 7&"(8
3. ! '()* %".+*8 3. ! '()* %".+*8
4. ! 0&'( 92.) 9&-$7*8 4. ! 0&'( 92.) 9&-$7*8
3. :* 5)2"0(); %0( $% $3 "0( &)-*8 3. :* 5)2"0(); %0( $% $3 "0( &)-*8

CorrecL form comes afLer a loL of speaklng, noL before. lL ls a resulL, noL a prerequlslLe. 1he mlddle
school mlsLakes of unlverslLy freshmen clearly aLLesL Lo Lhls. 1hose mlsLakes wlll endure unLll Lhey speak a
loL. A loL of speaklng requlres a way Lo LesL a loL of speaklng, and Lhus my quesL for a communlcaLlve LesL.

!"# %!(0#*!%

lor Lhe pasL flve years ln korean unlverslLy freshman Lngllsh classes, l have glven Lhree-person, 20-
mlnuLe conversaLlon speaklng LesLs for mld-Lerms and flnals Lo over 300 sLudenLs. 1hey are whaL l Lerm
advanced Lngllsh sLudenLs buL beglnnlng speakers. Class slze ranges from 13 Lo 21 sLudenLs and class Llme
ls Lhree hours per week, meeLlng Lwlce a week. no sLudenLs are Lngllsh ma[ors, and mosL hlgh-level
speakers exempL ouL.

!"# 6',%%
My freshman classes are prlmarlly palr and Lhree-person conversaLlons. ÞarLners are swlLched (llke ln
speed daLlng) every 10 mlnuLes or so. Speed daLlng has Lhe merlLs of J><K; (one Loplc), GE@7:AH (many
parLners) and @:I:A7A7>B (many parLners). 1yplcal conversaLlon Loplcs are used (such as <&-$7*; /((=(3+%;
:2'$(%; >?2)"%; @&,&"$23%) and each Loplc has approxlmaLely 40 advanced vocabulary and ldloms.
!#%! #2&'(!)&*
8ecause my classes are sLudenL-cenLered conversaLlons, so Loo should be Lhe LesLlng. 1he LesL lengLh
of 20 mlnuLes ls opLlmum. lL ls long enough Lo ensure auLhenLlc communlcaLlon (afLer any memorlzed
passages peLer ouL) and shorL enough LhaL Lranscrlblng Llme ls reasonable. LlghL groups of Lhree can be
LesLed ln Lhree hours of class, Lhe normal weekly class Llme, so Lhere ls no reason Lo have LesLs shorLer Lhan
20 mlnuLes.
lnlLlally Lhe LesL gradlng was my hollsLlc lmpresslon of Lhelr baslc speaklng ablllLy, welghLed wlLh Lhe
amounL of new Loplcal vocabulary used. Whlle l felL Lhls provlded sufflclenL accuracy, lL offered llLLle ln Lhe
way of acLlonable feedback. 1o Lhls end, recordlng and Lranscrlblng were lnLroduced.
When sLudenLs dld noL balk aL Lranscrlblng l had Lhem complle daLa: LoLal-words, average-lengLh-of-
uLLerance, and vocabulary. Speaklng ablllLy could be seen (Lhe LranscrlpL) and could be measured (LoLal-
words, and average-lengLh-of-uLLerance). lurLher, by comparlng mldLerm and flnal LesL daLa, lmprovemenL
could be measured. 1hls ls parLlcularly rewardlng Lo boLh sLudenLs and Leachers.

!"# !#%!
1. 1hree sLudenLs have a 20-mlnuLe conversaLlon on asslgned Loplcs from Lhe conversaLlon LexLbook.
2. SLudenLs are grouped by slmllar ablllLy for accuracy and falrness. 8esL frlends are never parLners.
3. 1he Leacher asks one quesLlon Lo sLarL Lhe conversaLlon, and Lhen observes.
4. 1he conversaLlon ls recorded on dlglLal recorder, and Lhen Lhe MÞ3 flle ls emalled Lo sLudenLs.
3. SLudenLs Lranscrlbe Lhe whole conversaLlon, buL complle only Lhelr own daLa (LoLal-words, average-
words-per-uLLerance, advanced vocabulary). 1hls Lakes abouL 3.3 hours.
6. WlLhln one week sLudenLs boLh emall and Lhen hand ln Lhelr LranscrlpL.

!/,*%6/)-)*4L !"# *#M .)//&/
1ranscrlblng provldes exLenslve self-
evldenL, self-correcLable feedback. (:*
5)2"0(); %0( $% $3 &)-*, requlres no Leacher
lnpuL.) lor Lhe same reason LhaL every
balleL sLudlo and healLh club has mlrrors,
every speaklng LesL should be Lranscrlbed.

!/,*%6/)1! 0,!,

Moreover, LranscrlpL daLa provldes
concreLe measures for ablllLy and
lmprovemenL. !EN9: O shows LhaL LoLal-
words-spoken has a very sLrong correlaLlon
Lo speaklng ablllLy, abouL .78. ln
comparlson, Lhe Amerlcan SA1 LesL (3.3
hours) has a correlaLlon Lo academlc
success of .33. Cn Lhe oLher hand, Lhe
average-words-per uLLerance has a low
correlaLlon, ln Lhe .20 range. (Some low-
level sLudenLs prepare a long dlalog, and
many hlgh-level speakers ofLen make shorL
lnLer[ecLlons (A(&77*; /0()(; /0*6).
llnal LesL parLners are always dlfferenL,
and grouped by LoLal-words spoken on Lhe
mld-Lerm LesL. 1hls prevenLs hlgh-ablllLy speakers from monopollzlng Lhe conversaLlon and glves lower-
ablllLy speakers greaLer opporLunlLy. 1ranscrlpL daLa clearly shows LesL performance, and enables clearly
falr parLner palrlngs.
llnal gradlng comblnes Lhe hollsLlc raLlng by Lhe lnsLrucLor, vocabulary counL and Lhe LoLal-word counL.
70° A8lLl1?: a hollsLlc score of Lhelr baslc speaklng ablllLy (quallLaLlve)
20° Þ8LÞA8A1lCn: number of advanced vocabulary words (quallLaLlve and quanLlLaLlve)
10° ÞL8lC8MAnCL: number of LoLal words spoken (quanLlLaLlve)

!EN9: P shows ln broad sLrokes an example of how Lhls
grade LrlangulaLlon spreads Lhe grades ouL, based on
performance, noL on errors. Shown are 3 sLudenLs wlLh Lhe
same baslc speaklng ablllLy score (63 ouL of 70, A-) and
dlfferlng vocabulary and LoLal-words scores. SLudenLs of Lhe
same &5$7$"* had Lhelr grades spread 23 polnLs by LesL
?()92)-&3,(. unllke LesL deslgners, classroom Leachers
requlre accuraLe &3+ wlde-spread grades. 1hls LesL provldes
boLh. 1he score ls a balance of !"#$#%& and '()*+),!-.(.
ln a Lyplcal class, Lhe ablllLy score spread ls 70 Lo 93. AfLer
facLorlng ln LesL performance (LoLal-words and vocabulary) Lhls 23-polnL spread ls wldened Lo 40 polnLs,
from 33 Lo 93.

.#,%(/)*4 ,-)')!5
1hls changes everyLhlng. 1hls LesL dlrecLly and ob[ecLlvely measures communlcaLlve ablllLy, raLher Lhan
lndlrecLly and sub[ecLlvely measurlng subcomponenLs of speaklng.
1. llrsL and foremosL, an accuraLe speaklng ablllLy grade can be rendered wlLhouL recourse Lo
1)&--&); ?)23.3,$&"$23; &,,.)&,*; 97.(3,*, and ,2-?)(0(3%$23 scorlng. 1here ls no need and llLLle
beneflL Lo Ledlously complle Lhose lndlrecL and lmpacLless lndlcaLors. (A sLudenL who speaks 1000
words ls lnvarlably beLLer across Lhe board Lhan a sLudenL who speaks 200 words.)
2. Second, lL ls LransparenLly falr. SLudenLs can clearly see all faceLs of Lhelr - and Lhelr classmaLes -
LesL performance.
3. 1hlrd, lL has greaL dlscrlmlnallLy. lL spreads Lhe grades ouL by performance, much more so, and
more preclsely, Lhan grammar, (" &7.
4. lourLh, lL measures how much sLudenLs do, noL how much Lhey do wrong. 8y rewardlng how much
sLudenLs do, lL moLlvaLes Lhem do Lo more.
uurlng Lhe LesL Leachers prlmarlly observe real-llfe college conversaLlons, secure ln Lhe knowledge LhaL Lhe
LranscrlpLs wlll provlde boLh exLenslve correcLlve feedback and accuraLe gradlng daLa.

.#,%(/)*4 ).1/&2#.#*!
1he LoLal words-counL and average-words-per-uLLerance also provlde a benchmark of currenL ablllLy
and a basellne for measurlng lmprovemenL. Whlle Lhe average-words-per-uLLerance does noL have a usable
value for gradlng speaklng ablllLy, lL ls exLremely useful for measurlng speaklng lmprovemenL. Longer
speaklng ls beLLer, smooLher and more confldenL speaklng. !EN9: Q shows Lhe daLa from Lhe cover sheeL of
a Lyplcal flnal LesL LranscrlpL.

1hls average lmprovemenL occurred ln every class, ln every semesLer, for flve years, wlLh over 300
sLudenLs. lurLher, Lhls lmprovemenL ls noL an ephemeral, one-Llme phenomenon. ln 2010, 14 sLudenLs
were ln my class ln boLh Lhe sprlng and fall semesLer. 1hey had four speaklng LesLs wlLh a LoLal of elghL
dlfferenL parLners. 1hey showed an average lncrease ln LoLal-words of 43° and average-uLLerance of 34°.
ls Lhls lncrease ln quanLlLy also an lncrease ln quallLy?

?es. A beglnnlng lce skaLer who can go 30°
longer wlLhouL falllng ls lmprovlng. A beglnnlng Lennls player who can rally 30° longer ls lmprovlng. 1hls ls
noL smoke and mlrrors, Lhls ls human naLure. 8eglnners - lncludlng LlL speakers - who do a loL lmprove a


1hls LesL ls pracLlcal. Lvery Leacher can glve a personal oplnlon score of baslc ablllLy and [oL down
vocabulary used. Lvery sLudenL can Lranscrlbe lL - ln less Llme Lhan Lhey spend sLudylng for a Lyplcal LesL.
1ranscrlblng adds one layer of complexlLy Lo Lhe LesL, buL lL provldes slx layers of beneflL: correcLlve
feedback, accuracy, Lransparency, wlde-spread grades, and a yardsLlck Lo measure lmprovemenL. LasLly and
mosL lmporLanLly, Lranscrlblng enables communlcaLlve LesLlng, whlch enables communlcaLlve classes.

!"# %&'(!)&*
My class and LesLlng are whaL l Lerm B23'()%&"$23C5&%(+ D317$%0. ConversaLlon ls Lhe class, Lhe acLlvlLy
and Lhe LesL. Accuracy ls lmproved as a byproducL of exLenslve speaklng and correcLlve feedback ls
provlded by Lranscrlblng. 1he englne of speaklng lmprovemenL ls speaklng, so sLep aslde and leL Lhem
speak. Pere ls a way Lo boLh measure and lmprove LhaL speaklng.