RESEARCH ARTICLE
CORRESPONDING AUTHOR:
S e lana Pine
Depa men o Neu ology,
Johns Hopkins Uni e si y,
Bal imo e, US; BCBL. Basque
Cen e on Cogni ion, B ain
and Language, Donos ia-San
Sebas ian, ES
[email p o ec ed]
KEYWORDS:
e o s; epai ; co ec ion;
yping; awa eness
TO CITE THIS ARTICLE:
Pine , S., & Noza i, N.
(2022). Co ec ion Wi hou
Consciousness in Complex
Tasks: E idence om Typing.
Jou nal o Cogni ion, 5(1):
11, pp. 1–14. DOI: h ps://doi.
o g/10.5334/joc.202
Co ec ion Wi hou
Consciousness in Complex
Tasks: E idence om Typing
SVETLANA PINET
NAZBANOU NOZARI
ABSTRACT
I has been demons a ed ha wi h p ac ice, complex asks can become independen
o conscious con ol, bu e en in hose cases, epai ing e o s is hough o emain
dependen on conscious con ol. This pape epo s wo s udies p obing conscious
awa eness o e epai s in nea ly 15,000 yping e o s collec ed om 145 pa icipan s
in a single-wo d yping- o-dic a ion ask. We p o ide e idence o subconscious epai s
by uling ou al e na i e accoun s, and epo wo se s o analyses showing ha a)
such epai s a e no con ined o a speci ic s age o p ocessing and b) ha hey a e
sensi i e o he inal ou come o epai . A hi d se o analyses p o ides a de ailed
compa ison o he imeline o ials wi h conscious and subconscious epai s, e ealing
ha he di e ence is con ined o he epai p ocess i sel . We p opose an accoun o
epai p ocessing ha accommoda es hese empi ical indings.
*Au ho a ilia ions can be ound in he back ma e o his a icle
2Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
INTRODUCTION
Moni o ing pe o mance o p e en ion, de ec ion, and co ec ion o e o s is an in eg al pa o
mos beha io s. I has been a gued ha highly p ac iced asks ely less on conscious p ocesses
ha con ol and egula e hem (Ande son, 1982; G. D. Logan, 1988). Bu can moni o ing and epai
p ocesses hemsel es become independen o conscious awa eness? The e is e idence ha e o
de ec ion can be ca ied ou subconsciously; o example, e o ela ed nega i i y (ERN), an EEG
ma ke o e o de ec ion, is also ound in subconsciously epo ed e o s, al hough i s magni ude
may be modula ed by consciousness (Nieuwenhuis, Ridde inkho , Blom, Band, & Kok, 2001; Wessel,
2012). Less is known abou he dependence o e o co ec ion on consciousness. Some e idence
sugges s ha co ec ions can be made wi hou any conscious ecollec ion o ha ing commi ed
an e o o ha ing a emp ed a epai . This e idence, howe e , comes om simple asks ha a e
hemsel es no subjec o consciousness, e.g., saccadic eye mo emen s (Nieuwenhuis e al.,
2001). I is hus an open ques ion whe he epai ing e o s in mo e complex asks can be ca ied
ou wi hou conscious awa eness. We answe his ques ion by examining epai s in yping.
WHY TYPING?
Typing is an in iguing ask combining highe -le el seman ico-lexical p ocesses wi h lowe -le el
mo o p ocesses (G. D. Logan & C ump, 2011; Pine & Noza i, 2018; Pine , Ziegle , & Ala io, 2016;
Yamaguchi, C ump, & G. D. Logan, 2013). On he one hand, he gene a i e na u e o language
p oduc ion limi s epe i i e beha io in yping. On he o he hand, he beha io can be highly
p ac iced o he deg ee ha implici yping pe o mance in expe ypis s is a be e han he
explici knowledge o e he ask de ails such as he absolu e and ela i e posi ions o he keys
(Liu, C ump, & G. D. Logan, 2010; Snyde , Ashi aka, Shimada, Ul ich, & G. D. Logan, 2014). This
combina ion makes yping a hallma k o a complex goal-o ien ed ask wi h an au oma ized
componen (see G. D. Logan, 2018, o an implemen a ion o such a “con olled au oma ic”
p ocess h ough a con ex e ie al and upda ing mechanism).
A he same ime, compa ed o a simila ly complex ask o spoken p oduc ion (<1% e o s),
yping is mo e e o -p one (~10–15%, Pine & Noza i, 2018). This p o ides many oppo uni ies
o obse ing epai s. The epai p ocess i sel is also unique in se e al ways: a) i is disc e e in
ha i equi es eplacing a whole segmen (i.e., a le e ) wi h a new one. b) I equi es e asing he
mis ake by he use o a unique ope a ion (p essing he backspace) be o e applying he epai . c)
Repai s can be implemen ed immedia ely a e he e o segmen . This equi es no only s opping
he cu en ac i i y bu i s esump ion om he igh poin a e epai , which makes acking he
poin o epai c i ical. In sho , apa om yping i sel being a complex gene a i e ask, epai ing
yping e o s is i sel a complex p ocess, equi ing s opping, dele ing, inse ing a new segmen ,
and esuming p oduc ion om a speci ic poin . Due o such complexi ies, e en models such as G.
D. Logan (2018), which p oposes an au oma ic componen o yping, main ain ha epai s mus
equi e conscious delibe a ion and applica ion o op-down con ol. This pape e isi s his posi ion.
WHAT IS AT STAKE?
Whe he moni o ing in a complex ask such as language p oduc ion does o does no ha e a
subconscious componen leads o e y di e en heo e ical accoun s. On he e o de ec ion
side, o example, he e ha e been signi ican heo e ical ad ances o e he pas decade
ha ha e o e ed subconscious al e na i es (Guen he , 2016; Hickok, 2012; Noza i, Dell, &
Schwa z, 2011; see Noza i, 2020 o a e iew) o he classic comp ehension-based moni o
(Le el , 1983). In con as , a clea mechanism o epai s in language p oduc ion is ye o be
es ablished (Gau in & Ha suike , 2020). An excep ion is he ex ension o Guen he ’s o wa d
model, which co ec s o he de ia ion be ween mo o plans and pe cep ual a ge s h ough
an in e se model (Guen he , 2016; Wolpe & Kawa o, 1998). As use ul as such a mechanism
is o adjus ing he low-le el aspec s o p oduc ion, i is unlikely o unde lie he epai o la ge
segmen s such as phonemes, le e s, and wo ds (see Noza i, 2020 o de ailed a gumen s).
This, in u n, calls o a epai mechanism applicable o highe le els o he p oduc ion sys em,
i.e., epai s o segmen s (e.g., le e s) and wo ds. Two gene al mechanisms ha e been p oposed
in he li e a u e: a) e-ins a ing he plan. G. D. Logan (2018) p oposes ha a yping epai
equi es e-ins a ing he ini ial wo d command be o e he au oma ic componen can ake o e
(G. D. Logan, 2018, p. 475). Since such a p ocess equi es conscious planning, i na u ally pu s
3Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
epai s ou side o he domain o subconscious p ocessing. b) Local swi ches. An al e na i e is
a p ocess ha does no equi e e-ins a ing he plan, bu ins ead makes local swi ches in he
au oma ic pa o he p oduc ion p ocess. An example is eplacing a segmen o a wo d wi h
he nex mos highly ac i a ed ep esen a ion upon he gene a ion o an e o signal (e.g.,
Noo eboom & Quené, 2020; Noza i, Ma in, & McCloskey, 2019). This p ocess could ope a e
wi hou conscious awa eness.
Cu en ly, he empi ical e idence o he deg ee o which conscious con ol is in ol ed in
epai s in language p oduc ion is mixed. On he one hand, he e is indi ec e idence poin ing
o a po en ial au oma ic componen o epai s: co ec ions can be e y quick and e icien
(Noo eboom & Quené, 2019, 2020), and s udies in some popula ions like young child en and
indi iduals wi h aphasia sugges ha such co ec ions may be ca ied ou wi hou explici
awa eness o e he e o o he epai p ocess (e.g., Cla k, 1978). On he o he hand, some
indings ha e been aken as e idence ha epai s a e subjec o a en ional con ol. Fo
example, when pa icipan s become mo e e o p one, hey end o co ec a highe p opo ion
o hei e o s (Le el , 1983; Noza i e al., 2019). Bu being subjec o con olled p ocessing does
no ule ou an unde lying au oma ic p ocess. In a seminal pape , Moo s and de Houwe (2006)
e iewed he a ious app oaches o de ining au oma ici y (e.g., ea u e-based s. cons uc -
based) and discussed he disag eemen e en among he p oponen s o he ea u e-based
app oach on whe he all o a subse o c i ical ea u es should be p esen o a p ocess o be
conside ed au oma ic, gi ing ise o he all-o -none s. decomposi ional iews o au oma ici y,
espec i ely. Following he logic o he decomposi ional iews, i is pe ec ly plausible o a
complex ope a ion o only pa ially mee he c i e ia o au oma ici y. While he e is some
disag eemen abou wha he key c i e ia a e, subconscious p ocessing equen ly makes he
lis ; i a ask can be pe o med wi hou conscious awa eness, i alls close o au oma ici y on
he spec um han asks ha equi e conscious p ocessing o comple ion.
While he e a e some implici ways o p obing consciousness o e e o s, such as using EEG
componen s ha ha e been empi ically linked o conscious judgmen s (e.g., la e posi i i y;
Pine & Noza i, 2020), he mos well-es ablished me hod has been he use o me acogni i e
judgemen s. This can ake se e al o ms: p essing a bu on only in he case o an e o
(Nieuwenhuis e al., 2001), classi ying ials as e o o co ec (Cha les e al., 2013), o speci ying
he ype o e o (e.g., which dis ac o was chosen, Di G ego io e al., 2016). The assump ion
he e is ha al hough one can ha e an implici sense o how one is pe o ming in a gi en ask, a
delibe a e me acogni i e judgmen abou pe o mance equi es consciousness. I his canno
be achie ed, he in o ma ion om he p ima y ask is deemed inaccessible o consciousness, a
leas o he le el equi ed o o ming he basis o a clea conscious judgmen .
THE CURRENT STUDY
This pape epo s wo s udies p obing conscious awa eness o e epai s in abou 15,000 yping
e o s om a single-wo d yping- o-dic a ion ask. I is impo an o no e ha he ques ion
is no whe he mos epai s a e subconscious, bu a he i any epai s a e subconscious. In
o he wo ds, unlike mos s udies whe e he p obabili y o an e en is o g ea in e es , wha
is c i ical he e is he possibili y o an e en . The eason is ha es ablishing he occu ence
o subconscious co ec ions calls o mechanisms ha a e inhe en ly di e en om cu en
p oposals o consciously es a ing he p oduc ion p ocess and in i es new heo e ical accoun s.
The c i ical poin , in u n, is o es ablish ha subconscious co ec ions, i any, a e a) eplicable,
and b) no he p oduc o al e na i e mechanisms such as o ge ing a conscious ac . This is he
app oach we ollow in his pape . Once es ablished, we epo h ee se s o analyses explo ing
key cha ac e is ics o subconscious epai s, and end by p oposing an accoun o co ec ion
p ocesses ha accommoda es he empi ical indings.
METHODS
CORPUS COMPOSITION
We analyzed a co pus o yping e o s collec ed om wo expe imen s (s udy 1 and s udy
2). The expe imen s a e simila in hei gene al s uc u e; hus, hey a e desc ibed unde one
me hodology, and hei da a a e me ged o he analyses. S udy 2 p o ides a eplica ion and
add esses some conce ns in s udy 1.
4Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
PARTICIPANTS
Pa icipan s we e ec ui ed ia Amazon Mechanical Tu k. They we e consen ed unde a p o ocol
app o ed by he Ins i u ional Re iew Boa d o Johns Hopkins School o Medicine and we e
compensa ed o hei pa icipa ion. Pa icipa ion eligibili y was de e mined by a sc eening es
adminis e ed be o e pa icipan s we e accep ed in o he s udy. This es had wo pa s. In he
i s pa , pa icipan s yped 15 wo ds o dic a ion (one a a ime), wi h a elaxed deadline
(5000 ms). In he second pa , hey yped 15 wo ds o dic a ion (one a a ime), unde a sho e
deadline (2000 ms). The beginning and end o he yping pe iod o each wo d was indica ed
by sho beeps. Pa icipan s had o each 80% accu acy on he i s pa , and inish yping a
leas 80% o ials unde he ime limi (wi h minimum 50% accu acy) on he second pa o
be accep ed in o he s udy. The sc eening phase con inued un il he p e-de e mined numbe
o pa icipan s was eached o each s udy (60 o s udy 1, and inc eased o 85 o s udy 2 o
inc ease he chance o inding unconscious co ec ion a emp s).
S udy 1 included 60 pa icipan s (25 males, 35 emales), aged on a e age 38.0 ±10.4 yea s old
( ange 21–63). All had comple ed high school and 27 had a bachelo ’s deg ee o highe . All bu
wo we e English na i e speake s. The wo who we e no , had lea ned English a he age o i e.
The sample had a mean yping speed o 75.3 ± 19.0 wo ds pe minu e ( ange 49–126), and an
accu acy o 93.3 ± 5.8% ( ange 70–100). All pa icipan s epo ed using se en o mo e inge s
o yping. Se en (11%) epo ed looking a hei inge s when yping and 22 (37%) epo ed
ha ing o mal yping aining. Pa icipan s had been yping o 21.0 ± 8 yea s ( ange 6–40) and
yped an a e age o 3.6 ± 2.6 hou s a day ( ange 0.5–12).
S udy 2 comp ised 85 pa icipan s (53 males, 32 emales), aged on a e age 35.3 ± 9.7 yea s
old ( ange 21–65). All had comple ed high school and 39 had a bachelo ’s deg ee o highe .
All we e English na i e speake s. The sample had a mean yping speed o 75.5 ± 15.0 wo ds
pe minu e ( ange 45–118), and an accu acy o 94.2 ± 5.5 % ( ange 76–100). Twen y-nine
pa icipan s epo ed ha ing o mal yping aining, and 21 epo ed looking a hei inge s
when yping. Ten pa icipan s epo ed using six o ewe inge s, which could be a sou ce o
conce n abou hei yping expe ise. Howe e , o e he whole sample, he numbe o inge s
pa icipan s epo ed using was no co ela ed wi h hei yping speed o accu acy. Pa icipan s
had been yping o 20.5 ± 8 yea s ( ange 2–42) and yped an a e age o 3.7 ± 2.6 hou s a day
( ange 0–10).
Eigh pa icipan s pa icipa ed in bo h s udies. They we e included in he analysis o each
indi idual da ase . Howe e , o analyze he whole co pus o e o s, only hei en y om
da ase 1 was included, so ha only unique subjec s we e p esen in he co pus. The inal
co pus comp ised da a om 136 pa icipan s.
MATERIALS
Fo bo h da ase s, s imuli we e 7 and 8-le e wo ds. S imuli o da ase 2 we e a subse o
s imuli used o da ase 1 (see P ocedu es below o numbe s). Log- ans o med wo d equency
was kep unde 3, based on he SUBTLEX da abase (B ysbae & New, 2009), o inc ease ask
di icul y. No plu al wo ds, compound wo ds o wo ds ha had homophones we e selec ed.
Audi o y s imuli we e eco ded by a na i e English speake .
PROCEDURES
The expe imen was p og ammed using he jsPsych lib a y (de Leeuw, 2015), embedded in
an HTML en i onmen . The Py hon lib a y psiTu k (Gu eckis e al., 2016) was used o handle
pa icipan s’ ec ui men and compensa ion. Pa icipan s pe o med he expe imen online
using hei own compu e and keyboa d.
Da ase 1 e e s o da a collec ed unde s udy 1. I was acqui ed o e wo sessions and comp ised
300 ials pe pa icipan (i was pa o a la ge expe imen comp ising 1200 ials). Da ase 2
(i.e., da a collec ed unde s udy 2) was acqui ed o e one session and con ained 600 ials pe
pa icipan . In bo h cases, he ial s uc u e was simila and as ollows (see
Figu e 1
): an audi o y
s imulus was p esen ed, ollowed immedia ely by a sho beep ha se ed as a cue o s a
yping he wo d. Pa icipan s hen had 1800 ms o ype be o e hey hea d a lowe pi ched beep
ha ma ked he deadline. We bu e ed an addi ional 500 ms a e he second beep o p ese e
5Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
some o he esponses yped a e he deadline. Pa icipan s could see wha hey yped on he
sc een in eal ime. Once he ime was up, a p omp asked hem o judge whe he hey had
made an e o o no by p essing ‘Y’ o yes, ‘N’ o no, and ‘A’ in case hey did no p oduce a
esponse. Ins uc ions s essed ha i hey had made an e o and co ec ed i , i s ill coun ed as
an e o and he app op ia e esponse would be ‘yes’. I hey did no submi an answe by 5000
ms, a p omp eminded hem o espond as e . Pa icipan s could ake a b eak e e y 50 ials.
Figu e 1 T ial s uc u e o
bo h da ase s: a wo d yping
pe iod (unde a ime limi ) is
ollowed by an e o p omp ,
and e en ually a co ec ion
p omp (in Da ase 2 only).
Da ase 2 was collec ed wi h modi ied ins uc ions o dec ease, o he ex en possible, he
occu ence o un epo ed co ec ions by chance. Two changes we e implemen ed: 1) The
esponse o ma was changed om p essing a Y/N bu on o yping ou he yes/no esponse.
This elimina es he chance o a mis ake due o pushing he w ong bu on. 2) A wo-s ep
p ocedu e was implemen ed, in which pa icipan s i s answe ed he ques ion “Did you make
an e o when you i s yped he wo d?”. I hey answe ed ‘yes’, a second p omp asked hem
“Did you y o co ec ?”. To gi e hem he bes sho a epo ing hei e o s, we wen one s ep
u he his ime: i hei esponses con ained a backspace, e en i hey answe ed “no” o he
i s ques ion, we asked hem he second ques ion. This s a egy elimina ed any chance ha a
misunde s anding abou he na u e o he ques ion may ha e con ibu ed o pa icipan s no
epo ing hei e o s.
DATA ANALYSES
Any wo d ha con ained ewe keys okes han he a ge wo d, and/o a keys oke no
expec ed in he a ge wo d was conside ed an e o . Reac ion imes (RTs) we e calcula ed om
he onse o he i s beep (which ollowed immedia ely a e he p esen a ion o he audi o y
wo d) o he i s keys oke. In e -keys oke in e als (IKIs) we e he ime in e als be ween
consecu i e keys okes and we e a e aged o e each wo d. Hi Ra es, Misses, Co ec Rejec ions
and False Ala ms we e calcula ed o each condi ion o each pa icipan , and SDT pa ame e s
(d’ and c i e ion) we e es ima ed using he psycho R package ( e sion 0.4.91, Makowski, 2018,
R e sion 3.3.3). RTs and IKIs abo e o below 3SD o he mean o each pa icipan ’s RT and
IKI dis ibu ions we e emo ed o he analysis. E o de ec ion a es we e compu ed based
on subjec i e epo s (me acogni i e p omp a he end o e e y ial). Since he use o a
backspace is i al in any co ec ion a emp s, we ha e de ined ou measu e o co ec ion as
such. Co ec ion a emp a es we e compu ed as he numbe o ials in which he backspace
key was p essed a leas once, di ided by he o al numbe o ials. Analyses we e un using
non-pa ame ic s a is ical es s o ca ego ical a iables (Wilcoxon sign- ank es s) and linea
mixed e ec models o con inuous a iables (package lme Tes s, e sion 3.0-1). Da a and
analysis sc ip s can be ound he e: h ps://os .io/jg3 m/.
RESULTS
One pa icipan (in s udy 2) ne e epo ed e o s, meaning he had a disc iminabili y index d’
o ze o and a high esponse c i e ion (c = 2.82). Since e o de ec ion beha io is ou a iable o
in e es he e, his pa icipan was excluded om u he analyses.
6Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
DATASETS DESCRIPTION
We analyzed a co pus o 14,841 yping e o s, 3,459 (20%) om da ase 1 and 11,382 (22.6%)
om da ase 2. E o de ec ion a es we e 69% and 68%, and co ec ion a es 29% and 37%,
in da ase s 1 and 2, espec i ely (
Figu e 2
), wi h no signi ican di e ences be ween he wo
da ase s. The mean IKI (i.e., ime be ween wo keys okes) in da ase 1 was 166.1 ms/keys oke,
anging om 106.5 o 235.1 ms/keys oke. The a e age RT (i.e., ime o he i s keys oke) was
332.9ms, anging om 198.7 o 543.5ms. In da ase 2, mean IKI was 161.9 ms/keys oke,
anging om 103.8 o 240.7 ms/keys oke. The a e age RT was 312.7ms, anging om 213.1
o 453.5ms. Nei he he mean IKIs, no he RTs we e signi ican ly di e en in he wo da ase s.
The RT analyses eplica ed he inding in bo h yping and spoken p oduc ion (C ump & G. D.
Logan, 2013) ha e o s (392.4 ± 103 ms) ha e signi ican ly longe RTs han co ec (318.1 ± 63
ms) wo ds (ß = –55.9, = –30.1, p < .001).
Figu e 2 Dis ibu ions o hi
a es o e o de ec ion and
e o co ec ion a emp o e
subjec s o each da ase .
Con idence in e als (95%) a e
p esen ed in squa e b acke s.
UNREPORTED CORRECTION ATTEMPTS
Un epo ed co ec ion a emp s we e de ined as ials ha we e no epo ed as con aining an
e o , e en hough a backspace was used o change a yped le e . The e we e 173 (19.2% o
900 esponses wi h a backspace) such ials in da ase 1 and 397 (10.3% o 3817 esponses
wi h a backspace) in da ase 2 (see
Table 1
).
Figu e 3
shows he a iabili y in he p opo ions o
un epo ed co ec ion a emp s be ween pa icipan s. They we e e iden in 63% o pa icipan s
in da ase 1. Despi e he measu es aken o educe he chance o missing a co ec ion a emp ,
82% o pa icipan s in da ase 2 s ill showed e idence o un epo ed co ec ion a emp s,
meaning ha swi ching Y/N o yped ou esponses did no elimina e un epo ed co ec ion
a emp s. Sepa a ion o p obing o e o de ec ion and co ec ion a emp in da ase 2 also
did no elimina e un epo ed co ec ion a emp s. Ou o 3404 epo ed e o s, 241 ials
wi h backspace (7%) we e s ill epo ed as no an a emp a co ec ion. As a compa ison,
co ec ion a emp s we e alsely epo ed on 825 ials wi hou backspace (19%). Finally,
clueing pa icipan s o he ac ha an e o has been made by asking hem abou co ec ion
a emp s on all e o ials wi h a backspace, whe he epo ed and un epo ed, did no
elimina e un epo ed co ec ion a emp s. Recall ha in da ase 2, i pa icipan s esponded
“no” o he i s ques ion abou e o commission bu had used he backspace, hey we e s ill
p esen ed wi h he second ques ion abou a emp ing o co ec . Ou o such 394 co ec ion
a emp s, 241 (61%) emained un epo ed.
DATASET 1 DATASET 2
ERROR
REPORTED
ERROR NOT
REPORTED
ERROR
REPORTED
ERROR NOT
REPORTED
Co ec ion a emp 727 (80.8%) 173 (19.2%) 3423 (89.7%) 394 (10.3%)
No co ec ion a emp 1647 (64.4%) 912 (35.6%) 3998 (56.1%) 3128 (43.9%)
Table 1 Raw coun s and
pe cen ages o ials wi h
co ec ion a emp s, b oken
down by whe he an e o was
epo ed o no . Cells in bold
cells co espond o un epo ed
co ec ion a emp s
(pe cen age compu ed ou o
he o al numbe o ials wi h
co ec ion a emp ).
7Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
Figu e 3 Dis ibu ions o he
pe cen age o un epo ed
co ec ion a emp s o e
subjec s o each da ase .
Median, Q1, and Q3 alues
a e plo ed as a boxplo , and
gi en below each g aph.
To summa ize, he e was clea e idence, ac oss wo da ase s, o co ec ion a emp s ha we e
no epo ed, sugges ing he possibili y ha such co ec ion a emp s may ha e been ca ied
ou wi hou conscious awa eness. While he manipula ions in da ase 2 exclude possibili ies
such as misunde s anding o ins uc ions and mis akes in esponse selec ion, hey canno
add ess a mo e heo e ical al e na i e, namely ha pa icipan s may ha e been conscious o
a co ec ion a emp bu o go en i by he ime a epo was due a he end o he ial. We
e alua e his “memo y ailu e” accoun using a posi ional analysis.
ARE UNREPORTED CORRECTION ATTEMPTS MEMORY FAILURES?
Gi en he p imacy and ecency e ec s in memo y (Ebbinghaus, 1913) and hei well-es ablished
ex ension o he phonological bu e in p oduc ion (Acheson & MacDonald, 2009; Page & No is,
2009), a memo y accoun o un epo ed epai s would p edic he lowes p obabili y o such
epai s (i.e., he lowes a e o o ge ing) a he beginning and end o he wo d, and a highe
p obabili y a he middle o he wo d. The obse ed da a a e plo ed in
Figu e 4
. Posi ion was
coded as he dis ance be ween he epai and he inal edge o he wo d, such ha posi ion
0 ma ks he las le e , posi ion 1 he penul ima e le e , and so on. A o mal analysis wi h
posi ion as a p edic o showed ha he a e age posi ion o he backspace in he wo d did no
di e signi ican ly be ween epo ed (M = 5.49) and un epo ed (M = 5.47) co ec ion a emp s,
Z = 1.04, p = .30. Impo an ly, as can be seen in he igu e, he pa e n does no esemble he
p edic ion o he memo y accoun : he p obabili y o no epo ing a co ec ion a emp was
nume ically highes in posi ions 7 and 0, i.e., a he beginning and end o he wo d, whe e such
p obabili y, i due o o ge ing, should be he lowes . To a ibu e his o memo y ailu e, one
mus assume e e sed p imacy and ecency e ec s. These da a ule ou memo y ailu e as a
iable explana ion o un epo ed co ec ion a emp s.
Figu e 4 S acked ba s o
epo ed (g ey) and un epo ed
(black) co ec ion a emp s
by posi ion. Posi ion is coded
as he dis ance om he inal
edge o he wo d, wi h posi ion
0 being he inal le e . Only
esponses ha had he same
numbe o cha ac e s as he
a ge wo d we e included,
whe he he inal ou come
was co ec o no .
8Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
Be o e we mo e on o he p ope ies o subconscious co ec ion a emp s, i is help ul o ule
ou ano he possibili y, namely ha pa icipan s may ha e had s o ed pa e ns o yping and
co ec ing, as a chunk, o ce ain wo ds ha hey au oma ically execu ed. This is easonable
o sho and equen wo ds ha a e p one o consis en e o pa e ns (e.g., “ he” o “ eh”).
Ou analysis showed ha 46 wo ds in hese da ase s we e subjec o only un epo ed co ec ion
a emp s. By design, none o hese wo ds we e sho e han 7 le e s, none we e unc ion wo ds,
and hei median log equency was 2, meaning ha hey we e no among he highe - equency
wo ds in he lexicon. Mo eo e , he posi ion o e o s on hese wo ds was no consis en ac oss
pa icipan s, sugges ing ha he e o pa e ns a e no s anda ds o highly p edic able o be
au oma ed. The combina ion o hese indings does no p o ide suppo o he idea o s o ed
e o + epai chunks ha could be quickly and subconsciously execu ed as an in eg al pa e n.
PROPERTIES OF UNREPORTED CORRECTION ATTEMPTS
The analyses epo ed so a ocused on es ablishing he exis ence o un epo ed co ec ion
a emp s and he subconscious na u e o such a emp s. The nex se ies o analyses ocus on
unde s anding he p ope ies o hese subconscious co ec ion a emp s.
S ages o p ocessing. Because yping in ol es se e al le els o ope a ion, yping e o s may a ise
om di e en s ages. Following p e ious wo ks in yping (e.g., F. A. Logan, 1999; Rumelha &
No man, 1982), G. D. Logan (2018) d aws a dis inc ion be ween inge selec ion e o s (e o
keys adjacen o he a ge keys) and con ex - e ie al e o s, i.e., hose which happen p io
o he launching o a inge mo emen command (o en no adjacen o he a ge key). We
compa ed 402 co ec ed adjacen e o s o 772 co ec ed non-adjacen e o s. The a e o
un epo ed co ec ion a emp s was 23% s. 18%, espec i ely, sugges ing ha un epo ed
a emp s we e no es ic ed o e o s a ising om a ce ain le el o p ocessing. This di e ence
be ween he wo was also no signi ican , Z = 0.92, p = 0.36.
Repai ou come. Co ec ion a emp s can esul in he co ec esponse (success ul epai s) o
no . O he 2023 co ec ion a emp s wi h co ec ou comes, 23% we e no epo ed. This was
signi ican ly highe han he 11% ou o 1835 o a emp s ha did no esul in he co ec
esponse, Z = –3.39, p < .001. This means ha pa icipan s we e mo e likely o ail o epo an
e o and i s a emp a co ec ion i he inal ou come had been co ec .
Timing o un epo ed co ec ion a emp s. As he i s -pass explo a ion o iming di e ences
be ween epo ed and un epo ed co ec ion a emp s, we es ed he di e ences be ween
immedia e and delayed epai s. Immedia e epai s (i.e., hose wi h a single backspace igh a e
he e o ) we e mo e common (2296) han he delayed epai s (i.e., wi h se e al backspaces
o dele e he le e s yped a e he e o ; 1558), and p oduced signi ican ly mo e un epo ed
co ec ion a emp s (19% s. 13%, espec i ely; Z = –2.96, p = 0.003). Since his inding sugges s
ha iming may be an impo an aspec o subconscious co ec ion a emp s, we ollowed up
his analysis wi h mo e de ailed analyses o he iming o epo ed s. un epo ed co ec ion
a emp s in immedia e epai s.
The inding o longe RTs o e o s. co ec wo ds was eplica ed a he le el o segmen s:
e o segmen s (234.94 ± 55 ms) had signi ican ly longe IKIs han p e-e o segmen s (171.78
± 35 ms; ß = –62.0, = –16.54, p < .001), sugges ing he same gene al dynamics a he le el o
wo ds and segmen s. The compa ison o mean IKIs (i.e., IKIs a e aged o e all segmen s o he
wo d) be ween epo ed (233.7 ± 85 ms) and un epo ed (206.2 ± 50 ms) co ec ion a emp s
showed signi ican ly longe imes o he o me (ß = 11.8, = 5.72, p < .001). To u he
in es iga e he p ocess om which his di e ence a ose, we di ided he imeline o a ial in o
ou egions: 1) P e-e o was de ined as all co ec keys okes be o e he e o keys oke, 2)
E o was de ined as he e o keys oke, 3) Repai was de ined as he backspace keys oke and
he ollowing keys oke, which we e analyzed sepa a ely. Finally, 4) Pos - epai was de ined as
any keys oke ollowing he epai egion.
Figu e 5
shows he du a ion o hese egions in ials wi h epo ed and un epo ed epai s and
Table 2
epo s he s a is ics. The esul s showed ha he di e ence was con ined o he Repai
egion. Bo h he backspace IKI and he new le e IKI we e signi ican ly as e o un epo ed
han epo ed co ec ion a emp s (backspace IKI: ß = 18.0, = 2.94, p = .0034; new le e IKI:
ß = 11.7, = 3.64, p < .001). Since epea ed compa isons (i.e., di e en es s in he ou zones)
9Pine and Noza i
Jou nal o Cogni ion
DOI: 10.5334/joc.202
inc ease he p obabili y o inding a signi ican e ec by chance, i is app op ia e o apply a
co ec ion o mul iple compa isons o hese esul s. Bo h e ec s emained signi ican a e
Bon e oni co ec ion o mul iple compa isons (co ec ed alpha-le el = .008). None o he o he
egions showed signi ican di e ences be ween ials wi h and wi hou epo ed co ec ion
a emp s. This pa e n o esul s shows ha he e ec is es ic ed o he epai p ocess.
Figu e 5 Timeline o ials
wi h and wi hou epo ed
co ec ion a emp s. The
IKI (o mean IKI i he e a e
se e al keys okes) is shown
o each egion. As e isks
indica es he egion whe e he
di e ence be ween epo ed
and un epo ed co ec ion
a emp s was signi ican .
REPORTED NOT
REPORTED
COHEN’S
d
S P 95%CI
BOOTSTRAP
Gene al Mean IKI (ms) 233.7 (84.9) 206.2 (49.7) 0.409 11.8 5.72 <.001 7.39 15.9
P e-e o Mean IKI be o e
e o (ms)
173.2 (39.0) 167.4 (52.8) 0.126 3.51 1.41 0.159 –1.61 8.59
E o E o IKI (ms) 238 (63.5) 227.3 (116.8) 0.119 6.38 1.47 0.142 –1.94 14.6
Repai Backspace IKI
(ms)
394.8 (104.6) 385 (153.6) 0.076 18.0 2.94 0.0034 5.54 30.1
IKI a e
backspace (ms)
182.3 (65.6) 163.8 (70.7) 0.271 11.7 3.64 <.001 5.2 18.5
Pos -
co ec ion
Mean IKI a e
epai (ms)
83.7 (36.5) 76.7 (47.5) 0.167 2.71 0.971 0.332 –2.79 8.1
Table 2 Resul s o s a is ical
analyses compa ing iming
measu es be ween epo ed
and un epo ed co ec ion
a emp s. All compa isons
we e un using mixed-e ec
linea models. Numbe s in he
pa en heses a e SD.
DISCUSSION
The pu pose o his s udy was o de e mine ha e o s can be epai ed subconsciously, and hus
o es he su iciency o accoun s ha ely on conscious ini ia ion o epai s. Using a la ge co pus
o yping e o s, we epo ed clea e idence o co ec ion (a emp s) ha we e no epo ed,
sugges ing he possibili y o epai ing an e o wi hou conscious awa eness. Speci ic manipula ions
in s udy 2 excluded possibili ies such as misunde s anding o he ins uc ions and mis akes in
esponse selec ion. We u he excluded he possibili y ha pa icipan s may ha e been conscious
o a co ec ion a emp bu ha e o go en i by he ime hey had o epo i , by showing ha he
posi ional e ec s we e incompa ible wi h a memo y ailu e accoun . These indings p o ide s ong
suppo o a mechanism o he exis ence o subconscious epai s. Nex , we in es iga ed h ee
p ope ies o conscious s. subconscious epai s o help us de ine a epai mechanism. These h ee
cha ac e is ics a e summa ized below, ollowed by a schema ic model ha accommoda es hem.
Cha ac e is ic 1) S age o p ocessing. Subconscious co ec ions we e no con ined o a ce ain
s age o he sys em om which e o s a ose. Bo h adjacen and non-adjacen e o s we e
subjec o subconscious co ec ions a compa able a es. The exis ence o subconscious epai s
in bo h e o ypes, which oughly map on o e o s o mo o planning s. segmen al encoding
in psycholinguis ic models, sugges s ha subconscious epai s a e no ied o a speci ic s age
o p ocessing. The epai model mus hus accommoda e a mechanism ha is amenable o
ep esen a ions a di e en s ages.
