 S aipsniai / A icles 273
CHARITONCHARITONIDIS
Independen  esea che
ORCIDid:o cid.o g/0000-0003-0298-2629
Fieldso  esea ch:wo d o ma ion,mul iwo dexp essions,
lexicalseman ics,wo d ecogni ion,emo ion.
DOI:doi.o g/10.35321/all90-11
EXPLORINGSCALEDAIC
WITHINENGLISHCLOSED
COMPOUNDS
Anglųkalbosužda ųjųjunginių
(sudu iniųžodžių)skalėsAIC y imas
ANNOTATION
TheAkaikeIn o ma ionC i e ion(AIC)isanes ablishedgoodness-o - i measu e o 
selec ingmodelsin heanalysiso empi icalda a.Howe e ,AICissensi i e osample
size.Au ho ’sp e ious esea chhasshown ha ScaledAIC,i.e.AICdi idedbysample
size,isane ec i e ool o assessingmodel i andhie a chizing eg essionmodels.The
p esen s udyexplo es u he p ope ieso  his a iable.Theobjec o in es iga iona e
66mul iple eg essionmodels e e ing o hep ocessingo closed(conca ena ed)English
compounds aken om Gagné e  al.’s (2019) La ge Da abase o  English Compounds
(LADEC).Inpa icula ,ScaledAICisjux aposed o heEnglishLexiconP ojec (ELP)
andB i ishLexiconP ojec (BLP)assou ceso  esponse imes, helexicaldecisionand
naming asks,compoundleng h,and anspa encyno ms.One-wayANOVA,maine ec s
analysis,andnon-pa ame ic es sa eusedasme hods.The indingssugges  ha Scaled
AICis esponsi e oexpe imen aldesign, hesou ceo  esponse imes,and helexical
decision and naming asks. A  he same ime, he esul s o  his s udy o e  empi ical
suppo  o  he alida iono me hodsemployedbyGagnée al.(2019).
 KEYWORDS:Englishcompounds,ScaledAIC,lexicaldecision,naming.
ANOTACIJA
Akaikės in o macijos k i e ijus (angl. AIC) y a pas o us modelių inkamumo ma as,
aikomasempi iniųduomenųanalizei.TačiauAICy ajau usim iesdydžiui.Anks esni
CHARITONCHARITONIDIS
274 Ac aLinguis icaLi huanicaXC
au o iaus y imai pa odė, kad skalės AIC, padaly as iš im ies dydžio, y a eiksminga
p iemonė modelio inkamumui į e in i i  eg esijos modeliams hie a chizuo i. Šiame
y imenag inėjamos olimesnėsšiokin amojoypa ybės.Ty imoobjek as–66daugialypės
eg esijos modeliai, susiję su užda ųjų (sudu inių) anglų kalbos junginių, paim ų iš
Gagné’ėsi ki ų (2019)Anglųkalbossudu iniųžodžių(junginių) didžiosiosduomenų
bazės(angl.LADEC),apdo ojimu.Pi miausiaAICsug e inamassuAnglųkalbosžodyno
p ojek u(angl.ELP)i B i ųkalbosžodynop ojek u(angl.BLP),kaipa sakolaiko,leksinių
sp endimųi į a dijimoužduočių,junginiųilgioi skaid umono mųšal iniai.Naudojami
me odai– ienpusėANOVA(angl.Analysiso  a iance),pag indinių ezul a ųanalizėi 
nepa ame iniai es ai.Iš ados odo,kad skalės AIC eaguojaįekspe imen inį p ojek ą,
a sakolaikošal inįi leksiniųsp endimųbeiį a dijimoužduo is.Tuopačiušio y imo
ezul a aisu eikiaempi inįpag indąGagné’ėsi ki ų(2019) aikomųme odųpa i inimui.
 ESMINIAIŽODŽIAI:anglųkalbosjunginiai(sudu iniaižodžiai),skalėsAIC, leksinis
sp endimas,į a dijimas.
1. THELARGEDATABASE
OFENGLISHCOMPOUNDS
(LADEC:GAGNÉETAL.2019)1
The La ge Da abase o  English Compounds (LADEC: Gagné e  al.
2019) is he la ges  exis ing da abase o  compound wo ds. I  con ains o e 
8000nonspaced(“closed”o “conca ena ed”)compounds(=nouns)selec ed
om a ioussou cesincluding,amongo he s, heCELEXda abase(Baayen
e al.1995), heEnglishLexiconP ojec (ELP;Balo ae al.2007), heB i ish
Lexicon P ojec  (BLP; Keulee s e  al. 2012), he B i ish Na ional Co pus
(BNC),andWo dne .F om he ullse o LADECen ies,7,804compounds
canbeuniquelypa sedin o wo eemo phemescons i uen s.2A as  a ie y
o compoundsisconside ed, o ins ancenoun-nouncompounds,e.g.bu e cup,
shipya d,compoundswi hasecondcons i uen de i ed oma e bals em,
e.g.pacemake , painkille ,e c.( o de ini ionso compoundclassesseeLiebe 
2004:46).The i s non-headcons i uen  e e s oawide angeo g amma ical
ca ego ies.Figu e1con ainsab ie sampleo LADECen ies.
Gagnée al.’s(2019)mul iple- eg essionmodelsincludeawide angeo 
p edic o (=independen ) a iables,suchascompoundleng h,big am equency
 1 Thissec ionwasadop ed omCha i onCha i onidis(2022)wi hsligh al e a ions.
2
LADECincludesplu also al eadylis edcompoundsassepa a een ies.
 S aipsniai / A icles 275
Explo ing Scaled AIC wi hin English Closed Compounds
a  he mo pheme bounda y, amily size, wo d equency, p obabili y and
associa ion ( ec o -based) measu es, emo ional/sen imen  no ms compu ed
ompa icipan  a ings,e c.Thelog esponse imes o  hecompounds om
ELP(lexicaldecision,naming)andBLP(lexicaldecision)a eusedasdependen 
a iables.Fo  hemos pa ,compoundleng h(numbe o cha ac e s)andlog
compound(=wo d) equency om heSUBTLEX-USco pus(B ysbae ,New
2009)3andBNC(BLP)a eusedascon ol a iables.InGagnée al.’s(2019)
models, he p edic o  a iables men ioned abo e had signi ican  e ec s on
lexicaldecisionandnaming imes.
FIGURE 1. LADECen ies:sample
a e li e
ai c a
ash ay
dayd eaming
dimwi
d awback
pacemake
padlock
painkille
backboa d
ballplaye 
bu e cup
ea hquake
egghead
eyeb ow
shipya d
shoelace
sho gun
ca e ake
cas away
c oss i e
o sp ing
ou cas s
o e d i e
ex book
h owback
u na ound
The p ima y ocus in Gagné e  al.’s (2019) s udy was placed on a ious
measu eso seman ic anspa ency.Gagnée al.(2019)askedpa icipan s o a e
compoundsconside inghowp edic able hemeaningo  hecompoundis om
i s pa s (meaning p edic abili y a ings, compound-based) and how much o 
hemeaningo eacho  hecons i uen sis e ainedin hecompound(meaning
e en ion a ings,cons i uen -based).Theau ho s ound ha  hedis ibu ion
o  anspa encies o  hesecondcons i uen wasmuchmo epeakedandhighe 
han hedis ibu iono  anspa encies o  he i s cons i uen (MC1:64.80[SD:
19.59] s.MC2:71.00[SD:16.46].N = 8115). Howe e , he a ing o  he
3
TheSUBTLEX-USco pusisa51-million- okenco pusbasedonsub i les omUS ilmsand
ele ision p og ams. Se e al ecen  s udies ha e p o ided e idence indica ing ha  equency
no msob ained omsub i leso mo iesand ele isionp og ams end obemo ee ec i e han
hosede i ed omp in ed ex swheni comes oexplaining hedi e encesinlexicalp ocessing
imeand,insomecases,accu acyamongna i espeake so  a iouslanguages(seeChene al.
2018:2and he e e ences he ein).
CHARITONCHARITONIDIS
276 Ac aLinguis icaLi huanicaXC
i s  cons i uen  was mo e s ongly co ela ed wi h he a ing o  he en i e
compound hanwas he a ing o  hesecondcons i uen (c1~cmp: =0.75,
p<.001 s.c2~cmp: =0.66,p<.001.N=429).4Mos no ably, hemeaning
e en ion a ing o  he i s cons i uen and hemeaningp edic abili y a ing
o  hecompoundp edic edall h ee ypeso  esponse imes,i.e.ELPlexical
decision,BLPlexicaldecision,andELPnaming imes.
Toconclude, hepeakedandhighe dis ibu iono  anspa encies o  he
second cons i uen  and he i s  cons i uen ’s be e  associa ion wi h he
compound’s meaning p edic abili y appea  o be immedia ely mapped on o
heheadope a ionsinEnglishcompounds.Thesecondcons i uen ,i.e. he
head,isauni whose anspa encyisenhancedca ego iallyandseman ically
(as o  heseman ic aspec , see he ela ionso  en ailmen andhyponymy).
The i s cons i uen ,i.e. hemodi ie ,is hemos c i ical ac o ines ablishing
compound e e ence.Asa esul ,i s anspa encyco a ieswi h he anspa ency
o  hecompoundmos s ongly.5
2. AKAIKEINFORMATIONCRITERION(AIC)
In 1973, Hi o ugu Akaike de eloped a me hod o es ima e he ela i e
expec a iono Kullback-Leible dis ance(Kullback1959)usingFishe ’smaximized
log-likelihood(Fishe 1922;seealsoAld ich1997).Thismeasu e,commonly
e e ed oas heAkaike In o ma ion C i e ion(AIC;Akaike1973),in oduceda
no el amewo k o selec ingmodelsin heanalysiso empi icalda a,ma king
asigni ican pa adigmshi (Bu nham,Ande son2002).
AICis ypicallycalcula edas ollows:–2lnL+2k,inwhich‘lnL’ e e s o
hemaximized/ ulllog-likelihoodo  hemodeland‘k’ e e s o henumbe 
o pa ame e sincluding hecons an .Asmalle se o p edic o sis ypically
associa edwi hmo ee icien models(modelswi halowe in o ma ionloss).
Thelowe (=mo enega i e) heAIC alue, hebe e  he i o  hemodel.In
hiscon ex ,AICpenalizes,asagoodness-o - i measu e, heuseo ala ge
numbe o p edic o s ha ,po en ially, esul inhighe AIC alues(see he‘+2k’
pa o  heAICequa ion).
4
S eige ’s(1980)z es showed ha  hisdi e encewassigni ican ,z=27.71,p<.0001(Gagné
e al.2019).
 5 By e e ing op e ious esea ch,Gagnée al.(2019) epo  ha “ hemodi ie ( he i s cons i uen 
inEnglish) ends oplayala ge  olein heease-o - ela ionselec iondu ing hep ocessingo 
compoundsandnounph ases.”
 S aipsniai / A icles 277
Explo ing Scaled AIC wi hin English Closed Compounds
AICissensi i e osamplesize.AICc,aco ec ed e siono AIC,inco po a es
samplesize h ough he o mula2k(k+1)/(n–k–1).Howe e ,i speci ically
add essessmall samplesizesand isno  ecommended o  models basedon
la gesamplesizessuchas ha inGagnée al.(2019).6I shouldbeno ed ha 
esea che ssuchasKenne hP.Bu nham&Da idR.Ande son(2002)dono 
o e ade ini i esolu ion o compa ingAIC alueso models i edonbo h
di e en andla gesamplesizes.7
In pa icula , Bu nham & Ande son (2002: 80–85, 334–335) p o ide a
comp ehensi ediscussiono  heimplica ionso unequalsamplesizes o model
compa ison.Theya gue ha employingin o ma ionc i e ia ocompa emodels
wi hdi e en samplesizescanlead omisleading esul s.Simila ly,asno edin
anonlinediscussionbyS e unko in2016(see e e encea e  hebibliog aphy),
all in o ma ion c i e ia a e based on he likelihood unc ion ha , in u n,
dependsonsamplesize.Speci ically,as hesamplesizeinc eases, helikelihood
dec eases.Consequen ly,in o ma ionc i e iawillalsoinc easeinsuchcases.8
3. PREVIOUSRESEARCH
InCha i onidis(2022) heAIC alues o 44mul iple eg essionmodelswi h
di e en combina ionso emo ion a iables( alence,a ousal,andconc e eness
o (a)wo dsand(b)wo dcon ex s)we edi idedbysamplesize(N) oyield
ScaledAIC(AIC/N) alues.9Subsequen ly, hese alueswe eu ilized oassess
 6 Fo  u he in o ma iononAICc, he eade is e e ed oBu nham&Ande son(2002:374–380).
7
Oneo  hesolu ions ha Bu nham&Ande son(2002)p opose e e s o he ans o ma iono  he
AIC alues o“Akaikeweigh s” ha a ede inedas“ he ela i elikelihoodo  hemodel,gi en he
da a”(Bu nham,Ande son2002:xiii;seealsoWagenmake s,Fa ell2004).
 8 A ailablea :h ps://s a s.s ackexchange.com/ques ions/94718/model-compa ison-wi h-aic-based-
on-di e en -sample-size [accessed 16.06.2023]. The eade  can comp ehend S e unko ’s
s a emen bysubs i u ingdi e en  alues o  he‘lnL’componen in heAICequa ion,while
main aining he‘2k’componen cons an .Adec easein helnL aluewill esul inahighe ,i.e.
in e io ,AIC alue.
 9 In heli e a u e,ScaledAICisalso e e ed oas“meanAIC”.Acco ding oS e unko (pe sonal
communica ion), hep ac iceo di iding heAkaikeIn o ma ionC i e ionby hesamplesize
isno no el.Fo ins ance,Has iee al.(2009:230–231)de ineAICinanon-canonicalmanne ,
employingNas hedenomina o in he o mula.While hisde ia ion om hecon en ional
AIC o mulaisno wi hou i sc i ics,i  emainsa p e alen app oach,as exempli iedbyi s
inclusionin hes a is icalso wa epackageS a a.Fo ins ance,S a a epo s“AICdi idedbyN”
ini smodelou pu ,ase idencedby a iousexamplesa ailableonline(G a i udeisex ended o
I.S e unko  o p o iding hisin o ma ion).

CHARITONCHARITONIDIS
278 Ac aLinguis icaLi huanicaXC
andcompa e hemodels’goodness-o - i .Theinse iono keyp edic o sin o
global, i.e.gene al,modelsshowed ha  heBLPlexicaldecision imescalled
o abe e goodness-o - i  han heELPlexicaldecision imes.The i o  he
ELPnamingmodels ellwi hin he angeo  hoseobse ed o  heELPand
BLPlexicaldecisionmodels.Mos no ably,con ex conc e eness o he second
cons i uen eme gedasasigni ican p edic o inallmodelswi hSUBTLEX-US
equency,ac osslexicaldecisionandnaming.
InCha i onidis(2024),allsigni ican coe icien s om heglobalmodelswi h
SUBTLEX-US equencywe ejux aposed o hehyponymy a iable(Gagné
e al.2020).I was ound ha modelsincludingbo hhyponymyandcon ex 
conc e eness o  hesecondcons i uen we ealwaysassocia edwi h helowes 
(=bes )ScaledAIC alueascompa ed ones ed, i.e. educed,modelsomi ing
ei he o  hese wo a iables.Thesubsequen lyappliedWald es sshowed ha 
nes edmodels,always e e ed oasigni ican  educ ion(=de e io a ion)o  he
coe icien o de e mina ion(R2).Tables1and2display heScaledAIC alues
and he esul so  heco espondingWald es s, espec i ely.
TABLE1. ScaledAIC alues o nes edmodelsomi inghyponymy(‘Model2’)
o con ex conc e eness o  hesecondcons i uen (‘Model3’) om ull
models(‘Model1’) op edic EnglishLexiconP ojec (ELP)lexical
decision(LD) imes,B i ishLexiconP ojec (BLP)lexicaldecision
imes,andELPnaming imes
Model Scaled AIC AIC N
ELPLD
1 -3.36281a-3557.85 1058
2-3.30375 -4169.334 1262
3-3.34845 -3700.038 1105
BLPLD
1 -3.79552a-2903.574 765
2-3.76618 -3920.592 1041
3-3.76718 -2987.37 793
ELPnaming
1 -3.58686b-7396.108 2062
2-3.54304 -8418.27 2376
3-3.58379 -7389.784 2062
 S aipsniai / A icles 279
Explo ing Scaled AIC wi hin English Closed Compounds
a. P edic o s: (Cons an ), hyponymy judgemen , leng h o  compound,
SUBTLEX-US equency, ep esen a ion alence(cmp),con ex conc e eness
(c2)
b. P edic o s: (Cons an ), hyponymy judgemen , leng h o  compound,
SUBTLEX-US equency,con ex  alence(cmp),con ex a ousal(c1),con ex 
a ousal(c2),con ex conc e eness(c2)
TABLE2. Wald es s o nes edmodelsomi inghyponymy(‘Model2’)o con ex 
conc e eness o  hesecondcons i uen (‘Model3’) om ullmodels
(‘Model1’) op edic EnglishLexiconP ojec (ELP)lexicaldecision
(LD) imes,B i ishLexiconP ojec (BLP)lexicaldecision imes,and
ELPnaming imes
Model R2 squa e F change d 1 d 2 p
ELPLD
1 .184a47.506 5 1052 .000
2-.004 4.562 1 1052 .033
3-.010 13.175 1 1052 .000
BLPLD
1 .211a40.479 5 759 .000
2-.013 12.091 1 759 .001
3-.015 14.007 1 759 .000
ELPnaming
1 .288b118.711 72054 .000
2-.003 8.286 1 2054 .004
3-.003 8.309 1 2054 .004
a. P edic o s:(Cons an ),hyponymyjudgemen ,leng ho compound,
SUBTLEX-US equency, ep esen a ion alence(cmp),con ex conc e eness
(c2)
b. P edic o s: (Cons an ), hyponymy judgemen , leng h o  compound,
SUBTLEX-US equency,con ex  alence(cmp),con ex a ousal(c1),con ex 
a ousal(c2),con ex conc e eness(c2)
Inconclusion, wodi e en e ec -sizemeasu es,namelyScaledAICand
R2,hie a chized hesame eg essionmodelsiden icallywhiledemons a ing
hesamep e e ence o  hebes model.Thus, he eiss onge idence ha  he
ScaledAICmeasu eisaquali a i e ool o assessingmodel i .
CHARITONCHARITONIDIS
280 Ac aLinguis icaLi huanicaXC
4. THEPRESENTSTUDY
Thep esen s udybuildsupon heau ho ’sp e ious esea chp esen edin
sec ion3.The esea chsubjec sa e66lexicaldecisionandnamingmodels o 
heEnglishclosed(conca ena ed)compoundsbuil byGagnée al.(2019).All
modelsincludeSUBTLEX-US equencyascon ol a iable.Ou objec i es
a e wo oldand uninpa allel.Fi s ,weassess hecha ac e is icso Gagnée 
al.’s(ibid.)models.Second,weexplo eessen ialp ope ieso  heScaledAIC
measu e.
The esea chques ionsa e:
1.IsScaledAICsensi i e o hemodeldesigninGagnée al.(2019)?Which
modelg oupsa e a ou ed?
2.Wha is heimpac o  hecon ol a iables‘compound equency’and
‘compoundleng h’onScaledAIC?
3.Howismo phological anspa ency ela ed oScaledAIC?
Ou s udyiss uc u edas ollows:Sec ion5p o idesano e iewo ou 
me hods.Sec ion6.1p o idesdesc ip i es a is ics o ScaledAIC e e ing
o hemodelsunde conside a ion.Emphasisisgi en o hepa ame ic e sus
non-pa ame iccha ac e is ics o modelca ego ies.Sec ion6.2explo es he
ela ionshipbe ween hesou ceo  esponse imesand helexicalp ocessing
asks.Sec ion6.3jux aposesScaledAIC o hecon ol a iables‘compound
equency’and‘compoundleng h’.Insec ion6.4 hesigni icancele elso  he
anspa encycoe icien s omGagnée al.’s(2019)modelsa emappedon o
heScaledAIC alues.Thekey indingsa esumma izedinsec ion7, ollowed
byadiscussiono  he esul sinsec ion8.
5. METHODS
Ou gene alme hodwas hecompa a i eanalysiso  hemainpa ame e sand
cha ac e is icso Gagnée al.’s(2019)models,usingScaledAICas hedependen 
a iable. Independen  a iables included sample cha ac e is ics (e.g. esponse
imesou ceand helexicalp ocessing asks),s udydesign(e.g.con ol a iables),
and hesigni icancele elo  anspa encycoe icien s,amongo he  ac o s.
Thespeci ic s a is icalme hodsemployed we eas ollows: (a)desc ip i e
s a is icspe aining omeansandmedians,alongwi h heapplica iono  he
Shapi o-Wilk es  oassess hecen al endency, a iabili y,anddis ibu ion
o ScaledAICac ossdi e en modelca ego iesandg oups(sec ions6.1and
6.2), (b) main e ec s analyses conduc ed o  he sou ce o  esponse imes
(ELP/BLP)and helexicalp ocessing asks(lexicaldecision/naming)(sec ion
 S aipsniai / A icles 281
Explo ing Scaled AIC wi hin English Closed Compounds
6.2),(c)dis inc ANOVAspe o medon esponse imesou ceand helexical
p ocessing asks,inco po a ingcompoundleng hasaco a ia e(sec ion6.3),
and (d) u iliza ion o  he K uskal-Wallis and he Jonckhee e-Te ps a es s
o explo e di e ences among he anks o  o dinally- ecodedcoe icien s o 
seman ic anspa ency(sec ion6.4).Fo mo ein o ma iononme hods, he
eade is e e ed o heanalysesinsec ions6.1–6.4.
6. ANALYSES
6.1. ScaledAIC s.modelca ego ies
The66AIC alues omGagnée al.’s(2019)mul iple- eg essionmodels
wi h SUBTLEX-US equency as a con ol a iable we e di ided by each
model’ssamplesize oyieldase o 66ScaledAIC alues.
Table3belowp o ides hedesc ip i es a is ics o ScaledAICandFigu e
2displays heco espondingboxplo  e e ing o heo de edse o  alues.10
The ewe enoou lie sin hesample.Theskewness(Sk)andku osis(Ku)
alueswe e ole able.11
The mean alue o  Scaled AIC was -3.50030. The s anda d de ia ion
was0.19052, ha is heobse a ionswe e ela i ely igh lyclus e eda ound
hemean.Theminimumandmaximum alueswe e-3.85502and-3.15754,
espec i ely.Themedian aluewas-3.55732,i.e.sligh lylowe  han hemean
alue.12Themiddle50%o  heda a angedbe ween-3.66575( i s qua ile,
Q1) and -3.30959 ( hi d qua ile, Q3). Acco dingly, he in e qua ile ange
(IQR)was0.35616.
 10 Thelowe o  i s qua ilelineo  hebox(Q1)ma ks hebounda ybelowwhich hebo om
25%o  heda aex ends.Simila ly, heuppe o  hi d qua ilelineo  hebox(Q3)ma ks he
bounda yabo ewhich heuppe 25%o  heda aex ends.Theshadeda eashows hebounda ies
o  hemiddle50%o  heda ao in e qua ile ange(IQR),whichcanbecompu edbysub ac ing
he i s qua ile om he hi dqua ile(Q3-Q1).Theho izon allineinside heboxshows he
mediano middle qua ile(Q2),i.e. he alue ha  allsin hemiddleo  heda ase .
 11 Wi h e e ence o he SPSS en i onmen , he alues be ween -1 and +1 o  skewness and
be ween -2 and +2 o  ku osis a e gene ally conside ed accep able o  no mal dis ibu ion
assessmen . I  is wo h no ing, howe e , ha  skewness and ku osis alone do no  p o ide a
conclusi e p oo  o  no mali y (see also he discussion on h ps://www. esea chga e.ne /pos /
Wha _is_ he_accep able_ ange_o _skewness_and_ku osis_ o _no mal_dis ibu ion_o _da a).
 12 Inlinewi h hispa e n, he ewasasmallamoun o posi i eskewin heda a(0.494).
CHARITONCHARITONIDIS
288 Ac aLinguis icaLi huanicaXC
Tode ec  hein luenceo compoundleng handcompound equencyon
ScaledAIC, he espec i e eg essioncoe icien swe e ecodedin oo dinal
alues acco ding o hei  posi i i y and signi icance le el, see Table 6. The
signi icancele elswe emappedon oo dinalscalesbecause hey ep esen ed
con en ionalcu -o poin sbasedon heexac signi icance alues.
TABLE6.O dinal ecodingcha  o  eg essioncoe icien s
Signi icance Posi i i y O dinal alues Desc ip ion
p<.001 nega i e -3 la genega i ee ec
p<.01 nega i e -2 mode a enega i ee ec
p<.05 nega i e -1 smallnega i ee ec
p>.05 nega i e/posi i e 0 non-signi ican e ec
p<.05 posi i e 1 smallposi i ee ec
p<.01 posi i e 2mode a eposi i ee ec
p<.001 posi i e 3la geposi i ee ec
In Gagné e  al.’s (2019) models, SUBTLEX-US equency was always
associa edwi hnega i e(=la ency- educing)coe icien swi hala gee ec ,p<
.001.Acco dingly,allcoe icien swe e ecodedas-3,a alue ha waspe ec ly
collinea wi h heou come a iable,ScaledAIC.Fo  his eason,SUBTLEX-US
equencywasexcluded om hep esen analysis.
As o compound leng h,allsigni ican  eg essioncoe icien s omGagnée 
al.’s(2019)modelshadala geposi i e(=la ency-inducing)e ec ,p<.001.
Incon as  o heSUBTLEX-US a iable,se e alnon-signi ican coe icien s
showedup.Gi en hesepa e ns,aca ego ical a iablewasc ea edwi h he
alues‘1’ o posi i ee ec (=in e e enceo compoundleng h)and‘0’ o no
e ec (=noin e e enceo compoundleng h).The esul ingsamplecon ained
39ScaledAIC alues.ThePea sonco ela ion es be weencompoundleng h
andScaledAICyieldedahighlysigni ican co ela ioncoe icien o 0.51,p=
.001,indica ingamode a e- o-s ongco ela ionbe ween he wo a iables.
Compoundleng hwasincludedasasingleindependen  a iableinalinea 
eg ession model. I  was ound ha  he p edic ed Scaled AIC mean o  no
in e e enceo compoundleng hwas3.611(= hein e cep ).Thein e e ence
o compoundleng h esul edinahighe (=in e io ) alueo -3.407(b=0.204,
p=.001).Figu e6belowillus a es hesepa e ns.Inanu shell,ScaledAIC
de e io a eswhencompoundleng hbecomes ele an wi hinmodels.

 S aipsniai / A icles 289
Explo ing Scaled AIC wi hin English Closed Compounds
FIGURE6.Compoundleng handScaledAIC
Weconduc edsepa a eANOVAs o  esponse imesou ce(ELP/BLP)and
ask pe o mance (lexical decision/naming), aking in o accoun  compound
leng h as a co a ia e. The p ima y objec i e was o iden i y dispa i ies in
means ha we ep e iouslyadjus ed oaccommoda e hecon ollinge ec so 
compoundleng h.
(a)ANOVA o  esponse imesou ce.BLPwasassigned he alue‘0’and
ELPwasassigned he alue‘1’.ThePea sonco ela ion es  e ealedas ong
collinea i ybe ween esponse imesou ceandcompoundleng h, =1(N=39).
The ollowinge idencesuppo sou  inding:Fi s , heELPg oupconsis en ly
showed signi ican  posi i e co ela ions wi h compoundleng h,indica inga
la ge e ec . Second, he BLP g oup consis en ly displayed non-signi ican 
co ela ionswi hcompoundleng h.17Consequen ly, hep edic edScaledAIC
mean o  esponse imesou cewas hesamewi ho wi hou compoundleng h
in heanalysis(M=3.407inbo hcases).Insumma y,compoundleng hdid
no ha easigni ican e ec onScaledAICwhen he esponse imesou cewas
included.
(b)ANOVA o  askpe o mance.Namingwasassigned he alue‘0’and
lexicaldecisionwasassigned he alue‘1’.ThePea sonco ela ion es  e ealed
anega i eco ela ionbe ween askandcompoundleng h,indica ingamode a e
e ec , =-.5,p=.001(N=39).This esul sugges s ha compoundleng his
mo e ele an  onaming han olexicaldecision.
 17 I shouldbeno ed ha 11o  he13BLPcoe icien swe enega i e.
CHARITONCHARITONIDIS
290 Ac aLinguis icaLi huanicaXC
When conside ing ask as he p ima y a iable, compound leng h was a
signi ican p edic o o ScaledAIC,F(1,145.030),p=.000.Simila ly,when
conside ingcompoundleng has hep ima y a iable, askwasasigni ican 
p edic o o ScaledAIC,F(1,125.30),p=.000.Thep edic edScaledAICmean
o  askalonewassigni ican lydi e en  om hep edic edScaledAICmean
whencompoundleng hwas akenin oaccoun (3.584 s.3.203, espec i ely).
Likewise, hep edic edScaledAICmean o compoundleng halone(3.584)
wassigni ican lydi e en  om hep edic edScaledAICmeanwhen he ask
was akenin oaccoun (-3.23).Summa izing,in e mso co a ia eadjus men ,
bo h helexicaldecision askandcompoundleng hp edic edin e io models.
6.4. ScaledAIC s. anspa encyno ms
This sec ion in es iga es he e ec  o  eg essioncoe icien s o seman ic
anspa encyinGagnée al.(2019)modelsonScaledAIC.These eg ession
coe icien swe ecodedon h eeo dinalscales,eachco esponding ooneo 
he h eemo phologicalle els,i.e.compound, i s cons i uen ,andsecond
cons i uen .Theo dinal ecodingcha canbe oundinTable6.
Table7belowdisplays hemediansand angeso  heo dinally- ans o med
anspa ency coe icien s o  all h ee mo phological le els. Τhe medians
p o ideuse ulin o ma ionabou  hecen al endencyanddispe siono o dinal
aluesandcanhelpin o manalysesbasedono dinal a iables.
TABLE7.O dinally- ans o med anspa encycoe icien s:Mediansand anges
Median Minimum Maximum
Compound -3 -3 -1
Fi s cons i uen 2-3 3
Secondcons i uen 0 -2 3
N=18
Ascanbeseen, hemedian o  hecompoundwas‘-3’, hemedian o  he
i s cons i uen was‘2’,and hemedian o  hesecondcons i uen was‘0’.
These indingssugges  ha inGagnée al.’s(2019)modelswi hSUBTLEX-US
equency, anspa ency o  he compound was associa ed wi h a la ge
nega i ee ec (sho e  esponse imes), anspa ency o  he i s cons i uen 
was associa ed wi h a mode a e posi i e e ec  (longe  esponse imes), and
anspa ency o  hesecondcons i uen didno ha easigni ican e ec o had
 S aipsniai / A icles 291
Explo ing Scaled AIC wi hin English Closed Compounds
anunce ain ole.Thehighe  anspa ency a ings o  hesecondcons i uen ,
epo edbyGagnée al.(ibid.),sugges aninhe en bias a ou ingi ,leading o
heo e all anspa encyo  hecompoundbeingdependen on he anspa ency
o  he i s cons i uen .In hiscon ex , heposi i e,la ency-inducing,median
o  he i s cons i uen indica esi smedia ing,pe haps e e ence-es ablishing,
olein his ela ionship(seealsosec ion1).I  emains o be demons a ed
whicha e heseman ic unc ions ha su icien ly ep esen ,inp ocessing e ms,
heinhe en biaso  hesecondcons i uen .18
The esea chques ion obeadd essednowiswhe he  heposi i i yand
signi icance le el o  anspa ency coe icien s in luence Scaled AIC. Ou 
me hodp ima ilyaimsa de ec ingo e i inge ec s.AsDanielJ.Na a o
&JayI.Myung(2005)a gue,o e i ingoccu swhen“acomplexmodelwi h
manypa ame e sandhighlynonlinea  o mcano en i da abe e  hana
simple model wi h ew pa ame e s e en i  he la e  gene a ed he da a”
(Na a o,Myung2005:1240).Acco dingly,ala genumbe o pa ame e sha e
hepo en ial ocap u enoiseo uniquecha ac e is icso  hea ailableda abu 
mayhinde  hemodel’sabili y ogene alize o new da a.AICmi iga es heissue
o o e i ingbyin oducingapenal yon heinclusiono nume ouspa ame e s
inamodel,see he‘+2k’pa o  heAICequa ioninsec ion2.
Rega ding heanalysis o ollow,i ispos ula ed ha modelsexhibi inghighe 
(=in e io )ScaledAIC aluesmaypossesssigni ican ,sys ema icallyde i ed,
coe icien s,i.e.coe icien s ha a e ele an acco ding o heLADECda ase 
alone.In hiscon ex ,ou conjec u esugges s ha acon as ing endmigh 
eme gein heconnec ionbe ween anspa encyandScaledAIC,ascompa ed
o heindica ionp o idedby hemediansinTable7.
In pa icula , lowe  (=be e ) Scaled AIC alues may be associa ed wi h
(a) posi i e coe icien s (longe  esponse imes) conce ning he whole
compound,(b)nega i ecoe icien s(sho e  esponse imes)conce ning he
i s  cons i uen , and (c) posi i e o  nega i e coe icien s (longe  o  sho e 
esponse imes, espec i ely)conce ning hesecondcons i uen .I shouldbe
no ed ha , ega ding hesecondcons i uen , hemedianinTable7sugges sno
e ec .
To answe  he esea ch ques ion, wo non-pa ame ic measu es will be
employed,i.e. heK uskal-Wallis es and heJonckhee e-Te ps a es .The
K uskal-Wallis es ,alsoknownas he‘H es ’,isanon-pa ame ic es basedon
 18 InCha i onidis(2024)i isa gued ha bo hhyponymyandcon ex conc e eness o  hesecond
cons i uen  a e signi ican  seman ic p edic o s in lexical decision and naming. The analysis
p esen ed he einshows ha includingbo ho  hesep edic o s esul sinanimp o emen in
ScaledAICandR2ascompa ed omodels ha omi ei he o  hese a iables.
CHARITONCHARITONIDIS
292 Ac aLinguis icaLi huanicaXC
hechi-squa edis ibu ion.I  equi es ha  hedependen  a iablebeo dinal
o con inuous.This es isdesigned ode e minewhe he  he ea esigni ican 
di e ences be ween he medians o  wo o  mo e g oups and is used as an
al e na i e oone-wayANOVA.Conce ning hep ocedu e, he alueso  he
con inuousdependen  a iable,i.e.ScaledAIC,we eo de ed omlowes  o
highes and hesco eswe eassigned anks.The esul ing ankswe een e ed
backin o heg oupso signi icancele el( heindependen  a iable)and he
anks o eachg oupwe esummed.The o mula o calcula ing‘H’in ol ed,
amongo he s,squa ing hesumo  anks o eachg oupand hendi iding his
aluebysamplesize.19Tables8–10con ain heinpu da aconside edand he
sumo  anks o eachg oup.20
TABLE8.Compound
Signi icancele els N Sumo Ranks
Scaled
AIC
1–smallnega i ee ec 1 2
2–mode a enega i ee ec 5 46
3–la genega i ee ec 12 123
To al 18
TABLE9.Fi s cons i uen
Signi icancele els N Sumo Ranks
Scaled
AIC
1–la geposi i ee ec 6 59
2–mode a eposi i ee ec 4 34
3–smallposi i ee ec 1 3
4–noe ec 1 2
5–smallnega i ee ec 1 10
6–mode a enega i ee ec 1 11
7–la genega i ee ec 452
To al 18
 19 Fo  he es o calcula ionsseeField(2009:561–562).
 20 Inall h ee ables, he o alsumo  anksisapp oxima ely171.I isequal o hesumo  hein ege s
om1 o18,seesamplesize(N).
 S aipsniai / A icles 293
Explo ing Scaled AIC wi hin English Closed Compounds
TABLE10.Secondcons i uen
Signi icancele els N Sumo Ranks
Scaled
AIC
1–la geposi i ee ec 6 59
2–smallposi i ee ec 1 14
3–noe ec 8 59
4–smallnega i ee ec 1 18
5–mode a enega i ee ec 221
To al 18
Be o edel ingin o he esul so  heK uskal-Wallis(H) es ,i isimpo an 
ono e ha  his es doesno p o idein o ma ionabou  hespeci icdi e ences
be weenindi idualg oups.Toadd ess hisissue, heJonckhee e-Te ps a(JT)
es wasaddi ionallyemployed.This es p o idedin o ma ionabou whe he 
hemedianso  heg oupsinc eased o  dec eased in he o de speci iedby
hecoding(=g ouping) a iable,speci ically omla geposi i ee ec  ola ge
nega i ee ec .Rega dingme hods, heJTs a is icwascon e edin oaz-sco e.
Aposi i ez- alueindica eda endo ascendingmedians, ha is hemedians
inc eased(=highe /in e io ScaledAIC)as he alueso  hecoding a iable
inc eased.Anega i ez- alueindica eda endo descendingmedians, ha is
hemediansdec eased(=lowe /be e ScaledAIC)as he alueso  hecoding
a iableinc eased.In he ollowing, he esul so  heK uskal-Wallis(H)and
Jonckhee e-Te ps a(JT) es sa egi enjoin ly.
(a)ScaledAIC o  hecompoundwasno signi ican lya ec edbysigni icance
le el,asde e minedby heK uskal-Wallis es (H(2)=2.226,p>.05).A end
o ascendingmedianswas oundcon i mingou o e i inghypo hesis,see he
nega i emedian o  hecompoundinTable7.This end,howe e ,wasno 
s a is icallysigni ican acco ding o heJonckhee e-Te ps a es (JT=50,z=
1.064,p>.05).
(b) Scaled AIC o  he i s  cons i uen  was no  signi ican ly a ec ed by
signi icancele el,asde e minedby heK uskal-Wallis es (H(6)=5.427,
p>.05).A endo ascendingmedianswas ound ejec ingou o e i ing
hypo hesis,see heposi i emedian o  he i s cons i uen inTable7.This
end,howe e ,wasno s a is icallysigni ican acco ding o heJonckhee e-
Te ps a es (JT=70,z=0.549,p>.05).
(c)ScaledAIC o  hesecondcons i uen wasno signi ican lya ec edby
signi icancele el,asde e minedby heK uskal-Wallis es (H(4)=4.607,p>
.05).A endo ascendingo descendingmedianswasno obse ed, ejec ing

CHARITONCHARITONIDIS
294 Ac aLinguis icaLi huanicaXC
ou  o e i ing hypo hesis. In pa icula , he z-s a is ic o  he Jonckhee e-
Te ps a es wasessen iallyze o,inacco dancewi h heze omedian o  he
secondcons i uen inTable7(JT=54,z=0.041,p>.05).
Summa izing,i canbein e ed ha  hesigni icancele elo  he anspa ency
coe icien sinGagnée al.’s(2019)modelswi hSUBTLEX-US equencydoes
no a ec  hemagni udeo ScaledAIC.This indingindi ec lysuppo s he
quali yo Gagnée al.’s(2019)modelswi h anspa encyp edic o s,speci ically
indica ing ha  heo e i inghypo hesis o  hesemodelsis no  enable. A
limi a iono  hep esen s udyis hesmallsamplesizeused,wi hN=18.To
con i mou  indings,mo e esea chisneededusingawide  angeo Scaled
AIC alues.
Toensu ecla i yandcomple enessinp esen ingou  esea chou comes,we
ha einco po a edadedica edsec ion ocusedonsumma izing hekey indings
o ou s udy.Fo  hiscomp ehensi eo e iew,pleasecon inue oSec ion7.
7. KEYFINDINGS
Table 11 below p esen s a comp ehensi e analysis o  model pe o mance
and ele an  a iables in he con ex  o  lexical decision and naming asks,
basedon he indingso Gagnée al.(2019).Eachsec iono  he abledel es
in ospeci icsubjec s, e ealingwhichmodelsa emos e ec i e.TheANOVA
and heK uskal-Wallis/Jonckhee e-Te ps a es s(sec ions6.3and6.4)we e
applieda e assigningnominal(o dinalo ca ego ical) alues o he eg ession
coe icien s omGagnée al’s(2019)models.Fo de ailson hespecial es s
applied,please e e  o he espec i esec ions.
TABLE11. ScaledAICwi hinEnglishclosedcompounds:Comp ehensi eanalysis
o modelpe o manceinlexicaldecisionandnaming asks(Gagnée al.
2019)
Subjec s S a is ics E alua ion Sec ion
Modelca ego ies Desc ip i es
No mali y es s
ELPlexicaldecision
BLPlexicaldecision
ELPnaming
NPAR/~
PAR/✓
PAR/✓
6.1
Response ime
sou ce
Lexicalp ocessing
ask
Maine ec s ELPlexicaldecision
BLPlexicaldecision
ELPnaming
~
✓
✓6.2
 S aipsniai / A icles 295
Explo ing Scaled AIC wi hin English Closed Compounds
Subjec s S a is ics E alua ion Sec ion
Con ol a iables ANOVA Compound
equency
Compoundleng h
✓
~
6.3
Seman ic
anspa ency
K uskal-Wallis
Jonckhee e-
Te ps a
Fi s cons i uen
Secondcons i uen
Compound
NOF/ns
NOF/ns
NOF/ns
6.4
PAR:pa ame icda a|NPAR:non-pa ame icda a| ✓:be e models
(lowe AIC)~:in e io models(highe AIC)|NOF/ns:noo e i ing/non-
signi ican  es
8. DISCUSSION
P e ious esea ch by Cha i onidis (2022, 2024) has demons a ed ha 
ScaledAICisa eliablegoodness-o - i measu e ha canbeemployedinmodel
selec ion,pe hapsincoope a ionwi ho he measu essuchas heWald es (see
sec ion3).Wi h e e ence oGagnée al.’s(2019)mul iple- eg essionmodels
wi h SUBTLEX-US equency, he p esen  analysis in oduced addi ional
p ope ieso  heScaledAICmeasu e.While alidconce nsha ebeen aised
ega ding he compa ison o  models i ed on di e en  sample sizes using
in o ma ionc i e ia(seesec ion2), he indingso  hiss udysugges  ha in
ce aincon ex s,ScaledAICcanindeedbea aluable ool o assessingmodel
i andhie a chizing eg essionmodels.Ou  esea chhasdemons a ed ha 
ScaledAICis esponsi e oexpe imen aldesign, esponse imesou ces,and
speci ic asks. Howe e , i  is essen ial o ecognize ha  he applicabili y o 
ScaledAICmaybecon ex -dependen ,andi su ili yshouldbee alua edon
acase-by-casebasis.
Be o ep oceeding o hep ima y indingso  hispape ,i isimpo an  o
add ess he esea chques ionsse upinsec ion4.
1. The dis ibu ions o  Scaled AIC alues, along wi h combina ions o 
di e en sou ceso  esponse imesandp ocessing asks,sugges  ha Scaled
AICe ec i elyiden i ies hep esenceo absenceo well-de inedunde lying
ac o sinexpe imen aldesignands a is icalmodelling.In hiscon ex ,BLP
lexicaldecisionandELPnamingexhibi eds onge p edic i epowe  o Scaled
AICe enunde con olledcondi ions.
2.Compound equencywasunexcep ionallyanega i ep edic o o Scaled
AIC,alwaysindica ingala gee ec .ELPlexicaldecisionconsis en lyshowed
CHARITONCHARITONIDIS
296 Ac aLinguis icaLi huanicaXC
signi ican  posi i e co ela ions wi h compound leng h p edic ing highe 
(=in e io )ScaledAIC alues.BLPlexicaldecisionconsis en lyshowednon-
signi ican  co ela ions wi h compound leng h. Bo h lexical decision and
compoundleng hp edic edin e io modelsinco a ia eadjus men .
3.Theposi i i yand hesigni icancele elo  anspa encycoe icien sin
Gagnée al.’s(2019)modelsdidno a ec  hemagni udeo ScaledAIC.This
indingimplies ha Gagnée al.’s(2019)modelswi h anspa encyp edic o s
dono in oduceo e i ingbias.
By e e encingspeci icsec ionso  heanalyses, hep ima y indingso  his
s udycanbesumma izedas ollows:
InSec ion6.1,ou analysis ocusedon hecompa isonbe weenScaledAIC
aluesac oss a iousmodelca ego ies.E ena e a emp ing he ans o ma ions
‘na u alloga i hm’and‘squa e oo ’on heabsolu e alues, heo e allScaled
AIC sample did no  con o m o a no mal dis ibu ion. Simila ly, he ELP
lexicaldecisionmodelsshowcasedanon-pa ame icdis ibu iono  hei Scaled
AIC alues.On hecon a y, heda a ela ed oBLPlexicaldecisionandELP
naming ollowedano maldis ibu ionpa e n.
In Sec ion 6.2, ou  ocus shi ed o examining he ela ionship be ween
ScaledAICand(a) hesou ceso  esponse imesand(b) askpe o mance.
In e es ingly, he angeso ScaledAIC alues o  heELPandBLPlexical
decision models did no  o e lap, signi ying hei  dis inc i eness. The es 
esul s e ealedsigni ican maine ec so bo h esponse imesou ceand ask
pe o manceonScaledAIC.No ably, hep edic i ecapabili yo ScaledAIC
wasbe e  o modelsassocia edwi h heBLPlexicaldecision imesand he
naming ask. These indingscon ibu e o he p ecisionande icacy o  he
espec i emodelssigni ican ly.
InSec ion6.3,ou explo a iondel edin o he ela ionshipbe weenScaled
AICand hecon ol a iables‘compound equency’and‘compoundleng h’.
Compound equencywasexcluded om heanalysisbecausei waspe ec ly
collinea wi hScaledAIC.On heo he hand,adeclineinScaledAIC alues
wasobse edwhencompoundleng hbecamea ele an  ac o wi hinmodels.
Concomi an ly,compoundleng hwasmos  ele an  o  henaming ask.
In e mso co a ia eadjus men ,bo h helexicaldecision askandcompound
leng hwe ep edic i eo in e io models.
InSec ion6.4,ou  ocuswasplacedon he ela ionshipbe weenScaledAIC
andseman ic anspa ency.The esea chques ionwaswhe he  heposi i i y
and hesigni icancele elo  anspa encycoe icien sinGagnée al.’s(2019)
modelshadanimpac onScaledAIC.Thep ima ygoalo ou me hodwas
ode ec po en ialo e i inge ec s.Wepos ula ed ha modelswi hin e io 
ScaledAIC aluesmigh possesssigni ican coe icien s ha hold ele ance
 S aipsniai / A icles 297
Explo ing Scaled AIC wi hin English Closed Compounds
acco ding o heLADECda ase alone.Whileweobse eda endo inc easing
(=in e io )ScaledAIC alues o aclus e o signi ican nega i ecoe icien sa 
hecompoundle el–aligningwi hou o e i inghypo hesis– heJonckhee e-
Te ps a es showed ha  his enddidno achie es a is icalsigni icance.
Inconclusion, heexplo a iono di e en pa ame e susingScaledAICasa
dependen  a iablehasillumina ed hedi e sewaysinwhichmodelca ego ies,
esponse ime sou ce, p ocessing asks, con ol a iables, and seman ic
anspa encyimpac  hegoodness-o - i o models.By ecognizing henuanced
ela ionshipsamong heseelemen s, esea che sa ebe e equipped omake
in o meddecisionsinmodelselec ion,adjus men s,andin e p e a ion.
REFERENCES
Akaike Hi o ugu 1973:In o ma ionTheo yandanEx ensiono  heMaximum
Likelihood P inciple. – Second In e na ional Symposium on In o ma ion Theo y, eds.
B.F.Csaki,B.N.Pe o ,Budapes :AcademiaiKiado,267–281.
Ald ichJohnR.1997:R.A.Fishe and heMakingo MaximumLikelihood1912–
1922.–S a is ical Science12(3),162–176.
Baayen Ha ald R., Piepenb ock Richa d, Gulike s Leon 1995: The
CELEX Lexical Da abase(Da ase ,Release2,CD-ROM),Linguis icDa aConso ium,
Uni e si yo Pennsyl ania.A ailablea :h ps://ca alog.ldc.upenn.edu.
Balo a Da id A., Yap Mel in J., Co ese Michael J., Hu chison
Kei hI.,Kessle B e ,Lo isBjo n,NeelyJamesH.,NelsonDouglas
L.,SimpsonG egB.,T eimanRebecca2007:TheEnglishLexiconP ojec .–
Beha io Resea ch Me hods39,445–459.
B ysbae  Ma c, New Bo is 2009: Mo ing Beyond Kuče a and F ancis: A
C i icalE alua iono Cu en Wo dF equencyNo msand heIn oduc iono aNew
andImp o edWo dF equencyMeasu e o Ame icanEnglish.–Beha io Resea ch
Me hods41,977–990.DOI:doi.o g/10.3758/BRM.41.4.977.
Bu nhamKenne hP.,Ande sonDa idR.2002:Model Selec ion and Mul imodal
In e ence: A P ac ical In o ma ion-Theo e ic App oach, 2ndedi ion,NewYo k:Sp inge .
DOI:dx.doi.o g/10.1007/b97636.
Cha i onidis Cha i on 2022:Con ex Conc e eness o  heSecondCons i uen 
SlowsDownCompound-Wo dP ocessing.–Lexis20.DOI:doi.o g/10.4000/lexis.6769.