Timed picture naming in seven languages Languages.pdf · reto for her generous assistance. Our...

Copyright 2003 Psychonomic Society Inc 344

Psychonomic Bulletin amp Review2003 10 (2) 344-380

Timed picture naming in seven languagesELIZABETH BATES

University of California San Diego La Jolla CaliforniaSIMONA DrsquoAMICO

University of Rome ldquoLa Sapienzardquo Rome Italyand University of Aquila LrsquoAquila Italy

THOMAS JACOBSENUniversity of Leipzig Leipzig Germany

ANNA SZEacuteKELYEotvos Lorant University Budapest Hungary

ELENA ANDONOVANew Bulgarian University Sofia Bulgaria

ANTONELLA DEVESCOVIUniversity of Rome ldquoLa Sapienzardquo Rome Italy

DAN HERRONUniversity of California San Diego La Jolla California

CHING CHING LUNational Yang Ming University Taipei Taiwan

and National Hsinchu Teachers College Hsinchu TaiwanTHOMAS PECHMANN

University of Leipzig Leipzig GermanyCSABA PLEacuteH

Lorant Eotvos University Budapest HungaryNICOLE WICHA and KARA FEDERMEIER

University of California San Diego La Jolla CaliforniaIRINI GERDJIKOVA

New Bulgarian University Sofia BulgariaGABRIEL GUTIERREZ

University of California San Diego La Jolla CaliforniaDAISY HUNG

National Yang Ming University Taipei Hsinchu TaiwanJEANNE HSU

National Yang Ming University Taipei Taiwanand National Tsing Hua University Hsinchu Taiwan

GOWRI IYERUniversity of California San Diego La Jolla California

KATHERINE KOHNERTUniversity of California San Diego La Jolla Californiaand University of Minnesota Minneapolis Minnesota

TEODORA MEHOTCHEVANew Bulgarian University Sofia Bulgaria

ARACELI OROZCO-FIGUEROAUniversity of California San Diego La Jolla California

andANGELA TZENG and OVID TZENG

National Yang Ming University Taipei Taiwan

Timed picture naming was compared in seven languages that vary along dimensions known to affect lexicalaccess Analyses over items focused on factors that determine cross-language universals and cross-language dis-parities With regard to universals number of alternative names had large effects on reaction time within andacross languages after targetndashname agreement was controlled suggesting inhibitory effects from lexical com-petitors For all the languages word frequency and goodness of depiction had large effects but objective pic-ture complexity did not Effects of word structure variables (length syllable structure compounding and ini-tial frication) varied markedly over languages Strong cross-language correlations were found in naminglatencies frequency and length Other-language frequency effects were observed (eg Chinese frequenciespredicting Spanish reaction times) even after within-language effects were controlled (eg Spanish frequenciespredicting Spanish reaction times) These surprising cross-language correlations challenge widely held as-sumptions about the lexical locus of length and frequency effects suggesting instead that they may (at least inpart) reflect familiarity and accessibility at a conceptual level that is shared over languages

TIMED PICTURE NAMING IN SEVEN LANGUAGES 345

Within and across languages word forms bear little orno resemblance to the concepts that they represent Thesame furry four-legged animal is called dog in EnglishHund in German perro in Spanish and cane in ItalianEven within the same language family (eg Romance)there are often striking differences in names for the sameconcept (eg butterfly is mariposa in Spanish farfallain Italian and papillon in French) Despite these well-known cross-language differences in the shape of wordsit is generally assumed that people access their mentallexicon in the same way in every natural language on thebasis of a universal architecture for word comprehensionand production This belief rests crucially on the as-sumption that the relationship between meaning andform is arbitrary Word forms have no effect on the pro-cess by which speakers move from concept to lexical se-lection and meanings have no effect on the shape ofwords or on the kind of processing required to map a se-lected concept onto its associated sound

In the present study we will investigate some of thesefundamental assumptions by presenting what is (to the bestof our knowledge) the first large-scale cross-linguisticstudy of timed picture naming (cf Bachoud-Levi DupouxCohen amp Mehler 1998) We will investigate universaland language-specific contributions to naming behavioracross seven languages (English German Spanish Ital-

ian Bulgarian Hungarian and Mandarin Chinese) thatvary markedly along lexical and grammatical parametersknown or believed to affect word retrieval This cross-linguistic strategy will shed light on several theoreticalquestions Will we find significant differences across lan-guages in ease of naming andor in the particular items thatare hard or easy to name Will such differences reflectcultural variations in the accessibility of concepts andorlinguistic variations in the properties of target namesWill cross-linguistic differences reflect the linguistic dis-tance that separates our seven languages (eg GermanicRomance and Slavic variants of Indo-European UralicSino-Tibetan) Will we find a universal set of predictorndashoutcome relationships that hold in every language (egeffects of frequency on naming latency) andor will wefind language-specific profiles in the relationship be-tween word structure and naming behavior (eg effectsof syllable structure in Chinese that have no equivalentin English) Finally our results will offer insights intothe levels at which word and picture properties have theireffects Specifically for those word properties that areassumed to apply only at the level of word form andorat the level of lemma selection (Caramazza 1997 Dell1990 Jescheniak amp Levelt 1994) we should find within-language correlations (eg English word frequenciespredict English reaction times [RTs] ) but no corre-sponding cross-language correlations (eg Chineseword frequencies should not predict English RTs) In con-trast if supposed word form effects are confounded by vari-ance from a deeper conceptual level (eg word frequencyis contaminated by conceptual familiarityaccessibility)we may f ind significant cross-language effects evenafter own-language properties are controlled (eg Chi-nese word frequency will predict English RTs even afterEnglish frequencies are controlled)

Timed picture naming was a good candidate for thiscross-linguistic strategy for several reasons Picturenaming is a widely used technique for the study of lexicalaccess (for reviews see Glaser 1992 Johnson Paivio ampClark 1996) and timed picture naming is one of the firstparadigms ever used to study real-time language pro-cessing from early studies by Cattell (1886) through thepioneering work of Wingfield (1967 1968) Lachman andcolleagues (Lachman 1973 Lachman Shaffer amp Henn-rikus 1974) and Snodgrass and colleagues (Sanfeliu ampFernandez 1996 Snodgrass amp Vanderwart 1980 Snod-grass amp Yuditsky 1996) On-line (timed) and off-line(untimed) picture-naming methods are also widely usedin studies of brain-injured patients (for reviews see Chenamp Bates 1998 Druks amp Shallice 2000 Goodglass1993 Murtha Chertkow Beauregard amp Evans 1999) andboth normal and language-impaired children (CycowiczFriedman Rothstein amp Snodgrass 1997 DrsquoAmico De-vescovi amp Bates 2001 Davidoff amp Masterson 1996Dockrell Messer amp George 2001 Nation Marshall ampSnowling 2001) More recently the neural basis of lex-ical retrieval has been explored in normal adults usingeither overt or covert picture naming in functional brainimaging (Damasio et al 2001 Hernandez Dapretto

This multifaceted collaborative project has required great effort bymany people The first four authors (listed in alphabetical order) tookprimary organization responsibility for the cross-language project as awhole including development of the scoring system organization of thedatabase andor statistical analysis and interpretation of cross-languageresults in the final stages of the project The next seven authors (in al-phabetical order) each took primary responsibility for launching andorsupervising data collection within each of their respective languagesites (Devescovi in collaboration with DrsquoAmico for Italian Herron incollaboration with Bates for English Pechmann in collaboration withJacobsen for German and Pleacuteh in collaboration with Szeacutekely for Hun-garian) The remaining authors (in alphabetical order) made importantcontributions at various stages of the project within their own-languagegroup The project as a whole was supported by a grant to EB (NIDCDR01 DC00216 ldquoCross-linguistic studies of aphasiardquo) Developmentaloffshoots of the project have been partially supported by Grants NINDSP5022343 and NIDCD P50 DC01289 The Bulgarian portion of theproject was also supported by a grant from the James McDonnell Foun-dation and support for travel during the final phases of manuscriptpreparation was provided by a grant from NATO (LSTCLG977502ldquoComparative Studies of Language Development and Reading in FourLanguagesrdquo) Each of the associated institutions has provided space fordata collection and personnel time We are grateful to the School of Hu-manities at the Universidad Autoacutenoma de Baja CaliforniandashTijuana forallowing us to use their subject pool and to Lourdes Gavaldoacuten de Bar-reto for her generous assistance Our thanks to Maria Polinsky for ad-vice on the classif ication of the seven languages (as outlined inTable 1) to Meiti Opie and Ayse Saygin for assistance with manuscriptpreparation to Robert Buffington and Larry Juarez for technical assis-tance in development of the norming procedures to Luigi Pizzamiglioand Nina Dronkers for assistance in locating picture stimuli and to Vic-tor Ferreira and David Balota for careful reading of and critical com-ments on the manuscript itself Any flaws and limitations that remainare not the responsibility of these good friends and advisors Corre-spondence concerning this article should be addressed to E Bates Cen-ter for Research in Language 0526 University of California SanDiego La Jolla CA 92093-0526 (e-mail batescrlucsdedu)

346 BATES ET AL

Mazziotta amp Bookheimer 2001) or event-related brainpotentials (Schmitt Muumlnte amp Kutas 2000 van Turen-nout Hagoort amp Brown 1997 1998 1999 Wicha BatesMoreno amp Kutas 2000)

In our own laboratories we have used picture namingfor several years to explore the time course of lexical ac-cess within and across natural languages (Bates et al2000) In some studies pictures have been named out ofcontext in randomized lists by monolingual adults(Bates Burani DrsquoAmico amp Barca 2001 Iyer Sac-cuman Bates amp Wulfeck 2001 Szeacutekely amp Bates 2000Szeacutekely et al 2003 Szeacutekely et al in press) by youngchildren (DrsquoAmico et al 2001) and by SpanishndashEnglishbilinguals across the lifespan (Hernandez et al 2001Hernandez Martinez amp Kohnert 2000 Kohnert 2000Kohnert Bates amp Hernandez 1999 Kohnert Hernan-dez amp Bates 1998) In other studies we have used pic-ture naming to study the effects of phrase and sentencecontext on lexical access in several languages Examplesinclude demonstrations of sentence-level semantic prim-ing in English speakers between 3 and 83 years of age(Roe et al 2000) the effects of grammatical gender onnoun retrieval in Spanish (Wicha et al 2000) Bulgarian(Kokinov amp Andonova unpublished) German (Hillertamp Bates 1996 Jacobsen 1999) and Italian (BentrovatoDevescovi DrsquoAmico amp Bates 1999) the effects of nom-inal classifiers on noun retrieval in Chinese (Lu HungTzeng amp Bates unpublished) and Swahili (Alcock ampNgorosho in press) and the differential effects of syn-tactic cues on retrieval of nouns versus verbs in English(Federmeier amp Bates 1997) and Chinese (Lu et al 2001)

Although the utility of picture naming is beyond disputeit has become increasingly clear to us that cross-linguisticcomparisons using this technique are limited by the ab-sence of comparable naming norms across the languagesof interest To meet this need we launched an internationalproject to obtain timed picture-naming norms across awide range of languages We have now collected object-naming norms (including both naming and latency) for520 black-and-white drawings of common objects in sevendifferent languages American English Spanish ItalianGerman Bulgarian Hungarian and Mandarin ChineseInitial results from this study are presented below but firstit will be useful to consider in more detail some of the the-oretical as well as the practical motivations for this work

Theoretical Basis for Cross-Linguistic Differences

Why would we expect any systematic differences inpicture naming aside from cultural differences of limitedinterest for psycholinguistic theories It is of coursewell known that languages can vary qualitatively in thepresenceabsence of specific linguistic features that arerelevant for lexical access (eg Chinese has lexical toneHungarian has nominal case markers and English hasneither) In addition languages can vary quantitativelyin the shape and magnitude of the lexical phonologicaland grammatical challenges posed by equivalent struc-

tures for real-time processing and learning For examplethe ldquosamerdquo lexical item (translation equivalents namesfor the same pictures) may vary in frequency from onelanguage to another This might occur for a variety ofreasons including cultural variations in the frequency oraccessibility of the concept as well as cross-languagevariations in the availability of alternative names for thesame concept

Holding frequency constant equivalent lexical phono-logical andor grammatical structures can also vary intheir reliability (cue validity) and processibility (cue cost)These two constructs figure prominently in the compe-tition model (Bates Devescovi amp Wulfeck 2001 Batesamp MacWhinney 1989 MacWhinney 1987) a theoreti-cal framework developed explicitly for cross-linguisticresearch on acquisition processing and aphasia Withinthis framework cue validity refers to the informationvalue of a given phonological lexical morphological orsyntactic form within a particular language whereas cuecost refers to the amount and type of processing associ-ated with the activation and deployment of that form(eg perceivability salience neighborhood density vsstructural uniqueness demands on memory or demandson speech planning and articulation)

The seven languages that we have selected for studypresent powerful lexical and grammatical contrasts withimplications for cue validity and cue cost in word re-trieval (summarized in Table 1) Hungarian is a Uraliclanguage (from the Finno-Ugric subclass) and is one ofthe few non-Indo-European languages in Europe Chi-nese is a Sino-Tibetan language with strikingly differentfeatures from all of the other languages in our study(eg lexical tone a very high degree of compoundingand no inflectional paradigms) The other five languagesare Indo-European although they represent differentsubclasses Bulgarian is a Slavic language Italian andSpanish are Romance languages German is the proto-typical Germanic language and English shares historyand synchronic features with both Germanic and Ro-mance languages The seven languages vary along a hostof grammatical and lexical dimensions that are known orbelieved to affect real-time word perception and produc-tion These include degree of word order flexibility (eghow many different orders of subject verb and objectare permitted in each language) and the range of contextsin which subjects andor objects can be omitted Boththese factors influence the predictability of form class(eg the probability that the next word will be a nounadjective or verb) and hence lexical identity These lan-guages also vary in the availability of morphologicalcues to the identity of an upcoming word includingnominal classifiers (in Chinese) and prenominal ele-ments that agree in grammatical gender (in Spanish Ital-ian German and Bulgarian) or case (in German andHungarian) Phonological features that also assist inword retrieval and recognition include lexical tone (in Chi-nese) vowel harmony (in Hungarian) and stress (in allthese languages except Chinese)


Tab

le 1

Cha

ract

eris

tics

of

the

Sev

en P

arti

cip

atin

g L

angu

ages

Lin

guis

tic F

eatu

res

Eng

lish

Ger

man

Ital

ian

amp S

pani

shB

ulga

rian

Hun

gari

anC

hine

se

Indo

-Eur

opea

nye

sye

sye

sye

sno

no

Lan

guag

e fa

mily

Ger

man

ic (

stro

ngG

erm

anic

Rom

ance

Slav

icU

ralic

Sino

-Tib

etan

infl

uenc

e of

Rom

ance

)(F

inno

-Ugr

ic)

Wor

d or

der v

aria

tions

lo

wm

ediu

mhi

ghhi

ghm

ediu

mm

ediu

m

Infl

ectio

nal m

orph

olog

ysp

arse

rich

rich

rich

rich

none

Om

issi

on o

f con

stitu

ents

in f

ree-

stan

ding

sen

tenc

esno

t per

mitt

edno

t per

mitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t and

obj

ect c

anbe

om

itted

Use

of c

ompo

undi

ngdagger

high

high

low

med

ium

med

ium

high

(gt80

o

f all

wor

ds)

Lex

ical

am

bigu

ity fo

r wor

ds o

ut o

f con

text

high

esp

ecia

lly f

orm

oder

ate

esp

ecia

llylo

w fo

r al

l cat

egor

ies

low

for

all c

ateg

orie

slo

w fo

r al

l cat

egor

ies

high

for n

ouns

ver

bs

noun

s an

d ve

rbs

for n

ouns

and

ver

bsdu

e to

infl

ectio

nal

due

to in

flec

tiona

ldu

e to

infl

ectio

nal

and

func

tion

wor

dsm

arki

ngm

arki

ngm

arki

ng

Mor

phol

ogic

al re

gula

rity

one

regu

lar a

ndm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le r

egul

ar

lexi

cal r

egul

arity

onl

ym

ultip

le ir

regu

lar

irre

gula

r an

dir

regu

lar

and

irre

gula

r an

dir

regu

lar

and

degr

ees

offo

rms

for p

lura

l and

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

prod

uctiv

ity in

past

tens

epr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

sco

mpo

und

form

atio

n

Gra

mm

atic

al c

ues

to w

ord

iden

tityDagger

form

cla

ssfo

rm c

lass

gen

der

form

cla

ss g

ende

rfo

rm c

lass

gen

der

form

cla

ss c

ase

form

cla

ss n

omin

alca

secl

assi

fier

s

Pros

odic

cue

s to

wor

d id

entit

yst

ress

stre

ssst

ress

stre

ssst

ress

vow

el h

arm

ony

lexi

cal t

one

Ort

hogr

aphy

and

ort

hogr

aphi

c re

gula

rity

alph

abet

ic h

ighl

yal

phab

etic

som

eal

phab

etic

hig

hly

alph

abet

ic h

ighl

yal

phab

etic

hig

hly

logo

grap

hic

one

op

aque

irre

gula

rir

regu

lari

ties

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

sylla

ble

map

s to

man

y ch

arac

ters

Ref

ers

to t

he n

umbe

r of

dif

fere

nt o

rder

s of

sub

ject

ver

b a

nd o

bjec

t th

at a

re p

ossi

ble

in t

he s

poke

n la

ngua

ge

dagger Ref

ers

to w

ords

that

are

com

pose

d of

oth

er f

ree-

stan

ding

wor

ds (

cont

ent w

ords

and

or

func

tion

wor

ds)

Dagger Am

ong

gram

mat

ical

cue

s to

wor

d id

enti

ty ldquo

form

cla

ss c

uesrdquo

ref

er to

wor

ds o

r ph

rase

s th

at r

elia

bly

dist

ingu

ish

betw

een

noun

s v

erbs

and

oth

er g

ram

mat

ical

cla

sses

as

in th

e di

ffer

-en

ce b

etw

een

ldquoI w

ent t

o th

e da

ncerdquo

ver

sus

ldquoI w

ant t

o da

nce

rdquo S

tudi

es h

ave

show

n th

at s

uch

form

cla

ss c

ues

like

gen

der

case

and

nom

inal

cla

ssif

iers

can

ldquopr

imerdquo

(fa

cilit

ate

or in

hibi

t) r

etri

eval

of

wor

dsfr

om d

iffe

rent

gra

mm

atic

al c

lass

es

348 BATES ET AL

These seven languages also vary in aspects of wordstructure that may influence the speed and accuracy ofword retrieval outside of a phrase or sentence contextThese include large variations in word length from En-glish (where monosyllabic content words are common)to Italian (where monosyllabic content words are rare)The languages also vary extensively in the use of com-pounding In Chinese more than 80 of all contentwords are compounds made up of two or more syllablesthat occur in many other Chinese words At the oppositeend of the spectrum compounding is relatively uncommonin Spanish or Italian with the other languages falling inbetween The languages in our data set differ in the amountof lexical homophony that they permit andor tolerateand in the number of different inflected forms that thelexical target might take Some of these differences arestraightforward and transparent with documented effectsOthers are more subtle and their effects on lexical accessare still unknown

In the present study we will demonstrate many similar-ities across languages in the factors that influence lexicalaccess (eg frequency) due in part to the fact that wordswere accessed out of context in the simplest possible (ci-tation) form Nevertheless some intriguing differenceshave already begun to emerge Under normal listeningconditions small cross-language differences in averageword frequency length or complexity may have little orno effect on naming behavior All languages have evolvedunder intense communicative pressures pushing them toldquobreak evenrdquo in the overall amount of processing requiredHence it is likely that any costs associated with (for ex-ample) a greater difference in length are compensatedfor by advantages in other parts of the system Howeverthe slightly greater demands associated with such cross-linguistic differences may be more evident under adverseprocessing conditionsmdashfor example in brain-injured pa-tients or in young children Hence Language A may ldquobreakdownrdquo before Language B in some situations Some ofthe relatively small but significant effects of word struc-ture that we will demonstrate below fall in that category

Theoretical Basis for Cross-Linguistic UniversalsJohnson et al (1996) have provided a review of picture-

naming studies that assumes on the basis of Paiviorsquos dual-code model (Paivio 1971) a minimum of three universalstages in word production that are tapped by the picture-naming task (1) analysis and recognition of the depictedobject or event (2) retrieval of one or more word formsthat express the recognized object or event in the speakerrsquoslanguage and selection of the preferred name and (3) plan-ning and execution of the selected name No other inter-vening stages or abstract levels between meaning andsound are assumed In contrast Leveltrsquos influential modelof word production (which has also relied heavily on picture-naming data Glaser 1992 Levelt 1989 LeveltRoelofs amp Meyer 1999) suggests an intervening stagedisputed by Johnson et al Levelt assumes four universalstages in word production (1) individuation of a target

concept (which could be named in more than one way)(2) selection of a word-specific lemma (the componentof a lexical item that contains definitional lexical andgrammatical content) (3) activation of the word form(an abstract characterization of the sound pattern associ-ated with a specific lemma with modifications appro-priate for the grammatical context) and (4) articulationof the motor program associated with a word form mod-ified to fit its articulatory context Depending on whichmodel ultimately proves to be correct it is reasonable toassume that natural languages are affected by universalconstraints on processing at each of these levels fromconceptualization (perceptual constraints in picture de-coding cognitive and social constraints on concept for-mation) to retrieval of lexical forms (effects of frequencyand complexity) to articulation (universal effects ofphonetic structure and length) The search for such uni-versals motivates the selection of measures for multi-variate analyses in this cross-linguistic study

Picture Decoding Visual Complexity andGoodness of Depiction

As was discussed by Johnson et al (1996) one ele-ment that may contribute to difficulty in picture decod-ing is visual complexity In most studies of picture nam-ing that have dealt with this possibility complexity hasbeen assessed by subjective ratings (eg Snodgrass ampYuditsky 1996) However Szeacutekely and Bates (2000)have shown that subjective ratings of complexity areconfounded with subjective judgments of familiarity Toobtain a measure of visual complexity without these con-founds they used the size of the digitized picture file asa predictor variable measured 10 different ways Theseindices proved to be highly correlated with each otherand with subjective ratings of complexity but were notrelated to frequency One of these measures (see the Ma-terials section) will be used in the present study on thebasis of the assumption that digitized file size should beindependent of culture-specific expectations about theldquobestrdquo way to represent a given concept

We will also make use of subjective goodness-of-depiction ratings by US college students who were askedto judge how well each picture illustrated the target namesthat emerged in our study of English (see the Method sec-tion) Although it would be preferable to have goodness-of-depiction ratings for all seven languages (a long-termgoal) the English ratings did prove to be an excellent pre-dictor of naming behavior across languages reflectingwhat may be universal properties of picture decoding

Conceptual AccessibilityA major goal of this cross-linguistic study was to un-

cover lexical concepts (as represented in our picturestimuli) that are equally accessible or inaccessible acrosslanguages For this purpose we have developed indices ofcross-language disparity and cross-language similarityfor both name agreement and latencies (see the Scoringsection below)


FrequencyOn the basis of hundreds of studies documenting word

frequency effects in lexical access we expected to findhigher name agreement and faster naming latencies formore frequent words (corresponding to word forms inthe Johnson et al model and to lemmas plus their asso-ciated word forms in the Levelt model) These frequencyeffects should be roughly equivalent in magnitude withineach of the languages under study However the locus ofthese frequency effects is another matter As Johnson et al(1996) noted conceptual accessibility and word frequencycan be hard to distinguish within a given language (is theword frequent because its concept is frequent or viceversa) Our cross-linguistic design offers a unique op-portunity to learn more about the locus of frequency ef-fects in picture naming If we find that frequency in Lan-guage A predicts name agreement andor RT not only inLanguage A but also in other languages as well it islikely that conceptual accessibility (in addition to wordform frequency) is contributing to the frequency effect(although this would of course not rule out frequency ef-fects at other levels as well see Balota amp Chumbley 1985for evidence of frequency effects when word naming isdelayed well past the point at which the word itself is rec-ognized) Conversely if frequency in Language A pre-dicts behavior better in Language A than in Language Bit is likely that word form frequency is contributing vari-ance beyond the variance accounted for by the familiar-ity of the underlying concept For all seven languagesobjective or (in one case) subjective measures of wordfrequency will be used to compare within- and across-language frequency effects

Word LengthIn previous studies (DrsquoAmico et al 2001 Szeacutekely et al

2003 Szeacutekely et al in press see Johnson et al 1996)modest correlations have been reported between wordlength and picture-naming behaviors reflecting lowername agreement and slower latencies for longer namesHowever length effects in picture naming appear to besubstantially smaller than length effects in word percep-tion tasks (Bates Burani et al 2001 Carr McCauleySperber amp Parmelee 1982) and are often wiped out inregression analyses removing the effects of confoundingvariables One of these confounds is the well-known neg-ative correlation between frequency and word length(ie Zipf rsquos law Zipf 1965) We may also anticipateconfounds between word length and word complexity(see below) as well as conceptual accessibility (morediff icult concepts are more likely to be named withlonger and more complex words) In our search for uni-versals we anticipated that word length effects would besmall and similar in direction for all the languages re-flecting a universal tendency to avoid longer wordsHowever we also anticipated cross-linguistic variationsin the size of word length effects which may interact inturn with cross-linguistic differences in word structure

One such example investigated below involves thefrequency of different word structure templates For ex-

ample a large majority of content words in Chinese aredisyllables Each of these syllables is represented by aseparate character in the written language and usuallyhas a distinct meaning (which may or may not be modi-fied when that syllable is combined with others to forma compound word) We know from previous studies ofChinese (Chen 1997 Chen amp Bates 1998) that wordstructure typicality has effects on naming that are par-tially independent of word length That is disyllables areeasier to recognize andor retrieve than other word typesincluding monosyllables In the same vein monosyllabiccontent words are common in English but rare in ItalianHence any general advantage that may accrue to mono-syllables because of their length should be greater in En-glish than it is in Chinese or Italian To explore the an-ticipated tradeoff between length and word structuretypicality within a cross-linguistic framework we willinclude a weighted measure of typicality in the presentstudy based on the number of target names that fall intoeach syllable length category (see the Method section)

Phonetic Structure and Word ComplexityFor all of the languages under study we include a di-

chotomous measure that reflects whether or not the tar-get name produced by our participants begins with africative or an affricate (ie initial frication) This vari-able is necessary because it is known that presence of africative (which contains white noise prior to or con-temporaneous with voicing) can slow down the detectionof word onset These frication effects are observed inpeople who participate in timed word perception studies(Bates Devescovi Pizzamiglio DrsquoAmico amp Hernandez1995) and they also affect word detection by a computer-controlled voice key in word production studies Hencewe knew it would be useful to track initial frication inour picture-naming study even if the relationship betweenconcepts and word forms was completely arbitrary andrandomly distributed across natural languages In addi-tion we knew in advance that these seven languages sharea certain number of cognates words that are physicallysimilar because they are drawn from a common sourceThis is true not only for the f ive Indo-European lan-guages in our study (English Spanish Italian Germanand Bulgarian) but also for Hungarian (a Uralic lan-guage) and Mandarin Chinese (a Sino-Tibetan language)Hence the modest nuisance variable initial frication maybe positively correlated over languages reflecting a mod-est degree of physical similarity in the word forms pro-duced within and across these seven languages

In the same vein some of the concepts illustrated byour picture stimuli tend to be encoded within and acrosslanguages by complex words (compounds or inflectedforms) and or by multiword phrases (eg ice creamcone) Some items may elicit complex names in everylanguage reflecting the composite nature of the depictedconcept For example we anticipated a tendency for alllanguages to include terms for fire and truck in the wordchosen to name a vehicle driven by people who put outfires In other cases complexity effects may be language

350 BATES ET AL

specific For example Italian tends to use periphrasticconstructions like macchina da stirare (literally machinefor ironing) or macchina da scrivere (literally machine forwriting) where other languages use a single simple word(eg iron) or a compound word (eg typewriter) Todeal with universal as well as language-specific effectsof word complexity we included a dichotomous code forword complexity in all multivariate analyses anticipat-ing that word complexity would be associated with lowername agreement and longer naming latencies Targetnames were classified as complex if they were com-pounds or multiword constructions andor if they con-tained a marked inflection (eg plurals or diminutives)We anticipated that complexity would be confoundedwith word length as well as with word frequency re-quiring regression designs to assess unique versus over-lapping contributions of all three constructs

The complexity measure was easily obtained for En-glish Spanish Italian German Bulgarian and Hungar-ian because (1) these languages do have bound inflec-tions and (2) orthographic conventions make it relativelyeasy to determine whether a word is a compound or amultiword utterance It was not possible to derive a sim-ple dichotomous measure of complexity for ChineseFirst there is no bound morphology in Chinese (undermost definitions) Second because Chinese writing islogographic rather than orthographic the distinction be-tween inflected forms compounds andor multiword ut-terances is controversial Third it is estimated that morethan 80 of all Chinese content words are compoundsthose multisyllabic words that are not decomposable intoseparate syllables with separate meanings tend to be for-eign loan words (eg the item in our data set that elicitedldquosandwichrdquo) Because a dichotomous measure of com-plexity in Chinese would be diff icult to obtain andwould in any case have an entirely different meaningfrom the corresponding division in the other six lan-guages under study Chinese was excluded from allanalyses using the complexity measure

Finally we discovered in the course of this investiga-tion that speakers sometimes produce the same targetname for more than one picture (see also Bates Buraniet al 2001 DrsquoAmico et al 2001 Szeacutekely et al 2003Szeacutekely et al in press) Although our seven target lan-guages vary in the extent to which this sharing strategyis used (see below) it is likely to occur more often forsome picture stimuli than others Specifically our stud-ies to date suggest that shared names are used more oftenfor items that are unfamiliar or difficult to decode How-ever the names that are used for this purpose also tendto be shorter and more frequent than those names thatemerge as the dominant (target) name only once This phe-nomenon poses an interesting challenge for multivariateanalyses Because harder items are more likely to sharetheir target names name sharing may be associated withslower RTs on the other hand because shared namestend to be shorter and higher in frequency name sharingcould lead to faster RTs We will try to disentangle thesecompeting possibilities in regression analyses

The latter two variables (complexity and name shar-ing) are examples of historical feedback across the lev-els postulated by both Paiviorsquos three-stage model andLeveltrsquos four-stage model Whether or not these rela-tionships between levels are modular within individualspeakerslisteners (eg the absence of feedback fromword form selection to lemma selection is an importantconstraint in Leveltrsquos theory) the coevolution of con-cepts and word forms across language history can resultin systematic correlations (Kelly 1992) In other wordseven though the relationship between meaning and wordform is arbitrary (ie words do not resemble their refer-ences) word forms can contain correlational cues tomeaning which may affect naming behavior in universalandor language-specific patterns

METHOD

ParticipantsAll the participants were native speakers of their respective lan-

guages (although amounts of second-language experience may varywith the culture) All were college students tested individually in auniversity setting (English speakers in San Diego Spanish speak-ers in Tijuana Mexico Italians in Rome Germans in Leipzig Bul-garians in Sofia Hungarians in Budapest and Mandarin Chinesespeakers in Taipei) There were 50 participants each in EnglishSpanish Italian Bulgarian Hungarian and Chinese 30 partici-pants were tested in German

An additional 20 US students participated in a goodness-of-depiction ratings task (see below) Because no adequate objectiveword frequency norms were available for Bulgarian a further set of20 Bulgarian students provided subjective ratings of frequency forthe target names obtained with our 520 object pictures rated on ascale of 1 to 7 from lowest to highest

MaterialsPicture stimuli for object naming were 520 black-and-white line

drawings of common objects Pictures were obtained from varioussources primarily US and British (listed in Table 2) including 174pictures from the original Snodgrass and Vanderwart (1980) setAlthough different sources were tapped to supplement the Snod-grass and Vanderwart set all were comparable in style The full setof picture candidates included more than 1000 items many of themoverlapping in their content and intended name Pilot studies (focusgroups) were carried out in the US for the selection of the final set(eg to choose the ldquobest beerdquo out of three different options) Itemselection was subject to several constraints including picture qual-ity visual complexity and potential cross-cultural validity of thedepicted item Nevertheless the fact that the picture set was origi-nally compiled for English is a limitation of this study and must bekept in mind in comparing results across languages

The final set of 520 was selected in the hope that each wouldelicit a distinct object name (although this did not always prove tobe the case even in English) They were scanned and stored digi-tally for presentation within the PsyScope Experimental ControlShell (Cohen MacWhinney Flatt amp Provost 1993) in 10 differentrandomized orders The black-and-white simple line drawings werescanned and saved as (300 3 300 pixels) Macintosh PICT file formateach in a separate file A demo version of the handmade softwareutility Image Alchemy 18 (Woehrmann Hessenflow Kettmann ampYoshimune 1994) was used to convert the stimuli to various graph-ics file formats Over 30 different f ile types and degrees of com-pression for the 520 object and 275 action pictures were computedand seven commonly used formats were selected according to theirrelation to subjective visual complexity and other variables One ofthese was the Joint Photographic Experts Group (JPEG) (with de-


fault Huffman coding and high qualityndashlow degree of compres-sion) The syntax used when converting the pictures in ImageAlchemy was -j98 (high quality refers to 98 on a scale from 1 to100) This format was selected for the visual complexity measure(see the Predictor Variables section below)

Pilot-naming studies indicated that normal adult participantswere able to complete 520 items in a single 45- to 60-min sessionwith occasional breaks In a paper on English only Szeacutekely et al(in press) found no significant correlation between accuracy andorder of presentation across the naming session However naminglatencies did slow gradually across the session justifying our deci-sion to use multiple randomized orders The same 10 orders wereused in every language (3 participants per list in German 5 partic-ipants per list in the remaining six languages)

ProcedureThe participants were tested individually in a dimly lit quiet

room Prior to the picture-naming task voice sensitivity was cali-brated for each participant who read a list of words with variousinitial-phoneme patterns (none of these was appropriate as a namefor a picture in the main experiment) They were instructed to namethe pictures that would appear on the screen as quickly as they couldwithout making a mistake and to avoid coughs false starts hesita-tions (eg ldquouhmmrdquo) articles or any other extraneous material(eg ldquoa dogrdquo or ldquoThatrsquos a dogrdquo) but to give the best and shortestname they could think of for the depicted object or action To fa-miliarize the participants with the experiment a practice set of pic-tures depicting geometric forms such as a triangle a circle and asquare were given as examples in object naming

To maximize comparability identical equipment was used in allseven languages During testing the participants wore headphoneswith a sensitive built-in microphone (adjusted to optimal distancefrom the participantrsquos mouth) that were connected to the CarnegieMellon button box an RT-measuring device with 1-msec resolutiondesign for use with Macintosh computers The pictures were displayedon a standard VGA computer screen set to 640 3 480 bit-depth reso-

lution (the pictures were 300 3 300 pixels) The participants viewedthe items centered from a distance of approximately 80 cm On eachtrial a fixation crosshatch ldquo+rdquo appeared centered on the screen for200 msec followed by a 500-msec blank interval The target pictureremained on the screen for a maximum of 3 sec (3000 msec) Thepicture disappeared from the screen as soon as a vocal response wasregistered by the voice key (at the same time a dot ldquordquo signaled voicedetectionmdasha clue for the error-coding procedure) If there was no re-sponse the picture disappeared after 3000 msec but another1000 msec were added to the total response window just in case thespeakers had initiated a response right before the picture disappearedHence the total window within which a response could be made was4000 msec The period between offset of one trial and onset of thenext was set to vary randomly between 1000 and 2000 msec Thisintertrial jitter served to prevent subjects from settling into a responserhythm that was independent of item difficulty

RTs were recorded automatically by the voice key in the CMUbutton box and served as critical outcome measures for statisticalanalysis For each of the 10 different randomized versions of theexperiment a printout served as a score sheet for coding purposesduring the experiment The experimenter took notes on the scoresheet according to an error-coding protocol (see details below) Al-ternative namings were also recorded manually on the score sheetNo pictures were preexposed or repeated during the test hence notraining of the actual targets occurred A short rest period was in-cluded automatically after 104 trials but the participants could askfor a pause in the experiment at any time Experimental sessionslasted 45 min on average and were tape-recorded for subsequentoff-line checking of the records

To obtain subjective ratings of goodness of depiction the same520 object pictures were presented to a separate group of 50 UScollege students The students were asked to rate how well the picturefit its dominant name (determined empirically from the Englishpicture-naming study see the Scoring section) on a 7-point scalefrom worst to best A keypress recording procedure allowed the par-ticipants to use a broad time interval for making their responsessince the stimulus remained on the screen until the participant re-sponded by pressing one of the keys representing the scales Aver-age ratings were calculated for each picture

ScoringOur scoring criteria were modeled closely on procedures adopted

by Snodgrass and Vanderwart (1980) with a few exceptions Thetarget name (dominant response) for each picture was determinedempirically in two steps

First error coding was conducted to determine which responsescould be retained for both naming and RT analyses The followingthree error codes were possible

1 Valid response referred to all the responses with a valid (cod-able) name and valid response times (no coughs hesitations falsestarts or prenominal verbalization such as ldquothatrsquos a ballrdquo) Anyword articulated completely and correctly was kept for the evalua-tion except for expressions that were not intended namings of thepresented object such as ldquoI donrsquot knowrdquo

2 Invalid response referred to all the responses with an invalidRT (ie coughs hesitations false starts or prenominal verbaliza-tions) or a missing RT (the participant did produce a name but itfailed to register with the voice key)

3 No response referred to any trial in which the participant madeno verbal response of any kind

Once the set of valid responses had been determined the targetname was defined as the dominant responsemdashthat is the name thatwas used by the largest number of participants In the case of ties (tworesponses uttered by exactly the same number of participants) threecriteria were used to choose one of the two or more tied responses asthe target (1) the response closest to the intended target (ie the hy-pothesized target name used to select stimuli prior to the experiment)

Table 2Sources of Object-Naming Stimuli

Source No

Snodgrass amp Vanderwart 19801 174Alterations of Snodgrass amp Vanderwart1 2Peabody Picture Vocabulary Test 19812 62Alterations of Peabody Picture Vocabulary Test 19812 8MartinezndashDronkers set3 39Abbate amp La Chappelle ldquoPictures Pleaserdquo 198445 168Max Planck Institute for Psycholinguistics6 20Boston Naming Test 19837 5Oxford ldquoOne Thousand Picturesrdquo8 25Miscellaneous 171S n o d g r a s s J G amp V a n d e r w a r t M ( 1 9 8 0 ) A s t a n d a r d i z e d s e t o f 2 6 0

p i c t u r e s N o r m s f o r n a m e a g r e e m e n t f a m i l i a r i t y a n d v i s u a l c o m p l e x i t y

J o u r n a l o f E x p e r i m e n t a l P s y c h o l o g y H u m a n L e a r n i n g amp M e m o r y 6

1 7 4 - 2 1 5

2Dunn Lloyd M amp Dunn Leota M (1981) Peabody Picture Vocabu-lary TestndashRevised Circle Pines MN American Guidance Service3Picture set used by Dronkers N (personal communication)4Abbate M S amp La Chapelle N B (1984b) Pictures please An ar-ticulation supplement Tucson AZ Communication Skill Builders5Abbate M S amp La Chapelle N B (1984a) Pictures please A lan-guage supplement Tucson AZ Communication Skill Builders6Max Planck Institute for Psycholinguistics Postbus 310 NL-6500 AHNijmegen The Netherlands7Kaplan E Goodglass H amp Weintraub S (1983) Boston NamingTest Philadelphia Lee amp Febiger8Oxford Junior Workbooks (1965) Oxford Oxford University Press

352 BATES ET AL

(2) the singular form if singular and plural forms were tied or (3) theform that had the largest number of phonological variants in common

Second all valid responses were coded into the following differ-ent lexical categories in relation to the target name using the samecriteria

Lexical Code 1 The target name (dominant response empiri-cally derived)

Lexical Code 2 Any morphological or morphophonological al-teration of the target name defined as a variation that shares theword root or a key portion of the word without changing the wordrsquoscore meaning Examples would include diminutives (eg bike forbicycle doggie for dog) pluralsingular alternations (eg cookieswhen the target word was cookie) reductions (eg thread for spoolof thread ) or expansions (eg truck for firemen for firetruck)

Lexical Code 3 Synonyms for the target name (which differ fromCode 2 because they do not share the word root or key portion ofthe target word) With this constraint a synonym was defined as aword that shared the same truth-value conditions as the target name(eg couch for sofa or chicken for hen)

Lexical Code 4 This category was used for all names that couldnot be classified in Codes 1ndash3 including hyponyms (eg animalfor dog) semantic associates that are not synonyms (eg cat fordog) partndashwhole relations at the visual-semantic level (eg fingerfor hand) and all frank visual errors or completely unrelated re-sponses

Name AgreementPercentage of name agreement was defined for each item as the

proportion of all valid trials (a codable response with a usable RT)on which the participants produced the target name (Lex1) Thenumber of alternative names for each picture (number of types) wasderived by simply counting number of different names provided onvalid trials including the target name In addition following Snod-grass and Vanderwart (1980) we also calculated the H statistic orH Stat (also called the U statistic) a measure of response agree-ment that takes into consideration the proportion of participantsproducing each alternative High H values indicate low name agree-ment and 0 refers to perfect name agreement Percentage of nameagreement measures for each item were based on the 4-point lexi-cal coding scheme For each item Lex1 refers to the percentage ofall codable responses with a valid RT on which the dominant namewas produced

Although the 520 object pictures were selected with the inten-tion of eliciting one unique target name for every picture this wasnot always the case Occasionally within a given language thesame name emerged as the dominant response for two or more pic-tures We treated this event as a dependent variable (called sames)assigning a score of 1 for each item that shared its target name(dominant response) with another picture in the data set for that lan-guage This assignment was made independently for each languageHence an item might have a unique name in English (esames = 0)but share its name in Bulgarian (bsames = 1) and so forth

Reaction TimeRT total refers to mean RTs across all valid trials regardless of

the content of that response RT target refers to mean latency fordominant responses only

Cross-Language Universality and DisparityThe cross-linguistic design of the present study permitted us to

derive some novel measures of universality and cross-language dis-parity for the 520 picture stimuli For this purpose name agreement(Lex1) and latencies to produce the target name were first convertedto z scores within each language (representing a continuum frombest to worst in name agreement and from fastest to slowest in tar-get RT) By using z scores we removed main effects of language onboth of these dependent variables and ensured that no single lan-

guage was contributing disproportionately to these estimates of uni-versality and disparity Estimates of universality were obtained byaveraging the z scores for each item across all seven languages forLex1 and RT target respectively Thus if an item tended to elicithigh agreement and fast RTs in all or most of the languages itwould have an average universal z score at the high-performanceend of each continuum (one for name agreement one for RT) Con-versely if an item tended to elicit low agreement and or slow RTsin all or most of the languages it would have an average universalz score at the low-performance end of each continuum Items thatproduced little consensus across the seven languages would tend tocluster in the middle of these two universality measures A thirdrelatively simple measure of universal tendencies was also con-structed for each item on the basis of the arithmetic average of thenumber of word types measure for each of the seven languages

To complement these measures of universality we also calcu-lated measures of cross-language disparity We began by comput-ing for each language the average z scores for the other six lan-guages (for Lex1 and for RT target respectively) So for examplethe Other-Language Z-score for English name agreement would bethe average of the z scores for German Spanish Italian BulgarianHungarian and Chinese In the same vein the Other-Language Z-score for Bulgarian RT target would be the average of the RTz scores for English German Spanish Italian Hungarian and Chi-nese With these statistics in hand we then calculated a differencescore for each language for agreement and RT respectively usingsimple subtraction in a direction that would indicate an advantagefor the language in question For example a positive Lex1 differ-ence score for German would indicate that Germans had highername agreement for that item than did the other six languages (onaverage) Similarly a negative RT target difference score for Chi-nese would indicate that Chinese speakers were relatively fast onthat item as compared with the speakers in the other six languages(on average) Using these difference scores we can investigate thefactors that confer a relative advantage (easy item) or disadvantage(hard item) within each individual language as compared with theothers in the study Finally the absolute values of these seven dif-ference scores were averaged for both Lex1 and RT target to pro-duce estimates of cross-language disparity in nameability (Lex1)and RT (RT target) respectively Thus items with high disparityscores are those that elicited more cross-language variation itemswith low disparity scores are those that elicited less variability and amore universal response (although such items could be universallygood or universally bad )

In addition to these measures of performance (dependent vari-ables) the target (dominant) names produced for each item werecoded along a number of dimensions that were believed to affectaccuracy and or latency in studies of lexical access (independentvariables) In the full database this list includes some variables thatare applicable (eg grammatical gender) or available (eg age-of-acquisition [AoA] norms) only for a subset of the languages For thepresent study we restricted our attention to the following indepen-dent or predictor variables that were available for all seven lan-guages (with two exceptions indicated below)

1 Visual complexity In addition to predictor variables associatedwith the target names an estimate of visual complexity was ob-tained for each picture based on file size in the JPEG format (seethe Materials section above for additional details see Szeacutekely ampBates 2000)

2 Goodness-of-depiction ratings were available only for US par-ticipants but this variable proved to be such a powerful predictoracross all seven languages that it was included in the present study

3 Word frequency of the target names was extracted for each lan-guage from written or spoken sources (because there were no fre-quency corpora available for Bulgarian subjective ratings of fre-quency were used) The sources included the following theCELEX database for English and German (Baayen Piepenbrock amp


Gulikers 1995) Alameda and Cuetos (1995) for Spanish DeMauro Mancini Vedovelli and Voghera (1993) for Italian Fuumlrediand Kelemen (1989) for Hungarian and the Chinese KnowledgeInformation Processing Group (1997) for Chinese

4 Word length in syllables 5 Syllable type frequency As we noted in the introduction the

seven languages vary markedly in the distribution of monosylla-bles disyllables and words with three or more syllables After thetarget names were ascertained for each language and their length insyllables was calculated we constructed a measure to reflect thefrequency of word types in the full corpus of 520 names for eachlanguage For example if the corpus for Language A comprised220 monosyllables 200 disyllables 80 three-syllable words 19four-syllable words and 1 five-syllable word each monosyllabicname received a score of 220 each disyllable received a score of200 each three-syllable word received a score of 80 each four-syllable target name received a score of 19 and the remaining five-syllable word received a score of 1

6 Word length in characters was available for all the languagesexcept Chinese

7 Initial frication was a dichotomous variable reflecting presenceabsence of a fricative or affricate in the initial consonant (0 = nofricative or affricate 1 = fricative or affricate)

8 Complex word structure was a dichotomous variable that wasassigned to any item on which the dominant response was an in-flected form a compound word or a periphrastic (multiword) con-struction This was available for all languages except Chinese

RESULTS AND DISCUSSION

All the analyses were conducted over items averagedacross the 50 subjects (30 for German) who participatedin the naming study for each language Hence languagewas treated as a within-item factor for all direct cross-linguistic comparisons Within all the relevant tables lan-guages are ordered to reflect hypothesized differences inlanguage distance from English and German (Ger-manic) to Spanish and Italian (Romance) to Bulgarian(Slavic) and then to the two non-Indo-European lan-guages Hungarian and Chinese Because these resultsare unusually complex (involving seven languages mul-tiple dependent variables and multivariate analyses) wehave tried to make the text more readable by placing sta-tistical details into tables wherever possible Hence thetext itself will focus on overall patterns that emerged

from these detailed results Some of the tables are re-ferred to in the text (in appropriate order) but for the sakeof economy they are excluded from the published textand are instead available for inspection (together withthose pictures that are publicly available) on our Web siteat httpwwwcrlucsdedu~aszekelyipnp7lgpnohtml

Part I Cross-Language Variations in NamingBehavior (Dependent Variables)

Part I presents cross-linguistic similarities and differ-ences in naming behavior (dependent variables) includingvarious measures of name agreement as well as latenciesto produce the target name This section also includesmultivariate analyses of the relationships among thesedependent variables within and across languages Thisis the place where we can ask our first three questionsabout cross-language disparities and cross-language uni-versals in naming behavior

Question I-1 To what extent will naming behaviorsvary across these seven languages If differences in nam-ing behavior are detected will they reflect a classic gra-dient of language distance (eg stronger correlationsamong Indo-European languages vs Hungarian and Chi-nese stronger correlations within specific Indo-Europeanlanguage families such as Romance)

Question I-2 Will cross-language correlations be sim-ilar for name agreement and reaction times reflectingbasic conceptual universals Or will we find strongercross-language similarities in the time required to re-trieve the dominant target name reflecting universal as-pects of processing that cannot be detected with nameagreement alone

Question I-3 Will the number of alternative names foreach picture slow down RTs (reflecting competitor ef-fects) even after percentage of agreement on the domi-nant name is controlled If such competitor effects canbe demonstrated will they qualify as a cross-languageuniversal evident within every language and in cross-language summary scores

Descriptive statistics A breakdown of the proportionof valid responses invalid responses and nonresponsesobserved within each language (means standard devia-

Table 3Summary Statistics for Correctness in the Different Languages

English German Spanish Italian Bulgarian Hungarian Chinese

valid response (F = 8956 p lt 001)Mean 961 947 932 920 892 941 893SD 60 96 103 109 111 82 119Range 60ndash100 17ndash100 34ndash100 18ndash100 20ndash100 22ndash100 22ndash100

no response (F = 3560 p lt 001)Mean 23 33 52 55 51 22 46SD 50 88 96 99 100 67 101Range 0ndash34 0ndash80 0ndash66 0ndash80 0ndash78 0ndash74 0ndash76

invalid response (F = 19164 p lt 001)Mean 15 20 16 25 57 37 61SD 23 30 21 28 45 33 42Range 0ndash16 0ndash20 0ndash14 0ndash14 0ndash32 0ndash20 0ndash22

354 BATES ET AL

tions and ranges) is presented in Table 3 One-way analy-ses of variance over items revealed significant cross-linguistic differences on all three variables In part thisreflected an overall advantage for English (with a meanof 961 valid responses with usable RTs) and a relativedisadvantage for Bulgarian and Chinese (at 892 and893 respectively) The fact that performance was bet-ter in English is not at all surprising since the items wereinitially developed and selected for an English-speakingaudience However this is not the whole story When thesame analyses of variance were repeated with Englishexcluded significant cross-language differences remainedfor the other six languages on all three measures in Table 3(with F values between 311 and 1767 all ps lt 001)Although there is certainly cross-language variation onall three measures the most important finding in Table 3is that this naming paradigm works in all seven lan-guages and also yields enough variation in nameabilitywithin each language to justify the multivariate approachthat we take for the remainder of the paper

Descriptive statistics for more detailed name agreementvariables within each language are presented in Table 4There were robust cross-linguistic differences on everymeasure (assessed by a one-way analysis of variance for

all continuous variables and by a k-level chi-square forthe binary variable same name) Performance was highestfor English speakers in every case (as we would expectsince the items were originally compiled from Englishsources) However significant cross-language differencesremained when the same analyses were repeated exclud-ing English (all ps lt 001) Chinese speakers were almostalways at a greater disadvantage (lower name agreementand more alternative names) But target name agreementwas high in every language group from a low of 719for Chinese to a high of 850 in English

Naming latencies for each language are presented inTable 5 for all valid trials (total RT) and for those trialson which speakers produced the target name (target RT)One-way analyses of variance over languages again re-vealed significant differences with the fastest RTs ob-served in English (M = 1041 msec) and the slowest RTsobserved in Bulgarian (M = 1254 msec) and Chinese(M = 1241 msec) However the range of RTs was verylarge within every language and the fastest RTs werequite comparable (ranging from 656 msec in English to768 msec in Bulgarian) Both the mean RTs and the min-imum RTs are comparable to values that have been re-ported in prior studies using timed picture naming

Table 4Summary Statistics for Name Agreement in the Different Languages


Number of types (F = 5843 p lt 001)Mean 335 514 415 439 382 416 547SD 228 342 291 285 256 296 363Range 1ndash18 17ndash217 1ndash17 1ndash20 1ndash14 1ndash21 1ndash21

H statistics (F = 4665 p lt 001)Mean 067 076 086 095 084 091 116SD 061 068 072 073 065 073 079Range 0ndash290 0ndash328 0ndash290 0ndash347 0ndash270 0ndash352 0ndash357

Lex 1 dominant (F = 3283 p lt 001)Mean 850 811 800 770 802 780 719SD 164 199 204 216 204 213 233Range 28ndash100 21ndash100 17ndash100 12ndash100 13ndash100 13ndash100 11ndash100

Lex 2 phonetic variance (F = 2164 p lt 001)Mean 37 44 32 49 41 71 85SD 87 100 84 104 98 129 124Range 28ndash100 21ndash100 17ndash100 12ndash100 13ndash100 13ndash100 11ndash100

Lex 3 synonym (F = 1178 p lt 001)Mean 24 32 42 52 25 43 16SD 77 84 101 110 77 102 55Range 28ndash100 21ndash100 17ndash100 12ndash100 13ndash100 13ndash100 11ndash100

Lex 4 erroneous (F = 2919 p lt 001)Mean 90 114 127 129 33 106 180SD 124 164 162 164 174 162 198Range 28ndash100 21ndash100 17ndash100 12ndash100 13ndash100 13ndash100 11ndash100

Same name ( c 2 = 912 p lt 001)Mean () 46 83 121 87 129 140 196Range 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1

Since data were collected from only 30 subjects in the German language the number of alternative typeswere calculated as (raw type number) 3 (5030)


Again when the cross-language analyses were conductedwith English excluded significant main effects of lan-guage remained

Correlations and regressions within languages fordependent variables Tables 6ndash12 (on our Web site athttpwwwcrlucsdedu~aszekelyipnp7lgpnohtml)present correlations among eight dependent measureswithin each language including a summary score forpercent correct that conflates Lexical Codes 1 2 and 3(targets + morphophonological variants + synonyms)This ldquolenientrdquo score for name agreement has been usedin other picture-naming studies and sometimes yieldsdifferent results as compared with the more conserva-tive measure in which name agreement refers to produc-tion of the target name only (see Szeacutekely et al in pressfor details) In the present study the results for these twoversions of name agreement were quite similar hencewe will restrict our attention to the strict (Lex1) measureof name agreement for most of the analyses that followMost important for present purposes inspection of Ta-bles 6ndash12 indicates that there are striking similaritiesacross languages in the pattern of correlations Of coursesome of these correlations are inevitable because vari-ables such as percentage of target names and the H sta-tistic stand in a partndashwhole relationship Others are moreinteresting including high negative correlations betweentarget name agreement (Lex1) and target RT The leastobvious statistics in Tables 6ndash12 involve small but sig-nificant correlations with the binary variable same nameIn general these results suggest that speakers tend toproduce the same name for more than one picture whenthey find themselves in difficulty a strategy that is leastlikely in English and most likely in Chinese but may re-

flect similar processes in all seven languages (for furtherdiscussion of this point see Szeacutekely et al 2003)

To complement the language-specific statistics in Ta-bles 6ndash12 Table 13 presents correlations among the fivesummary measures of universality and disparity aver-aged across the seven languages The cross-languagepattern is similar to the relationships seen within indi-vidual languages However the coefficients for thesesummary variables are notably higher than the corre-sponding coefficients for each individual language Forexample the correlation between target agreement andtarget RT ranges between 255 and 267 in the seven in-dividual languages but the corresponding relationshipbetween universal agreement and universal RT is 274Similarly the correlations between target RT and num-ber of alternative names range between +61 and +76within individual languages but the cross-language re-lationship between target RT and average number oftypes is +81 This result is not trivial In summarizingover seven different language types we might have lostinformation based on specific language forms (egitem-specific effects of frequency in German or length inHungarian) that contributes to the magnitude of the cor-relations within each language If such form-specificfacts played a major role in the correlations among nam-ing measures we would expect these coefficients to dropmarkedly when cross-language summary variables areused Instead we seem to be picking up power and reli-ability by averaging across languages reflecting stronguniversal trends that are greater than the sum of the parts

Finally Table 13 shows that cross-language universalsare inversely correlated with corresponding measures ofcross-language disparity Although this is not surprising

Table 5Summary Statistics for Mean Reaction Time in the Different Languages


RT total (F = 13676 p lt 001)Mean 1041 1130 1168 1163 1254 1105 1241SD 230 281 280 270 283 281 319Range 656ndash1843 663ndash2397 711ndash2063 694ndash2580 768ndash2373 659ndash2300 686ndash2389

RT target (F = 11500 p lt 001)Mean 1019 1101 1139 1133 1217 1071 1200SD 211 273 262 264 261 268 312Range 656ndash1823 663ndash3117 711ndash2392 694ndash2831 768ndash2273 659ndash3139 686ndash2403

Table 13Correlation Matrix of Cross-Language Dependent Variables

Cross-Language Cross-Language Cross-Language Cross-LanguageName Agreement Target RT Number of Types Naming Disparity

Cross-language name agreement ndashCross-language target RT 274Dagger ndashCross-language number of types 289Dagger 181Dagger ndashCross-language naming disparity 260Dagger 140Dagger 150Dagger ndashCross-language target RT disparity 253Dagger 166Dagger 155Dagger 251DaggerDaggerp lt 01

356 BATES ET AL

(indeed it was inevitable) it is interesting that the rela-tionships are lower than we might expect For examplethe correlation between universal name agreement andcross-language naming disparity is 260 This meansthat items high in disparity tend to be low in agreementbut it also means that the two measures share only 36of their variance To explore the relationship betweenuniversality and disparity further we examined scatter-plots for two pairs of variables universal name agree-ment versus naming disparity (Figure 1) and universaltarget RTs versus RT disparity (Figure 2) Both of thesescatterplots show that good items are very good indeed(high in agreement low in RT and low in disparity onboth measures) However bad items can be universallybad (low agreement large RTs and low disparity on bothmeasures) or they can elicit quite diverse results (bad foronly a subset of languages with high disparity)

The raw correlations in Tables 6ndash13 suggest anotherquestion of considerable theoretical importance Is name-ability a single dimension that bears a singular relation-ship to naming latency or can we distinguish betweennameability (reflected in proportion of participants pro-ducing the dominant name) and competition from alter-native names To deal with this question we conductedstepwise regression analyses for target RT in each lan-guage using name agreement (Lexical Code 1) andnumber of types as predictors Each regression was con-ducted twice giving each predictor an opportunity toenter the equation last in order to assess the amount ofunique variance contributed by that measure when theother was controlled We used number of types instead ofthe H statistic for this purpose because the H statistic(which is weighted for the number of participants choos-ing each alternative) and Lexical Code 1 are correlatedso highly that they are likely to cancel each other out

Table 14 summarizes the results for these stepwiseanalyses within each language and it also presents thecorresponding results when cross-language summaryvariables are used In all seven languages the two naming

measures together accounted for a substantial amount ofthe variance in naming latencies (from a low of 392 inGerman to a high of 58 in Spanish) In all seven casesnumber of word types contributed significant and substan-tial variance to naming latencies after name agreementwas controlled (from a low of 60 in German to a highof 171 in English) In all cases the direction of thatcontribution was positive indicating that RTs sloweddown as a function of the number of alternative namesthat were available independent of name agreement forthe dominant response This result is compatible withmodels in which alternative names exert a competitiveeffect on word retrieval (Caramazza 1997 Cutting ampFerreira 1999 Humphreys Riddoch amp Quinlan 1988Levelt et al 1999 Roelofs 1992 Schriefers Meyer ampLevelt 1990) In all seven languages target name agree-ment also made a significant contribution after numberof word types was controlled always in a negative di-rection (indicating that higher agreement is associatedwith faster RTs) However these contributions were con-siderably smaller (from a barely significant low of205 in English to a high of 244 in Italian) Hencewe may conclude that name agreement and alternativenames exert at least partially independent effects on thepicture-naming process but that the inhibitory effect of al-ternative names is larger than the facilitative effect of nameagreement

These appear to be universal effects although theyvary in magnitude from one language to another In thisregard the regression using cross-language summaryvariables is particularly informative Using universal RTas the dependent variable (ie the cross-language averagefor z score target-naming latencies) the cross-languageaverages for number of types and for name agreementaccount together for a significant 661 of the RT vari-ance ( p lt 001) This statistic is (once again) larger thanthe corresponding f igures obtained within individuallanguages indicating that we have gained in power andreliability by summarizing across languages The cross-language average for number of word types made a large

Figure 1 Scatterplot of name agreement disparity scores plot-ted against name agreement z scores (both averaged over lan-guages)

Figure 2 Scatterplot of reaction time (RT) disparity scoresplotted against RT z scores (both averaged over languages)


and significant contribution (+121 p lt 01) to uni-versal naming latencies when universal name agreementwas controlled at a level midway between the uniquecontributions observed in each individual languageHowever in contrast with the analyses conducted withineach individual language (in which name agreement al-ways made a small but significant independent contri-bution) the summary score for name agreement had nosignificant effect when the summary score for numberof types was entered into the equation first In otherwords the most important determinant of RTs within andacross languages appears to be the number of alterna-tive names that are available rather than the level ofagreement that was reached for the target name itself

Correlations and regressions across languages fordependent variables Tables 15ndash18 (on our Web site athttpwwwcrlucsdedu~aszekelyipnp7lgpnohtml)present cross-language correlations for name agreementnumber of alternative names same names and targetname agreement

For name agreement all correlations are significant( p lt 01) suggesting that there was some cross-languagegeneralization in the relative difficulty (nameability) ofthese 520 picture stimuli We might have expected themagnitude of these correlations to follow a typologicalmetric of language distance but there was relatively lit-tle evidence for such a gradient Indeed the highest cor-relations often involve Hungarian (which is not Indo-European) with other European languages Correlationswith Chinese do tend to be lower but not enough to sug-gest that this language is an outlier

Table 16 summarizes correlations across languagesfor number of alternative names Coefficients for num-ber of word types are in the lower triangle coefficientsfor the H statistic are in the upper triangle Again all ofthese correlations are robust ( p lt 01) and moderate insize In other words there is cross-language generality inthe tendency for some items to elicit multiple names

Table 17 presents correlations across languages forthe dichotomous variable same name Although thesecoefficients are smaller than the other naming variablesthey are all highly significant ( p lt 01) and all in the

same direction This result provides support for our hy-pothesis that speakers tend to share generic names fordifficult itemsmdashwhich are in many cases the sameitems from one language to another

Finally cross-language correlations for naming laten-cies are presented in Table 18 All these correlations aresignificant ( p lt 01) and they are also very large Themagnitude of these correlations is striking in comparisonwith the more modest coefficients observed for variousmeasures of name agreement (Tables 15ndash17) Howeverthere is no obvious typological gradient underlying thesecorrelations and no evidence that any individual languageis an outlier Instead these results suggest that naming la-tencies are influenced by universal stages and processesshared by the widely varying languages in our sampleThese shared processes are more evident in the time re-quired to produce a name than they are in the extent towhich speakers agree on the names that they produce

Part I Interim SummaryTo summarize the results so far let us return to the

three main questions posed in Part IQuestion I-1 To what extent will naming behaviors

vary across these seven languages If differences innaming behavior are detected will they reflect a classicgradient of language distance (eg stronger correlationsamong Indo-European languages vs Hungarian andChinese stronger correlations within specific Indo-European language families such as Romance)

The seven languages under study here did differ sig-nificantly on all indices of name agreement and in thetime required to retrieve and produce those names En-glish had the advantage on all the measures that weusedmdashnot a surprising result since these pictures weredesigned for use in American or British English studiesand were compiled for the present study in an English-speaking context However all cross-language compar-isons remained robustly significant when English wasexcluded from the analysis The lowest scores for nameagreement and the longest RTs were typically observed inChinese andor Bulgarian However the cross-languagedifferences in performance observed so far provide little

Table 14Regressions of Naming Behavior on Naming Latencies Within Each Language and AcrossLanguages (Cross-Language Average Z Score Reaction Times [RTs]) Total Percentage of

Variance Accounted for and Unique Contributions of Each Variable on the Last Step

Total Unique Variance Unique VarianceLanguages Variance From Name Agreement From No of Types

English 473Dagger 205dagger +171DaggerGerman 392Dagger 221Dagger + 60DaggerSpanish 580Dagger 207Dagger +136DaggerItalian 460Dagger 244Dagger +59DaggerBulgarian 525Dagger 217Dagger +116DaggerHungarian 510Dagger 210Dagger +119DaggerChinese 489Dagger 219Dagger +87DaggerCross-language Z score RT average 661Dagger ns +121Daggerdaggerp lt 05 Daggerp lt 01

358 BATES ET AL

evidence for a typological gradient That is there is lit-tle evidence here for greater similarities in performancewithin language families or for greater disparities be-tween language families

Question I-2 Will cross-language correlations besimilar for name agreement and RTs reflecting basicconceptual universals Or will we find stronger cross-language similarities in the time required to retrieve thedominant target name reflecting universal aspects ofprocessing that cannot be detected with name agreementalone

There were moderate to strong cross-language corre-lations for all of the dependent variables suggesting thatat least some of these 520 items tend to be hard or easyin every language But correlations were substantiallylarger in the RT data suggesting that this timed picture-naming paradigm taps into universal stages andor uni-versal processes that transcend nameability itself Theseresults were even stronger for our cross-language sum-mary variable as compared with correlations withineach individual language This is we must reiterate anontrivial result If language-specif ic differences inword properties were playing a major role in our resultswithin-language correlations among naming behaviorsought to be stronger than across-language correlationsThe fact that our novel cross-linguistic dependent vari-ables worked so well suggests that we gained in powerand reliability by pooling results over languages

Question I-3 Will the number of alternative namesfor each picture slow down RTs (reflecting competitoreffects) even after percentage of agreement on the dom-inant name is controlled If such competitor effects canbe demonstrated will they qualify as a cross-languageuniversal evident within every language and in cross-language summary scores

In all seven languages the number of alternativenames for each picture contributed substantial varianceto RTs after name agreement was controlled Whencross-language summary scores were used results wereeven stronger This universal result is compatible withmodels that emphasize competitor effects during nameretrieval

We turn now to cross-language similarities and differ-ences in the properties of those words that emerged em-pirically as target names (also called dominant responseor Lexical Code 1) in relation to each other and to prop-erties of the picture stimuli

Part II Word and Picture Properties(Independent Variables)

Part II presents cross-linguistic findings for the stim-ulus properties (independent variables) that characterizethe target pictures (visual complexity and goodness ofdepiction) and the target names elicited in each language(eg frequency length and word structure) Multivari-ate analyses will focus on similarities and differences inthe pattern of relationships among independent variableswithin and across languages In this section we can ad-

dress the following four questions about cross-languageuniversals and cross-language disparities

Question II-1 How much will languages differ in theproperties of target names that are elicited by the same520 pictures of common objects

Question II-2 Will picture properties (objective com-plexity and goodness of depiction) be independent ofeach other and will they be independent of the proper-ties of target names within each language If we do findpicturendashname confounds (eg complex pictures elicitmore complex names) will these be language specificor will they generalize across languages in the study

Question II-3 Will we replicate Zipfrsquos law across allthe languages in the study That is will frequency be as-sociated universally with longer words And will thesame law extend to other properties of word structure(eg canonical syllable structure)

Question II-4 Will we find correlations in word char-acteristics across the seven languages in this study Ifthe link between meaning and word form is truly arbi-trary we would not expect to find cross-language corre-lations in the properties of target picture names (eg fre-quency length or word structure) But if instead we dofind strong cross-language correlations in word proper-ties (eg word frequency in Language A correlates withword frequency in Language B) a case can be made thatthese correlations reflect (at least in part) universals at adeeper level in the picture-naming process (eg con-ceptual accessibility)

Descriptive statistics Table 19 provides descriptivestatistics for each language for five properties of the tar-get names that emerged for each picture stimulus in thatlanguage length in syllables length in characters (ex-cluding Chinese) word frequency (based on a differentmetric for every language including subjective fre-quency ratings in Bulgarian vs log natural frequencyfrom available corpora for the other six) initial frication(a binary variable) and word complexity (also a binaryvariable not available for Chinese) One-way within-item analyses of variance were conducted on the twolength measures chi-square (k-level) analyses were con-ducted on the two binary variables The seven frequencyindices were not compared directly because they werebased on very different counts

As Table 19 shows there were robust cross-linguisticdifferences in every relevant category reflecting differ-ences in word structure of the sort that were described inthe introduction (see summary in Table 1) The shortestwords were produced in English (M = 174 syllablesrange 1ndash5) and the longest words were produced in Ital-ian (M = 292 syllables range 1ndash8) with the other lan-guages distributed in between at roughly even intervalsThere were also marked variations in the probability thattarget names will begin with a fricative or affricate froma low of 123 in Spanish to a high of 348 in Hun-garian a factor that could skew naming latencies de-tected by a voice key There were substantial differencesin word complexity as well ranging from 81 of the


words in Bulgarian to 338 of the words in GermanThis last finding does not correspond to any obvious gra-dient of inflectional complexity and indeed examinationof the target words themselves suggests that inflectedforms (diminutives plurals) were relatively rare Ratherthe cross-language variation in complexity appears to re-flect variations in the use of compounds andor multi-word target names (by this measure almost all of thewords in Chinese would have been scored as complex)

Table 20 provides a more detailed breakdown oflength in syllables showing the number and percentage(out of 520) of the target names for each language thatfell into each of eight syllable-length bins It is clearfrom this table that monosyllables were the modal re-sponse in English (comprising 457 of the words that

emerged as the dominant response) contrasting withlows of 08 in Italian and 23 in Spanish Disyllableswere (as was expected) the modal response in Chinesewhere they comprise 70 of all target names Disylla-bles were also the modal response in German (498)Spanish (438) Bulgarian (537) and Hungarian(512) albeit at lower levels In Italian trisyllableswere the modal response (400) although disyllableswere not far behind (379) To the extent that wordlength does play a role in picture naming we may expectthis variable to contribute to cross-language differencesin naming behavior Because we anticipated large cross-language differences in syllable type we constructed ameasure of syllable type frequency for each language(see the Method section) Effects of length syllable type

Table 19Summary Statistics for Independent Variables Within Each Language

English German Spanish Italian Hungarian Bulgarian Chinese

Length in syllables (F = 16950 p lt 001)Mean 174 213 276 292 228 240 209SD 083 087 096 100 097 092 060Range 1ndash5 1ndash6 1ndash7 1ndash8 1ndash8 1ndash7 1ndash5

Length in characters (F = 2908 p lt 001)Mean 589 673 648 707 607 629

notSD 222 274 214 250 228 223 relevantRange 2ndash15 2ndash19 3ndash17 2ndash20 2ndash19 3ndash17

Word frequency (no statistical comparison)Mean 250 201 290 116 138 425 336SD 157 150 173 143 193 109 201Range 0ndash740 0ndash662 0ndash832 0ndash620 0ndash684 15ndash68 0ndash1056

Frication () ( c2 = 8534 p lt 001)Mean 281 275 123 248 348 271 258Range 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1

Word complexity () (X 2 = 2469 p lt 001)Mean 163 338 85 96 185 81 notRange 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1 0ndash1 relevant

Bulgarian frequency was determined on the basis of subjective ratings and is presented in raw figures

Table 20Number and Percentage of Target Names Within Each Language Falling Into Each Category of Syllable Length (Out of 520 Items)

No of Syllables English German Spanish Italian Bulgarian Hungarian Chinese

Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL

and other word structure variables on naming behaviorwill be described at the end of Part III

Correlations within languages Tables 21ndash27 (onour Web site at httpwwwcrlucsdedu~aszekelyipnp7lgpnohtml) summarize correlations among thepredictor variables (a separate table for each language)including the measure of syllable type frequency de-signed to capture the cross-language differences in wordstructure that are evident in Table 20

As was expected there were significant correlationsamong the properties of target words Length in sylla-bles and length in characters were of course highly cor-related but there was some variation The lowest corre-lation was in German (which has an especially high ratioof consonant clusters) and the highest was in Italian (anorthographically transparent language with very fewconsonant-final words) Length in syllables was signifi-cantly and negatively correlated with syllable type fre-quency reflecting the rarity of words with more thanfour syllables in all of these languages However themagnitude of this correlation varied markedly over lan-guages from a very low correlation in Chinese (wherethe overwhelming majority of words are disyllables) to acorrelation approaching 10 in English (a near-perfectrelationship between number of syllables and frequencyof each word type) As was anticipated there were signif-icant confounds between length and complexity althoughthese confounds varied in magnitude from measure tomeasure and from language to language In particularcorrelations in every language provide evidence forZipf rsquos law the long-noted tendency for languages to re-serve shorter words for more frequent concepts How-ever these relationships also varied in magnitude overlanguages The remaining correlations in Tables 21ndash27(eg with initial frication) were small and largely non-significant but there were enough of them to warrant in-clusion in regression analyses

Goodness-of-depiction ratings and visual complexitywere selected to measure properties of the picture stim-uli The relationship between these two pictorial mea-sures was significant but very small indicating thatmore complex pictures tend to be rated a bit more highlyBoth picture measures also proved to be largely indepen-dent of the properties that characterize their dominantnames although there were a few exceptions Keepingin mind that goodness ratings were provided by English-speaking raters we note that good pictures tended to be

named with words that have atypical syllable structurein Spanish and Bulgarian ( p lt 05) with trends in thesame direction for Hungarian and Chinese ( p lt 10)This peculiar finding may reflect (at least in part) theborrowing of English-based words and concepts in someof the languages under study (eg words like igloo andEskimo) Visual complexity was weakly associated withlonger words and with atypical word structure in En-glish Spanish and Chinese suggesting that speakers arecoming up with ad hoc constructions to describe some ofthe more complex pictures Visual complexity was alsoweakly but positively associated with word complexityin English and in Spanish Finally there was a small butsignificant positive association between visual complex-ity and word frequency in Chinese The primary lessonto take away from these weak and sporadic effects is thatpicture and word properties are largely independent butthere are enough small confounds lurking about to jus-tify inclusion of the picture variables in all regressionanalyses

Correlations across languages The last issue underPart II pertains to cross-language correlations for the majorlexical variables in this study another way to approachthe question of universal factors in picture namingTable 28 presents correlations for the various frequencyindices Even though these indices are all quite different(representing subjective ratings in Bulgarian and esti-mates from corpora that vary in size and modality for theother six languages) all possible cross-correlations werepositive robust and significant ranging from a low of+37 (between Bulgarian and Chinese and between Bul-garian and Spanish) to a high of +66 (between Germanand Hungarian) None of the languages was consistentlyan outlier Hence it seems that common forces are atwork in determining target name frequencies acrossthese seven languages despite the different ways thatfrequency was assessed Following Johnson et al (1996)we speculate that at least some of the variance in theseseven word frequency measures is contributed by aspectsof conceptual accessibility or familiarity that all seven cul-tures have in common and may not be as direct an indexof lexical frequency (lemma andor word form frequency)as investigators sometimes assume when they use mea-sures of this kind We will return to this point later

Table 29 summarizes cross-language correlations forword length in number of syllables (lower triangle) andin number of characters (upper triangle) These findings

Table 28Correlations Among Word Frequency Measures Across Languages

EN GE SP IT BU HU CH

English ndashGerman 165Dagger ndashSpanish 156Dagger 156Dagger ndashItalian 154Dagger 155Dagger 154Dagger ndashBulgarian 142Dagger 143Dagger 137Dagger 153Dagger ndashHungarian 157Dagger 166Dagger 153Dagger 157Dagger 148Dagger ndashChinese 153Dagger 156Dagger 143Dagger 147Dagger 137Dagger 146Dagger ndashDaggerp lt 01


are even more surprising than the results for frequencygiven above All possible cross-correlations of wordlength were positive and robust ( p lt 01) ranging froma low of +20 for length in syllables between Bulgarianand Chinese to a high of +53 for length in characters be-tween Spanish and Italian The cross-language correla-tions for frequency given above were explainable in termsof cross-language commonalities in the familiarity or ac-cessibility of the pictured concepts But why should wefind any correlations over languages in word length Wesuggest that this result is a secondary effect of Zipfrsquos lawShort words tend to be reserved for more frequent con-cepts and conceptual frequency is shared across theseven languages under study here (we will return to thispoint later in regression analyses within and across lan-guages)

Finally Table 30 presents the cross-correlations amongthese languages for initial frication (a binary variable)Some of these correlations failed to reach significanceand those that did reach significance were modest insize as compared with the frequency and length coeffi-cients reported above But all of the significant correla-tions were in the same direction ranging from a low of+09 (between Spanish and Bulgarian) to a high of +33(between English and German) The real surprise here isthe fact that there were any significant correlations at allWhy should these seven languages from a range of dif-ferent language families tend to assign target nameswith the same (or similar) initial consonant Both qual-itative and quantitative assessment of the 520 target namesfor each language (details at httpwwwcrlucsdedu~aszekelyipnp7lgpnohtml) provides an answer All ofthese languages share at least a few cognates includingnot only the words that evolved from common ancestors

(eg Germanic or Romance) but also some words bor-rowed directly from English (the base language forwhich the initial set of picture stimuli was derived) It islikely that we would find such modest positive correla-tions for any phonological or phonetic coding that is sen-sitive to cognate status We chose frication not for thispurpose but because initial fricatives can affect detectionof RTs by a voice key Nevertheless the correlational re-sult is useful because it alerts us to the many factors thatcan drive cross-language similarities and differences innaming behavior

Part II Interim SummaryTo summarize the results for Part II let us return to

the four main questions it posedQuestion II-1 How much will languages differ in the

properties of target names that are elicited by the same520 pictures of common objects

We found many significant cross-linguistic differ-ences in target words elicited by our common set of pic-ture stimuli In particular target names in these sevenlanguages differed significantly across multiple param-eters of word structure (length in syllables or ortho-graphic characters syllable type frequency and com-plexity) In view of these differences it is all the moresurprising (and gratifying) that we found so many cross-linguistic universals in naming behavior in the results forPart I

Question II-2 Will picture properties (objectivecomplexity and goodness-of-depiction) be independentof each other and will they be independent of the prop-erties of target names within each language If we dofind picturendashname confounds (eg complex pictureselicit more complex names) will these be language spe-

Table 29Correlations Among Length Measures Across Languages


English ndash 145Dagger 143Dagger 140Dagger 129Dagger 138DaggerGerman 144Dagger ndash 135Dagger 143Dagger 147Dagger 138DaggerSpanish 140Dagger 130Dagger ndash 153Dagger 139Dagger 128DaggerItalian 142Dagger 138Dagger 149Dagger ndash 151Dagger 139DaggerBulgarian 130Dagger 142Dagger 135Dagger 145Dagger ndash 130DaggerHungarian 136Dagger 139Dagger 128Dagger 140Dagger 131Dagger ndashChinese 137Dagger 134Dagger 125Dagger 132Dagger 120Dagger 139Dagger ndash

NotemdashUpper triangle length measured in characters lower triangle in syllables Daggerp lt 01

Table 30Correlations of Initial Frication (01) Across Languages


English ndashGerman 133Dagger ndashSpanish 124Dagger ns ndashItalian 129Dagger 112Dagger 122Dagger ndashBulgarian 125Dagger ns 109dagger 123Dagger ndashHungarian 120Dagger ns 108 120Dagger 121Dagger ndashChinese 111dagger ns ns 109dagger 108 ns ndash

p lt 1 daggerp lt 05 Daggerp lt 01

362 BATES ET AL

cific or will they generalize across languages in thestudy

Picture properties were largely independent of targetword properties in every language although there weresome small confounds that could affect the raw correla-tions between independent and dependent variables jus-tifying their inclusion in the multivariate analyses thatwe are about to present (Part III)

Question II-3 Will we replicate Zipf rsquos law across allthe languages in the study That is will frequency be as-sociated universally with longer words And will the samelaw extend to other properties of word structure (egcanonical syllable structure)

There do appear to be some universal patterns in therelationships among word properties especially the well-established correlation between length and frequency re-ferred to as Zipf rsquos law This oft-cited relationship alsoextended to relationships between frequency and otheraspects of word structure that have been examined farless often in the psycholinguistic literature (eg wordcomplexity and syllable structure)

Question II-4 Will we find correlations in word char-acteristics across the seven languages in this study If thelink between meaning and word form is truly arbitrarywe would not expect to find cross-language correlationsin the properties of target picture names (eg frequencylength or word structure) To the extent that we do findstrong cross-language correlations in word properties (egword frequency in Language A correlates with word fre-quency in Language B) a case can be made that thesecorrelations reflect universals at a deeper level in thepicture-naming process (eg conceptual accessibility)

The most surprising finding in Part II lies in the exis-tence and magnitude of the cross-language correlationsfor all these predictor variables including weak but sig-nificant correlations for initial frication as well as largercorrelations for frequency and length These results sug-gest that lexical frequency and surface characteristics ofword structure may reflect processes at a deeper levelrooted in conceptual accessibility as well as in languagehistory This brings us to Part III where we will exam-ine the effects of lexical and pictorial properties on nam-ing behavior

Part III Effects of Word and Picture Propertieson Naming Behavior

Part III emphasizes predictorndashoutcome relationshipswithin and across languages Because there are signifi-cant correlations among many of the predictor variableswe will emphasize regression analyses within each lan-guage (eg the contributions of frequency length initialfricatives and properties of the picture stimuli to nameagreement and naming latencies) and regressions acrosslanguages (eg the extent to which frequency in Lan-guage A predicts naming behavior in Language B)Three critical questions will be addressed in Part III

Question III-1 What word and picture properties pro-vide the best predictors of naming behavior Are nameagreement and latency affected by the same variables or

by different variables Are the same patterns observedin every language or do some languages display specificpredictorndashoutcome relationships that are not universallyobserved

Question III-2 Will we observe cross-language corre-lations between predictors and outcomes For examplewill frequency or length in Language A predict RTs inLanguage B To the extent that this is the case we must(again) conclude that these word properties are contam-inated by variance at a deeper level in the naming pro-cess (eg associated variations in picture characteristicsandor conceptual accessibility) We will pursue theseparticular questions with some novel cross-linguisticvariables including other-language frequency versusown-language frequency

Question III-3 What are the contributions of language-specific word structure properties to naming behaviorFor example in languages that favor multisyllabic wordsdoes syllable structure play a role that is not observed inlanguages in which short words prevail

Correlations within languages Tables 31ndash37 areavailable at our Web site (at httpwwwcrlucsdedu~aszekelyipnp7lgpnohtml) summarizing raw correla-tions within each language between eight predictor vari-ables and six outcome variables The eight predictor variableswere goodness-of-depiction visual complexity word fre-quency initial frication length in syllables syllable typefrequency length in characters and word complexity (thelatter two are not available for Chinese) Six outcome vari-ables were chosen from the larger set of correlated nam-ing measures because they are partially dissociable andthus offer distinct perspectives on naming behavior per-centage of target name agreement (ie dominant responseor Lexical Code 1) number of types same name targetname RT naming disparity scores (where a positive scoreindicates relatively higher agreement in that language ascompared with the other six) and target RT disparityscores (where a negative score indicates relatively fasterRTs in that language as compared with the other six)

Because Part II showed substantial collinearity amongthe predictor variables (within and across languages)these correlational results must be followed up with mul-tiple regressions summarized below We refer readersdirectly to the correlation tables (Tables 31ndash37) for indi-vidual results and will restrict ourselves here to a fewsummary statements

First goodness of depiction and word frequency werethe strongest predictors of naming behavior in all lan-guages With a few exceptions items with higher pictureratings and higher word frequencies were associatedwith higher name agreement fewer alternative namesand faster RTs

Second objective visual complexity and initial frica-tion had very few effects on any of the dependent vari-ables although there were enough small and sporadic ef-fects to justify including these two measures in allregression analyses

Third effects of the various length and word structuremeasures were usually significant but quite modest and


they varied in magnitude over languages Because ofconfounds among the different length variables these re-lationships are difficult to interpret (see the comparativeanalyses of word structure below) The general tendencyis that longer more complex andor less typical wordforms (ie unusual syllable structure) are associatedwith lower agreement on target names more alternativenames more use of shared names and longer RTs

Fourth the evidence in Tables 31ndash37 also helps to char-acterize same names a dichotomous variable that cap-tures those cases in which target names are used for morethan one item These shared names tend to be signifi-cantly shorter less complex more typical in their wordstructure (ie number of syllables) and higher in fre-quency in every language This result is compatible withthe idea that speakers resort to common multipurposeword forms when they are ldquostuckrdquo on difficult items (asreported for English by Szeacutekely et al 2003)

Because of the surprising evidence uncovered in Part IIregarding cross-language correlations in frequency andlength we also decided to construct summary variablesto reflect universal trends in frequency and length Twoprincipal-component factor analyses were conducted oneon the frequency measures for all seven languages andanother on length in syllables for all seven languages(length in characters and word complexity were not usedfor this purpose because these measures are not avail-able for Chinese) In each factor analysis a single factoremerged with an eigenvalue greater than one corre-sponding to universal frequency and universal lengthrespectively Table 38 summarizes correlations of thesetwo factor scores as well as the shared measures of good-ness of depiction and visual complexity with the fivecross-language measures of naming behavior describedearlier

The results were very clear First objective visual com-plexity bore no relationship to any of the cross-languagesummary variables for naming behavior Second eventhough the ratings were obtained from English ratersonly goodness of depiction bore the largest and mostconsistent relationship to cross-language naming behav-ior (all ps lt 01) from a low of 214 with naming dis-parity (indicating less disparity over languages forhighly rated pictures) to a high of 257 with averagenaming latencies (indicating universally faster RTs forhighly rated pictures) Third the cross-linguistic fre-

quency index correlated significantly with all summaryvariables for naming performance (all ps lt 01) from alow of 223 for disparity in name agreement (less dis-parity for items that elicit high-frequency words acrosslanguages) to a high of 244 for universal naming la-tency (universally faster responses for items that elicitwords of greater average frequency) Fourth the resultsfor universal length were weaker than the results for uni-versal frequency but all correlations were significantand in the predicted direction (lower name agreementslower RTs and more cross-language disparity for itemsthat elicit words of greater average length)

Regressions of predictors on dependent variablesTo disentangle the many confounds among our predictorvariables reported in Part II we turn next to regressionanalyses of predictorndashoutcome relationships These analy-ses were conducted on the same six dependent variablesdescribed above percentage of target name agreementnumber of types the binary variable same name targetRT naming disparity and RT disparity A series of step-wise analyses were conducted on each variable withineach language using the five basic predictor measuresthat all seven languages share length in syllables initialfrication word frequency goodness of depiction and vi-sual complexity Each analysis was repeated five timespermitting each predictor to enter into the equation lastin order to assess its unique contribution to the outcomewhen the others were controlled Tables 39ndash42 summarizethe results across languages with one table for each ofthe six dependent variables

Name agreement As is summarized in Table 39 thefive variables together accounted for a significant pro-portion of the variance in name agreement ( p lt 01) in allseven languages ranging from a low of 80 in Bulgar-ian to a high of 198 in English

In all the languages the single largest unique contri-bution to name agreement came from goodness of de-piction with positive contributions on the last step rang-ing from a low of +49 in Italian to a high of +145in English (the language in which these ratings were de-rived) The direction of these findings is not surprisingName agreement tends to be higher for those pictures thatare rated as good representations of the English targetname However the fact that ratings by English speakerswork so well across widely different languages did comeas a surprise suggesting that English goodness-of-

Table 38Correlations of Independent With Dependent Cross-Language Summary Variables

Average Average Average Disparity in Disparity inName Target Number of Name Target

Agreement Latencies Types Agreement Latencies

Cross-language length factor 211dagger 118Dagger 110dagger 119Dagger 120DaggerCross-language frequency factor 125Dagger 244Dagger 227Dagger 223Dagger 235DaggerGoodness of depiction 141Dagger 257Dagger 250Dagger 214Dagger 225DaggerVisual complexity ns ns ns ns nsdaggerp lt 05 Daggerp lt 01

364 BATES ET AL

depiction ratings may reflect a relationship between thepicture and its associated concept rather than betweenthe picture and its English target name

In all of the languages except Hungarian there werealso unique positive contributions of word frequency onthe last step indicating higher name agreement for morefrequent words There were no significant contributionsof visual complexity or initial frication in any languagewhen these variables were entered on the last step

The most uneven findings for name agreement werethose involving length in syllables This variable made aunique positive contribution to name agreement in Chi-nese (+09 p lt 05) but a unique negative contributionto name agreement in Spanish (214 p lt 01) Italian(218 p lt 01) and Hungarian (215 p lt 01)whereas it made no significant contribution at all toname agreement in English German or Bulgarian Wesuspect that these variations are influenced by the largedifferences in word structure including word complex-ity and typicality of syllable structure (as describedabove) We will return to this issue shortly in more de-tailed analyses that compare all length-related wordstructure measures

Number of types As is summarized in Table 40 thefive measures together also accounted for a significantamount of variance in number of types in every language( p lt 01) and at levels consistently higher than the cor-responding values for name agreement The total vari-ance ranged from a low of 133 in Italian to a high of263 in English

In every language the strongest unique contributionto number of alternative names came from goodness ofdepiction These ratings made large and robust negative

contributions on the last step ranging in magnitude froma low of 288 in Italian to a high of 2221 in En-glish (which is of course the language in which theseratings were derived) In other words fewer alternativenames were produced in every language for those pic-tures that were judged as good representations of En-glish target names

The unique contribution of frequency to alternativenames failed to reach significance in Bulgarian or Hun-garian In the other five languages frequency contribu-tions on the final step were significant ranging from alow of 214 ( p lt 05) in Spanish to a high of 262( p lt 01) in Chinese Hence there is an (almost) univer-sal tendency to produce fewer alternative names whenthe target name is higher in frequency

There were no unique contributions of visual com-plexity or initial frication to alternative names in any lan-guage There were also very few significant effects oflength in syllables to number of types restricted to smallbut significant contributions on the last step in Spanish(+11 p lt 05) and Italian (+09 p lt 05) suggesting thatalternative names are used more often in these languagesif the target name is relatively long To some extent thismay reflect our instructions to participants who wereasked to produce the shortest name that came to mindpreferably a single word However we suspect that theseresults are in part reflections of the variations in wordstructure (including typical syllable length) explored inmore detail below

Same names The results of regression analyses onthis variable are summarized in Table 41 They arelargely compatible with the notion that same names (useof the same word for more than one picture) reflects a

Table 39Regressions of Five Major Independent Variables on Name Agreement

Predictors English German Spanish Italian Bulgarian Hungarian Chinese

Goodness of depiction 1145Dagger 169Dagger 161Dagger 149Dagger 170Dagger 166Dagger 1105DaggerVisual complexity ns ns ns ns ns ns nsFrequency 132Dagger 134Dagger 108dagger 129Dagger 108dagger ns 162DaggerLength in syllables ns ns 214Dagger 218Dagger ns 215Dagger 109daggerInitial frication 105 ns ns ns ns ns nsTotal 198Dagger 114Dagger 92Dagger 114Dagger 80Dagger 101Dagger 192Dagger

NotemdashTotal variance and percentage of unique variance accounted for by each predictor after all the othersare controlled Plusminus indicates direction of contribution p lt 1 daggerp lt 05 Daggerp lt 01

Table 40Regressions of Five Major Independent Variables on Number of Types


Goodness of depiction 2221Dagger 2130Dagger 2111Dagger 288Dagger 2136Dagger 2135Dagger 2138DaggerVisual complexity ns 105 ns ns ns ns nsFrequency 218Dagger 222Dagger 214dagger 224Dagger ns 205 262DaggerLength in syllables ns ns 111dagger 109dagger ns ns nsInitial frication ns ns ns ns ns ns nsTotal 263Dagger 167Dagger 149Dagger 133Dagger 143Dagger 147Dagger 223Dagger



strategy that speakers use when they find it difficult tocome up with a picture description However all of theseeffects are modest Table 41 shows that the five predic-tors explained significance variance in same names insix of the seven languages although these joint contri-butions were rather small They ranged from nonsignif-icance in Italian to a high of 77 in Hungarian ( p lt01)

Recall that the biggest independent contributions toname agreement and number of types came from good-ness of depiction In contrast goodness of depiction hadvery little effect on name sharing in any language (seeTable 41) suggesting that name sharing occurs not be-cause the picture itself is hard to recognize but becausethe concept it represents is difficult to encode The biggestcontributor to name sharing on the final step was wordfrequency although these results also varied over lan-guages Frequency made a small but significant positivecontribution on the last step in five languages (EnglishSpanish Italian Bulgarian and Hungarian) These val-ues ranged from a low of +08 ( p lt 05) in English (thelanguage with the lowest incidence of name sharing at45) to a high of +38 ( p lt 01) in Hungarian (whichhas the second highest incidence of name sharing at140) These results suggest that speakers tend to re-sort to certain high-frequency names when they areldquostuckrdquo for an answer on difficult items in line with pre-vious findings for English only (Szeacutekely et al 2003)

However we were surprised to find a significant con-tribution in the opposite direction for Chinese (211p lt 01) the language with the highest incidence of namesharing (at 196 close to one fifth of the picture stim-uli) That is shared words in Chinese tend to be lower infrequency when other variables are controlled This re-

sult suggests that the phenomenon of name sharing maybe more complex than we thought in our earlier work onEnglish The difference may stem from the extensive useof compounding in Chinese which involves a kind ofname sharing at the sublexical level (Lu et al 2001)That is a ldquotypicalrdquo Chinese noun is made up of two ormore sublexical elements each with an independentmeaning (although that meaning is often modified whenit is combined with other words) Each of these syllablesmay occur in many different words so that their sub-lexical frequencies are very high However for the mostcommon and productive word templates in this languagefrequencies at the whole-word level can be relatively low(reflecting a common pattern over languages for highlyproductive patterns to have high type frequency but lowtoken frequency) When Chinese speakers are trying tofind a name for especially difficult pictures they may useproductive compounding to produce an expression thatis low in frequency at the whole-word level even thoughit is made up of common (generic) elements that are easyto retrieve at the sublexical level This would be similarto a situation in which English speakers produce con-structions such as flying machine when they are unableto retrieve a target word such as helicopter

Most of the remaining effects on shared names weresmall and variable reflecting cross-linguistic variations inword structure that we will consider in more detail later

Naming latencies Table 42 presents joint and uniquecontributions of the same five variables to target-naminglatencies In contrast with the modest to moderate resultsobtained in comparable analyses of name agreementnumber of types and same names these five predictorvariables explained a substantial amount of the variancein naming time in every language Values ranged from a

Table 41Regressions of Five Major Independent Variables on Same Name


Goodness of depiction ns ns ns ns 109dagger 105 nsVisual complexity ns 207 ns ns ns 105 215DaggerFrequency 108dagger 105 117Dagger 108dagger 136Dagger 138Dagger 211daggerLength in syllables ns 211dagger 206 ns 221Dagger ns 237DaggerInitial frication ns ns ns ns ns ns nsTotal 28dagger 40Dagger 37Dagger 18 ns 72Dagger 77Dagger 272Dagger


Table 42Regressions of Five Major Independent Variables on Target Reaction Time


Goodness of depiction 2317Dagger 2228Dagger 2198Dagger 2198Dagger 2245Dagger 2206Dagger 2246DaggerVisual complexity ns 106dagger ns ns 106dagger ns nsFrequency 268Dagger 256Dagger 232Dagger 282Dagger 271Dagger 236Dagger 2101DaggerLength in syllables ns ns 110Dagger 106dagger ns 109dagger nsInitial frication ns 108dagger ns ns 107dagger ns nsTotal 433Dagger 335Dagger 263Dagger 308Dagger 324Dagger 284Dagger 396Dagger

NotemdashTotal variance and percentage of unique variance accounted for by each predictor after all the othersare controlled Plusminus indicates direction of contribution daggerp lt 05 Daggerp lt 01

366 BATES ET AL

low of 263 of the variance in Spanish to a high of433 in English

In every language the largest contributions by farcame from goodness-of-depiction ratings When theseratings were entered into the equation on the last stepthey made negative contributions ranging from 2198in Spanish and Italian ( p lt 01) to 2317 in English( p lt 01) In other words pictures that were judged byEnglish raters to be better representations of their En-glish target names were associated with substantiallyfaster reaction times in every language in the study

The unique contributions of word frequency to nam-ing times were smaller but quite robust in every lan-guage ranging from 232 in Spanish to 2101 inChinese (all ps lt 01) Hence we may conclude that thefacilitative effect of word frequency on picture-naminglatencies is a cross-language universal This is true de-spite wide differences in the methods by which word fre-quency was assessed (although as we shall see belowthe locus of this effect is far from clear)

The remaining effects were small and sporadic vary-ing from one language to another Visual complexitymade small positive contributions to naming latencies inGerman and Bulgarian (+06 p lt 05 in both cases)suggesting that more complex pictures take longer toname But the same effect did not reach significance inthe other languages Initial frication was also associatedwith significantly slower RTs in German (+08 p lt05) and Bulgarian (+09 p lt 05) but this effect alsofailed to reach significance in the other languages It isnot at all obvious to us why these two languages were theonly ones to show the anticipated slowing of voice de-tection for words that begin with white noise AlthoughGerman and Bulgarian both have relatively high propor-tions of fricative-initial words (275 and 271 re-spectively) proportions are even higher in English(281) and Hungarian (348) which did not showthis effect The answer may lie in idiosyncrasies of wordstructure that have escaped us here but are related indi-rectly to initial frication For example fricatives are usedin some languages in the formation of consonant clus-ters which are often present in words with complex de-rivational morphology and (by extension) complex se-mantics Finally there were small but significant uniquecontributions of length in syllables for only three of the

seven languages +10 ( p lt 01) in Spanish +06( p lt 05) in Italian and +09 ( p lt 05) in HungarianThese were all in the expected direction with longer RTsfor longer words But this relationship is clearly not uni-versal at least not in its simplest form That is longerwords do not always take longer to produce As we shallsee shortly results for length vary over languages de-pending on the measure of length and word structure thatwe use

Predictors of cross-language disparity For the fourdependent variables discussed so far regressions yieldedinformation about universal effects on naming behaviorFor the two measures of disparity that we turn to next re-gressions will tell us instead about the factors that leadto language differencesmdashthat is to a relative advantageor disadvantage for each language as compared with theother six The results for regressions on naming dispar-ity are presented in Table 43 the results for disparity inRTs are presented in Table 44 In contrast with the rela-tively strong and consistent f indings across Tables39ndash42 few of the results in Tables 43 and 44 reachedsignificance In other words these five predictors arenot very effective in explaining language differenceseven though they are quite effective in explaining lan-guage universals (see Tables 43 and 44 for details)

Universal predictors We also conducted a series ofuniversal regressions using four cross-language predic-tors (goodness of depiction visual complexity cross-language frequency and cross-language length) and fivecross-language summary variables for naming behavior(averages over languages for name agreement numberof types and naming latency averages of the absolutevalues for disparity scores in name agreement and nam-ing latency) Details for these analyses are summarizedin Table 45 For all five cross-linguistic dependent vari-ables the four cross-language predictor variables ac-counted together for a robust and significant proportionof the variance ranging from lows of 182 ( p lt 01) forboth of the disparity scores to a high of 49 ( p lt 01) inthe cross-language average for naming RTs

Goodness of depiction emerged once again as thestrongest universal predictor of naming behavior (not sur-prising since it was also the strongest predictor withineach individual language) This variable was associatedwith universally higher name agreement (+159) fewer

Table 43Regressions of Five Major Independent Variables on Naming Disparity Scores for Each Language


Goodness of depiction 122Dagger ns ns 211dagger ns ns nsVisual complexity ns 207 ns ns ns ns nsFrequency ns 108dagger ns 110dagger ns ns 116DaggerLength in syllables ns 105 213dagger 220Dagger ns 206 109daggerInitial frication ns ns ns ns ns ns nsTotal 128dagger 19 16 ns 254Dagger 07 ns 14 ns 234Dagger

NotemdashTotal variance and percentage of unique variance accounted for by each predictor after all the othersare controlled positive score indicates a relative advantage for that language Plusminus indicates directionof contribution p lt 1 daggerp lt 05 Daggerp lt 01


alternative names (2236) and faster RTs for the targetname (2288) Its effects on the two disparity mea-sures were much smaller but the direction of these con-tributions indicates that higher picture ratings are asso-ciated with less disparity in name agreement (217p lt 01) and less disparity in RTs (256 p lt 01)Hence the slight advantage that goodness-of-depictionratings provide for English participants is more than off-set by the overall advantage that good pictures conferacross all seven languages

Objective visual complexity had no detectable effecton any of these cross-language summary variables eventhough it did have sporadic effects within individual lan-guages Hence the robust universal effect of goodness ofdepiction derives not from the objective complexity ofthese pictures but from how well each picture representsits associated concept

Universal frequency (the average of z score frequen-cies across all seven languages) made significant uniquecontributions on the last step for all five cross-languagesummary variables resulting in higher name agreementfewer alternative names faster RTs and less disparityover languages in both agreement and RT

Finally the cross-language summary score for targetword length made a modest but significant contributionto disparity in name agreement (+11 p lt 01) indi-cating that there is more disparity in naming for thoseitems that tend to elicit longer descriptions Universallength made no other unique contributions on the laststep which suggests to us that cross-language correla-tions in length are indeed (as we suggested above) re-

flections of Zipf rsquos law When the confound between uni-versal length and universal frequency is controlled inthis regression analysis the universal effects of length are(with the single exception just noted) largely eliminatedNote that this conclusion pertains only to universalcross-language effects As we will see shortly language-specific effects of length and other aspects of word struc-ture do sometimes persist within specific languages sur-viving even after frequency is controlled

To summarize the results so far for Part III correlationand regression analyses involving picture properties in-dicate that goodness of depiction is the single best pre-dictor of naming behaviors in every language even afterother variables are controlled This is true despite thefact that goodness ratings (which were collected in En-glish) are associated with a slight advantage for Englishspeakers (as reflected in naming and RT disparity scores)In contrast objective visual complexity has very few ef-fects on naming behavior and no effect at all at the levelof cross-language summary variables It is associatedwith some small and sporadic effects within individuallanguages but the direction of those effects can vary(ie greater detail is sometimes associated with morename agreement but greater complexity is occasionallyassociated with slower RTs) In general we may con-clude that picture difficulty is a cognitive effect that iscaptured reasonably well by goodness-of-depiction rat-ings and appears to be largely independent of visualcomplexity at least as we have measured it here

With regard to the predictive value of word propertiesword frequency is the second-best predictor of naming

Table 44Regressions of Five Major Independent Variables on

Target Reaction Time Disparity Scores for Each Language


Goodness of depiction 224Dagger ns ns 109dagger ns ns nsVisual complexity 206 107 ns ns 108dagger ns nsFrequency ns ns ns 221Dagger ns ns 214DaggerLength in syllables ns ns 107 116Dagger 108dagger 106 nsInitial frication ns 115Dagger ns ns 124Dagger ns nsTotal 35Dagger 23dagger 16 ns 164Dagger 139Dagger 10 ns 121

NotemdashTotal variance and percentage of unique variance accounted for by each predictor after all the othersare controlled negative score indicates a relative advantage for that language Plusminus indicates directionof contribution p lt 1 daggerp lt 05 Daggerp lt 01

Table 45Regressions of Cross-Language Independent Variables on Cross-Language Summary Scores for

Naming Behavior Total Percentage of Variance Accounted for and Unique Contributions of EachPredictor on the Final Step


Predictors Agreement Latencies Types Agreement Latencies

Goodness of depiction 1159Dagger 2288Dagger 2236Dagger 217Dagger 256DaggerVisual complexity ns ns ns ns nsCross-language length factor ns ns ns 111Dagger nsCross-language frequency factor 136Dagger 2123Dagger 243Dagger 219Dagger 271DaggerTotal variance accounted for 1226Dagger 1490Dagger 1303Dagger 182Dagger 182DaggerDaggerp lt 01

368 BATES ET AL

behavior (after goodness of depiction) resulting in higheragreement fewer alternative names and faster RTswithin and across languages What remains to be seen iswhether these frequency effects are language specific(reflecting the ldquotruerdquo frequency of target names withineach individual language) or whether they reflect aspectsof conceptual familiarity or accessibility that hold upacross languages and are relatively independent of thetarget words themselves To pursue this issue further weturn now to some specific comparisons of own-languagefrequency and other-language frequency For these analy-ses we will restrict our attention to the two most impor-tant dependent variables in the study name agreementand latencies to produce the target name

Cross-language word frequency effects In Part IIwe uncovered some surprisingly high cross-language cor-relations among our word frequency measures despitemarked differences in the way that frequency norms werecollected in each language This result leads us to askwhether word frequency effects on naming behavior arelanguage specific or whether they are in fact interchange-able measures of a common source of variance includinguniversal variations in the familiarity and accessibilityof the concepts illustrated in our 520 pictures

To investigate this issue we constructed a novel set ofother-frequency variables For each language a factoranalysis was conducted on the frequency measures for theother six languages and the first principal componentfrom that analysis was used as a single estimate of other-language frequency For example the other-languagefrequency factor for English is based on the first princi-pal component from factor analyses of frequency mea-sures for German Spanish Italian Bulgarian Hungar-ian and Chinese In each of these factor analyses (onefor each language) only one factor emerged with aneigenvalue greater than one which testifies further to thehypothesis that a single underlying construct (eg con-ceptual accessibility) is driving these effects Table 46summarizes the cross-language correlations of word fre-quency with name agreement within and across all sevenlanguages Table 47 provides corresponding statistics forthe cross-language relationship of word frequency to nam-ing times In both tables we include correlations withuniversal frequency (the first principal component for all

seven frequency measures) as well as other-language fre-quency (the first principal component for the six remain-ing languages when the target language is excluded)

For name agreement 54 of the 63 correlations inTable 46 were significant and all of them were positiverepresenting higher name agreement for items with morefrequent names Most important for our purposes herewithin-language correlations (frequency in Language Awith name agreement in Language A) were often lowerthan cross-language correlations (frequency in Lan-guage A with name agreement in Language B) In En-glish German and Chinese the own-language correla-tion was numerically larger than both the other-languagecorrelation and the universal frequency correlation Inthe remaining four languages (Spanish Italian Bulgar-ian and Hungarian) own-language frequency actuallyresulted in numerically smaller correlations than bothuniversal frequency and other-language frequency

Corresponding results for naming times were evenstronger (Table 47) All 63 within- and across-languagecorrelations were significant (at p lt 01) all were largerthan the corresponding values for name agreement and allwere in the negative direction (indicating faster latenciesfor higher frequency items) In every language exceptChinese the other-language frequency and universal fre-quency correlations were higher than own-language frequency To some extent these results may reflect weak-nesses in the frequency measures for some of the indi-vidual languages For example Bulgarian frequency es-timates are based on subjective ratings and the SpanishItalian Bulgarian and Hungarian frequency corpora arerelatively small as compared with the corpora on whichEnglish German and Chinese frequencies are basedNevertheless the magnitude and consistency of all thecross-language frequency effects suggests to us thatpicture-naming behavior is driven at least in part byfacts about conceptual accessibility and familiarity thatare shared across languages independent of the targetname that is actually used

A more rigorous test of this hypothesis can be obtainedwith regression analyses in which the unique contribu-tions of own-language frequency and other-language fre-quency are compared Tables 48 and 49 present results ofthese regressions for name agreement and target RTs re-

Table 46Correlations of Word Frequencies With Name Agreement Within and Across Languages

Frequencies From English German Spanish Italian Bulgarian Hungarian Chinese

English 121Dagger 117Dagger 122dagger 129Dagger 116Dagger 121Dagger 109daggerGerman 108dagger 120Dagger 118Dagger 125Dagger 113Dagger 117Dagger 107Spanish 112Dagger 110Dagger 114Dagger 119Dagger ns 116Dagger 111DaggerItalian 108dagger 111Dagger 115Dagger 122Dagger 109dagger 110dagger 106Bulgarian ns 107 116Dagger 110dagger 109dagger ns 106Hungarian 108dagger 111Dagger 113Dagger 119Dagger 106 112Dagger nsChinese 113Dagger 113Dagger 120Dagger 121Dagger 110dagger 120Dagger 126DaggerOther-language frequency 111Dagger 115Dagger 122Dagger 127Dagger 112Dagger 119Dagger 109daggerUniversal frequency 114Dagger 117Dagger 122Dagger 128Dagger 113Dagger 119Dagger 112Dagger



spectively In these analyses we also controlled forgoodness of depiction visual complexity initial frica-tion and length in syllables to maximize comparabilityto the regressions reported earlier in Part III Each analy-sis was repeated six times allowing each predictor anopportunity to enter into the equation last Althoughthere were a few small changes for other variables (egsome small effects that disappeared with the addition ofa new variable) for present purposes we will restrict ourattention to the unique contributions of own- versus other-language frequency (the reader may refer to Tables 48and 49 for details regarding the other measures)

For name agreement own-language frequency madesignificant and unique positive contributions on the laststep in English (+33) German (+16) and Chinese(+65 all ps lt 01) For the remaining languages the ef-fects of own-language frequency were wiped out whenother-language frequency was entered into the equationfirst Note that English German and Chinese are the lan-guages in which frequencies are based on especially largewritten corpora which may provide more reliable esti-mates of word frequencies in their respective languages

In contrast contributions to name agreement on thelast step for other-language frequency varied over lan-guages in direction as well as in magnitude Significantpositive contributions of other-language frequency wereobserved on the last step in Spanish (+24) Italian(+19) and Hungarian (+16 all ps lt 01) In thesethree languages these were the only frequency effects ob-

served In Bulgarian no significant effects of either own-or other-language frequency were observed suggestingthat these two indices were explaining the same variancein this language It is interesting to note in this regardthat Bulgarian is the only language in which frequencywas assessed by subjective ratings The most peculiar re-sults were observed in Chinese When own-language fre-quency was entered into the equation first a small butsignificant negative relationship emerged with other-language frequency (207 p lt 05) In other words forChinese speakers name agreement was actually some-what lower for pictures that were named with high-frequency words in other languages This may reflect avariety of factors including a competition between pro-duction of Chinese terms and production of foreign loanwords that are frequent in some of the other Europeanlanguages but somewhat less accessible in Chinese

In the corresponding analyses of naming times resultswere stronger and more consistently in favor of other-language frequencies When own-language frequencywas entered into the equation last significant negativecontributions emerged for English (207) Chinese(241) and Bulgarian (221) The emergence ofBulgarian in the RT analyses is interesting since this isthe only language in which frequency was measured bysubjective ratings It suggests that frequency ratings (asopposed to frequency counts) may have a degree of ldquostay-ing powerrdquo for prediction of RTs that they do not have forname agreement in this language

Table 47Correlations of Word Frequencies From Each Language With Naming Latencies Within and Across Languages


English 234Dagger 231Dagger 237Dagger 239Dagger 228Dagger 229Dagger 233daggerGerman 230Dagger 232Dagger 235Dagger 238Dagger 227Dagger 232Dagger 229DaggerSpanish 227Dagger 228Dagger 224Dagger 233Dagger 221Dagger 226Dagger 225DaggerItalian 228Dagger 222Dagger 227Dagger 233Dagger 218Dagger 220Dagger 219DaggerBulgarian 231Dagger 225Dagger 234Dagger 232Dagger 227Dagger 222Dagger 228DaggerHungarian 227Dagger 227Dagger 227Dagger 233Dagger 222Dagger 227Dagger 225DaggerChinese 231Dagger 229Dagger 235Dagger 234Dagger 226Dagger 228Dagger 239DaggerOther-language frequency 238Dagger 235Dagger 242Dagger 245Dagger 230Dagger 233Dagger 234DaggerUniversal frequency 239Dagger 236Dagger 241Dagger 245Dagger 231Dagger 235Dagger 237Daggerdaggerp lt 05 Daggerp lt 01

Table 48Regressions on Name Agreement Within Each Language Using Both Own-Language Frequency

and Other-Language Frequency as Predictors


Goodness of depiction 1147Dagger 169Dagger 160Dagger 144Dagger 167Dagger 164Dagger 1106DaggerVisual complexity ns ns ns ns ns ns nsLength in syllables ns ns 207dagger 213Dagger ns 213Dagger 107daggerInitial frication 105 ns ns ns ns ns nsOwn-language frequency 133Dagger 116Dagger ns ns ns ns 165DaggerOther-language frequency 205 ns 124Dagger 119Dagger ns 116Dagger 207daggerTotal variance 203Dagger 115Dagger 115Dagger 133Dagger 184Dagger 117Dagger 199Dagger


370 BATES ET AL

Most important for our purposes here when other-language frequencies were entered into the RT equationlast including a control for own-language frequencysignificant unique contributions of other-language fre-quency were present in all seven languages These val-ues were all in the same direction ranging from a low of214 in Chinese to a high of 2106 in Spanish (allps lt 01) It is clear from these results that word frequencyreflects universal factors that go beyond the frequency ofspecific word forms within each language

If it is the case that these cross-language frequency ef-fects apply at a conceptual level we can make a strongprediction Facilitating effects of cross-language fre-quency (and perhaps own-language frequency) shouldbe observed not only for RTs for producing the targetname but also for the time required to produce any le-gitimate name for the same concept The number ofitems receiving Lexical Codes 2 and 3 was relativelysmall (see Table 4) and some items (in some languages)elicited no instances of morphophonological variantsandor synonyms Hence our test of this prediction restson relatively noisy data with a smaller number of itemsWith that caveat in mind we examined the correlationsof own- versus other-language frequencies with RTs forall cases in which speakers produced a response receiv-ing Lexical Codes 2 or 3 These results are presented inTable 50 for each language

Although many of these correlations failed to reachsignificance the pattern in Table 50 is very clear RTs toproduce an alternative name are faster if the target nameis high in frequency This result is most evident for other-language frequencies where it holds for f ive out of

seven languages in Lexical Code 2 (morphophonologicalvariants) and six out of seven languages on Lexical Code 3(synonyms with no morphological overlap) But the re-sults are in the same direction for own-language fre-quencies where they reach significance for three out ofseven languages on Lexical Code 2 and two of the sevenlanguages on Lexical Code 3 In fact of the 28 possiblecorrelations in Table 50 17 are significant and 24 are inthe predicted negative direction This is the opposite ofwhat we would expect if frequency effects in picturenaming occur at the level of name selection That is ahigh-frequency target name ought to inhibit its competi-tors and yet we find that any name is easier to producefor those pictures that elicit a high-frequency dominantresponse These results suggest that at least some of thevariance in the frequencyRT relationship is due to thefamiliarity and accessibility of the pictured concept in-dependently of the name that speakers finally retrieveand produce for that concept

It has been suggested to us that at least some of theother-language frequency effects reported here might bean indirect reflection of cognatesmdashthat is words thatshare similarity of form as well as meaning across twoor more languages Specifically it has been suggestedthat cross-language frequency effects may be driven bywords that are the same on a variety of levels The defi-nition of a cognate is neither straightforward nor uncon-troversial Some cases are obvious (eg television in En-glish and televisione in Italian) but others are moreopaque and can be recognized only by speakers whoknow the phonological and morphological relationshipsbetween the two languages (eg refrigerator in English

Table 49Regressions on Naming Latencies Within Each Language Using Both Own-Language Frequency



Goodness of depiction 2307Dagger 2231Dagger 2192Dagger 2178Dagger 2233Dagger 2200Dagger 2243DaggerVisual complexity ns 106dagger ns ns 107dagger ns nsLength in syllables ns ns ns ns ns 106dagger nsInitial frication ns 109Dagger ns ns 106dagger ns nsOwn-language frequency 207dagger ns ns ns 221Dagger ns 241DaggerOther-language frequency 228Dagger 234Dagger 2106Dagger 273Dagger 220Dagger 238Dagger 214DaggerTotal variance 461Dagger 368Dagger 368Dagger 382Dagger 344Dagger 322Dagger 410Dagger


Table 50Correlations of Own-Language Frequency and Other-Language Frequency With Reaction Times

for Morphophonological Variants (Lexical Code 2) and Synonyms (Lexical Code 3)


Own frequencyLexical Code 2 ns ns ns 217Dagger 221Dagger ns 222DaggerLexical Code 3 216 223Dagger ns 219Dagger ns ns 216

Other-language frequencyLexical Code 2 217dagger ns 226Dagger 223Dagger 225Dagger 208 232DaggerLexical Code 3 228Dagger 227Dagger 225Dagger 233dagger 228Dagger 221Dagger ns



vs frigorifero in Italian) Hence a comprehensive cod-ing of cognate status would require expert linguisticjudgments applied to phonemically coded words as op-posed to automatic coding procedures applied to ortho-graphic representations Nevertheless to obtain a roughestimate of between-language overlap among the targetnames elicited in this study we quantified cognates byapplying a trigram overlap search algorithm (write to usdirectly at aszekelycrlucsdedu or see our Web site athttpwwwcrlucsdedu~aszekelyipnp7lgpnohtml fordetails) For any given pair of languages a score greaterthan 1 would reflect an average of at least one trigramoverlap for each pair of target names (out of 520 pairs)a score of 25 would represent a situation in which oneshared trigram overlap was identified for approximatelyevery four pairs of target words and so forth

Table 51 summarizes the amount of trigram overlapidentified across each pairing of the seven languages inour study as well as the average overlap each languageshared with the other six combined Keeping in mind thelimits of a cognate metric that is based solely on orthog-raphy (rather than phonology) and is applied from left toright without regard to morphological structure (egroots vs inflections) or language history Table 51 doesprovide interesting results Specifically it provides ourfirst evidence for the metric of language distance that weexpected to find (but did not find) with our other nam-ing variables The highest overlaps were observed be-tween closely related languages (Italian vs Spanish andEnglish vs German) and the lowest overlaps involved

Hungarian and Chinese (which come from distinct lan-guage families) To determine whether this measure ofcognate status might affect the other-language frequencyeffects reported above we repeated the regressions onRT in Table 49 but added cognate status (amount ofinterlanguage overlap) to the list of variables that werecontrolled All of the significant results for own-languagefrequency and other-language frequency remained ex-actly the same Hence the cross-language frequency ef-fects reported above are not simply by-products of cog-nate status We reiterate however that this is an issuethat requires a more detailed inquiry Even for the crudecognate metric employed here interesting differencesare observed across languages in the relationship be-tween cognate status frequency and word structure vari-ables (details are provided on our Web site at httpwwwcrlucsdedu~aszekelyipnp7lgpnohtml)

Cross-language and language-specific effects ofword structure In contrast with word frequency wordlength appears to play a relatively minor role in namingbehavior However length is also the only factor thatcontributes consistently to cross-language disparity Be-cause length in syllables is only one aspect of language-specific word structure we will end Part III with someexploratory analyses of universal length followed bymore detailed and language-specific comparisons of dif-ferent measures of word structure

To investigate the issue of universal length we followedthe model described above for word frequencies and con-structed for each language a measure of other-language

Table 51Cognate Status Physical Similarity Between Target Names in Average Number of Overlapping

Orthographic Trigrams Across All Pairs of Languages


English ndashGerman 065 ndashSpanish 048 029 ndashItalian 056 044 101 ndashBulgarian 030 048 031 044 ndashHungarian 019 030 018 027 033 ndashChinese 002 002 002 001 002 000 ndashMean overlap with all other languages 037 036 038 046 031 021 002

NotemdashAll languages 30

Table 52Correlations of Length in Syllables From Each Language

With Name Agreement Within and Across Languages

Length in Syllables English German Spanish Italian Bulgarian Hungarian Chinese

English 209dagger ns 207 215Dagger 207 208dagger nsGerman 206 ns 208dagger 211Dagger 208dagger ns nsSpanish 207 ns 213Dagger 211Dagger ns 208dagger nsItalian 206 ns 210Dagger 218Dagger ns ns nsBulgarian ns 110dagger ns 208dagger ns 215Dagger 108daggerHungarian 210dagger ns 208dagger 215Dagger 209dagger 215Dagger nsChinese na ns ns 215Dagger ns 208dagger nsOther-language length 208dagger ns 209dagger 219Dagger ns ns nsUniversal length 209dagger ns 211Dagger 220Dagger 208 209dagger ns


372 BATES ET AL

length That is we conducted principal-component fac-tor analyses on length in syllables in the six remaininglanguages when the target language was removed Sofor example the other-language length measure for Ital-ian was the first principal component for factor analysesof length in syllables in English German Spanish Bul-garian Hungarian and Chinese In all seven of theseanalyses only one factor emerged with an eigenvaluegreater than 1 providing further evidence that some kindof unitary latent variable underlies all these measures oflength Table 52 summarizes correlations with namingagreement in each language for the individual lengthmeasures in every language for the universal length fac-tor described in Part II and for all seven other-languagelength factor scores Table 53 presents the correspondingcorrelations for naming latency The results were muchweaker than the analogous findings for word frequency(in Tables 46 and 47) both within and across languagesand the findings were especially weak for name agree-ment Nevertheless the general trend in Tables 52 and53 confirms that effects of length on naming behaviorare often just as strong across languages (eg length inLanguage A with RTs in Language B) as they are withinlanguages (eg length in Language A with RTs in Lan-guage A) Hence there does appear to be a universal fac-tor underlying the relationship between length and nam-ing behavior

In the explorations of frequency described above wecarried out regression analyses comparing the uniquecontributions of own-language frequency versus other-language frequencies We decided not to pursue such astrategy for own-language length versus other-languagelength because we already know from analyses reported

earlier (Table 45) that most of the effects of universallength are wiped out when universal frequency is con-trolled In other words the cross-linguistic componentin our length measures seems to be a by-product of thewell-known association between length and frequency(Zipf rsquos law) However there was one important excep-tion to this generalization (in Table 45) Cross-languagelength did make a significant unique contribution to dis-parity in name agreement Hence variations in wordlength can lead to language-specific advantages or dis-advantages in name retrieval

Within-language effects of word structure The re-mainder of Part III is devoted to a more detailed explo-ration of language-specific effects involving four par-tially dissociable measures of word structure length insyllables syllable type frequency length in charactersand word complexity (the latter two measures are notavailable for Chinese) Although these effects are smallthey comprise some of the most consistent findings inthis study for cross-linguistic differences in lexical re-trieval Furthermore such effects may be magnified indevelopmental andor clinical studies of word produc-tion under nonoptimal conditions Hence we believethat they merit consideration

First we conducted two rounds of regressions one forname agreement (Table 54) and the other for target nam-ing times (Table 55) examining the contribution of each ofthe four word structure variables after goodness of depic-tion visual complexity initial frication and frequency arecontrolled These analyses can tell us which of the fourpredictors (if any) does the best job of capturing length-related variance for each language when it is working asthe only length-related variable Tables 54 and 55 show

Table 53Correlations of Length in Syllables From Each Language on Naming Latencies Within and

Across Languages


English 116Dagger 118Dagger 112Dagger 119Dagger 115Dagger 113Dagger 112DaggerGerman 118Dagger 116Dagger 114Dagger 113Dagger 114Dagger 111Dagger 114DaggerSpanish 112Dagger 114Dagger 113Dagger 116Dagger 108dagger 109dagger 106Italian 119Dagger 113Dagger 116Dagger 114Dagger ns ns nsBulgarian 115Dagger 114Dagger 108dagger ns ns ns nsHungarian 113Dagger 111Dagger 109dagger ns ns 118Dagger 112DaggerChinese 114Dagger 119Dagger 108dagger 119Dagger 113Dagger 112Dagger 117DaggerOther-language length 114Dagger 114Dagger 114Dagger 122Dagger 116Dagger 111dagger 111DaggerUniversal length 116Dagger 115Dagger 115Dagger 122Dagger 115Dagger 113Dagger 113Dagger


Table 54Unique Variance Contributed by Each Length Predictor to Name Agreement on the Last Step


Length in syllables ns ns 214Dagger 218Dagger ns 215Dagger 109daggerSyllable frequency type ns ns 115Dagger 110dagger ns 124Dagger 106Length in characters 208dagger ns 213Dagger 218Dagger 207dagger 215Dagger naWord complexity 215Dagger ns 236Dagger 224Dagger ns 227Dagger na

NotemdashEach in separate regressions after goodness of depiction visual complexity word frequency and ini-tial frication are controlled Plusminus indicates direction of contribution p lt 1 daggerp lt 05 Daggerp lt 01


that all of these length-related f indings are relativelymodest in size Their primary interest lies in the consid-erable variation observed over languages in the presencedirection and magnitude of word structure effects Forexample naming behavior in German (both agreementand latency) seems to be impervious to any of these wordstructure variables after other factors are controlled ForEnglish and Bulgarian a few word structure effectsreach significance but they are weak and sporadic Incontrast almost all of the word structure variables madea contribution on the last step for Spanish Italian andHungarian

Since word structure appears to be important for atleast some of these languages we pursued the issue fur-ther by conducting regressions in which both length insyllables and frequencies of syllable type were used aslength-related predictors after other variables were con-trolled (ie goodness of depiction visual complexityfrequency and initial frication) To the best of our knowl-edge this is the first time that effects of word structuretypicality have been assessed in a cross-linguistic frame-work in any modality including picture naming Each ofthese word structure measures (which are available forall seven languages) was given an opportunity to enterinto the equation last The results (summarized in Tables56 and 57 for name agreement and RTs respectively)provide further evidence for language specificity

For name agreement these length-related measureseither canceled each other out or had no effect to beginwith in English German and Bulgarian In contrastItalian showed a small negative effect of length (208p lt 05) whereas Hungarian showed a small positive ef-fect of syllable type (+13 p lt 01) Chinese showedyet another pattern with facilitative effects on name

agreement from length (+12 p lt 01) as well as fromsyllable type (+09 p lt 05) As we have noted earlierthe helpful effects of length in Chinese probably reflectthe overwhelming predominance of disyllables in thatlanguage

For naming latency none of the languages showed in-hibitory effects of length after all the other variables (in-cluding syllable type) were controlled However facili-tative effects of syllable type were observed in Spanish(208 p lt 05) Italian (206 p lt 05) Hungarian(207 p lt 05) and especially Chinese (220 p lt01) Although these effects are small they indicate thatit is easier in some languages to retrieve words that cor-respond to the dominant word structure template This isa novel f inding that could prove important in futurecross-language studies

To push the limits of our findings on word structureone final round of analyses was conducted in which allfour word structure variables were used together in thesame equation for all languages except Chinese (forwhich length in characters and word complexity mea-sures were not available) These analyses are summa-rized in Tables 58 and 59 Once again each word struc-ture measure was given an opportunity to enter into theequation last after controlling for the other word struc-ture measures and for goodness of depiction visual com-plexity initial frication and frequency Because it wouldbe difficult for any variable to make a unique contribu-tion after this large list of factors is entered into the equa-tion we also ran a set of analyses in which the four wordstructure variables were entered together as a block onthe last step to see if their joint contribution is greater insome languages than it is in others These results are alsoincluded in Tables 58 and 59 The main message to take

Table 55Unique Variance Contributed by Each Length Predictor to Naming Latencies on the Last Step


Length in syllables ns ns 110Dagger 106dagger ns 109dagger nsSyllable frequency type ns ns 219Dagger 212Dagger 205dagger 214Dagger 220DaggerLength in characters 105dagger ns 110Dagger 106dagger 111Dagger 116Dagger naWord complexity ns ns 120Dagger ns ns 109dagger na

NotemdashEach in separate regressions after goodness of depiction visual complexity word frequency and ini-tial frication are controlled Plusminus indicates direction of contribution daggerp lt 05 Daggerp lt 01

Table 56Regressions on Name Agreement Using Both Length in Syllables and Syllable Type Frequency


Goodness of depiction 1145Dagger 170Dagger 163Dagger 149Dagger 170Dagger 170Dagger 1108DaggerVisual complexity ns ns ns ns ns ns nsInitial frication 105 ns ns ns ns ns nsWord frequency 132Dagger 135Dagger 108dagger 129Dagger 108dagger 105 163DaggerLength in syllables ns ns ns 208dagger ns ns 112DaggerSyllable type frequency ns ns ns ns ns 113Dagger 109daggerTotal 198Dagger 1116Dagger 194Dagger 1114Dagger 181Dagger 114Dagger 1201Dagger

NotemdashTotal variance and percentage of unique variance accounted for by each predictor on last stepPlusminus indicates direction of contribution p lt 1 daggerp lt 05 Daggerp lt 01

374 BATES ET AL

away from these analyses is that every language shows aunique pattern of word structure contributions to picturenaming

For name agreement word structure variables made asignificant joint contribution on the last step in EnglishSpanish Italian and Hungarian For naming latenciesthe same word structure variables made significant jointcontributions in Spanish Italian Bulgarian and Hun-garian The pattern of specific predictors was different inevery language For example German proved to be im-pervious to length-related effects separately or togetherfor either dependent variable For the remaining lan-guages the best predictor of name agreement on the laststep was length in characters for English and Bulgarianand word complexity for Spanish and Italian whereassyllable type and word complexity had equal effects forHungarian For naming latencies the best predictors onthe last step were length in characters for English andBulgarian word complexity for Spanish and syllabletype frequency for Italian and Hungarian

These detailed word structure effects suggest that con-tributions of word structure to naming behavior are lan-guage specific There are good reasons to believe thatsuch patterns will be even more pronounced in studies ofword reading word recognition and comprehensionmdashan issue worth pursuing further in studies specificallydesigned for each language

Part III Interim SummaryTo summarize the findings for Part III we return to

the three main questions posed for this section earlier on

Question III-1 What word and picture properties pro-vide the best predictors of naming behavior Are nameagreement and latency affected by the same variables or bydifferent variables Are the same patterns observed inevery language or do some languages display predictorndashoutcome relationships that are not universally observed

In all seven languages the two best predictors of nam-ing behavior were goodness of depiction (assumed to bea property of the pictures) and word frequency (typicallyassumed to be a property of the target names) These fac-tors were important for both name agreement and nam-ing latency but effects were especially strong for latencyIn contrast with these big predictors effects of other in-dependent variables were relatively small tended to varyfrom one language to another and were not consistent insize direction or the dependent variable on which theyhad their primary effect The weak contribution of initialfrication is particularly interesting in this regard becausephoneme onset properties are known to have a massiveeffect on RTs for word naming which also involves avocal response (Spieler amp Balota 1997)

Question III-2 Will we observe cross-language corre-lations between predictors and outcomes For examplewill frequency or length in Language A predict RTs inLanguage B If this is the case what will happen when wecompare other-language frequency versus own-languagefrequency and other-language length versus own-languagelength

We did indeed find robust cross-language predictorndashoutcome relationshipsmdashthat is effects of word proper-ties in Language A on naming behavior in Language B

Table 57Regressions on Target Reaction Time Using Both Length in Syllables and Syllable Type Frequency


Goodness of depiction 2316Dagger 2227Dagger 2204Dagger 2198Dagger 2248Dagger 2210Dagger 2254DaggerVisual complexity ns 106dagger ns ns 106dagger ns nsInitial frication ns 108dagger ns ns 106dagger ns nsWord frequency 266Dagger 252Dagger 232Dagger 282Dagger 271Dagger 239Dagger 2108DaggerLength in syllables ns ns ns ns ns ns nsSyllable type frequency ns ns 208dagger 206dagger ns 207dagger 220DaggerTotal 1433Dagger 1335Dagger 1271Dagger 1314Dagger 1327Dagger 1291Dagger 1415Dagger

NotemdashTotal variance and percentage of unique variance accounted for by each predictor on last stepPlusminus indicates direction of contribution daggerp lt 05 Daggerp lt 01

Table 58Regressions on Name Agreement Using All Four Length Metrics

Predictors English German Spanish Italian Bulgarian Hungarian

Goodness of depiction 1151Dagger 169Dagger 161Dagger 145Dagger 170Dagger 170DaggerVisual complexity ns ns ns ns ns nsWord frequency 112Dagger 132Dagger 112Dagger 124Dagger 108dagger nsInitial frication 107dagger ns ns ns ns nsLength in syllables ns 106 ns ns 105 nsSyllable type frequency ns ns ns ns ns 108daggerLength in characters 207dagger ns ns ns 210dagger nsWord complexity 204 ns 225Dagger 211dagger ns 208daggerFour length metrics entered together (23Dagger) (ns) (41Dagger) (30Dagger) (ns) (37Dagger)Total variance 221Dagger 120Dagger 119Dagger 126Dagger 91Dagger 123Dagger



To push this finding as far as we could we developednovel summary measures of universal frequency andother-language frequency and compared the predictivevalue of these measures with own-language frequencyIn all seven languages other-language frequencies provedto be robust predictors of RT even after own-languagefrequencies (as well as all the other word and pictureproperties) were controlled When regressions were runin the opposite direction own-language frequencies alsocontributed signif icant variance after other-languagefrequencies (and other word and picture properties) wereentered into the equation However this residual effect ofown-language frequency reached significance for nameagreement only in English Chinese and German andreached significance for RTs only in English Chineseand Bulgarian These are also the languages that have theldquobestrdquo frequency measures (English German and Chi-nese are based on especially large written corpora Bul-garian is based on subjective frequency ratings of thesort that sometimes outperform objective estimates offrequency Balota Pilotti amp Cortese 2001) All of theseresults remained when analyses were repeated control-ling for a rough (orthographic) estimate of cognate sta-tus Hence the other-language frequency effect is notcoming entirely from words that look the same acrossthese seven languages Although we are not denying arole for word form frequency in picture naming the robusteffects of other-language frequency even after own-fre-quency is controlled suggests that frequency effects in pic-ture naming are driven (at least in part) by deeper univer-sal factors such as conceptual accessibility or familiarity

To test this hypothesis further we asked whether fre-quency estimates (own language and other language)would facilitate RTs not only for the target name itself(the name used to derive own-language frequency) butalso for the time required to produce alternative names(ie synonyms and morphological variants) If own- andother-language frequencies apply at the lexical level(lemma or word form) high-frequency target namesought to inhibit production of alternative names Insteadwe found exactly the opposite Own- and other-languagefrequencies were negatively correlated with RTs not onlyfor the target name itself but also for the time requiredto produce synonyms or morphological variants These

results strongly suggest that some (although perhaps notall) of the frequency effects on picture-naming timescomes from a conceptual level that is shared across lan-guages facilitating any of the names that speakers areable to find for that concept

Similar cross-language effects were found for wordlength (ie length in Language A occasionally predictsnaming behavior in Language B) However these cross-language length results were smaller than the cross-language results reported above for own- and other-language frequency and the universal effect of lengthdisappears when the universal frequency factor is con-trolled Hence we tentatively conclude that the cross-language effects of length are by-products of Zipf rsquos law(ie the confound between frequency and length inevery language)

Question III-3 What are the contributions of language-specific word structure properties to naming behaviorFor example in languages that tend to favor multisyllabicwords does syllable structure play a role on word re-trieval that is not observed in languages in which shortwords prevail

In contrast with the universal effects of goodness ofdepiction and frequency word structure effects weresmall and variable over languages But they do exist andthey can be seen for several different aspects of wordstructure This includes some novel effects that have notbeen described in the literature on real-time languageprocessing such as the frequency of a particular syllabletemplate In short word structure does matter but it ap-pears to matter in ways that vary from one language toanother

SUMMARY CONCLUSIONS AND FUTUREDIRECTIONS

We set out to investigate similarities and differences intimed picture naming across seven languages that varyalong phonological lexical and grammatical dimen-sions that are known or suspected to play a role in lexi-cal access We have uncovered ample evidence for cross-linguistic differences but the bulk of our findingssuggest strong similarities in the factors that influenceretrieval and production of words within this paradigm

Table 59Regressions on Target Reaction Time Using Four Length Metrics


Goodness of depiction 2315Dagger 2227Dagger 2201Dagger 2197Dagger 2246Dagger 2208DaggerVisual complexity ns 106dagger ns ns 107dagger nsWord frequency 256Dagger 270Dagger 236Dagger 279Dagger 269Dagger 234DaggerInitial frication ns 105 ns ns ns nsLength in syllables ns ns ns ns 205dagger nsSyllable type frequency ns ns 208dagger 205dagger ns 206daggerLength in characters 106dagger ns ns ns 111Dagger 105daggerWord complexity ns ns 111dagger ns ns nsFour length metrics entered together (ns) (ns) (30Dagger) (12Dagger) (17Dagger) (23Dagger)Total variance 439Dagger 337Dagger 282Dagger 314Dagger 338Dagger 298Dagger

NotemdashTotal variance and percentage of unique variance accounted for by each predictor on last stepp lt 1 daggerp lt 05 Daggerp lt 01

376 BATES ET AL

Although this study was not designed to decide amongcurrent theories of word production in general the re-sults have implications for those theories and suggest anumber of directions for future research

In Part I we focused on similarities and differences inname agreement and RTs measured in several differentways The results included an across-the-board advan-tage for English which is not surprising since our 520picture stimuli were assembled in an English-speakingenvironment But significant main effects of languageremained on all variables when English was excludedfrom the analysis In all of the tables comparing resultsfor individual languages we organized the data to reflecta hypothesized continuum of language distance fromEnglish to German (two Germanic languages) to Span-ish and Italian (two Romance languages) to Bulgarian(Slavic) and then to Hungarian and Chinese (the twonon-Indo-European languages in our sample) Howeverwe found relatively little evidence in favor of a languagedistance metric Although Chinese and Bulgarian tendedto show slower RTs andor lower name agreement noneof the seven languages was an obvious outlier on any ofour behavioral measures

Correlations among the dependent variables were alsoquite similar within individual languages including thefinding that number of alternative names predicts nam-ing latencies after percentage of agreement is controlledThis result is compatible with theories of naming that as-sume competition among alternative names prior to lex-ical selection

All of the major dependent variables were signifi-cantly correlated across languages suggesting that hardor easy items tend to be hard or easy for everyone Theseconclusions were also supported by analyses usingcross-language summary variables (eg average nameagreement z scores average RT z scores or averagenumber of alternative names) In fact correlations usingthese summary variables were higher than the corre-sponding correlations within any individual languagessuggesting that we gained power and reliability by aver-aging our results If relationships among dependent vari-ables were driven in large measure by language-specificdetails this should not have occurred

In our view the most significant result in Part I is thefinding that cross-language correlations are much higherfor RTs than for any other measure of naming behaviorThis result underscores the value of a timed picture-naming paradigm which appears to be sensitive to uni-versal stages andor processes that are not detected withoff-line naming measures

In Part II we focused on similarities and differences inthe target names (dominant response or Lexical Code 1)that emerged in each language Target word characteris-tics proved to be largely independent of picture charac-teristics (ie visual complexity and goodness of depic-tion) but there were enough small confounds to warrantinclusion of both word and picture characteristics in allthe subsequent analyses As was expected we foundsubstantial cross-language differences in word structure

(length frequency of different word templates and ini-tial frication) However there were also significant cor-relations across languages on almost all measures of tar-get word characteristics including frequency lengthand initial frication In other words target names that arefrequent long or fricative initial in one language tend tohave the same characteristics in other languages as wellThe small frication effect reflects the presence of cog-nates across all of these languages (including unrelatedlanguages such as English Hungarian and Chinese)and it is likely that similar correlations would appear ifwe had looked at another class of initial phonemes (ini-tial frication was used because initial white noise canslow down detection of RTs by a voice key)

The lengthndashfrequency confound within languages re-flects a well-known tendency for languages to assignshorter words to more frequent concepts (ie Zipf rsquoslaw) Our cross-language results show that Zipf rsquos lawmanifests itself both within and across languages re-flecting (we propose) cross-linguistic similarities in thefrequency familiarity andor accessibility of the con-cepts illustrated by our picture stimuli We quantifiedthis notion further by developing cross-language sum-mary measures for both length and frequency includinguniversal frequency and universal length (based on thefirst principal component of factor analyses across allseven languages) together with some novel measures ofother-language frequency and other-language length foreach individual language

In Part III we examined the relationships betweenpredictor and outcome variables both within and acrosslanguages Goodness-of-depiction ratings proved to bethe best predictor of all naming measures in every lan-guage although these ratings did work slightly better forEnglish (the language in which they were compiled) Incontrast our objective measure of visual complexity hadalmost no effect on any aspect of naming behaviorHence we may conclude that goodness of depiction re-flects cognitive factors in picture naming (and pictureevaluation) that are largely independent of objective vi-sual complexitymdashat least as we have measured it here(see Laws Leeson amp Gale 2002 for evidence that Eu-clidean overlap can slow down naming times under someconditions) Ratings had been obtained from a separatesample of English participants who were asked to ratehow well each picture illustrates the empirically derivedEnglish target name The fact that these ratings predictednaming behavior in seven different languages suggeststhat our English raters were evaluating how well eachpicture illustrates the intended concept rather than thetarget word itself

Word frequency was the second-best predictor ofnaming behavior replicating a well-known result in thepicture-naming literature (eg Oldfield amp Wingfield1964 1965) More surprising were the strong correla-tions that appeared across languages between frequencyand naming behavior (eg Chinese frequencies predictnaming behavior in Spanish) In fact regression analy-ses showed that other-language frequencies made a sig-


nificant contribution to RTs when own-language fre-quencies were controlled in every language A similarresult was observed for within- versus cross-language ef-fects of length although this result appears to be a by-product of the findings for frequency (evidence for theuniversality of Zipf rsquos law)

These effects of cross-language frequency (and theshadow effects of cross-language length) may be the mostimportant findings in this cross-linguistic study becausethey force us to reconsider the locus of frequency effectsin theories of picture naming It is typically assumed thatfrequency effects are lexical in nature reflecting thenumber of times that a speakerlistener is likely to haveencountered a particular word form However in theirreview of the picture-naming literature Johnson et al(1996) warn us that this assumption is not always war-ranted because word frequency may index the frequencyfamiliarity andor accessibility of the concept underly-ing that word A number of studies have shown that fre-quency effects are not observed in picture processing un-less lexical access is required (eg Caramazza CostaMiozzo amp Bi 2001 Griffin 2001 Kroll amp Potter 1984Meyer Sleiderink amp Levelt 1998) This would suggestthat frequency effects are not obligatory at the level ofpicture decoding per se So where do these effects takeplace

It has been suggested to us that our other-language fre-quency effects might constitute true word form frequencyeffects via an indirect route Frequency norms differ insize coverage reliability and the context in which theywere acquired Even within a single language one fre-quency measure sometimes ldquotrumpsrdquo another contribut-ing significant variance to RTs after the first measure iscontrolled From this point of view it is possible that ourother-language frequency measures are serving as an es-pecially reliable index of word form frequency withineach individual language We did indeed find that goodfrequency measures based on large corpora (in EnglishGerman and Chinese) andor subjective ratings (Bul-garian) survived controls for other-language frequencywhereas weaker frequency measures based on smallercorpora (Italian and Hungarian) disappeared when other-language frequency was controlled This suggests thatthe differential reliability of frequency measures is in-deed playing a role But other aspects of our findingssuggest that something else is going on as well Specifi-cally we found that target word frequency was associ-ated not only with faster RTs for the target word itselfbut also with faster RTs for alternative names (synonymsand morphological variants of the target word) This re-sult suggests that the frequency effects are occurring notat the level of word selection (where a high-frequencytarget should prove to be a fierce competitor slowingdown RTs for alternative names) but at some point dur-ing the process by which a picture is recognized wellenough to narrow down the lexical search These concep-tual effects also seem to percolate up to other levels ofword form as well reflected in the correlations amongfrequency and word structure variables (eg length syl-

lable type complexity) At the very least our resultssuggest that frequency effects are not restricted to theword form level There may be at least two kinds of fre-quency variance in play conceptual accessibility (whichmakes it easier to find any name for the picture in anylanguage) and word form frequency (which is sensitiveto variations in the reliability of the corpus from whichthe frequency measure is taken) We are currently tryingto sort out these two possibilities by conducting cross-linguistic studies of word reading for comparison withthe results presented here on picture naming using thesame target words Insofar as word reading tends to begoverned primarily by form-based factors whereas pic-ture naming is affected to a greater degree by conceptualfactors variations in the existence and magnitude of other-language frequency effects between reading and picturenaming may provide useful information about the pro-posed mix of form-based and concept-based frequencyeffects

Although we did find cross-linguistic effects of wordstructure the existence and magnitude of these effectsvaried markedly from one language to another Reliableand independent contributions of length and or wordcomplexity were observed in some languages (especiallyItalian and Spanish) but were weak or nonexistent in oth-ers (eg German) We also found independent effects oflength in syllables and the relative frequency of differentsyllable templates (eg reliably faster RTs for disylla-bles in Chinese) Although all of these word structure ef-fects were relatively small (as compared with the largeand universal effects of goodness of depiction and fre-quency) they merit further investigation Among otherthings the small language-specific effects observed innormal adults may be magnified in speakerlistenerswho are working under a disadvantage including youngchildren in the course of language learning older adultsworking with diminished processing resources and in-dividuals with congenital or acquired linguistic cogni-tive andor perceptualndashmotor deficits

On the basis of these results we are expanding our in-quiries in several new directions including cross-languagecomparisons of action and object naming (eg Szeacutekelyet al 2003 for English) cross-language comparisons ofword reading and auditory word repetition (using thesame target words as those elicited in the present study)and developmental studies (eg DrsquoAmico et al 2001Roe et al 2000) We are also expanding our database toinclude new predictor variables This will include a cross-linguistic exploration of AoA effects using adult ratingsof AoA as well as objective measures taken from stud-ies of early language development (see DrsquoAmico et al2001 Iyer et al 2001) Preliminary results suggest thatcross-language AoA effects may be even larger than thecross-language frequency effects reported here raisingfurther concerns about conceptual versus lexical inter-pretations of predictorndashoutcome relationships We arealso expanding our database within each language topermit a more detailed exploration of word structure ef-fects using measures that are appropriate for each indi-

378 BATES ET AL

vidual language This would include further studies ofChinese in which we take advantage of the rich sub-lexical structure of compound words (which make upmore than 80 of the words in this language) Withinand across languages the same materials are now beingapplied across clinical populations (including adultaphasics and children with language impairments) com-plemented by new initiatives in human brain imaging

Finally we are now in a better position to conduct thekinds of cross-linguistic studies of context effects thatmotivated us to obtain cross-linguistic norms in the firstplace The main effects of language on naming variablesthat we reported here (eg Tables 3 and 4) suggest thatsome languages are ldquoslowerrdquo than others However inview of the intense communicative pressures under whichall languages evolve it is quite unlikely that such a resultwill generalize when lexical access is studied in a dis-course context Every language has developed its ownset of tradeoffs For example a historical move towardgreater length is often compensated for by contextualcues that anticipate the point at which a word can be rec-ognized (eg gender agreement cues that short-circuitthe need for Italian listeners to wait for the gender mark-ings that fall at the end of all nouns) The kinds of normsthat we have developed and explored in the present studyfor lexical access out of context can be used to assesscross-linguistic variations in the contributions of seman-tic and grammatical cues to nature and timing of wordrecognition and production when these processes areembedded in a phrase sentence or discourse context

REFERENCES

Abbate M S amp La Chapelle N B (1984a) Pictures please A lan-guage supplement Tucson AZ Communication Skill Builders

Abbate M S amp La Chapelle N B (1984b) Pictures please An ar-ticulation supplement Tucson AZ Communication Skill Builders

Alameda J R amp Cuetos F (1995) Diccionario de frecuencias delas unidades linguumliacutesticas del castellano [Frequency dictionary forlexical items in Castilian Spanish] Oviedo Servicio de Publicacionesde la Universidad de Oviedo Available at httpwwwswanacukcalscalsresvlibraryjsg99ahtm

Alcock K J amp Ngorosho D (in press) Grammatical noun classagreement processing in Kiswahili Language amp Speech

Baayen R H Piepenbrock R amp Gulikers L (1995) The CELEXLexical Database (Release 2) [CD-ROM] Philadelphia Universityof Pennsylvania Linguistic Data Consortium [Distributor]

Bachoud-Levi A Dupoux E Cohen L amp Mehler J (1998)Where is the length effect A cross-linguistic study of speech pro-duction Journal of Memory amp Language 39 331-346

Balota D A amp Chumbley J I (1985) The locus of word-frequencyeffects in the pronunciation task Access andor production Journalof Memory amp Language 24 89-106

Balota D A Pilotti M amp Cortese M J (2001) Subjective fre-quency estimates for 2938 monosyllabic words Memory amp Cogni-tion 29 639-647

Bates E Burani C DrsquoAmico S amp Barca L (2001) Word readingand picture naming in Italian Memory amp Cognition 29 986-999

Bates E DrsquoAmico S Jacobsen T Szeacutekely A Andonova EDevescovi A Herron D Lu C-C Pechmann T Pleh CWicha N Y Y Federmeier K D Gerdjikova I Gutierrez GHung D Hsu J Iyer G Kohnert K Mehotcheva T

Orozco-Figueroa A Tzeng A amp Tzeng O (in press) Timedpicture naming in seven languages Psychonomic Bulletin amp Review

Bates E Devescovi A Pizzamiglio L DrsquoAmico S amp Hernan-dez A (1995) Gender and lexical access in Italian Perception ampPsychophysics 57 847-862

Bates E Devescovi A amp Wulfeck B (2001) PsycholinguisticsA cross-language perspective Annual Review of Psychology 52369-398

Bates E Federmeier K Herron D Iyer G Jacobsen T Pech-mann T DrsquoAmico S Devescovi A Wicha N Orozco-Figueroa A Kohnert K Gutierrez G Lu C-C Hung DHsu J Tzeng O Andonova E Gerdjikova I MehotchevaTSzeacutekely A amp Pleacuteh C (2000) Introducing the CRL InternationalPicture-Naming Project (CRL-IPNP) Center for Research in Lan-guage Newsletter (Vol 12 No 1) La Jolla University of CaliforniaSan Diego

Bates E amp MacWhinney B (1989) Functionalism and the compe-tition model In B MacWhinney amp E Bates (Eds) The crosslin-guistic study of sentence processing (pp 3-76) New York Cam-bridge University Press

Bentrovato S Devescovi A DrsquoAmico S amp Bates E (1999) Theeffect of grammatical gender and semantic context on lexical accessin Italian Journal of Psycholinguistic Research 28 677-693

Caramazza A (1997) How many levels of processing are there inlexical access Cognitive Neuropsychology 14 177-208

Caramazza A Costa A Miozzo M amp Bi Y C (2001) Thespecific-word frequency effect Implications for the representationof homophones in speech production Journal of Experimental Psy-chology Learning Memory amp Cognition 27 1430-1450

Carr T H McCauley C Sperber R D amp Parmelee C M(1982) Words pictures and priming On semantic activation con-scious identification and the automaticity of information process-ing Journal of Experimental Psychology Human Perception amp Per-formance 8 757-777

Cattell J (1886) The time to see and name objects Mind 11 63-65Chen S (1997) Intra-lexical nounndashverb dissociations Evidence from

Chinese aphasia Unpublished doctoral dissertation University ofSouthern California

Chen S amp Bates E (1998) The dissociation between nouns andverbs in Brocarsquos and Wernickersquos aphasia Findings from ChineseAphasiology 12 5-36

Chinese Knowledge Information Processing Group (1997) Acad-emia Sinica Balanced Corpus (WWW Version 30) Taipei Instituteof Information Science Academia Sinica Available at httpwwwsinicaedutwftms-binkiwish

Cohen J MacWhinney B Flatt M amp Provost J (1993)PsyScope An interactive graphic system for designing and control-ling experiments in the psychology laboratory using Macintosh com-puters Behavior Research Methods Instruments amp Computers 25257-271

Cutting J C amp Ferreira V S (1999) Semantic and phonologicalinformation flow in the production lexicon Journal of ExperimentalPsychology Learning Memory amp Cognition 25 318-344

Cycowicz Y M Friedman D Rothstein M amp Snodgrass J G(1997) Picture naming by young children Norms for name agree-ment familiarity and visual complexity Journal of ExperimentalChild Psychology 65 171-237

Damasio H Grabowski T J Tranel D Ponto L L B HichwaR D amp Damasio A R (2001) Neural correlates of naming actionsand of naming spatial relations NeuroImage 13 1053-1064

DrsquoAmico S Devescovi A amp Bates E (2001) Picture naming andlexical access in Italian children and adults Journal of Cognition ampDevelopment 2 71-105

Davidoff J amp Masterson J (1996) The development of picturenaming Differences between nouns and verbs Journal of Neurolin-guistics 9 69-84

Dell G S (1990) Effects of frequency and vocabulary type on phono-logical speech errors Language amp Cognitive Processes 5 313-349

De Mauro T Mancini F Vedovelli M amp Voghera M (1993)


Lessico di frequenza dellrsquoitaliano parlato [Frequency lexicon of spo-ken Italian] Milano ETASLIBRI

Dockrell J Messer D amp George R (2001) Patterns of namingobjects and actions in children with word-finding difficulties Lan-guage amp Cognitive Processes 16 261-286

Druks J amp Shallice T (2000) Selective preservation of namingfrom description and the ldquorestricted preverbal messagerdquo Brain ampLanguage 72 100-128

Dunn Lloyd M amp Dunn Leota M (1981) Peabody Picture Vo-cabulary TestndashRevised Circle Pines MN American Guidance Ser-vice

Federmeier K D amp Bates E (1997) Contexts that pack a punchLexical class priming of picture naming Center for Research in Lan-guage Newsletter (Vol 11 No 2) La Jolla University of CaliforniaSan Diego

Fuumlredi M amp Kelemen J (1989) A mai magyar nyelv szeacutepproacutezaigyakorisgi szoacutetra 1965ndash1977 [Prose frequency dictionary of con-temporary Hungarian language 1965ndash1977] Budapest AkadeacutemiaiKiadoacute

Glaser W R (1992) Picture naming Cognition 42 61-105Goodglass H (1993) Understanding aphasia San Diego Academic

PressGriffin Z (2001) Gaze durations during speech reflect word selec-

tion and phonological encoding Cognition 82 B1-B14Hernandez A E Dapretto M Mazziotta J amp Bookheimer S

(2001) Language switching and language representation in SpanishndashEnglish bilinguals An fMRI study NeuroImage 14 510-520

Hernandez A E Martinez A amp Kohnert K (2000) In searchof the language switch An f MRI study of picture naming in SpanishndashEnglish bilinguals Brain amp Language 73 421-431

Hillert D amp Bates E (1996) Morphological constraints on lexicalaccess Gender priming in German (Tech Rep 9601) La Jolla Uni-versity of California San Diego Center for Research in Language

Humphreys G W Riddoch M J amp Quinlan P T (1988) Cascadeprocesses in picture identification Cognitive Neuropsychology 567-103

Iyer G Saccuman C Bates E amp Wulfeck B (2001) A study ofage-of-acquisition (AoA) ratings in adults Center for Research inLanguage Newsletter (Vol 13 No 2) La Jolla University of Cali-fornia San Diego

Jacobsen T (1999) Effects of grammatical gender on picture and wordnaming Evidence from German Journal of Psycholinguistic Re-search 28 499-514

Jescheniak J-D amp Levelt W J M (1994) Word frequency effectsin speech production Retrieval of syntactic information and ofphonological form Journal of Experimental Psychology LearningMemory amp Cognition 20 824-843

Johnson C J Paivio A amp Clark J M (1996) Cognitive compo-nents of picture naming Psychological Bulletin 120 113-139

Kaplan E Goodglass H amp Weintraub S (1983) Boston NamingTest Philadelphia Lee amp Febiger

Kelly M (1992) Using sound to solve syntactic problems The roleof phonology in grammatical category assignments PsychologicalReview 99 349-364

Kohnert K J (2000) Lexical skills in bilingual school-age childrenCross-sectional studies in Spanish and English Unpublished doc-toral dissertation University of California San Diego and San DiegoState University

Kohnert K J Bates E amp Hernandez A E (1999) Balancingbilinguals Lexical-semantic production and cognitive processing inchildren learning Spanish and English Journal of Speech Languageamp Hearing Research 42 1400-1413

Kohnert K J Hernandez A E amp Bates E (1998) Bilingual per-formance on the Boston Naming Test Preliminary norms in Spanishand English Brain amp Language 65 422-440

Kroll J amp Potter M (1984) Recognizing words pictures and con-cepts A comparison of lexical object and reality decisions Journalof Verbal Learning amp Verbal Behavior 23 39-66

Lachman R (1973) Uncertainty effects on time to access the internallexicon Journal of Experimental Psychology 99 199-208

Lachman R Shaffer J amp Hennrikus D (1974) Language and

cognition Effects of stimulus codability name-word frequency andage of acquisition on lexical reaction time Journal of Verbal Learn-ing amp Verbal Behavior 13 613-625

Laws K Leeson V amp Gale T (2002) The effect of ldquomaskingrdquo onpicture naming Cortex 38 137-148

Levelt W J M (1989) Speaking From intention to articulationCambridge MA MIT Press

Levelt W J M Roelofs A amp Meyer A S (1999) A theory of lex-ical access in speech production Behavioral amp Brain Sciences 22 1-38 69-75

Lu C-C Bates E Hung D Tzeng O Hsu J Tsai C-H ampRoe K (2001) Syntactic priming of nouns and verbs in ChineseLanguage amp Speech 44 437-471

MacWhinney B (1987) The competition model In B MacWhinney(Ed) Mechanisms of language acquisition (pp 249-308) HillsdaleNJ Erlbaum

Meyer A Sleiderink A amp Levelt W J M (1998) Viewing andnaming objects Eye movements during noun phrase productionCognition 66 B25-B33

Murtha S Chertkow H Beauregard M amp Evans A (1999)The neural substrate of picture naming Journal of Cognitive Neuro-science 11 399-423

Nation K Marshall C M amp Snowling M J (2001) Phonolog-ical and semantic contributions to childrenrsquos picture-naming skillEvidence from children with developmental reading disorders Lan-guage amp Cognitive Processes 16 241-259

Oldfield R C amp Wingfield A (1964) The time it takes to name anobject Nature 202 1031-1032

Oldfield R C amp Wingfield A (1965) Response latencies in nam-ing objects Quarterly Journal of Experimental Psychology 17 273-281

Paivio A (1971) Imagery and verbal processes New York HoltRinehart amp Winston

Roe K Jahn-Samilo J Juarez L Mickel N Royer I amp Bates E(2000) Contextual effects on word production A lifespan studyMemory amp Cognition 28 756-765

Roelofs A (1992) A spreading-activation theory of lemma retrievalin speaking Cognition 42 107-142

Sanfeliu M C amp Fernandez A (1996) A set of 254 SnodgrassndashVanderwart pictures standardized for Spanish Norms for nameagreement image agreement familiarity and visual complexity Be-havior Research Methods Instruments amp Computers 28 537-555

Schmitt B M Muumlnte T F amp Kutas M (2000) Electrophysiolog-ical estimates of the time course of semantic and phonological en-coding during picture naming Psychophysiology 3 473-484

Schriefers H Meyer A S amp Levelt W J M (1990) Exploringthe time course of lexical access in language production Picture-word interference studies Journal of Memory amp Language 29 86-102

Snodgrass J G amp Vanderwart M (1980) A standardized set of260 pictures Norms for name agreement familiarity and visual com-plexity Journal of Experimental Psychology Human Learning ampMemory 6 174-215

Snodgrass J G amp Yuditsky T (1996) Naming times for the Snod-grass and Vanderwart pictures Behavior Research Methods Instru-ments amp Computers 28 516-536

Spieler D H amp Balota D A (1997) Bringing computational mod-els of word naming down to the item level Psychological Science 8411-416

Szeacutekely A amp Bates E (2000) Objective visual complexity as a vari-able in studies of picture naming Center for Research in LanguageNewsletter (Vol 12 No 2) La Jolla University of California SanDiego

Szeacutekely A DrsquoAmico S Devescovi A Federmeier K HerronDIyer G Jacobsen T Areacutevalo A Vargha A amp Bates E(2003) Timed action and object naming Manuscript submitted forpublication

Szeacutekely A DrsquoAmico S Devescovi A Federmeier K HerronDIyer G Jacobsen T amp Bates E (in press) Timed picture nam-ing Extended norms and validation against previous studies Behav-ior Research Methods Instruments amp Computers

380 BATES ET AL

van Turennout M Hagoort P amp Brown C (1997) Electrophys-iological evidence on the time course of semantic and phonologicalprocesses in speech production Journal of Experimental Psychol-ogy Learning Memory amp Cognition 23 787-806

van Turennout M Hagoort P amp Brown C (1998) Brain activ-ity during speaking From syntax to phonology in 40 millisecondsScience 280 572-574

van Turennout M Hagoort P amp Brown C (1999) The timecourse of grammatical and phonological processing during speakingEvidence from event-related brain potentials Journal of Psycholin-guistic Research 28 649-676

Wicha N Y Y Bates E Moreno E amp Kutas M (2000) Gram-matical gender modulates semantic integration of a picture in a Span-ish sentence Psychophysiology 37 (Suppl 1) S104

Wingfield A (1967) Perceptual and response hierarchies in objectidentification Acta Psychologica 26 216-226

Wingfield A (1968) Effects of frequency on identification and nam-ing of objects American Journal of Psychology 81 226-234

Woehrmann M H Hessenflow A N Kettmann D amp YoshimuneP H (1994) Image alchemy Freemont CA Handmade Software

Zipf G K (1965) Human behavior and the principle of least effort Anintroduction to human ecology New York Hafner

(Manuscript received April 25 2002revision accepted for publication September 9 2002)


Within and across languages word forms bear little orno resemblance to the concepts that they represent Thesame furry four-legged animal is called dog in EnglishHund in German perro in Spanish and cane in ItalianEven within the same language family (eg Romance)there are often striking differences in names for the sameconcept (eg butterfly is mariposa in Spanish farfallain Italian and papillon in French) Despite these well-known cross-language differences in the shape of wordsit is generally assumed that people access their mentallexicon in the same way in every natural language on thebasis of a universal architecture for word comprehensionand production This belief rests crucially on the as-sumption that the relationship between meaning andform is arbitrary Word forms have no effect on the pro-cess by which speakers move from concept to lexical se-lection and meanings have no effect on the shape ofwords or on the kind of processing required to map a se-lected concept onto its associated sound

In the present study we will investigate some of thesefundamental assumptions by presenting what is (to the bestof our knowledge) the first large-scale cross-linguisticstudy of timed picture naming (cf Bachoud-Levi DupouxCohen amp Mehler 1998) We will investigate universaland language-specific contributions to naming behavioracross seven languages (English German Spanish Ital-

ian Bulgarian Hungarian and Mandarin Chinese) thatvary markedly along lexical and grammatical parametersknown or believed to affect word retrieval This cross-linguistic strategy will shed light on several theoreticalquestions Will we find significant differences across lan-guages in ease of naming andor in the particular items thatare hard or easy to name Will such differences reflectcultural variations in the accessibility of concepts andorlinguistic variations in the properties of target namesWill cross-linguistic differences reflect the linguistic dis-tance that separates our seven languages (eg GermanicRomance and Slavic variants of Indo-European UralicSino-Tibetan) Will we find a universal set of predictorndashoutcome relationships that hold in every language (egeffects of frequency on naming latency) andor will wefind language-specific profiles in the relationship be-tween word structure and naming behavior (eg effectsof syllable structure in Chinese that have no equivalentin English) Finally our results will offer insights intothe levels at which word and picture properties have theireffects Specifically for those word properties that areassumed to apply only at the level of word form andorat the level of lemma selection (Caramazza 1997 Dell1990 Jescheniak amp Levelt 1994) we should find within-language correlations (eg English word frequenciespredict English reaction times [RTs] ) but no corre-sponding cross-language correlations (eg Chineseword frequencies should not predict English RTs) In con-trast if supposed word form effects are confounded by vari-ance from a deeper conceptual level (eg word frequencyis contaminated by conceptual familiarityaccessibility)we may f ind significant cross-language effects evenafter own-language properties are controlled (eg Chi-nese word frequency will predict English RTs even afterEnglish frequencies are controlled)

Timed picture naming was a good candidate for thiscross-linguistic strategy for several reasons Picturenaming is a widely used technique for the study of lexicalaccess (for reviews see Glaser 1992 Johnson Paivio ampClark 1996) and timed picture naming is one of the firstparadigms ever used to study real-time language pro-cessing from early studies by Cattell (1886) through thepioneering work of Wingfield (1967 1968) Lachman andcolleagues (Lachman 1973 Lachman Shaffer amp Henn-rikus 1974) and Snodgrass and colleagues (Sanfeliu ampFernandez 1996 Snodgrass amp Vanderwart 1980 Snod-grass amp Yuditsky 1996) On-line (timed) and off-line(untimed) picture-naming methods are also widely usedin studies of brain-injured patients (for reviews see Chenamp Bates 1998 Druks amp Shallice 2000 Goodglass1993 Murtha Chertkow Beauregard amp Evans 1999) andboth normal and language-impaired children (CycowiczFriedman Rothstein amp Snodgrass 1997 DrsquoAmico De-vescovi amp Bates 2001 Davidoff amp Masterson 1996Dockrell Messer amp George 2001 Nation Marshall ampSnowling 2001) More recently the neural basis of lex-ical retrieval has been explored in normal adults usingeither overt or covert picture naming in functional brainimaging (Damasio et al 2001 Hernandez Dapretto

This multifaceted collaborative project has required great effort bymany people The first four authors (listed in alphabetical order) tookprimary organization responsibility for the cross-language project as awhole including development of the scoring system organization of thedatabase andor statistical analysis and interpretation of cross-languageresults in the final stages of the project The next seven authors (in al-phabetical order) each took primary responsibility for launching andorsupervising data collection within each of their respective languagesites (Devescovi in collaboration with DrsquoAmico for Italian Herron incollaboration with Bates for English Pechmann in collaboration withJacobsen for German and Pleacuteh in collaboration with Szeacutekely for Hun-garian) The remaining authors (in alphabetical order) made importantcontributions at various stages of the project within their own-languagegroup The project as a whole was supported by a grant to EB (NIDCDR01 DC00216 ldquoCross-linguistic studies of aphasiardquo) Developmentaloffshoots of the project have been partially supported by Grants NINDSP5022343 and NIDCD P50 DC01289 The Bulgarian portion of theproject was also supported by a grant from the James McDonnell Foun-dation and support for travel during the final phases of manuscriptpreparation was provided by a grant from NATO (LSTCLG977502ldquoComparative Studies of Language Development and Reading in FourLanguagesrdquo) Each of the associated institutions has provided space fordata collection and personnel time We are grateful to the School of Hu-manities at the Universidad Autoacutenoma de Baja CaliforniandashTijuana forallowing us to use their subject pool and to Lourdes Gavaldoacuten de Bar-reto for her generous assistance Our thanks to Maria Polinsky for ad-vice on the classif ication of the seven languages (as outlined inTable 1) to Meiti Opie and Ayse Saygin for assistance with manuscriptpreparation to Robert Buffington and Larry Juarez for technical assis-tance in development of the norming procedures to Luigi Pizzamiglioand Nina Dronkers for assistance in locating picture stimuli and to Vic-tor Ferreira and David Balota for careful reading of and critical com-ments on the manuscript itself Any flaws and limitations that remainare not the responsibility of these good friends and advisors Corre-spondence concerning this article should be addressed to E Bates Cen-ter for Research in Language 0526 University of California SanDiego La Jolla CA 92093-0526 (e-mail batescrlucsdedu)

346 BATES ET AL










Tab

le 1

Cha

ract

eris

tics

of

the

Sev

en P

arti

cip

atin

g L

angu

ages

Lin

guis

tic F

eatu

res

Eng

lish

Ger

man

Ital

ian

amp S

pani

shB

ulga

rian

Hun

gari

anC

hine

se

Indo

-Eur

opea

nye

sye

sye

sye

sno

no

Lan

guag

e fa

mily

Ger

man

ic (

stro

ngG

erm

anic

Rom

ance

Slav

icU

ralic

Sino

-Tib

etan

infl

uenc

e of

Rom

ance

)(F

inno

-Ugr

ic)

Wor

d or

der v

aria

tions

lo

wm

ediu

mhi

ghhi

ghm

ediu

mm

ediu

m

Infl

ectio

nal m

orph

olog

ysp

arse

rich

rich

rich

rich

none

Om

issi

on o

f con

stitu

ents

in f

ree-

stan

ding

sen

tenc

esno

t per

mitt

edno

t per

mitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t and

obj

ect c

anbe

om

itted

Use

of c

ompo

undi

ngdagger

high

high

low

med

ium

med

ium

high

(gt80

o

f all

wor

ds)

Lex

ical

am

bigu

ity fo

r wor

ds o

ut o

f con

text

high

esp

ecia

lly f

orm

oder

ate

esp

ecia

llylo

w fo

r al

l cat

egor

ies

low

for

all c

ateg

orie

slo

w fo

r al

l cat

egor

ies

high

for n

ouns

ver

bs

noun

s an

d ve

rbs

for n

ouns

and

ver

bsdu

e to

infl

ectio

nal

due

to in

flec

tiona

ldu

e to

infl

ectio

nal

and

func

tion

wor

dsm

arki

ngm

arki

ngm

arki

ng

Mor

phol

ogic

al re

gula

rity

one

regu

lar a

ndm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le r

egul

ar

lexi

cal r

egul

arity

onl

ym

ultip

le ir

regu

lar

irre

gula

r an

dir

regu

lar

and

irre

gula

r an

dir

regu

lar

and

degr

ees

offo

rms

for p

lura

l and

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

prod

uctiv

ity in

past

tens

epr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

sco

mpo

und

form

atio

n

Gra

mm

atic

al c

ues

to w

ord

iden

tityDagger

form

cla

ssfo

rm c

lass

gen

der

form

cla

ss g

ende

rfo

rm c

lass

gen

der

form

cla

ss c

ase

form

cla

ss n

omin

alca

secl

assi

fier

s

Pros

odic

cue

s to

wor

d id

entit

yst

ress

stre

ssst

ress

stre

ssst

ress

vow

el h

arm

ony

lexi

cal t

one

Ort

hogr

aphy

and

ort

hogr

aphi

c re

gula

rity

alph

abet

ic h

ighl

yal

phab

etic

som

eal

phab

etic

hig

hly

alph

abet

ic h

ighl

yal

phab

etic

hig

hly

logo

grap

hic

one

op

aque

irre

gula

rir

regu

lari

ties

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

sylla

ble

map

s to

man

y ch

arac

ters

Ref

ers

to t

he n

umbe

r of

dif

fere

nt o

rder

s of

sub

ject

ver

b a

nd o

bjec

t th

at a

re p

ossi

ble

in t

he s

poke

n la

ngua

ge

dagger Ref

ers

to w

ords

that

are

com

pose

d of

oth

er f

ree-

stan

ding

wor

ds (

cont

ent w

ords

and

or

func

tion

wor

ds)

Dagger Am

ong

gram

mat

ical

cue

s to

wor

d id

enti

ty ldquo

form

cla

ss c

uesrdquo

ref

er to

wor

ds o

r ph

rase

s th

at r

elia

bly

dist

ingu

ish

betw

een

noun

s v

erbs

and

oth

er g

ram

mat

ical

cla

sses

as

in th

e di

ffer

-en

ce b

etw

een

ldquoI w

ent t

o th

e da

ncerdquo

ver

sus

ldquoI w

ant t

o da

nce

rdquo S

tudi

es h

ave

show

n th

at s

uch

form

cla

ss c

ues

like

gen

der

case

and

nom

inal

cla

ssif

iers

can

ldquopr

imerdquo

(fa

cilit

ate

or in

hibi

t) r

etri

eval

of

wor

dsfr

om d

iffe

rent

gra

mm

atic

al c

lass

es

348 BATES ET AL





















350 BATES ET AL





METHOD
























Source No




1 7 4 - 2 1 5


352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL










346 BATES ET AL










Tab

le 1

Cha

ract

eris

tics

of

the

Sev

en P

arti

cip

atin

g L

angu

ages

Lin

guis

tic F

eatu

res

Eng

lish

Ger

man

Ital

ian

amp S

pani

shB

ulga

rian

Hun

gari

anC

hine

se

Indo

-Eur

opea

nye

sye

sye

sye

sno

no

Lan

guag

e fa

mily

Ger

man

ic (

stro

ngG

erm

anic

Rom

ance

Slav

icU

ralic

Sino

-Tib

etan

infl

uenc

e of

Rom

ance

)(F

inno

-Ugr

ic)

Wor

d or

der v

aria

tions

lo

wm

ediu

mhi

ghhi

ghm

ediu

mm

ediu

m

Infl

ectio

nal m

orph

olog

ysp

arse

rich

rich

rich

rich

none

Om

issi

on o

f con

stitu

ents

in f

ree-

stan

ding

sen

tenc

esno

t per

mitt

edno

t per

mitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t and

obj

ect c

anbe

om

itted

Use

of c

ompo

undi

ngdagger

high

high

low

med

ium

med

ium

high

(gt80

o

f all

wor

ds)

Lex

ical

am

bigu

ity fo

r wor

ds o

ut o

f con

text

high

esp

ecia

lly f

orm

oder

ate

esp

ecia

llylo

w fo

r al

l cat

egor

ies

low

for

all c

ateg

orie

slo

w fo

r al

l cat

egor

ies

high

for n

ouns

ver

bs

noun

s an

d ve

rbs

for n

ouns

and

ver

bsdu

e to

infl

ectio

nal

due

to in

flec

tiona

ldu

e to

infl

ectio

nal

and

func

tion

wor

dsm

arki

ngm

arki

ngm

arki

ng

Mor

phol

ogic

al re

gula

rity

one

regu

lar a

ndm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le r

egul

ar

lexi

cal r

egul

arity

onl

ym

ultip

le ir

regu

lar

irre

gula

r an

dir

regu

lar

and

irre

gula

r an

dir

regu

lar

and

degr

ees

offo

rms

for p

lura

l and

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

prod

uctiv

ity in

past

tens

epr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

sco

mpo

und

form

atio

n

Gra

mm

atic

al c

ues

to w

ord

iden

tityDagger

form

cla

ssfo

rm c

lass

gen

der

form

cla

ss g

ende

rfo

rm c

lass

gen

der

form

cla

ss c

ase

form

cla

ss n

omin

alca

secl

assi

fier

s

Pros

odic

cue

s to

wor

d id

entit

yst

ress

stre

ssst

ress

stre

ssst

ress

vow

el h

arm

ony

lexi

cal t

one

Ort

hogr

aphy

and

ort

hogr

aphi

c re

gula

rity

alph

abet

ic h

ighl

yal

phab

etic

som

eal

phab

etic

hig

hly

alph

abet

ic h

ighl

yal

phab

etic

hig

hly

logo

grap

hic

one

op

aque

irre

gula

rir

regu

lari

ties

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

sylla

ble

map

s to

man

y ch

arac

ters

Ref

ers

to t

he n

umbe

r of

dif

fere

nt o

rder

s of

sub

ject

ver

b a

nd o

bjec

t th

at a

re p

ossi

ble

in t

he s

poke

n la

ngua

ge

dagger Ref

ers

to w

ords

that

are

com

pose

d of

oth

er f

ree-

stan

ding

wor

ds (

cont

ent w

ords

and

or

func

tion

wor

ds)

Dagger Am

ong

gram

mat

ical

cue

s to

wor

d id

enti

ty ldquo

form

cla

ss c

uesrdquo

ref

er to

wor

ds o

r ph

rase

s th

at r

elia

bly

dist

ingu

ish

betw

een

noun

s v

erbs

and

oth

er g

ram

mat

ical

cla

sses

as

in th

e di

ffer

-en

ce b

etw

een

ldquoI w

ent t

o th

e da

ncerdquo

ver

sus

ldquoI w

ant t

o da

nce

rdquo S

tudi

es h

ave

show

n th

at s

uch

form

cla

ss c

ues

like

gen

der

case

and

nom

inal

cla

ssif

iers

can

ldquopr

imerdquo

(fa

cilit

ate

or in

hibi

t) r

etri

eval

of

wor

dsfr

om d

iffe

rent

gra

mm

atic

al c

lass

es

348 BATES ET AL





















350 BATES ET AL





METHOD
























Source No




1 7 4 - 2 1 5


352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL











Tab

le 1

Cha

ract

eris

tics

of

the

Sev

en P

arti

cip

atin

g L

angu

ages

Lin

guis

tic F

eatu

res

Eng

lish

Ger

man

Ital

ian

amp S

pani

shB

ulga

rian

Hun

gari

anC

hine

se

Indo

-Eur

opea

nye

sye

sye

sye

sno

no

Lan

guag

e fa

mily

Ger

man

ic (

stro

ngG

erm

anic

Rom

ance

Slav

icU

ralic

Sino

-Tib

etan

infl

uenc

e of

Rom

ance

)(F

inno

-Ugr

ic)

Wor

d or

der v

aria

tions

lo

wm

ediu

mhi

ghhi

ghm

ediu

mm

ediu

m

Infl

ectio

nal m

orph

olog

ysp

arse

rich

rich

rich

rich

none

Om

issi

on o

f con

stitu

ents

in f

ree-

stan

ding

sen

tenc

esno

t per

mitt

edno

t per

mitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t can

be

omitt

edsu

bjec

t and

obj

ect c

anbe

om

itted

Use

of c

ompo

undi

ngdagger

high

high

low

med

ium

med

ium

high

(gt80

o

f all

wor

ds)

Lex

ical

am

bigu

ity fo

r wor

ds o

ut o

f con

text

high

esp

ecia

lly f

orm

oder

ate

esp

ecia

llylo

w fo

r al

l cat

egor

ies

low

for

all c

ateg

orie

slo

w fo

r al

l cat

egor

ies

high

for n

ouns

ver

bs

noun

s an

d ve

rbs

for n

ouns

and

ver

bsdu

e to

infl

ectio

nal

due

to in

flec

tiona

ldu

e to

infl

ectio

nal

and

func

tion

wor

dsm

arki

ngm

arki

ngm

arki

ng

Mor

phol

ogic

al re

gula

rity

one

regu

lar a

ndm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le re

gula

rm

ultip

le r

egul

ar

lexi

cal r

egul

arity

onl

ym

ultip

le ir

regu

lar

irre

gula

r an

dir

regu

lar

and

irre

gula

r an

dir

regu

lar

and

degr

ees

offo

rms

for p

lura

l and

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

ldquoin-

betw

eenrdquo

(par

tially

prod

uctiv

ity in

past

tens

epr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

spr

oduc

tive)

form

sco

mpo

und

form

atio

n

Gra

mm

atic

al c

ues

to w

ord

iden

tityDagger

form

cla

ssfo

rm c

lass

gen

der

form

cla

ss g

ende

rfo

rm c

lass

gen

der

form

cla

ss c

ase

form

cla

ss n

omin

alca

secl

assi

fier

s

Pros

odic

cue

s to

wor

d id

entit

yst

ress

stre

ssst

ress

stre

ssst

ress

vow

el h

arm

ony

lexi

cal t

one

Ort

hogr

aphy

and

ort

hogr

aphi

c re

gula

rity

alph

abet

ic h

ighl

yal

phab

etic

som

eal

phab

etic

hig

hly

alph

abet

ic h

ighl

yal

phab

etic

hig

hly

logo

grap

hic

one

op

aque

irre

gula

rir

regu

lari

ties

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

tran

spar

ent

regu

lar

sylla

ble

map

s to

man

y ch

arac

ters

Ref

ers

to t

he n

umbe

r of

dif

fere

nt o

rder

s of

sub

ject

ver

b a

nd o

bjec

t th

at a

re p

ossi

ble

in t

he s

poke

n la

ngua

ge

dagger Ref

ers

to w

ords

that

are

com

pose

d of

oth

er f

ree-

stan

ding

wor

ds (

cont

ent w

ords

and

or

func

tion

wor

ds)

Dagger Am

ong

gram

mat

ical

cue

s to

wor

d id

enti

ty ldquo

form

cla

ss c

uesrdquo

ref

er to

wor

ds o

r ph

rase

s th

at r

elia

bly

dist

ingu

ish

betw

een

noun

s v

erbs

and

oth

er g

ram

mat

ical

cla

sses

as

in th

e di

ffer

-en

ce b

etw

een

ldquoI w

ent t

o th

e da

ncerdquo

ver

sus

ldquoI w

ant t

o da

nce

rdquo S

tudi

es h

ave

show

n th

at s

uch

form

cla

ss c

ues

like

gen

der

case

and

nom

inal

cla

ssif

iers

can

ldquopr

imerdquo

(fa

cilit

ate

or in

hibi

t) r

etri

eval

of

wor

dsfr

om d

iffe

rent

gra

mm

atic

al c

lass

es

348 BATES ET AL





















350 BATES ET AL





METHOD
























Source No




1 7 4 - 2 1 5


352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL










348 BATES ET AL





















350 BATES ET AL





METHOD
























Source No




1 7 4 - 2 1 5


352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL




















350 BATES ET AL





METHOD
























Source No




1 7 4 - 2 1 5


352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL



























Source No




1 7 4 - 2 1 5


352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL










352 BATES ET AL









































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL































354 BATES ET AL




























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL























356 BATES ET AL
























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL


























358 BATES ET AL































Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL

























Monosyllables 236 118 12 4 53 90 61(457) (227) (23) (08) (102) (173) (117)

Disyllables 201 259 228 197 279 266 364(387) (498) (438) (379) (537) (512) (700)

3 syllables 63 104 179 208 140 110 86(121) (200) (344) (400) (269) (212) (165)

4 syllables 18 36 75 76 35 41 8(35) (69) (144) (146) (67) (79) (15)

5 syllables 2 2 19 24 5 9 1(04) (04) (37) (46) (10) (17) (02)

6 syllables 0 1 6 6 6 3 0(00) (02) (12) (12) (12) (06) (00)

7 syllables 0 0 1 4 2 0 0(00) (00) (02) (08) (04) (00) (00)

8 syllables 0 0 0 1 0 1 0(00) (00) (00) (02) (00) (02) (00)

360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL










360 BATES ET AL




























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL



























362 BATES ET AL































364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL























364 BATES ET AL

































366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL

























366 BATES ET AL































368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL




























368 BATES ET AL


























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL
























370 BATES ET AL

































372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL


























372 BATES ET AL







Across Languages























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL

























374 BATES ET AL































376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL























376 BATES ET AL


















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL

















378 BATES ET AL



REFERENCES

















































































380 BATES ET AL



























































380 BATES ET AL










Timed picture naming in seven languages Languages.pdf · reto for her generous assistance. Our...

Documents

Transcript of Timed picture naming in seven languages Languages.pdf · reto for her generous assistance. Our...