12
What ‘‘Mice Trap’’ tells us about the mental lexicon q Carolyn J. Buck-Gengler, a,c, * Lise Menn, b,c and Alice F. Healy a,c a Department of Psychology, University of Colorado at Boulder, Boulder, CO 80309-0345, USA b Department of Linguistics, University of Colorado at Boulder, Boulder, CO 80309-0295, USA c Institute of Cognitive Science, University of Colorado at Boulder, Boulder, CO 80309-0344, USA Accepted 4 December 2003 Available online 8 February 2004 Abstract Level ordering has proven inadequate as a morphological theory, leaving unexplained the experimental results taken to support it as a component of innate grammaryoung childrenÕs acceptance of irregular plurals in English compounds. The present study demonstrates that these results can be explained by slower access to the grammatically preferred singulars of irregular nouns when compounds are created on-line from plural stimuli. Experiments on English noun–noun compound production and on production of either singular or plural forms from the same or opposite form confirmed that more irregular than regular plurals were used in compounds, and showed that producing irregular singulars from plurals was slower than producing regular singulars. Plural re- sponses were also slower when cue and required response number differed. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Inflectional morphology; Irregular noun plurals; Noun–noun compounds; Innateness; Level ordering 1. Introduction The issue of regularity within inflectional morphology has been an especially important area of research and debate about the nature of language representation and even the innateness of language. A central aspect of this debate concerns whether the linguistic description in- volving a strict separation between regular and irregular word formation is supported experimentally as a psy- chological property of the mental lexicon. The present study reexamines an early experiment (Gordon, 1985) on regular and irregular English noun morphology, which has often been cited as support for dual mechanism models (rules for regulars; rote memory or an associative network for irregulars), innateness of grammar, and level ordering of morphology (e.g., Ki- parsky, 1982). In the experiments presented here, we examine the possibility of an alternative, on-line pro- cessing explanation for GordonÕs results. The regular/irregular distinction for English noun morphology can be summarized as follows. Most nouns take the regular orthographic -(e)s plural, phonologi- cally /-s, -z, - iz/. However, a number of nouns do not fit that pattern. Mass/non-count nouns have no plural form (e.g., corn). Irregular nouns include nouns with zero plural (e.g., sheep), nouns of foreign origin that use foreign plurals (e.g., stimulus/stimuli, analysis/analyses, q The research presented in this paper represents the main findings from the first three experiments of the first authorÕs dissertation, which was conducted under the joint supervision of the other two authors. More details may be found in the dissertation. We thank James Kole, Erica Wohldmann, and Katrina Raybun for their assistance in scoring the responses in Experiment 2, Jon Roberts and Ernest Mross for programming assistance, David Underwood for preparing the picture stimuli, and Holly Krech Thomas for recording the spoken words used in Experiment 3. We also thank two anonymous reviewers for insightful comments and suggestions, George Figgs for searching CHILDES for noun–noun compounds, Charles Judd for statistics advice, and Brian MacWhinney for advice on comparing frequencies. This research was supported in part by Army Research Institute Contracts DASW01-99-K-0002 and DASW01-03-K-0002 and Army Research Office Grant DAAG55-98-1-0214 to the University of Colorado. This research was also supported in part by a Student Research Award from the Institute of Cognitive Science (University of Colorado at Boulder). * Corresponding author. Fax: 1-303-492-8895. E-mail address: [email protected] (C.J. Buck- Gengler). 0093-934X/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.bandl.2003.12.001 Brain and Language 90 (2004) 453–464 www.elsevier.com/locate/b&l

What “Mice Trap” tells us about the mental lexicon

Embed Size (px)

Citation preview

Brain and Language 90 (2004) 453–464

www.elsevier.com/locate/b&l

What ‘‘Mice Trap’’ tells us about the mental lexiconq

Carolyn J. Buck-Gengler,a,c,* Lise Menn,b,c and Alice F. Healya,c

a Department of Psychology, University of Colorado at Boulder, Boulder, CO 80309-0345, USAb Department of Linguistics, University of Colorado at Boulder, Boulder, CO 80309-0295, USA

c Institute of Cognitive Science, University of Colorado at Boulder, Boulder, CO 80309-0344, USA

Accepted 4 December 2003

Available online 8 February 2004

Abstract

Level ordering has proven inadequate as a morphological theory, leaving unexplained the experimental results taken to support it

as a component of innate grammar—young children�s acceptance of irregular plurals in English compounds. The present study

demonstrates that these results can be explained by slower access to the grammatically preferred singulars of irregular nouns when

compounds are created on-line from plural stimuli. Experiments on English noun–noun compound production and on production

of either singular or plural forms from the same or opposite form confirmed that more irregular than regular plurals were used in

compounds, and showed that producing irregular singulars from plurals was slower than producing regular singulars. Plural re-

sponses were also slower when cue and required response number differed.

� 2003 Elsevier Inc. All rights reserved.

Keywords: Inflectional morphology; Irregular noun plurals; Noun–noun compounds; Innateness; Level ordering

1. Introduction

The issue of regularity within inflectional morphology

has been an especially important area of research and

qThe research presented in this paper represents the main findings

from the first three experiments of the first author�s dissertation, whichwas conducted under the joint supervision of the other two authors.

More details may be found in the dissertation. We thank James Kole,

Erica Wohldmann, and Katrina Raybun for their assistance in scoring

the responses in Experiment 2, Jon Roberts and Ernest Mross for

programming assistance, David Underwood for preparing the picture

stimuli, and Holly Krech Thomas for recording the spoken words used

in Experiment 3. We also thank two anonymous reviewers for

insightful comments and suggestions, George Figgs for searching

CHILDES for noun–noun compounds, Charles Judd for statistics

advice, and Brian MacWhinney for advice on comparing frequencies.

This research was supported in part by Army Research Institute

Contracts DASW01-99-K-0002 and DASW01-03-K-0002 and Army

Research Office Grant DAAG55-98-1-0214 to the University of

Colorado. This research was also supported in part by a Student

Research Award from the Institute of Cognitive Science (University of

Colorado at Boulder).* Corresponding author. Fax: 1-303-492-8895.

E-mail address: [email protected] (C.J. Buck-

Gengler).

0093-934X/$ - see front matter � 2003 Elsevier Inc. All rights reserved.

doi:10.1016/j.bandl.2003.12.001

debate about the nature of language representation and

even the innateness of language. A central aspect of this

debate concerns whether the linguistic description in-

volving a strict separation between regular and irregular

word formation is supported experimentally as a psy-

chological property of the mental lexicon.

The present study reexamines an early experiment(Gordon, 1985) on regular and irregular English noun

morphology, which has often been cited as support for

dual mechanism models (rules for regulars; rote memory

or an associative network for irregulars), innateness of

grammar, and level ordering of morphology (e.g., Ki-

parsky, 1982). In the experiments presented here, we

examine the possibility of an alternative, on-line pro-

cessing explanation for Gordon�s results.The regular/irregular distinction for English noun

morphology can be summarized as follows. Most nouns

take the regular orthographic -(e)s plural, phonologi-

cally /-s, -z, -�iz/. However, a number of nouns do not fit

that pattern. Mass/non-count nouns have no plural

form (e.g., corn). Irregular nouns include nouns with

zero plural (e.g., sheep), nouns of foreign origin that use

foreign plurals (e.g., stimulus/stimuli, analysis/analyses,

454 C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464

datum/data), and a small number of nouns that havesynchronically idiosyncratic plurals based on earlier

systematic changes that have been lost over time (e.g.,

mouse/mice, foot/feet). Semi-regular nouns (mostly end-

ing in /f/) add the -(e)s but change the final consonant

voicing (e.g., wife/wives, leaf/leaves); they were not used

in the present study or in Gordon�s (1985).Regularity/irregularity has consequences extending

beyond the choice of inflected form. In English noun–noun compound formation, the first noun must usually

be singular, but plurals of irregular nouns are more

acceptable in that position (e.g., Gordon, 1985; Kipar-

sky, 1982; Pinker, 1994, 1999). So, even when talking

about more than one rat or toy, one would not say *rats

catcher or *toys box. However, mice catcher is far more

acceptable. This dichotomy has been explained within

the theory of level ordering of morphology. Level or-dering is a feature of the theory of Lexical Phonology,

which is a generative attempt to formally capture and

unify many recurring patterns in the morphology and

phonology of the world�s languages. ‘‘The motivation

for setting up an independent stratum [level] in Lexical

Phonology is usually a convergence of morphological

and phonological facts. . .; a set of affixes. . .undergoesone set of phonological rules but not another’’ (Kaisse &Shaw, 1985, p. 10). These (morphological and phono-

logical) facts in turn result from events in the history of a

language over millennia.

In level ordering, assembly of words proceeds at

several ‘‘levels.’’ At Level 1, a base form (for almost all

English nouns, the singular) or another memorized form

(such as the irregular plural) is retrieved from the mental

lexicon. Unproductive derivations, especially those thatimpose phonological or stress changes in the stem, are

also found at Level 1. At Level 2, compounds are

formed and productive derivational affixes, such as un-,

-less, and -ly, are added. At Level 3, after compound

formation, regular affixes like the regular plural are

added. The normative English pattern that disprefers

*rats trap is explained by saying that the regular plural

rats is created too late (at Level 3) to be placed inside acompound (at Level 2). However, irregular plurals, be-

ing retrieved from memory at Level 1, are easily incor-

porated during compound formation, and thus should

be optionally allowed. Problems with exceptions to the

rule (e.g., civil rights commission, public works depart-

ment) have been explained by appeals to semantic and

syntactic factors (Kiparsky, 1982; Pinker, 1999).

Lexical Phonology is an account of linguistic struc-ture; it makes no claims about storage versus on-line

processing, nor about the formation of novel words and

compounds. Although level ordering has been shown to

have serious limitations for English and other languages

to which it has been applied (Bauer, 1990; Booij, 1992,

2002; Fabb, 1988; Haskell, MacDonald, & Seidenberg,

2003; Hay, 2002; Ramscar, 2002), some developmental

psycholinguists (Clahsen, 1999; Marcus, Brinkmann,Clahsen, Wiese, & Pinker, 1995) continue to rely on it or

its close relatives, perhaps for lack of an attractive

generative alternative.

1.1. Gordon (1985) and children’s compound formation in

English

Gordon (1985) explored compound formation in alanguage production experiment with English-speaking

children aged 3–5. The children were introduced to a

puppet with a voracious appetite (Cookie Monster).

Various objects that Cookie Monster might eat were

shown to each child. Each time the child was first asked

to name the object, then was asked, ‘‘What do you call

someone who eats X?’’ The children were trained to

answer with a noun–noun compound in the form of X-eater. The training items were English mass nouns,

which have no plurals in their basic senses. For each test

noun, the singular and plural of the noun were elicited in

turn by showing the child first one and then several of

the objects represented by the noun, and asking the child

to name what he or she saw. Whichever form of plural a

child gave (e.g., for the plural of mouse: mice, mouses,

mices) was used in the compound elicitation. The maintest words were five irregular nouns (mice, teeth, geese,

men, and feet) semantically matched with five regular

nouns (rats, beads, ducks, babies, and hands).

Gordon (1985) found that for regular nouns (and for

irregular nouns that were overregularized, e.g., mouses),

close to 100% of the compounds were formed with the

singular form for X in X-eater (e.g., rat-eater). On the

other hand, 90% of the irregular nouns for which a childhad produced the correct irregular plural received ir-

regular plural responses (e.g., mice-eater). In other

words, if the plural was ‘‘regular’’ (whether correct or

not), the singular was produced, but if the plural was

irregular, the plural was (almost always) produced.

These results matched the prediction of level ordering

that use of the plural would be optional for irregulars.

Gordon attributed the high rate of plural response forirregulars (90% rather than the random 50% rate that

the term ‘‘optional’’ suggests) to the fact that the chil-

dren always heard the plural of the first noun of the

compound just before they had to produce the com-

pound. Indeed, the linguistic notion of optionality

makes no predictions about the relative frequency of

choice among options or the factors that may influence

that choice.Gordon (1985) claimed that these results were strong

support for the innateness of the underlying grammar

(level ordering), because he also asserted, based on an

examination of compounds found in the Ku�cera and

Francis (1967) word list, that children only rarely hear

irregular plurals in compounds. Thus, children would

not have the opportunity to learn that irregular plurals

1 The orthographic change to ‘‘ies’’ for regular nouns ending in ‘‘y’’

is not likely to cause those regulars special problems: Phonologically,

which is how the words are first learned, there is no difference;

orthographically, this change is regular for this subclass of nouns.

Whether this subclass is treated differently remains to be tested

empirically.

C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464 455

are optionally allowed in compounds. He argued that ifa pattern cannot be learned from the input, then it must

be an ‘‘innate structural property of the lexicon’’ (Gor-

don, 1985, p. 73): Because children show a distinction

between regular and irregular nouns in this structure,

level ordering must be innate. This conclusion about

innateness has been repeated frequently elsewhere

(Clahsen, 1999; Hoff, 2001; Pinker, 1991, 1994, 1999).

However, we suggest that these findings are not dueto the operation of an innate grammar—a claim which is

already dubious in the face of purely linguistic counter-

evidence to the theory of level ordering already men-

tioned. Rather, they can be accounted for by on-line

processing demands interacting with a ‘‘soft’’ constraint:

The first element of a compound is singular, regardless

of whether the semantics has stimulated the retrieval of

a plural referent (e.g., toys, cookies). Thus, if a com-pound such as cookie jar is called for in a context where

plural cookies has been mentioned (‘‘Can we have some

cookies?’’), the primed cookies must be inhibited in fa-

vor of the singular before responding (‘‘Yes, they are in

the cookie jar.’’). We hypothesize that it is harder to

access the singular of an irregular noun from its plural

than it is to access the singular of a regular from its

plural. We will show that this is the case for adults;assuming that processing is similar in adults and chil-

dren, our study will also account for Gordon�s (1985)

findings. The ‘‘constraint’’ formulation does not require

invoking innate knowledge, because young children

have plenty of examples of this first-element-singular

pattern in their early years (e.g., toy box, raisin box, jelly

bean jar, cookie jar) and could well have adduced the

constraint by age 3, the youngest age that Gordon tes-ted. Children have also heard violations of that pattern

when the first noun has no singular (e.g., pluralia tantum

nouns such as clothes and eyeglasses); the CHILDES

database (MacWhinney, 2000) contains such forms in

language addressed to young children. Nouns like

clothes are not irregular, but forms like clothes-hanger

and glasses-case show that the first-element-singular

constraint is violable.As mentioned, in Gordon�s (1985) experiment the

children were always primed with the plural of the first

noun of the compound, right before they had to produce

the compound. On-line processes would therefore need

to retrieve the singular and/or inhibit the repetition of

the plural form to construct the noun–noun compound.

Either of these procedures would be more difficult for

irregular nouns—difficult enough that sometimes theseprocesses would not succeed before the output was

made. This difficulty should show up in two ways: First,

the already observed predilection of both adults and

children to produce (or at least accept) the irregular

plural noun in compounds more often than the regular

plural (predicted by both Gordon�s and our explana-

tions), and second, a longer time to produce such a

compound with a singular first noun when given an ir-regular plural as input (predicted by our explanation).

Indeed, for words that have a non-transparent rela-

tionship between their singular and plural forms, it

should be harder in general to produce the singular

when the plural is already in mind (or, for that matter,

to produce the plural when the singular is in mind).1

In summary, we make the following predictions: (a)

Irregular plurals will be produced in compounds moreoften than regular plurals; (b) it will take longer to

produce such a compound with a singular first noun

when given an irregular plural as input than it would

with other inputs; and (c) it will be harder in general to

produce either a singular or a plural form of a noun

when the opposite form is already in mind, especially for

words that have a formally non-transparent singular–

plural relationship.Three experiments examined these predictions. The

first two involved eliciting noun–noun compounds in

English; in Experiment 2 response times were also col-

lected. The third experiment used a task in which either

the singular or the plural of a noun was evoked by

means of a picture (a simple line drawing), and a num-

ber cue was given to signal which of the two forms

(singular or plural) was supposed to be produced;sometimes the evoked form and the required response

matched in number, and sometimes they did not match.

This task went beyond forming compounds in that both

singular and plural forms were explicitly required, rather

than just the singular (which is implicitly required in

English compound formation); thus participants� speedand accuracy changing the form in both directions, ra-

ther than just one, could be tested.

2. Experiments 1 and 2

The purpose of Experiments 1 and 2 was to test the

hypothesis that in the compound formation task it is

harder (and thus takes more time) to produce a singular

form immediately after seeing the plural form of an ir-regular noun than it is after seeing the singular form of

an irregular noun or either form of a regular noun. The

pattern of responses (both experiments) and the time to

respond (Experiment 2) were examined to address this

question. These experiments, though similar to that of

Gordon (1985), differed in several ways, one of which

was that whereas Gordon only prompted the children

with plurals in the X-eater questions, in Experiments 1

Table 1

Stimulus nouns (singular and plural forms) used in Experiments 1 and 2, by type

Irregular noun Regular noun

Semantic match Form match

Singular Plural Singular Plural Singular Plural

mouse micea rat ratsa nail nails

tooth teetha bead beadsa tape tapes

foot feeta hand handsa hat hats

goose geesea duck ducksa bell bells

man mena baby babiesa letter letters

louse lice fly flies knight knights

child children doll dolls chain chains

ox oxen horse horses ax axes

woman women monkey monkeys watch watches

fungus fungi fern ferns frog frogs

a From Gordon (1985).

3 See Gordon (1985) for a discussion of how semantically matched

nouns were selected.4

456 C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464

and 2 both singulars and plurals were used in the

prompts.

2.1. Method

2.1.1. Participants

Twelve members of the Boulder community partici-

pated in Experiment 1 as volunteers; 24 University ofColorado students participated in Experiment 2 in par-

tial fulfillment of introductory psychology course re-

quirements. All were native speakers of English.

2.1.2. Materials

2.1.2.1. Target nouns. Three types of target nouns were

used (Table 1): irregular, semantically matched regular,

and form matched regular. We used the five irregularnouns and their semantic matches from Gordon (1985),

and added five more irregular nouns and their regular

semantic matches for greater generalizability across

items. A non-semantically related noun was also mat-

ched with each irregular noun. This third set of nouns

was chosen such that each regular noun started with a

phoneme that had acoustic onset characteristics similar

to the first phoneme of the corresponding irregularnoun. These restrictions were imposed in anticipation of

Experiment 2, in which the participants pronounced the

responses aloud and the response times were based on

the acoustic properties of the leading phoneme.2 Stimuli

in the experiment included both the singular and the

plural forms of each noun. A noun set is defined to be

the six nouns, singular and plural, associated with a

2 See Kessler, Treiman, and Mullennix (2002) for a review of

problems with voice key measurements. Precautions in creating the

stimuli for Experiments 1 and 2 were similar to the ones they

recommend.

particular irregular noun (e.g., the mouse noun set is

mouse, mice, rat, rats, nail, and nails).

All nouns were imageable and concrete. Semantic

match nouns were chosen to be semantically or per-

ceptually similar to the irregular noun with which they

were paired.3

In Experiment 1, 10 nouns of each type were used,

both singular and plural, for a total of 60 target nouns.In Experiment 2, all 60 nouns were used in trials, but

trials for the woman and fungus sets were later treated as

fillers (due to the difficulty of scoring the woman/women

trials and for balance considerations), resulting in 48

target nouns.

The lists were equated for frequency and word

length.4 To verify that the lists did not differ on fre-

quency, the lists were compared on lemma frequency,frequency of singulars, and frequency of plurals. Un-

paired t tests (both on the 10-set lists used in Experiment

1 and on the 8-set lists used in Experiment 2) indicated

no significant difference in frequency between the ir-

regular list and either matched regular list. Word length

was compared on number of letters, syllables, and

phonemes. Again, unpaired t tests indicated no signifi-

cant difference between the irregular list and eithermatched regular list.

2.1.2.2. Target trials. Fill-in-the-blank sentence frames

were used to elicit noun–noun compounds. Agentive

Two measures of frequency were used: lemma and form. Lemma

frequency is the frequency of all forms of the ‘‘word’’ for that part of

speech (Francis & Ku�cera, 1982). For a noun, the lemma includes the

singular, plural, possessive, and any special spellings or forms. If a

particular form (e.g., fly) is both a noun and a verb, only the count for

the noun was used. The form frequency used was the sum of the

frequencies for each sub-entry for the singular and the plural of each

noun lemma given in Francis and Ku�cera. Thus, frequencies used here

for singular and plural sum to the lemma frequency.

Fig. 1. Sentence frames and examples of stimuli in Experiments 1 and 2. Note. Exp., Experiment; Figure shows actual formatting of sentence frame

used in each Experiment.

C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464 457

(Experiment 1) and container (both experiments) frames

were used (see Fig. 1). Agentive frames resulted in

compounds ending in follower/catcher/painter/watcher,

and container frames resulted in compounds ending in

bowl/box/crate/tub.

2.1.3. Design

The dependent variables were the proportion of sin-

gular out of singular and plural responses and (in Ex-

periment 2) the time to produce a singular response (for

trials in which the response was singular). Factors (all

within-participant) were noun type (irregular, semantic

match regular, form match regular), stimulus nounnumber (singular, plural), and sentence frame (agentive,

container; Experiment 1 only). In Experiment 1, each

noun was seen once in the singular form and once in the

plural, for a total of 60 responses. In Experiment 2, each

of the 48 target nouns, both singular and plural forms,

was paired with each of the 4 containers, for a total of

192 scored responses. Results were analyzed with a

2� 2� 3 (Experiment 1) or a 2� 3 (Experiment 2) re-peated measures analysis of variance.

2.1.4. Procedure

2.1.4.1. Experiment 1. Participants saw 60 fill-in-the-

blank sentence frames and completed noun–noun com-

pounds by writing the first noun of the compound,

based on the target noun in the sentence. Sentences were

presented 10 to a page; each noun was used once as atarget, with either the agentive frame or the container

frame (counterbalanced across participants; each saw 30

of each type of frame).

2.1.4.2. Experiment 2. First, participants were familiar-

ized with the individual target, container, and practice

words and the response procedure; they saw each word

twice in two different pseudorandom orders for 152

training trials. They were then tested on fill-in-the-blanksentence frames to which they responded with noun–

noun compounds; responses were spoken, with response

times collected by a computer using a voice key. Stimuli

were presented on iMac computers using PsyScope

(Cohen, MacWhinney, Flatt, & Provost, 1993); re-

sponses were tape-recorded. Each target noun was

paired with each of the four containers. Test trials elic-

iting compounds were intermixed with single-wordtrials. There were 240 compound-forming trials inter-

spersed with 360 single-word trials. The intended re-

sponse for compound trials was the whole compound,

using a form of the target noun and the container as the

two nouns of the compound. Responses were scored

from the tape recording based on the grammatical

number of the target noun: singular, plural, or other (if

a word other than the singular or plural of the targetnoun was used); trials scored as ‘‘other’’ were excluded

from the data analyses.

Trials were self-paced. For each trial, whether single

word or compound-forming, first a black dot was

shown. When the participant was ready to proceed, he

or she pressed the space bar causing the dot to disappear

and the stimulus to be shown. The stimulus word or

sentence was centered on the computer screen usingblack 12 pt Chicago font. When the participant re-

sponded, the computer registered the response via the

voice key; the time was recorded; the stimulus disap-

peared and was replaced by a red asterisk indicating that

Fig. 2. Proportion of singular responses by noun type and grammatical

number of the stimulus noun in Experiment 1. Note. Error bars pro-

vided in each figure show standard errors of the mean. The reader

should note, however, that the standard errors of the mean indicate the

between-subjects variability, so they actually do not inform us about

the comparisons between the means involving within-subject factors.

458 C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464

the response was registered; and then the black dotsignaling the next trial appeared. If the participant did

not speak loudly enough for the response to register, the

stimulus did not disappear, and he or she would have to

respond again to get the asterisk feedback. Those trials

(apparent on the sound recording) were eliminated from

the analyses.

2.2. Results

In general, results not reaching the .05 level of sig-

nificance are not reported for any of the experiments.

(For more complete results including additional analy-

ses not reported here, please see Buck-Gengler, 2003.)

The stimulus items in each of the Experiments are

taken from a natural language, and not randomly gen-

erated or chosen. Nevertheless, ANOVAs were com-puted both with participants (F1) and with word set (F2)as the random effect. A composite min F 0 (Clark, 1973;Raaijmakers, Schrijnemakers, & Gremmen, 1999) was

computed on the basis of F1 and F2 for each analysis.

Effects are reported as significant based on min F 0;however, a note is made in cases where min F 0 does notreach significance but one or both of F1 and F2 are sig-

nificant at the a¼ .05 level. When means are reported(either in the text or in a figure), they are from the par-

ticipant analysis. The min F 0 tests should indicate whe-

ther effects are specific to the items or might generalize,

but they should not be taken to imply that items were

selected randomly, because in most cases there was ac-

tually very little choice of items that fit all requirements.

2.2.1. Experiment 1

The main question of Experiment 1 was: Under what

circumstances did participants not use the singular form

of the stimulus noun? As one would expect, the singular

response was given significantly less often overall when

the stimulus noun was plural (M ¼ :904) than when

the stimulus noun was singular (M ¼ :988)[min F 0ð1; 15Þ ¼ 8:75, p ¼ :010; F1ð1; 11Þ ¼ 10:52,MSE ¼ :024, p ¼ :008; F2ð1; 9Þ ¼ 52:01, MSE ¼ :003,p < :001]. Also, as expected, the response was singular

significantly less often when the stimulus noun was an

irregular noun (M ¼ :842) than when the stimulus noun

was either a semantic match regular (M ¼ 1:000) or a

form match regular (M ¼ :996) [min F 0ð2; 39Þ ¼ 8:82,p < :001; F1ð2; 22Þ ¼ 14:35, MSE ¼ :027, p < :001;F2ð2; 18Þ ¼ 22:92, MSE ¼ :011, p < :001]. Post hoc

Newman–Keuls tests on the analysis with participants asthe random effect confirmed that the proportion of

singular responses for irregular nouns differed from

those for regular nouns, but the response proportions

did not differ between the two types of regular nouns.

Most interestingly, the interaction (see Fig. 2)

between number and noun type was significant

[min F 0ð2; 31Þ ¼ 7:92, p ¼ :002; F1ð2; 22Þ ¼ 9:72, MSE ¼

:024, p < :001; F2ð2; 18Þ ¼ 42:61, MSE ¼ :003, p < :001]:When the trial contained an irregular plural stimulus

noun, the response was singular only 72.0% of the time,

whereas the singular was used between 96.5 and 100% of

the time for the remaining five combinations of number

and noun type.Sentence frame did not interact significantly with any

other factors by the min F 0 statistic; however, the in-

teraction of sentence frame and noun type was signifi-

cant for both the analysis with participants and the

analysis with word sets as the random effect [min F 0ð2;39Þ ¼ 2:81, p ¼ :073; F1ð2; 22Þ ¼ 5:97, MSE ¼ :004, p ¼:009, F2ð2; 18Þ ¼ 5:30,MSE ¼ :002, p ¼ :016]. This resultreflects the fact that the agentive frame (e.g., X-watcher)had more singular responses than the container frame

(e.g., X-box) for irregular noun trials [agentiveM ¼ :877,container M ¼ :808] but not for trials with either type

of regular noun [agentive, semantic match M ¼ 1:000,agentive, form match M ¼ :992, container, semantic

match M ¼ 1:000, container, form match M ¼ 1:000].This semantic effect is interesting but not of primary

interest here; see Pinker (1999, pp. 181–186) for discus-sion of semantic factors affecting the acceptability of

plurals as first elements in compounds.

2.2.2. Experiment 2

Experiment 2 examined response time (RT) as well as

response preferences. Does it take longer to produce

a singular first noun response when the stimulus is an

irregular plural than when it is a regular plural, anirregular singular, or a regular singular?

2.2.2.1. Proportion of trials with singular first noun

responses. The first analysis examined how many re-

sponses used a singular first noun (out of responses that

C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464 459

were either singular or plural), no matter what thegrammatical number of the stimulus. Whereas .921 of all

semantic match trials and .932 of all form match trials

were responded to with the singular form of that noun,

only .604 of all trials with an irregular target noun

were responded to with the singular form

[min F 0ð2; 50Þ ¼ 73:45, p < :001; F1ð2; 46Þ ¼ 127:17,MSE ¼ :013, p < :001; F2ð2; 14Þ ¼ 173:87, MSE ¼ :003,p < :001]. Of trials with a singular target noun, .992 ofthe responses were also singular, whereas only .646 of

the trials with a plural target noun were responded to

with the singular form [min F 0ð1; 25Þ ¼ 69:68, p < :001;F1ð1; 23Þ ¼ 72:91, MSE ¼ :053, p < :001; F2ð1; 7Þ ¼1573:46, MSE ¼ :001, p < :001].

Again, what is interesting is the interaction between

these two factors [min F 0ð2; 55Þ ¼ 68:64, p < :001;F1ð2; 46Þ ¼ 107:78, MSE ¼ :014, p < :001; F2ð2; 14Þ ¼189:04, MSE ¼ :003, p < :001]. As can be seen in

Fig. 3A, when the trial contained an irregular plural

target noun, the response was the singular form only

22.3% of the time (individual participants ranged from 0

to 87.1% singular responses). For all five of the other

combinations, the singular form was used between 85.7

Fig. 3. Results for Experiment 2. (A) Proportion of singular response

by noun type and grammatical number of the stimulus noun. (B)

Response times in ms for singular responses.

and 99.2% of the time. This result confirms earlierfindings, notably those of Gordon (1985), that irregular

plurals are readily produced orally as the first noun of

noun–noun compounds in response to this type of elic-

itation frame.

2.2.2.2. RTs to singular first noun responses. The RT

analysis examined the time to respond in ms with a

singular form to sentences containing either singular orplural irregular target nouns (e.g., to respond mouse box

when shown either ‘‘a BOX for transporting a MOUSE

is a ______’’ or ‘‘a BOX for transporting MICE is a

______’’), compared to the time to respond with a sin-

gular form to sentences containing either a singular or

plural regular target noun (e.g., to respond rat box when

shown either ‘‘a BOX for transporting a RAT is a

______’’ or ‘‘a BOX for transporting RATS is a______’’). This analysis included only those trials in

which the response was singular; thus fewer trials con-

tributed to the irregular plural cell than to the other cells

(as seen in Fig. 3A). Seven participants responded with a

plural 100% of the time when given an irregular plural

stimulus, and their data were excluded from the RT

analysis. To minimize the problem of skewness of RT

data, the RTs reported from each participant in each cellare the median RTs for that cell. Because medians were

used, no special action was taken with outliers.

RTs for semantic match (695ms) and form match

(691ms) trials were significantly faster than those for the

irregular trials (809ms) [min F 0ð2; 44Þ ¼ 9:71, p < :001;F1ð2; 32Þ ¼ 15:90, MSE ¼ 961, p < :001; F2ð2; 14Þ ¼24:93, MSE ¼ 272, p < :001]. RTs for singular-stimulus

trials (687ms) were significantly faster than for plural-stimulus trials (777ms) [min F 0ð1; 22Þ ¼ 14:52, p ¼ :001;F1ð1; 16Þ ¼ 24:13, MSE ¼ 8658, p < :001; F2ð1; 7Þ ¼36:48, MSE ¼ 2633, p < :001]. Again, the key result is

that the interaction between these factors, seen in

Fig. 3B, was significant [min F 0ð2; 45Þ ¼ 7:9, p ¼ :001;F1ð2; 32Þ ¼ 12:82, MSE ¼ 9303, p < :001; F2ð2; 14Þ ¼20:56, MSE ¼ 2550, p < :001]. The irregular plural

stimuli had the slowest RT; all other forms were re-sponded to much more quickly. As in Experiment 1,

post hoc tests confirmed that there was no difference in

either measure between the two kinds of regular nouns.

2.3. Discussion

Gordon�s (1985) result that plural is used more often

when the immediately preceding noun is an irregularplural than a regular plural (or a singular) was replicated

in Experiments 1 and 2; the proportion of plural re-

sponses to irregular plural stimuli was greater in the

faster oral response task of Experiment 2 compared

to the slower written response task of Experiment 1. In

fact, seven participants in Experiment 2 always

responded with the plural in that case. Moreover, in

Table 2

Irregular and matched regular nouns, both singular and plural forms,

used as stimuli in Experiment 3

Irregular noun Regular noun

Singular Plural Singular Plural

child children car cars

foot feet fork forks

goose geese gun guns

louse lice letter letters

man men match matches

mouse mice moon moons

ox oxen owl owls

tooth teeth tree trees

460 C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464

Experiment 2 we found that even when the singular wassuccessfully produced, it took longer when the preceding

noun was an irregular plural than for any other type of

noun. Therefore, preference for producing irregular

plurals as first elements of compounds in such elicitation

tasks can be explained by processing difficulty: goose is

harder to access from geese than duck is from ducks (as

might be predicted by, e.g., Allen & Badecker, 2002, or

Levelt, Roelofs, & Meyer, 1999). The delay in producing‘‘goose crate’’ as a response to ‘‘a crate for carrying

geese is a ____’’ is not predicted by an account that

simply translates the ordered steps of a generative model

into real time operations. Indeed, since Savin and

Perchonock (1965), attempts to translate generative

models directly into sequences of psycholinguistic op-

erations have been shown to be inadequate.

5 Because responses were typed, accuracy in Experiment 3 is a

different type of measure from the proportion of singular responses

measure in Experiments 1 and 2. Most errors in Experiment 3 were

either typos or errors that might indicate that the participant first

intended one form but then changed mid-word and corrected to the

other form. Typing the complete wrong form without typographical

error was extremely rare.

3. Experiment 3

Experiments 1 and 2 showed that in the context of

compound formation it was not only less likely but also

slower to produce a singular from a plural when the

stimulus was irregular than when it was regular. Ex-

periment 3 tests the more general hypothesis that it ismore difficult to produce either the singular or the plural

when the opposite form is in mind, and that this diffi-

culty is greater with irregulars than with regulars. This

hypothesis does not bear directly on level ordering, but

on the plausibility of our alternative account for the use

of irregular plural modifiers in compounds.

3.1. Method

3.1.1. Participants

Sixteen University of Colorado students participated

in Experiment 3 in partial fulfillment of introductory

psychology course requirements. All were native

speakers of English and had not participated in the

earlier experiments.

3.1.2. Materials

The stimulus nouns used in Experiment 3 (Table 2)

were the singulars and plurals of the eight irregular

nouns used in Experiment 2, along with eight regular

nouns matched on length, frequency, and first letter, for

a total of 32 words. Lists were equated on length (in

letters and syllables) and frequency as in Experiments 1

and 2, and were found by unpaired t tests to not differ onthose factors. Pictures of both one and several of each

item were used as stimuli. (Most ‘‘plural’’ pictures were

of two of the item, although some had three or four of

the item; the picture for teeth was of a set of teeth.) In the

main task, each picture was accompanied by a number

cue (either ‘‘one ____’’ or ‘‘four ____’’) to indicate that

the required response was either singular or plural.

3.1.3. Design

The main dependent measures were accuracy, initia-

tion time, and response time per letter. Accuracy was

defined as typing all letters correctly with no backspac-

ing and with the same grammatical number as the re-

sponse number cue. Only trials scored as accurate were

included in the response time analyses.5 Initiation time

was the time in ms taken to type the first letter of theresponse, measured from the time the stimulus was

presented on the screen. RT per letter was computed by

taking the total time in ms to type the word (measured

from the time the picture and number cue were pre-

sented on the screen through the time to respond with

the final letter of the word, but not including the ENTERENTER

key) and dividing by the number of letters in the word;

this adjusted measure allows the direct comparison ofRTs for words of different lengths.

Each analysis is a 2� 2� 2 repeated measures anal-

ysis of variance. The three factors, required response

(singular or plural), match condition (does the gram-

matical number of the word or picture part of the

stimulus match the required response number), and

noun type (regular, irregular), were varied within par-

ticipants in the participant analysis. In the item analysis,required response and match condition were varied

within item, and noun type was varied between items.

3.1.4. Procedure

First, participants were familiarized with the set of

pictures and their associated words. They saw each

picture separately with its word, and typed the word; the

entire set was repeated twice in two different randomorders.

C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464 461

In the main task each stimulus picture was crossedwith each response number cue, giving a total of 64

trials. A trial consisted of a picture cue and a number

cue; the participant typed the word associated with the

picture but matching the number cue (required re-

sponse) in grammatical form. Thus, if a picture of four

trees was shown along with ‘‘one ____,’’ the required

response was ‘‘tree.’’ In half of the trials, the gram-

matical number of the picture and required responsematched, and in half they did not. See Fig. 4 for ex-

amples of main task trials.

Stimuli were presented on an iMac computer using a

program written in RealBasic, and responses were

typed; pressing the ENTERENTER key concluded the entry.

(Half of the participants also heard the word matching

the picture pronounced, and the other half did not;

however, this factor did not influence any preliminary

Fig. 4. Example screen shots of the main task in Experiment 3. (A)

Match trial—both picture and required response number are plural. (B)

Mismatch trial—picture is plural, required response number is singular.

analyses and therefore was dropped from the analysesreported here.)

3.2. Results

As in Experiment 2, medians of response times were

used to eliminate any effects of skewness of response

time data.

3.2.1. Accuracy

The overall proportion of correct responses was .856.

None of the main effects were significant. The only in-

teraction approaching significance by the min F 0 statistic(actually significant in both the participants and the

items analyses) for this measure was between noun type

and response number, as shown in Fig. 5A [min F 0ð1;21Þ ¼ 3:76, p ¼ :066; F1ð1; 15Þ ¼ 18:87, MSE ¼ :009,p < :001; F2ð1; 14Þ ¼ 4:69, MSE ¼ :019, p ¼ :048]. For

irregular nouns, plural responses were more accurate

than singular responses, whereas for regular nouns, the

opposite was true. These differences were also signifi-

cant by paired t tests [irregular nouns: tð15Þ ¼ �4:038,p ¼ :001; regular nouns: tð15Þ ¼ 2:837, p ¼ :013]. No

other interactions were significant.

3.2.2. Initiation time

Regular nouns (M ¼ 1083ms) were initiated more

quickly than irregular nouns (M ¼ 1272ms) [min F 0ð1;22Þ ¼ 9:17, p ¼ :006; F1ð1; 15Þ ¼ 38:41, MSE¼ 29,613,

p < :001; F2ð1; 14Þ ¼ 12:05, MSE¼ 42,566, p ¼ :004],and responses were initiated more quickly when the

numbers matched (M ¼ 1142ms) than when they did

not (M ¼ 1212ms) [min F 0ð1; 25Þ ¼ 4:26, p ¼ :0496;F1ð1; 15Þ ¼ 6:00, MSE¼ 26,315, p ¼ :027; F2ð1; 14Þ ¼14:73, MSE ¼ 5884, p ¼ :002]; singular responses (M ¼1160ms) tended to be initiated faster than plural

responses (M ¼ 1194ms), although this difference was

not significant by min F 0 and only marginally significant

by participants and items [min F 0ð1; 29Þ ¼ 1:88, p ¼:181; F1ð1; 15Þ ¼ 3:97, MSE ¼ 9399, p ¼ :065; F2ð1; 14Þ¼ 3:57, MSE¼ 10,081, p ¼ :080]. None of the interac-tions were significant.

3.2.3. Time per letter

The letters of regular nouns (405ms) were typed

faster than those for irregular nouns (459ms), but this

effect was significant by participants only [min F 0ð1; 16Þ¼ 2:52, p ¼ :132; F1ð1; 15Þ ¼ 46:36, MSE ¼ 2040, p <:001; F2ð1; 14Þ ¼ 2:67, MSE¼ 23,522, p ¼ :125].6 Extraresponse time per letter was needed when the gram-

matical numbers did not match (448ms) beyond the

time taken when they did match (416ms) [min F 0

6 In general, when the participant analyses but not the item

analyses were significant, the results held for some of the items but did

not generalize across items.

Fig. 5. Results for Experiment 3. (A) Interaction between noun type

and required response number for accuracy measure. (B) Interaction

between noun type and required response for response time per letter

measure. (C) Interaction (marginal) between noun type and match

condition for response time per letter measure.

462 C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464

ð1; 29Þ ¼ 6:14, p ¼ :019; F1ð1; 15Þ ¼ 12:29, MSE ¼ 2557,p ¼ :003; F2ð1; 14Þ ¼ 12:29, MSE ¼ 1332, p ¼ :004].There was no difference in the typing speed of letters in

singular versus plural responses.

Noun type interacted significantly (by participants

only) with required response number as can be seen in

Fig. 5B [min F 0ð1; 17Þ < 1; F1ð1; 15Þ ¼ 5:98, MSE ¼2908, p ¼ :027; F2ð1; 14Þ < 1]. For irregular nouns, it

took less time on average to type letters of the singularsthan of the plurals; but for regular nouns, the letters

were typed more quickly on average in plurals than in

singulars. Individual t tests comparing the times for

singular and plural for each noun type reveal that for

irregulars, the 15ms difference between the singular and

plural forms was not significant, whereas for regulars,

the 31ms difference between singulars and plurals was

significant [tð15Þ ¼ 3:683, p ¼ :002]. One reason for thisdifference is that the time for the extra -s on regular

plurals was always shorter than the overall average time

per letter, because of the contribution of the response

time for the first letter (which was always longer). By

comparing panels A and B of Fig. 5, we see that sing-

ulars are both slower and more accurate than plurals for

regulars, whereas the opposite is true for irregulars,

hinting at the possibility of a speed-accuracy trade-off. However, there is no other evidence of such a

tradeoff because there are no other significant effects of

accuracy.

As noted, the main effects of noun type and match

were both significant (although noun type was signifi-

cant only by participants). The interaction between these

two factors approached significance, but only by par-

ticipants [min F 0ð1; 14Þ < 1; F1ð1; 15Þ ¼ 3:23, MSE ¼987, p ¼ :092; F2ð1; 14Þ < 1]. As can be seen in Fig. 5C,

the extra response time needed when cue and response

numbers did not match was greater for irregular nouns

(41ms) than for regular nouns (21ms). Both differences

were significant by paired t tests [irregular: tð15Þ ¼ 3:56,p ¼ :003; regular: tð15Þ ¼ 2:30, p ¼ :036].

3.3. Discussion

The main result of this experiment is that additional

time is required to generate the correct response when

the picture number and response number do not match

relative to when they do; this additional time to respond

tends to be even longer for irregular nouns than it is for

regular nouns. The interaction between match and reg-

ularity was reduced in Experiment 3 when the task wasmade bi-directional to isolate the process of producing

the opposite form from a given form. It is likely that the

task demand of going in both directions made it hard

enough for regulars that the difference between match

and mismatch showed up in Experiment 3 for regulars

where it did not in Experiment 2. These results are not as

strong as those of Experiment 2, but still are compatible

with, and add to, the results from Experiments 1 and 2.Irregular noun singulars and plurals differ more from

each other than do regular noun singulars and plurals,

and they do so in a variety of ways. If that difference has

any effect at all in on-line processing, it is more likely to

make it harder to access the opposite form when it dif-

fers more, and more unpredictably, resulting in longer

response times.

C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464 463

4. General discussion

The experiments reported here and our interpreta-

tions of them lead us to the following conclusions:

1. The response pattern reported by Gordon (1985) was

previously interpreted as the automatic result of prop-

erties of an innate lexical structure, because it would

otherwise require children to use an unlearnable op-

tional rule—‘‘form all noun–noun compounds with asingular first element, except that the plural may op-

tionally be used for irregular first nouns.’’ The option-

ality of the rule was argued to be unlearnable because

of lack of relevant input.We argue that this English re-

sponse pattern, which we replicated in adults, can be

equallywell accounted for by a soft constraint—‘‘avoid

plural as the modifying element in a noun–noun com-

pound,’’ abbreviated *Npl +N. There is ample evi-dence for this constraint in language addressed to

children under three years of age. There are also many

instances of violations of the constraint in children�s in-put (e.g., clothes-hamper). The constraintmakes no ref-

erence to regularity or irregularity.

2. Complying with *Npl +N in a situation where Npl

has been aroused by hearing it, as in Experiment 2,

is slower for irregular than for regular nouns, sug-gesting either that (a) there is a competition between

the plural (e.g., geese) and the singular (e.g., goose)

for irregulars only, and/or (b) that the production

system is delayed while waiting for the irregular sin-

gular to be activated. Occasional second thoughts ev-

ident in the spoken protocols (‘‘lice box or louse

box’’) also suggest that the RT delay might result

from competition and mutual inhibition between‘‘lice-box’’ and ‘‘louse-box.’’

3. However, producing a plural modifier, thus violating

*Npl +N when Npl has just been heard (a violation

that almost never happens unless Npl is irregular)

does not delay output (Buck-Gengler, 2003). The

arousal of the singular of an irregular plural must

therefore be relatively slow, because otherwise we

would expect competition between the singular andthe plural to delay the response in the violation case

as well as in the compliance case. Therefore, the ma-

jor reason for the RT delay in experimental settings

appears to be slow arousal of the singular of an irreg-

ular plural, rather than competition between the al-

ready aroused plural and the singular.

4. The written responses (Experiment 1) yield far fewer

irregular plural first elements than oral responses (Ex-periment 2), even though the oral responding was

self-paced. Conscious self-monitoring may be in-

volved in compliance with the constraint when plural

irregular nouns are involved; as mentioned, in the

oral response task, several speakers made self-correc-

tions (which were not counted as responses). Note

again the extreme variability among participants in

Experiment 2, who ranged from only 12.9 to 100% ir-regular plural responses.

5. Typed naming responses to a pictured single or multi-

ple-image stimulus with a simultaneous number cue

(Experiment 3) were slower when the number cue

did not match the number of images. This finding

suggests that semantic arousal of noun number (sin-

gular or plural) from looking at an image is at least

as rapid as interpreting the words ‘‘one’’ or ‘‘four.’’6. The additional slowing for irregular singular-picture-

to-typed-plural-name or plural-picture-to-typed-

singular name suggests even more strongly that the

semantic arousal of an irregular noun specifies the

lemma fully marked for number, increasing the plau-

sibility that the plural of a modifying noun may be

aroused semantically (in spite of the grammatical

constraint preferring the singular form) in ordinarycommunication, as well as in experimental settings.

A reviewer suggests that these extra slow number-

mismatch responses could be due to the design, where

3/4 of the responses (the number-match responses

and the regular-noun mismatch responses) can be ini-

tiated by using the following strategy: Start to pro-

duce a stem form that might be either the plural or

the singular (e.g., duck-), and inhibit the plural suffixbefore it is typed. This potential option is not avail-

able for any of our irregulars except oxen (and chil-

dren when typing), because all the other irregulars

we used have a vowel change on their second pho-

neme. The same strategy—begin production without

worrying about the number, then inhibit the plural

affix if it is not appropriate—might possibly apply

to the much faster oral responses in Experiment 2.But this strategy does not invalidate the conclusion

that accessing irregular forms opposite in number

from the form initially aroused takes extra time.

A factor not tested explicitly (due to the nature of the

English plural paradigm), but which is intimately tied up

in English with the regular/irregular dichotomy, is the

overwhelming frequency of the regular plural morpheme

itself. The frequency of the regular morpheme, coupledwith its ease of segmentability, probably contributes a

great deal to regular singulars being so easy to produce

(Bybee, 1995), rather than a difference in regular versus

irregular, or rule use versus memory lookup. In other

languages (e.g., German) there is more than one para-

digm for forming plurals, and what is ‘‘regular’’ may be

fairly rare compared to some of the other paradigms

(Marcus et al., 1995). Languages such as German aretherefore better suited to tease out the contribution of

frequency from that of matching.

Do the differences between irregular and regular

nouns found here argue in favor of a strict dichotomy

between the two types of words? By no means. First,

nothing here argues against the possibility that fre-

quently used inflected forms are stored, or that linkages

464 C.J. Buck-Gengler et al. / Brain and Language 90 (2004) 453–464

between the stem and the plural affix are strengthenedeach time a word is used with that inflection. Second, we

have deliberately avoided semi-regulars like wife and

leaf, mentioned earlier. It is quite possible that a suffi-

ciently sensitive design would show that their behavior is

intermediate between the regulars and the irregulars

(see, e.g., Haskell et al., 2003). If intermediate behavior

exists, a model that posits a dichotomy between storage

of inflected forms and on-line attachment of inflectionalendings (whether or not frequently used inflected forms

might be stored) would not be adequate; one would need

a model where the predictability of the form relation-

ships between singular and plural could play a role.

What does the production of ‘‘mice trap’’ (and

‘‘mouse trap’’) tell us about the mental lexicon? We see

these results as putting constraints on the theories that

are used to explain the dichotomy between regular andirregular inflections. It is no longer sufficient to ascribe

the results of compound formation to level ordering or

similar generative theories. Any explanation must also

provide for differences in processing time. Processing

time differences, as well as the choice in form, may be

due to differences in segmentability or overlap between

the forms; morpheme frequency also may play a role.

None of these explanations depend on regularity. Re-search in other languages, with other paradigms (e.g.,

verb past tense), and with other kinds of tasks is nec-

essary to clarify the contributions of these factors.

In conclusion, the preference for irregular plural as

first element of noun–noun compounds can be explained

by a processing factor (the accessibility of the singular

from the plural) that affects all speakers. No appeal to

innate grammar is required to explain the similaritybetween child and adult performance on this aspect of

linguistic behavior.

References

Allen, M., & Badecker, W. (2002). Inflectional regularity: Probing the

nature of lexical representation in a cross-modal priming task.

Journal of Memory and Language, 46, 705–722.

Bauer, L. (1990). Level disorder: The case of -er and -or. Transactionsof the Philological Society, 88, 97–110.

Booij, G. (1992). Compounding in Dutch. Rivista di Linguistica, 4, 37–

59.

Booij, G. (2002). The morphology of Dutch. New York: Oxford

University Press.

Buck-Gengler, C. J. (2003). Processing inflectional morphemes: Effects

of segmentation difficulty, regularity, and frequency. Unpublished

Doctoral Dissertation, University of Colorado, Boulder, CO.

Bybee, J. L. (1995). Regular morphology and the lexicon. Language

and Cognitive Processes, 10, 425–455.

Clahsen, H. (1999). Lexical entries and rules of language: A multidis-

ciplinary study of German inflection. Behavioral and Brain

Sciences, 22, 991–1060.

Clark, H. H. (1973). The language-as-fixed-effect fallacy: A critique of

language statistics in psychological research. Journal of Verbal

Learning and Verbal Behavior, 12, 335–359.

Cohen, J. D., MacWhinney, B., Flatt, M., & Provost, J. (1993).

PsyScope: An interactive graphics system for designing and

controlling experiments in the psychology laboratory using Mac-

intosh computers. Behavior Research Methods, Instruments, &

Computers, 25, 257–271.

Fabb, N. (1988). English suffixation is constrained only by selectional

restrictions. Natural Language and Linguistic Theory, 6,

527–539.

Francis, W. N., & Ku�cera, H. (1982). Frequency analysis of English

usage: Lexicon and grammar. Boston: Houghton Mifflin.

Gordon, P. (1985). Level-ordering in lexical development. Cognition,

21, 73–93.

Haskell, T. R., MacDonald, M. C., & Seidenberg, M. S. (2003).

Language learning and innateness: Some implications of Com-

pounds Research. Cognitive Psychology, 47, 119–163.

Hay, J. (2002). From speech perception to morphology: Affix ordering

revisited. Language, 78, 527–555.

Hoff, E. (2001). Language development (2nd ed.). Belmont, CA:

Wadsworth/Thompson Learning.

Kaisse, E. M., & Shaw, P. A. (1985). On the theory of lexical

phonology. Phonology Yearbook, 2, 1–30.

Kessler, B., Treiman, R., & Mullennix, J. (2002). Phonetic biases in

voice key response time measurements. Journal of Memory and

Language, 47, 145–171.

Kiparsky, P. (1982). From cyclic phonology to lexical phonology. In

H. van der Hulst & N. Smith (Eds.), The structure of phonological

representations (Part I) (pp. 131–175). Dordrecht, The Nether-

lands: Foris.

Ku�cera, H., & Francis, W. N. (1967). Computational analysis of

present-day American English. Providence: Brown University

Press.

Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of

lexical access in speech production. Behavioral and Brain Sciences,

22, 1–75.

MacWhinney, B. (2000). The CHILDES project: Tools for analyzing

talk (3rd ed.). Mahwah, NJ: Erlbaum.

Marcus, G. F., Brinkmann, U., Clahsen, H., Wiese, R., & Pinker, S.

(1995). German inflection: The exception that proves the rule.

Cognitive Psychology, 29, 189–256.

Pinker, S. (1991). Rules of language. Science, 253, 530–535.

Pinker, S. (1994). The language instinct. New York: W. Morrow.

Pinker, S. (1999). Words and rules. New York: Basic Books.

Raaijmakers, J. G. W., Schrijnemakers, J. M. C., & Gremmen, F.

(1999). How to deal with ‘‘The language-as-fixed-effect fallacy’’:

Common misconceptions and alternative solutions. Journal of

Memory and Language, 41, 416–426.

Ramscar, M. (2002). Learning language from the input: Why innate

constraints aren’t implicated in compounding. Paper presented at the

Eighth Annual Conference on Architectures and Mechanisms for

Language Processing, Tenerife, Canary Islands, Spain. Also

presented at the 43rd Annual Meeting of the Psychonomic Society,

Kansas City, MO, November 2002.

Savin, H. B., & Perchonock, E. (1965). Grammatical structure and the

immediate recall of English sentences. Journal of Verbal Learning

and Verbal Behavior, 4, 348–353.