7/28/2019 Wurm (1997)

1/24

JOURNAL OF MEMORY AND LANGUAGE 37, 438461 (1997)

ARTICLE NO. ML972524

Audito ry P ro c es s ing of P refixed Eng lis h Wo rd s

Is B oth C ontinuous a nd Deco mpos itiona l

Lee H. Wurm

State University of New York at Stony Brook

Two experiments compared continuous and discontinuous models of word recognition. Partici-

pants heard prefixed words whose full-form and root uniqueness points (UPs) differed, in either

a gating or lexical decision paradigm. Identification points and reaction times were analyzed

using multiple regression. Full-form UPs predicted performance better than root UPs did. Full-

form frequency measures had reliable facilitative relationships with performance while root

frequency measures were not consistently significant. Prefix frequency had a reliable, inhibitory

effect. Judged prefixedness, semantic transparency, and prefix likelihood were related to perfor-mance, alone or in interaction. The results provide evidence for both kinds of word recognition

procedures. A model is proposed with two parallel recognition routines: a whole-word routine

and a decompositional routine that considers only unbound roots that can combine with the

prefix in question. A preliminary rating study provides stimulus values on several dimensions

and can be used as a database by other researchers. 1997 Academic Press

Spoken-word recognition is complex, and access of such morphologically complex

words might occur. First, complex wordsparticularly little is known about cases where

words are composed of more than one mor- might be recognized without decomposing

them into their constituent morphemes. Lexi-pheme. There are two ways in which lexicalcal entries would correspond to whole words.

Alternatively, complex words might be de-This research was supported by Augmentation Awardcomposed, with individual morphemes repre-for Science and Engineering Research Training Grant

93NL174 from the Air Force Office of Scientific Re- sented lexically.search, Grant R01 MH51663-03 (to Arthur G. Samuel) Each of these access strategies correspondsfrom the National Institute of Mental Health, and National

to one of the main schools of thought amongResearch Service Award 1 F32 MH11721-01, also from

psychologists. I will call one of these the dis-NIMH. This article is based on the authors doctoral dis-continuous (or decomposition) approach. Insertation (State University of New York at Stony Brook).

A portion of this work was presented at the 37th annual models of this type, complex words must bemeeting of the Psychonomic Society (Chicago, IL, Octo- decomposed into root and affixes prior to lexi-ber 31November 3, 1996). I thank Arthur Samuel, Mark

cal access. Lexical access can proceed onlyAronoff, Susan Brennan, Richard Gerrig, John Robinson,

via the root because complex words do notand Cynthia Connine, all of whom made comments onhave lexical entries. Although there have beenan earlier version of this paper. I also thank Donna Kat for

her invaluable help with computer programming. Arthur several discontinuous models suggested (e.g.,Aron, David Cross, and the Statistics Lunch group at Cutler, Hawkins, & Gilligan, 1985; Jarvella &Stony Brook provided a great deal of help with data analy-

Meijers, 1983; MacKay, 1978; Morton, 1969;sis, and Harald Baayen helped in the computation of mor-

1979), this approach has been chiefly associ-pheme frequencies. Annmarie Cano and Douglas Vakoch ated with Taft and his colleagues (1979a;provided emotional support and help with many of themore tedious aspects of this project. Marcus Taft, Dennis 1981; 1985; Taft & Forster, 1975; Taft, Ham-Norris, and an anonymous reviewer made very helpful bly, & Kinoshita, 1986).comments on an earlier version of this paper. The main finding from the work of Taft and

Correspondence concerning this article should be ad-his colleagues is that nonword decisions takedressed to Lee H. Wurm, Department of Psychology,longer when the nonwords carry a prefix thanState University of New York at Binghamton, Bingham-

ton, NY 13902-6000. E-mail: lwurm@binghamton.edu. when they do not, and they take longer still

4380749-596X/97 $25.00Copyright 1997 by Academic Press

All rights of reproduction in any form reserved.

7/28/2019 Wurm (1997)

2/24

439AUDITORY PROCESSING OF PREFIXED WORDS

when the stems of the prefixed nonwords ples, as the extent of the idiosyncracies of all

languages is coming to light.are genuine stems in English. Whether or not

the stems of such nonwords are genuine En- The Cohort model (Marslen-Wilson &

Welsh, 1978; Marslen-Wilson, 1984), al-glish stems makes no difference when the

nonwords do not begin with genuine prefixes. though not specifically designed to address is-

sues relating to morphological complexity, isThese stems are not even recognizable as

stems because there is no prefix to strip off. a continuous processing model. Networkmodels, too, have most often aligned them-Bergman, Hudson, and Eling (1988) also

found support for the affix-stripping model selves with this view (e.g., McClelland & El-

man, 1986; Norris, 1994). Other arguments(Taft, 1981; 1985; Taft & Forster, 1975).

Taft (1994) addressed some of the more for a continuous processing model have been

made by Henderson, Wallis, and Knightpersistent criticisms of his earlier work, and

the results completely supported the earlier (1984), Rubin, Becker, and Freeman (1979),

and Tyler, Marslen-Wilson, Rentoul, andconclusions in favor of an affix-stripping ap-

proach. Still, Taft (1994) modified his position Hanney (1988).

An important third class of model is drivensomewhat, writing that an interactive-activa-tion model provided the best explanation of by stress (Cutler, 1976; Cutler & Norris, 1988;

Grosjean & Gee, 1987). According to the met-morphological processing. In Tafts new

model, prefixes are independent activation rical segmentation strategy (MSS) of Cutler

and Norris (1988), a strong syllable initiatesunits, separate from roots (i.e., prefixed words

are stored in decomposed form), but there is a lexical search. The unstressed syllables im-

mediately surrounding the strong syllable areno prefix-stripping per se. In this type of

scheme the equivalent of a prefix-stripping checked by an acoustic pattern-matching rou-

tine, to see if they can combine with theprocedure is a part of the access process.

The opposing general approach is to access stressed syllable. While not involving affix-stripping per se, a stress-driven model is anmorphologically complex words on a strictly

continuous, or left-to-right, basis. I will also example of a discontinuous model; words are

not processed in a strict left-to-right manner.use the term full-listing to refer to models of

this type, because on this type of account, all Furthermore, stress-driven and affix-stripping

models make similar predictions in manywords are listed in the lexicon. Words are ac-

cessed as complete units, whether or not they cases. Affixes will play a peripheral role in

lexical access because they are typically un-contain affixes. The lexicon in this case would

contain separate entries for the words stressed. In fact, stressed prefixes could cause

processing difficulties by triggering a lexicalcover, uncover, covering, and so on,resulting in considerable redundancy (e.g., look-up on the prefix (or the prefix plus the

initial phoneme[s] of the root) rather than onverbs in Finnish and Georgian can take thou-

sands of distinct surface forms that are essen- the root itself.

There are also mixed models that incorpo-tially the same vocabulary item although they

differ by inflection [Anderson, 1988]). This rate both a continuous and a discontinuous

approach. For example, Bergman et al. (1988)redundancy is the reason people first proposed

a decompositional approach; the storage re- suggested a possibility based on the race

model of Cutler and Norris (1979): the speechquired by a morphemic lexicon is much

smaller than that required if all items are to processor might attempt to access morpholog-

ically complex words both as full-forms andbe listed. However, Bybee (1988; see also

Sandra, 1994) believes that the emphasis on as analyzed separate parts. Data will be pre-

sented below that are suggestive of this kindstorage efficiency is misguided, given the

huge capacity of the human brain. She also of model; other variations on this idea have

also been proposed (e.g., Anshen & Aronoff,notes that linguists are relaxing their insis-

tence on maximally efficient storage princi- 1981; Caramazza, Laudanna, & Romani,

7/28/2019 Wurm (1997)

3/24

440 LEE H. WURM

1988; Frauenfelder & Schreuder, 1992; concerning prefix likelihood and semantic

transparency. The predicted identificationSchriefers, Friederici, & Graetz, 1992).

In recent years, researchers have begun to points (IPs) for continuous and discontinuous

processing models were directly tested. Theexamine the possible effects of linguistic and

computational variables on complex word effects of prefix likelihood, degree of pre-

fixedness, and semantic transparency were ex-processing. Using the crossmodal priming

technique, Marslen-Wilson, Tyler, Waksler, amined for their individual contributions tothe recognition process, and importantly, forand Older (1994) found that derivationally

suffixed words primed and were primed by their joint or interactive contributions. This

has not previously been done and, as we willtheir roots, but only if they were semantically

transparent (i.e., only if there was an obvious see, there are interesting and theoretically im-

portant interactions between these variablessemantic relationship between the two, as in

government and govern). Marslen-Wil- (and others). Furthermore, in the present study

these independent variables were treated inson et al. (1994) concluded that semantics de-

termines which words are morphologically re- ways that are more natural and powerful than

in previous research.lated to each other (as well as which wordsare complex). This was the first study to show The main goal of the preliminary rating

study was to provide information to be usedthe importance of semantic transparency, a

variable that has interested linguists for some in interpreting experimental data from the sub-

sequent experiments. For each stimulus, val-time.

Laudanna, Burani, and Cermele (1994) ues were obtained on several different dimen-

sions of interest, including prefix likelihood,found that a pair of little-examined variables

affect lexical decision performance for visu- judged prefixedness, semantic transparency,

root morpheme frequency, prefix frequency,ally presented Italian nonwords: the number

of word-types beginning with a given prefix and neighborhood size. This study also pro-vides a database for other researchers to use;and the success rate of prefix-stripping for that

prefix (i.e., the proportion of encountered these values have not previously been avail-

able, and some of the computations are time-words beginning with that letter string that are

in fact truly prefixeda quantity I will refer consuming.

to as the prefix likelihood). Although thePRELIMINARY RATING STUDYtwo variables are highly correlated, Laudanna

et al. (1994) found that prefix likelihood was This study consisted of two parts, the gen-

eral procedure of which was the same; partici-the more important one. They concluded that

a word beginning with a prefix that has a high pants were presented with auditory stimuli andgave a rating of some kind. In Part I, partici-prefix likelihood is likely to be stored and ac-

cessed in decomposed form. pants rated the stimuli on semantic transpar-

ency, and in Part II, they rated the stimuli onThe movement of some researchers toward

the theoretical middle ground in recent years prefixedness.

The semantic transparency of an item mightis a major advance. Researchers are recogniz-

ing the possibility that some words may be influence the perceptual strategies used to pro-

cess it. Words like rebuild have an obviousdecomposed while others may not. In accord

with this, researchers have realized that in or- compositional meaning (re plus build)

that is lacking from words like prepare.der to specify the conditions under which vari-

ous lexical access strategies may be used, they However, it is only within the very recent past

that researchers have recognized this, and themust be concerned with concepts like seman-

tic transparency. These were largely ignored few researchers to study this variable have

artificially dichotomized it. One of the pur-in early research efforts.

The present study used prefixed English poses of this rating study was to get continu-

ous semantic-transparency ratings for stimu-words in an attempt to clarify recent findings

7/28/2019 Wurm (1997)

4/24

441AUDITORY PROCESSING OF PREFIXED WORDS

lus items, since no published norms exist. UP of distaste is the /aV/, in spite of the

existence of the related word distasteful.These ratings were used as predictors of per-

formance in the two main experiments of this Full-forms were digitized at a sampling rate

of 10 KHz, low-pass filtered at 4.8 KHz, andstudy.

A second purpose was to get ratings of the stored in disk files. Stimuli were spoken by a

female native English speaker who was notprefixedness of the items. Which words are

prefixed? is just as much a psychological familiar with the purpose of the study. Rootswere digitally spliced away from the full-question as a linguistic one. Noncircular defi-

nitions in this area are uncommon, and lin- forms using visual and auditory inspection,

and the resulting roots sounded quite natural.guists have not been able to reach a consensus

about exactly what is or is not an affix. How- For the prefixedness ratings, two stimulus

lists were created. A randomly-determinedever, even if linguists were to reach such a

consensus, it is not clear that their definition half of the full-forms were assigned to list A

and half were assigned to list B. The rootwould have psychological relevance.

corresponding to a given full-form was as-

signed to the opposite list. For example, ifMethodrebuild was in list A, build was in list

Participants. Twenty-one students from the B. For the semantic transparency ratings, theDepartment of Psychology subject pool at the same stimuli were used, but they were ar-State University of New York at Stony Brook ranged in pairs that consisted of a full-formprovided prefixedness ratings. Thirty students and its root (e.g., rebuild and build).from the same subject pool provided seman- Procedure. Participants were tested intic-transparency ratings. All participants were groups of one to four. Stimuli were presentednative speakers of English who received over headphones in a sound-attenuating cham-

course credit for their participation. ber. For the prefixedness ratings, participantsMaterials. Seventy-two prefixed and pseu- were randomly assigned to hear either list A

doprefixed words were selected from a or B. After each word, participants had 2500150,000-word computerized phonetic diction- ms to give a rating. A Likert scale with anchorary (Moby Pronunciator 1.01, 1989) that met points Not at all prefixed (1) and Verythe following criteria: (1) carried a root that prefixed (8) was used. Participants were freeis a free-standing monomorphemic English to decide for themselves what prefixedword or homophonous with one; (2) began means; no example or further instructionswith a string of phonemes that constitutes an were given. For each list, words were pre-

English prefix; (3) varied (as a set) as much sented in a different random order.as possible along a continuum from high to For the semantic-transparency ratings, onlow semantic transparency; and (4) had full- each trial, participants heard a full-form andform and root uniqueness points (UPs) that its corresponding root and were asked to ratediffered by at least one phoneme (the UP is . . . how related in meaning the two wordsthe point in the acoustic signal where the word are on an 8-point Likert scale. Anchor pointsin question diverges from all other words in were labeled Not at all related (1) anda languagesee Marslen-Wilson, 1984; Mar- Very related (8). On half of the trials, theslen-Wilson & Welsh, 1978). full-form was presented first, and on half the

In determining UPs, I searched the same root was presented first. Participants had 4000computerized phonetic dictionary (Moby Pro- ms to make their rating. Item pairs were pre-nunciator 1.01, 1989) used for initial stimulus sented in a random order.selection. I excluded any suffixed forms that

Results and Discussionwere related (by inflection or derivation) to the

word in question (see, for example, Marslen- Median judged prefixedness values are

listed for each stimulus in Appendix A. TheWilson, 1984; Tyler et al., 1988). Thus, the

7/28/2019 Wurm (1997)

5/24

442 LEE H. WURM

ratings distribution had moderate negative Schreuder and Baayen (1994) explained the

requirement of reasonably high semantic-skew, so the recommended reflect and

square-root transformation was performed transparency, which is almost always missing

from words carrying bound roots, this way:(Tabachnick & Fidell, 1989). Thus, items that

participants rated highly prefixed had lower The only way in which the language learner

can discover that a certain string actually is atransformed prefixedness scores. This trans-

formed measure is the one that will be used stem (free or bound) is when that string occursin at least one semantically fully transparentin all analyses reported in this study.

Median semantic-transparency values are [italics added] combination (p. 360). Even

in the case of very common bound roots thatalso listed for each stimulus in Appendix A.

Because the distribution of transparency combine with several different prefixes (e.g.,

-mit or -vent), any reliable covariancescores was positively skewed, the square-root

of the transparency scores was used in all anal- between form and meaning (the importance

of which is also discussed by van Orden, Pen-yses reported in this study.

nington, & Stone, 1990) is likely to be noticed

Calculation of Other Regressor Variables only by experts in linguistics. In addition,Henderson (1985) asserted that the morpho-Several other variables were used to predict

participants performance in Experiments 1 logical complexity of words with bound roots

is synchronically meaningless; their affixesand 2. Prefix likelihoods for each prefix were

calculated in the following way (I used pro- are fossils and not psychologically relevant

at all. In short, although they appear to benunciation rather than spelling in selecting

words). I first identified for each prefix all of prefixed, they are not (the data of Marslen-

Wilson et al. [1994], which I have alreadythe words from Websters Third New Interna-

tional Dictionary of the English Language discussed, agree with this conclusion).

I computed prefix likelihood, then, as a ra-(1993) that were, in fact, truly prefixed. Aword was considered truly prefixed if the pre- tio: the numerator was the summed frequency

(from Francis and Kucera, 1982) of the trulyfix contributes to the meaning and syntax of

the full-form. This definition implies three cri- prefixed words beginning with a given pho-

netic string, and the denominator was theteria: First, when the prefix is removed, what

remains must be a freestanding word. Second, summed frequency of all words beginning

with that string in which removal of the puta-the meaning of the full-form must be reason-

ably transparent. Finally, the semantic rela- tive prefix leaves a pronounceable syllable or

syllables. The decision to consider syllabifi-tionship should be fairly constant across com-

binations; for example, truly prefixed words cation is also consistent with the criteria ofSchreuder and Baayen (1994). For example,beginning with re- will mean roughly to

do (whatever the stem means) again (M. Ar- although coat begins with co-, this word

was not considered a prefix-stripping failureonoff clarified these issues for me in a per-

sonal communication, November 30, 1994). because the remainder of the word (simply the

phoneme /t/ in this case) does not constituteNote that words carrying bound roots are

excluded from the above definition, even a syllable.

The notion behind prefix likelihood is thatthough such words can subjectively appear to

be prefixed (e.g., supersede, which is not decomposing words may make lexical access

less efficient, if the majority of words begin-counted as truly prefixed by the above criteria,

had a high prefixedness rating [see Appendix ning with a given prefix are not in fact prefixed

words. Values for the prefixes used in thisA]). In some cases, the classification of a root

as either bound or unbound is somewhat am- study are listed in Appendix B and ranged

from the theoretical minimum of .000biguous, but the exclusion of words with

bound roots from the calculation of prefix (hyper-) to the theoretical maximum of

1.000 (twi-). To illustrate what these num-likelihood is justified along several lines. First,

7/28/2019 Wurm (1997)

6/24

443AUDITORY PROCESSING OF PREFIXED WORDS

bers mean, consider the prefix dis-, which nominator means a sparse lexical neighbor-

hood. Recognition is expected to be easier ifhas a prefix likelihood of .092. This means

that 9.2% of the time a reader or listener en- the surrounding lexical neighborhood is

sparsely populated, because items would havecounters a word beginning with dis, that

word is prefixed. At the upper extreme, all relatively little competition from neighbors

(e.g., Goldinger, Luce, & Pisoni, 1989; Luce,instances of words beginning with twi-

(phonetically, /twai/) are truly prefixed (words 1987; Luce, Pisoni, & Goldinger, 1990).Neighborhood sizes for each prefix are listedsuch as twice do not count as failures be-

cause removing the prefix leaves the nonsylla- in Appendix B. The neighborhood size of

pro-, for example, is the sum of the fre-bifiable phonetic segment /s/).

With few exceptions, these prefix-likeli- quencies of the more than 500 words begin-

ning with that sequence.hood values are quite low (M .19). This

value is quite comparable to the .17 reported Several different frequency measures were

also needed. Individual full-form and root to-by Schreuder and Baayen (1994) for the seven

English prefixes they examined, using the ken frequencies were taken directly from

Francis and Kucera (1982). The correspond-CELEX database (Baayen et al., 1993; Bur-nage, 1990). On the face of it, prefix likeli- ing family frequencies were calculated by

summing values from Francis and Kucerahoods like these would suggest that a decom-

position strategy has such a low payoff that it (1982) for each word and its morphological

relatives (family frequencies were includedwould make lexical access less efficient,

which is what those authors concluded. As because a consensus as to which frequency

measure is more important, token or family,will become clear, their conclusion may have

been premature. has not been reached). A continuous model

predicts that root frequency will have no roleThe research of Laudanna et al. (1994), in

which the notion of prefix likelihood (what in word recognition, and a decompositionalmodel predicts that root frequency will ex-they called the success rate) was initially

explored, was discussed earlier. Although they plain a large proportion of the variance (Taft

[1979b] found that both full-form and rootconcluded that prefix likelihood was more im-

portant, they also found that the number of frequencies influence performance).

Morpheme frequencies for each root andwords beginning with a given prefix was re-

lated to performance. This is essentially an each prefix had to be calculated in a different

way. The first step in calculating root frequen-unweighted measure of the density of each

words lexical neighborhood. Lexical neigh- cies was a non-position-specific string search

for each root in the Birmingham/Cobuild cor-borhood is a metaphor for understanding howwords are organized or stored in memory. For pus (18 million tokens) of the CELEX data-

base (Baayen et al., 1993; Burnage, 1990).example, each of the four words in English

that begin with the phonemic sequence /twai/ Frequencies were then summed across all

cases where that root was a morpheme (e.g.,have only the other three as neighbors in lexi-

cal space. They reside in a sparsely populated lead is a morpheme in the word mis-

lead, but not in the word plead). Thisneighborhood. Some neighborhoods have

hundreds or thousands of inhabitants (e.g., measure is similar to the family frequency

measure already calculated for each root, theover 500 words begin with the phonemic se-

quence pro-). key difference being the importance of posi-

tion. The root-morpheme frequency measureThe denominator of the prefix-likelihood

ratio is a frequency-weighted measure of included cases where the morpheme in ques-

tion was not the first morpheme of a complexneighborhood density. Regardless of the par-

ticular value of the prefix likelihood, a large word, whereas the family frequency measure

for each root did not. For example, the familydenominator means that a word resides in a

dense lexical neighborhood, and a small de- frequency measure for lead included words

7/28/2019 Wurm (1997)

7/24

444 LEE H. WURM

TABLE 1 EXPERIMENT 1

SUMMARY STATISTICS FOR FREQUENCY MEASURES This experiment tested the predictions of

continuous and discontinuous models with re-Frequency measure M (SD) Range

spect to the time at which word identification

Full-form measures should occur. A strictly continuous model pre-Tokena 8 (26) 0195 dicts that identification should occur at theFamilya 22 (66) 0529 full-form UP, while a strict affix-stripping

Morphemic measuresmodel predicts that identification cannot occur

Token (root)a 121 (188) 0 807before the root UP. The inclusion of severalFamily (root)a 227 (354) 0 1786

Morpheme (root)b 235 (409) 0 2379 different frequency measures, some of whichPrefixb 1572 (4358) 121963 are predicted to be irrelevant by a continuous

model, was intended to shed additional lightNote. All values are per million tokens.

on the issue. Finally, this experiment testeda From Francis and Kucera (1982).whether or not variables such as prefix likeli-b From the CELEX database (Baayen, Piepenbrock, &

van Rijn, 1993; Burnage, 1990). hood, semantic transparency, judged pre-fixedness, and stress influence the identifica-

tion of morphologically complex words.such as lead, leading, and leader

(among others). The root frequency measureMethod

included these words plus words such as

mislead. Root-morpheme frequency values Participants. Thirty-eight students from the

Department of Psychology subject pool at theare shown in Appendix A.

Prefix frequencies were computed in essen- State University of New York at Stony Brook

participated. All were native speakers of En-tially the same way. Counts of all words in the

Birmingham/Cobuild corpus beginning with glish with normal hearing. Participants re-ceived course credit or cash for their participa-each prefix string were obtained. From these,

the frequencies for those cases that were, in tion.

Materials. The 72 prefixed and pseudopre-fact, instances of prefixation were summed

(e.g., preview counts but preen does fixed words from the preliminary rating study,

along with their associated roots, were used.not). Prefix frequencies are shown in Appen-

dix B. These items are listed in Appendix A.

Procedure. The full-forms were randomlySummary statistics for all of the frequency

measures are shown in Table 1. Distributions divided into two sets of 36. The full-forms

from one set, along with the roots correspond-of prefix-likelihood values and all six fre-quency measures had severe positive skew. ing to the unpresented full-forms, were pre-

sented to one group of participants. A differentFor all of them, the logarithm of each value

was used in the analyses. group of participants heard the other half of

the stimuli.This rating study provided values on several

variables that were chosen as predictors of Participants in groups of one to four lis-

tened to stimuli over headphones in a sound-word recognition performance. The focus of

Experiment 1 shifted away from the stimuli attenuating chamber. For roots, the first gate

consisted of the first 50 ms of the word, theper se and toward the recognition performance

of participants. Experiment 1 provided the first second gate consisted of the first 100 ms of

the word, and so on, until the entire stimulusdirect tests of the competing models: Would

root or full-form UPs be more closely related was presented. For full-forms, the first gate

was between 50 and 99 ms long, dependingto participants IPs? Would morphemic vari-

ables like prefix likelihood, prefixedness, or on the duration of the words prefix (I wanted

to have the acoustic onset of one of the gatessemantic transparency be related to IPs?

Which frequency measures would matter? of the full-form coincide exactly with the

7/28/2019 Wurm (1997)

8/24

445AUDITORY PROCESSING OF PREFIXED WORDS

acoustic onset of corresponding root). For ex- comes: Were the items selected appropriately

(randomly, representatively, or whatever elseample, because the full-form belong had a

duration of 723 ms and the root long had appropriately might mean)? The interested

reader can refer to Cohen and Cohen (1983).a duration of 645 ms (giving a prefix duration

of 78 ms), the first gate of belong was 78 The analysis of IPs proceeded at three dif-

ferent levels, the first of which was a directms. Gate #2 then began at the acoustic onset

of the root. comparison of the two competing classes of models. The models make specific predictionsAfter each gate, participants were given

7000 ms to write down what they thought the that can be tested directly I simply deter-

mined which of the two UPs (full-form vsword was. Then the next gate, consisting of

the already-heard portion plus 50 ms more, root) better predicted performance. The sec-

ond level of analysis involved what I will callwas heard, and so on. Stimuli were presented

in a random order. prior effects (prefix stress and the six fre-

quency measures). I use the term prior ef-Results and Discussion fects because of the importance of account-

ing for these variables before making anyOne item (ultrasound) was discardedfrom all analyses because of digitization prob- strong claims about factors such as prefix like-

lihood and semantic transparency. The prior-lems. A response was considered an error if

the participant never did correctly identify the effects variables provided another test of the

two classes of modelsdecompositionalpresented word. Two participants data were

excluded because of very high error rates models predict that root measures should be

at least as important as full-form measures.(22% and 25%all other participants had er-

ror rates less than 5%). The IP, the point at Conversely, continuous models predict that

only full-form frequency measures shouldwhich the participant correctly identified the

presented word without subsequently chang- matter. Stress, too, is irrelevant from thestandpoint of a strictly continuous model. Theing his or her mind, was found for each trial.

Trials on which the participant never did ar- final level of analysis looked at what I will

call decomposition variables. These includerive at the correct word comprised 3.9% of

the data. These trials were discarded. prefix likelihood, prefixedness, and semantic

transparency, along with their interactionsIPs were analyzed using multiple regres-

sion. The independence of observations as- (with each other, with prefix and root fre-

quency measures, and with stress). Any sig-sumed by linear regression models did not

hold in the current experiment (i.e., each par- nificant effects from level 2 (the prior effects)

were partialed from this analysis becauseticipant provided more than one observation).In repeated-measures regression analyses, this those variables had moderate to strong corre-

lations with the variables in this level. Theis controlled by the inclusion ofN0 1 dummy

variables (35 in this case) that represent the analysis will proceed in exactly this way for

Experiment 2, as well. Since this level of theparticipants. In addition, item analyses cannot

be conducted in a repeated-measures regres- analysis contained a fairly large number of

statistical tests (19), p-values were adjustedsion design. The regressor variables (e.g., pre-

fixedness, semantic transparency, and so on) using a Bonferroni correction.

UP analyses. The mean IP for the full-are intrinsic, nonvarying aspects of the items

themselveseach item is essentially its own forms in this experiment was 441 ms (SE

4.12 ms), which is much closer to the meancondition. This is reflected in the large df

value in the denominator, which equals the full-form UP (480 ms) than to the mean root

UP (714 ms, measured within the full-formnumber of participants times the number of

critical stimuli, minus the number of incorrect tokens). One would expect actual performance

in a gating experiment to be slightly earliercritical trials and the number of previous fac-

tors in the model. The relevant question be- than the theoretically critical point because of

7/28/2019 Wurm (1997)

9/24

446 LEE H. WURM

TABLE 2guessing; on a certain proportion of the trials,

participants will stumble onto the correct re- RESULTS OF FREQUENCY ANALYSES (EXPERIMENT 1)sponse by chance before they have heard input

Regressor Sign F(1, 1212)sufficient to isolate a unique word candidate.

On average, then, IPs were fairly close to theFull-form measures

values predicted by a full-listing model andLog token frequency (full-form) 0 10.42*

preceded the decomposition values by an av- Log family frequency (full-form) 0 12.08**Morpheme measureserage of 273 ms. This result is problematic

Log token frequency (root) / 1for discontinuous models.Log family frequency (root) / 1Regression analyses supported this conclu-Log morpheme frequency (root) / 1.37

sion. Two analyses were performed. After en-Log prefix frequency / 25.81***

tering the N-1 dummy variables, the second

* p .01.step in each analysis was to enter the relevant** p .001.UP and see whether it produced a significant

*** p .0001.increase in the total explained variance. Both

UPs were significant predictors of perfor-mance (F(1,1213) 131.68 and F(1,1213)

44.00 for full-form UPs and root UPs, respec- stress]) was associated with later IPs. This

could be easily explained if IPs were measuredtivelyboth ps .0001).

Both full-form and root UPs can be signifi- from word onset: stressed prefixes have longer

durations. However, IPs were measured fromcant because they were correlated with each

other (r[70] .57, p .001). When the the full-form UP. The MSS (Cutler & Norris,

1988), while not addressing such variables asshared variance between them was accounted

for, however, only the full-form UPs were sig- semantic transparency, predicts that lexical

access should be root-driven when items havenificant. The partial correlation between full-form UPs and IPs, controlling for participants unstressed prefixes. This would mean later IPs

for those items because root UPs were alwaysand for root UPs, was .26 (p .001), while

the correlation between root UPs and IPs dis- later than full-form UPs. The observed rela-

tionship runs in the opposite direction.appeared when participants and full-form UPs

were partialed (r .01, p .66). As noted However, the MSS makes a second predic-

tion that is consistent with the observed rela-above, this result supports continuous models

over discontinuous ones. tionship: an effect like this could appear if

participants mistakenly treated stressed pre-Therefore, in subsequent analyses, IPs were

measured not from word onset but from the fixes as the first portion of roots. An attemptto access a root morpheme, using the prefixfull-form UP of each stimulus. This was nec-

essary because full-form UPs were correlated as the assumed first syllable (because it is

stressed), will fail. For example, participantswith other variables under investigation. Any

conclusion about the importance of word fre- might hear the initial portion of the word co-

pilot and mistakenly access words likequency (for example) that did not account for

the fact that each word has its own UP would cope. Items like these might have late IPs,

because participants are likely to perseveratebe meaningless.

Prior effects. The next level of the analysis on a mistaken hunch in a task like gating.

Which frequency measures were related toassessed the role of prefix stress and the fre-

quency measures. After the dummy variables IPs? Six regression analyses were performed.

In each one, the dummy variables were en-were entered, prefix stress was also found to

be a significant predictor of IPs (F(1,1213) tered first, along with prefix stress, and then

one of the frequency measures described4.77, p .05). Higher prefix stress (this was

an ordinal variable taking values of 0 [un- above. As can be seen in Table 2, both of the

full-form frequency measures had significantstressed], 1 [secondary stress], or 2 [primary

7/28/2019 Wurm (1997)

10/24

447AUDITORY PROCESSING OF PREFIXED WORDS

facilitative relationships with IPs. This is the

type of relationship generally expected: higher

frequency values were associated with earlier

IPs. Of the four morpheme-relevant measures,

only prefix frequency was a significant pre-

dictor of performance, and it had an inhibitory

relationship with the dependent variable: morecommon prefixes led to slower performance.

This effect has a natural explanation in terms

of neighborhood density prefix frequency

was significantly associated with the density

measure described earlier (r[70] .66, p

.001). This suggests that earlier IPs should be

found when prefix frequency is low because

these items reside in relatively sparse lexical

neighborhoods.Decomposition variables. This group con-

sisted of prefix likelihood, judged pre-

fixedness, and semantic transparency, as well

as interactions involving them. Continuous-FIG. 1. Mean identification point (IP) as a function of

processing models predict that none of thesesemantic transparency and prefix likelihood, in millisec-

analyses will be significant. As will be seen, onds (ms). An IP of 0 corresponds to the full-form unique-that prediction does not hold. Prefix likelihood ness point of each stimulus.turned out to play a major role in mediating

the effects of other variables.A separate analysis was conducted for each were partialed. In addition, the main effects of

the variables making up the interaction wereregressor. As the first step, the dummy vari-

ables, prefix stress, prefix frequency, and full- partialed. For the three-way interaction, the

two-way interactions were also partialed.form family frequency (the stronger of the two

significant full-form measures) were entered. These steps are all necessary to satisfy as-

sumptions of the general linear model (Co-Then the variable being considered was en-

tered. hen & Cohen, 1983).

For ease of interpretation, readers shouldHigher values of semantic transparency

were associated with earlier IPs (F(1,1210) note three things about the figures that willbe shown. First, untransformed prefixedness51.98, p .001). This finding supports de-

compositional models because the semantic ratings are shown so that values can be inter-

preted more naturally (transformed valuestransparency of a morphological combination

is irrelevant from a continuous perspective. were used in the statistical tests). Second, neg-

ative IPs reflect the fact that, on average, iden-There was also a trend toward the same kind

of IP advantage for higher levels of pre- tification of the items took place some 39 ms

before the full-form UP. Third, although thefixedness (F(1,1210) 8.97, p .10). By

itself, prefix likelihood had virtually no effect Y-axes show mean IPs based on median splits,

readers are reminded that none of the predictoron IPs (F(1,1210) 1).

Four analyses were conducted to assess the variables were dichotomized in the analyses;

this is simply the best way to illustrate theinteractions between these three variables

(three possible two-way interactions and the nature of each interaction.

The interaction between prefix likelihoodthree-way interaction). In all cases, the

dummy variables, prefix stress, prefix fre- and semantic transparency was significant and

is shown in Fig. 1 (F(1,1208) 22.84, p quency, and family frequency of the full-form

7/28/2019 Wurm (1997)

11/24

448 LEE H. WURM

.001). As can be seen in the figure, higher line for highly prefixed items is competition:

a high-frequency root may compete for activa-semantic transparency was associated with

earlier IPs but only for items with high-likeli- tion with the full-form that contains it, de-

laying recognition of the full-form. Schreuderhood prefixes. As with the significant main

effect of semantic transparency, this interac- and Baayen (1994) assert that this type of

competition is possible, and further evidencetion supports a decompositional processing

account. for it will be presented below.If we consider high values on decomposi-

EXPERIMENT 2tion variables to be cues associated with suc-

cessful morphemic access, then the earliest The gating paradigm has some distinct ad-

vantages. First of all, researchers can be pre-IPs on the graph indicate that multiple cues

result in enhanced performance. The latest IPs cise about how much acoustic information is

available at each response point. In addition,on the graph seem to reflect a cost incurred

when the processing system deals with items gating responses provide a picture of how the

group of word candidates forms and is nar-that, according to one variable (e.g., prefix

likelihood), should be decomposed but ac- rowed down. In spite of this, gating has beencriticized for various reasons, the most im-cording to another (e.g., semantic transpar-

ency) should not. portant being that listeners hear multiple repe-

titions of each stimulus and that responses areThe three-way interaction was also signifi-

cant (F(1,1204) 13.32, p .01) due to the not speeded. These criticisms have been con-

vincingly addressed by several studies (e.g.,fact that the two-way interaction between se-

mantic transparency and prefix likelihood Cotton & Grosjean, 1984; Marslen-Wilson,

1984; Salasoo & Pisoni, 1985; Tyler, 1984;shown in Fig. 1 was more pronounced the

lower judged prefixedness was. This may be Tyler & Marslen-Wilson, 1986; Tyler & Wes-

due to the moderately high correlation be-tween prefixedness and semantic transparency

(r[70] .58, p .01).

Twelve additional regression analyses were

conducted, each of which looked at the inter-

action between two of the following variables:

prefix stress, prefix frequency, root morpheme

frequency, prefix likelihood, prefixedness, and

semantic transparency. As a first step, the

dummy variables, prefix stress, prefix fre-quency, and family frequency for the full-form

were partialed, along with the two variables

whose interaction was being assessed. The in-

teraction term was entered next. It might be

that morphemic variables that do not account

for much variance overall (i.e., their main ef-

fects are not significant) are nonetheless im-

portant for certain kinds of words (i.e., they

interact).

One of these interactions was significant,

providing further evidence of decompositional

processing. This was the root frequency 1FIG. 2. Mean identification point (IP) as a function of

prefixedness interaction, shown in Fig. 2root frequency and judged prefixedness, in milliseconds

(F(1,1208) 9.11, p .05). One interesting (ms). An IP of 0 corresponds to the full-form uniquenesspoint of each stimulus.possibility concerning the positive slope of the

7/28/2019 Wurm (1997)

12/24

449AUDITORY PROCESSING OF PREFIXED WORDS

sels, 1983; 1985), but it is always wise to distributions as the other word stimuli, in

terms of word frequency, number of pho-ensure that a given result is not due to the

peculiarities of an experimental paradigm. Ex- nemes and syllables, and stress. To discourage

strategic responding, these words were of aperiment 2 tests the same factors as Experi-

ment 1, with a different methodology (audi- variety of different morphological types (mo-

nomorphemic, suffixed, and prefixed, includ-tory lexical decision). Participants heard each

stimulus uninterrupted and intact, and their ing both bound and unbound roots). Stress wasapproximately balanced for words and non-word/nonword responses were made as

quickly as possible. words. A practice list of similar composition

was used prior to the main experiment. ThisMethod list consisted of 24 stimuli.

Procedure. Participants in groups of one toParticipants. Participants were 110 students

from the Department of Psychology subject four listened to stimuli over headphones in a

sound-attenuating chamber, and stimulus pre-pool at the State University of New York at

Stony Brook. All were native speakers of En- sentation was randomized for each group of

participants. Half of the participants heard listglish with normal hearing. Participants re-ceived course credit or cash for their participa- A and then list B; the order was reversed for

the other half. On each trial, a participanttion.

Materials. The words used in this experi- heard an auditory stimulus and made a

speeded lexical decision by pressing buttonsment are listed in Appendix C. The full-forms

from Experiment 1 that carried a two-syllable on a response board with his/her dominant

hand. Participants pushed one button forprefix were dropped in the interest of unifor-

mity. Three hundred and sixty stimuli were words and another button for nonwords.

divided into two lists of 180. Each list con-

Results and Discussiontained: 30 of the full-forms from Experiment1; the roots of the other 30 full-forms from Participants were excluded if they had an

error rate greater than 15% or a mean RTExperiment 1; 30 filler words; 30 filler non-

words; and 60 morphological nonwords. greater than 1000 ms on either of the two lists

they heard. Twelve participants were excludedThese last 60 nonwords were of four different

types: no obvious morphemic structure (e.g., by these criteria. Analyses were conducted on

responses of the remaining 98 participants.pangort); genuine English prefix with a

nonword root (e.g., prezelp); nonprefix The lexical decision task is essentially a

kind of familiarity judgment. Participantswith a real word root (e.g., grevent); and

genuine English prefix with a real word root have some familiarity threshold above whichthey respond Word and below which they(e.g., precorrect). Fifteen of each type were

contained in each list. respond Nonword. Table 3, which shows

mean RTs for each type of word as a functionIn order to use the data on all 60 of the full-

forms, it was necessary to assess the effect of whether it was in the first list a participant

heard or in the second, can be understood ifof List 1 presentation on List 2 performance.

Certain stimulus items were included with this one assumes that because of the overall higher

familiarity after some experience with the var-in mind. First, the filler items (30 words and

30 nonwords) were identical in each list and ious kinds of stimuli (particularly the quite

novel nonwords), participants familiaritywere included to assess the effect of exact

repetition. Second, a subset of the participants thresholds were higher in List 2. The familiar-

ity interpretation explains the 50-ms facilita-(44/110) listened to the stimuli described

above plus an additional 40 words, 20 in each tion for filler words, which was significant in

an ANOVA by subjects (F1(1,97) 54.84, pof the two lists. These words were included

to assess practice effects. .001) and by items (F2(1,29) 39.38, p

.001). These words were exactly repeated andFiller words were chosen from the same

7/28/2019 Wurm (1997)

13/24

450 LEE H. WURM

TABLE 3 onset, the importance of the two UPs in pre-

dicting performance was assessed as before.MEAN RTs (ms) AS A FUNCTION OF WORD TYPE ANDLIST ORDER (EXPERIMENT 2) As in the gating data, both were significant

(F(1,5332) 120.85, p .0001, andList heard F(1,5332) 55.16, p .0001, for the full-

form and root UPs, respectively). The partialWord type First Second Difference

correlation coefficients lead to the same inter-Fillersa 637 587 050** pretation as before: the correlation betweenFull-formsb 603 607 /4 full-form UPs and RTs, partialing root UPsRootsb 489 486 03 and participants, was .12 (p .001), and thePractice-effect wordsc 656 684 /28

analogous partial correlation for root UP,

though significant, was much smaller (r .03,a These items were exactly repeated in both lists.b Any effect here would be essentially a repetition prim- p .05). Thus, as in the gating analyses, RTs

ing effect, caused by prior presentation of a shared root for the subsequent levels of the analysis weremorpheme. For example, if a given participant heard re- measured from the full-form UP of each stim-

build in the first list, build would be in the second ulus.list (and vice versa).Prior effects. Prefix stress was marginallyc This is strictly a practice effect, as none of the items

in this group was repeated. These items were heard by related to RTs (F(1,5332) 3.04, p .10);only a subset of the participants (44 out of 110). items with stressed prefixes tended to have

p .10 for subjects, p .05 for items. somewhat slower RTs (consistent with the** p .001 for subjects and items.

gating result). Table 4 shows the results of the

frequency measure analyses. The four mor-

phemic measures had inhibitory relationships

with RT, while the two full-form measuresthus highly familiar. The familiarity-threshold

account can also explain the 28-ms cost for had the expected facilitative relationships. Asdiscussed in Experiment 1, the effect of prefixpractice-effect words, which were new and

thus unfamiliar (significant by items, marginal frequency could be interpreted in light of that

variables correlation with neighborhood den-by subjectsF1(1,97) 3.70, p .10;

F2(1,39) 7.04, p .05). Full-forms and sity. The root frequency effects cannot be ex-

plained in this way but may indicate competi-roots more or less broke even, due to the raised

familiarity thresholds canceling the moderate tion between roots and the full-forms that con-

tain them (as discussed by Schreuder andincrease in familiarity (all four F-ratios 1).

The corresponding effects for the nonwords Baayen [1994] and in connection with Fig. 2

averaged 031 ms, with the negative sign re-flecting the importance of increased familiar-

TABLE 4ity for nonwords.

Data from the first and second list each par-RESULTS OF FREQUENCY ANALYSES (EXPERIMENT 2)

ticipant heard were pooled, and first or sec-

Regressor Sign F(1, 5332)ond list heard was used as a regressor in the

analyses. However, this variable will not beFull-form measuresdiscussed further; it did not account for a sig-

Log token frequency (full-form) 0 9.72*nificant proportion of the variance nor did it

Log family frequency (full-form)0

36.06***interact with any of the variables under study. Morpheme measuresLog token frequency (root) / 48.43***Analysis of the word data parallels the analy-Log family frequency (root) / 26.40***ses for Experiment 1. The same regressorsLog morpheme frequency (root) / 30.96***were used and the underlying logic was identi-Log prefix frequency / 30.59***

cal, but the dependent variable was RTs in-

stead of IPs. * p .01.*** p .0001.UP analyses. Measuring RTs from word

7/28/2019 Wurm (1997)

14/24

451AUDITORY PROCESSING OF PREFIXED WORDS

Fig. 1, at least as far as the slopes of the two

lines are concerned. In this figure, however,

performance on low prefix-likelihood items

relative to high prefix-likelihood items was

somewhat poorer than it was in the gating

experiment. This interaction supports decom-

positional models.Five of the next set of 12 interactions were

significant, as well. All of these significant

interactions involved either prefix likelihood

or prefix stress (or both). Figures 4 and 5 show

the interactions between prefix stress and pre-

fixedness and between prefix stress and se-

mantic transparency (F(1,5327) 10.57, p

.05, and F(1,5327) 57.87, p .001, respec-

tively). Figure 4 shows that when prefixednesswas low, the stress value of the prefix was

largely irrelevant. Theoretically, for low pre-

fixedness values, listeners would not process

the prefix and root separately. For highly pre-FIG. 3. Mean reaction time (RT) as a function of seman-

fixed items, one would expect to observe atic transparency and prefix likelihood, in milliseconds

RT advantage (recalling the main effect of(ms).prefixedness); however, this advantage was

erased for those highly prefixed items that also

happened to have stressed prefixes. This sug-of the present study). Such root effects supportdecompositional models, while the UP results gests that stressed prefixes fool the lexical

and the full-form frequency effects fit the pre-

dictions of a continuous processing model (as

was true in Experiment 1).

From the analyses conducted to this point,

then, it was determined that prefix frequency,

token frequency of the root, and family fre-

quency of the full-form would be partialed

from the further analyses. Again, the follow-ing p-values have been adjusted using a Bon-

ferroni correction.

Decomposition variables. The results for

the decomposition variables mirrored those

from the gating experiment and once again

suggest decompositional processing. Higher

levels of prefixedness (F(1,5329) 35.46, p

.001) and semantic transparency (F(1,5329)

66.60, p .001) were associated with faster

RTs, and once again prefix likelihood was not

significant.

As in the gating experiment, the prefix like-

lihood 1 semantic transparency interaction

was significant (F(1,5327) 10.75, p .05). FIG. 4. Mean reaction time (RT) as a function of prefixstress and judged prefixedness, in milliseconds (ms).Figure 3 shows this interaction and resembles

7/28/2019 Wurm (1997)

15/24

452 LEE H. WURM

high-frequency roots (Fig. 6) or prefixes (Fig.

7) was heightened by a potentially-disruptive

(i.e., fully-stressed) prefix.

Figure 8 shows the prefix frequency 1 pre-

fix likelihood interaction (F(1,5328) 11.22,

p .05). The inhibitory effect of prefix fre-

quency was strong for items with low prefixlikelihoods, while there was only a weak ef-

fect of prefix frequency for items with high

prefix likelihoods. One interpretation of this

pattern is that the perceptual system strips off

prefixes when the prefix likelihood is above

some threshold value. Prefix frequency cannot

have much of an effect if the prefix has been

stripped away, which would be the case for

items very high on prefix likelihood. The de-

layed word recognition for high-frequency

prefixes is consistent with the observation

made previously about high-density lexicalFIG. 5. Mean reaction time (RT) as a function of prefix neighborhoods.

stress and semantic transparency, in milliseconds (ms). As I noted in Experiment 1, none of these

variables is predicted to have any relevance by

continuous processing models (and obviously,access system, drawing processing away from

no interactions between them are predicted).the root, a possibility that was discussed ear- Therefore, all of the significant interactionslier.

Figure 5 shows the interaction between se-

mantic transparency and prefix stress. Given

the strong correlation between prefixedness

and semantic transparency, the expectation

was that this figure would resemble Fig. 4. It

did, except for items low on semantic trans-

parency with fully stressed prefixes. As pre-

dicted by the MSS (Cutler & Norris, 1988),strongweak words that are functionally mo-

nomorphemic (low on transparency) enjoyed

a RT advantage.

Figures 6 and 7 show the prefix stress inter-

actions with root frequency (F(1,5327)

10.82, p .05) and prefix frequency

(F(1,5328) 12.92, p .01). The two show

the same general pattern. For items with low-

frequency roots or prefixes, prefix stress had

a weak facilitative association with RTs, while

for items with high-frequency roots or pre-

fixes, prefix stress had a stronger, inhibitory

association. These effects fit the speculation

made earlier about the disruptiveness of words FIG. 6. Mean reaction time (RT) as a function of prefixstress and root frequency, in milliseconds (ms).with strong prefixes; the disruptiveness of

7/28/2019 Wurm (1997)

16/24

453AUDITORY PROCESSING OF PREFIXED WORDS

Discontinuous processing is implied by the

results involving the decomposition variables

and root frequency. These variables are com-

pletely irrelevant from the standpoint of the

continuous, left-to-right processing model im-

plicated by the UP and full-form frequency

results, yet all of these variables were signifi-cant either as main effects or in interactions.

In particular, it is clear that prefix likelihood

affects word-recognition performance, in

combination with other variables. Such results

cannot be accommodated within a strictly con-

tinuous framework.

Conditional Root UPs (CRUPs)

The original prediction concerning the de-composition variables was that they would be

positively related to IPs and RTs. Items high

on prefixedness or semantic transparency, ac-

cording to the prediction, would be moreFIG. 7. Mean reaction time (RT) as a function of prefixlikely to be decomposed into morphemes andstress and prefix frequency, in milliseconds (ms).accessed via their roots. They would therefore

have later IPs and slower RTs, because root

UPs were always later in the stimuli than full-

constitute evidence for decompositional pro- form UPs.cessing.

GENERAL DISCUSSION

The results of this study indicate that the

auditory processing of prefixed words has

both continuous and decompositional charac-

teristics. The results of the two main experi-

ments with respect to the full-form vs root

UPs, as well as the outcomes of the full-formfrequency analyses, support left-to-right pro-

cessing models. Although the reliable inhibi-

tory relationship between prefix frequency and

performance, observed in both experiments,

was not specifically predicted by either class

of model, it is best viewed as a by-product

of continuous processing. Decompositional

models would predict that if prefix frequency

were going to be used for word recognition,

higher frequency would be associated with

better performance, not worse (higher fre-

quency, as I have noted, is generally associ-

ated with faster processing). Therefore, this

result also favors continuous processing mod- FIG. 8. Mean reaction time (RT) as a function of prefixfrequency and prefix likelihood, in milliseconds (ms).els, though only mildly.

7/28/2019 Wurm (1997)

17/24

454 LEE H. WURM

The fact that the observed effects were in words do not count (-cretion and -crep-

ant are not unbound roots).the opposite direction (as well as the overall

Such a morphemic process could be a his-earliness of IPs in Experiment 1, which oc-torical by-product of the way language iscurred on average 39 ms before the full-formused. Speakers coin new words as they needUP) may be explained by a model that hasthem, essentially continuously (Henderson,two parallel but nonindependent processes, a

1985). Writers do this, too: Baayen (1994)whole-word process and a morphemic one,found that approximately 15 new wordsbut only if the morphemic process is selectiveending in -ity appear each month in Theabout the candidates it considers.New York Times. Although they are per-At word onset, both processes in such afectly intelligible, they are nonwords in thatmodel would begin and run in parallel. Thethey have never been seen before by readerswhole-word process simply checks the accu-and are not in any dictionary. Evidencemulating input against the lexicon, in exactlycomes from more anecdotal sources, asthe same way as the class of continuous mod-well: two examples I have recently heardels I have described throughout this paper.(one in the hallway and one on the eveningThis process will achieve a match at the full-news) are They were interrupted *mid-form UP. I will describe the operation of thesong by a power outage, and Violencemorphemic process in the following para-threatens to *re-Balkanize the region.graphs.These nonwords were quite easy to in-

The morphemic process, running in paral-terpret, and the speakers presumably knew

lel, strips any prefix and then attempts tothat they would be.

make a lexical match using the portion ofAronoff (1976) has noted that new word

the signal beginning just after the prefix. Theforms tend to be quite highly transparent while

relevant UP for this process depends on the older ones may not be because of semanticroot rather than the full-form, but it is not

drift. I have argued above that unbound rootsthe root UP I have been discussing through-

are required for high semantic transparency.out this paper. The process I am describing

To the extent that this argument is on theis selective in that, in attempting to match

mark, a decompositional process that onlythe input to the lexicon, it considers only

considers unbound roots would prove ex-unbound roots that attach to this prefix. The tremely useful in the recognition of newlymorphemic process will achieve a match at coined forms, which we encounter extremelywhat I will call the conditional root UP, or often (Henderson, 1985).

CRUP. A words CRUP is the root UP given In order to prevent the perceptual systemthis particular prefix. from committing prematurely to the wrong

Usually, the CRUP of a prefixed word will word, the perceptual system would have tobe the same point as the full-form UP. How- wait for verification from the whole-word pro-ever, there are many exceptions to this general cess before ultimately committing to a deci-rule. For example, consider the word dis- sion. Otherwise, the perceptual system mightcredit. The full-form UP of this word is the decide at the CRUP that the word being heardsecond /d/, because of words such as discre- is discredit, when in fact it is discretiontion. The root UP of discredit is the /t/, (for example). However, even though the per-because credible is still a competitor prior ceptual system would not be allowed to fullyto that point. The CRUP of discredit, commit to discredit until discretion hasthough, is the /r/: the only words still consis- been ruled out, facilitated recognition perfor-tent with discr- are discretion, dis- mance would reflect an activation boost re-crepant, and their morphological relatives, ceived by the root credit at the CRUP (abut because the morphemic process in this boost that was not received by bound roots

like -cretion). Furthermore, it is possiblemodel only considers unbound roots, those

7/28/2019 Wurm (1997)

18/24

455AUDITORY PROCESSING OF PREFIXED WORDS

that other kinds of computations can begin not predict RTs in Experiment 2, because such

strategic responding was unlikely: that experi-once a CRUPs-based hypothesis has been gen-

erated, while the system is waiting for the ment not only used a different methodology,

but also contained a variety of different typeswhole-word response. These might include

accelerated access of meaning, integration of of words and nonwords (suffixed, prefixed

[with bound as well as unbound roots], andthe lexical hypothesis with a sentence repre-

sentation, and/or access of a grammatical cate- monomorphemic). The CRUPs result forExperiment 2 was just like that for Experimentgory.

As a post hoc check of this account, all of 1: CRUPs significantly predicted RTs

(F(1,5332) 54.11, p .0001), doing a betterthe words were analyzed to see if their

CRUPs differed from their full-form UPs. Of job than root UPs but a somewhat worse job

than full-form UPs.the 71 full-forms in Experiment 1, nine

(12.7%) had CRUPs that were earlier than There are two findings in the literature that

may relate to CRUPs. Schriefers, Zwitserlood,their full-form UPs, as in the above example

of discredit. For the remaining 62 words, and Roelofs (1991) found that prefixed words

were identified earlier than unprefixed wordsthe CRUP was the same point as the full-form UP. This suggests that even if prema- with identical predicted recognition points (in

Dutch). For example, both the prefixed wordture commitment to an incorrect lexical hy-

pothesis cannot be avoided in the way I just opstaan and the root staan have the fi-

nal /n/ as their UP, but in a gating experimentdescribed (i.e., even if the morphemic pro-

cess does not wait for the whole-word pro- participants identified the prefixed words an

average of 37 ms earlier (compared to the 39cess), the occurrence of such errors would be

fairly low; perhaps the computational advan- ms advantage found in Experiment 1 of this

study). This general prefixation advantagetages outweigh the occasional cost. The mean

CRUP across all 71 words was 456 ms, a was replicated in two subsequent experimentsand could not be explained by either class ofvalue that corresponds to the mean IP in Ex-

periment 1 (441 ms) even better than the model. It is not possible to say whether

CRUPs would explain this effect, but the pos-mean full-form UP (480 ms) did.

However, there was conflicting evidence re- sibility is intriguing. Taft (1988) also found

what appears to be a general prefixation ad-garding CRUPs. In a regression analysis for

CRUPs just like that for full-form UPs and vantage in lexical decision times, although he

provided relatively little methodological de-root UPs, although CRUPs did account for a

significant increase in the explained variance tail. To explain his nonword data, he sug-

gested an activate and check model that is(F(1,1213) 81.52, p .0001, for the gatingexperiment), the effect was not as large as that similar to the smart morphemic half of the

dual-route model suggested in this paper, butfor full-form UPs. Still, this notion is deserv-

ing of some direct investigation with stimuli one that considers bound rather than unbound

roots. The model did not apply, however, toexplicitly chosen to test it, because of the close

overall correspondence between CRUPs and his word data.

CRUPs may also help explain the prefixIPs and because several of the results in this

study are most easily interpretable from the stress interactions from Experiment 2 of the

current study. Specifically, if the basic stress-perspective of a selective parallel model.

One could argue that the CRUPs result is driven model is modified so that it is sensitive

to the decomposition variables and so that itartifactual and due to the combination of the

gating task (where prefixes might be argued operates via CRUPs rather than root UPs,

there seem to be two possible predicted out-to be processed completely and independently

of what follows them) and the stimulus set comes. One of these is illustrated fairly well

by Fig. 4: prefix stress should not matter for(there were no stimuli with bound roots or

nonroots). If this were the case, CRUPs should items low on prefixedness (for example), but

7/28/2019 Wurm (1997)

19/24

456 LEE H. WURM

should begin to affect RTs as prefixedness in- modate the current results, if modified to re-

flect the importance of prefix likelihood, forcreases. The RTs that should stand out as

faster (because of the facilitative potential of example.

Cutler et al. (1985) argue in favor of a de-CRUPs) are those for words high on pre-

fixedness with unstressed prefixes. compositional serial autonomous model,

based on process considerations rather than

CONCLUSION storage and efficiency considerations. Theybelieve that a strict prefix-stripping accountMost existing models of word recognition

have too rigidly insisted on either continuous is probably wrong, but that decomposition of

prefix and stem is a strategy that seems rou-processing with total disregard for decomposi-

tional variables or on strict decompositional tinely available to the language processor.

However, one aspect of their framework thatprocessing in every instance. The current

study demonstrates that both theoretical posi- would seem not to fit the current results is

their insistence that listeners compute thetions are wrong; there is evidence for both

full-form processing and at least some decom- meaning of stems before affixes. While this

makes sense in many situations, particularlyposition, the latter perhaps involving CRUPs.In hindsight, the insistence of some research- those involving suffixing (e.g., sad /

-ness), the current results indicate that pre-ers on one or the other kind of mechanism

seems difficult to understand. The ease with fixes are dealt with very early in the recogni-

tion process. Prefix likelihood should not playwhich people coin and understand new words,

the idea of semantic drift, and the flexibility of a role in the early stages of recognition if stem

processing has to be completed first.the perceptual system demonstrated in various

tasks should have suggested earlier that there Network models (e.g., McClelland & El-

man, 1986; Norris, 1994) might be able tomight be both types of processing mechanism.

In addition, as pointed out in the Introduction, accommodate the current results without hav-ing two separate processes. Such models arewhile nonword data from studies such as Taft

et al. (1986) suggest an important role for often claimed to have the ability to extract

structural (e.g., that re- is a separable unit)roots, this does not imply that the system al-

ways uses root access for all words. or distributional (e.g., relating to prefix likeli-

hoods) information based on covariance ofAs I have noted throughout this paper, the

Cohort model (Marslen-Wilson, 1984; Mar- form and meaning. Van Orden et al. (1990)

provide a useful discussion of this issue.slen-Wilson & Welsh, 1978) cannot accom-

modate the results having to do with the de- A final interesting point concerns the fact

that the structure of the input itself determinescomposition variables. In addition, the sug-gested competition between high-frequency the specifics of perceptual processing. It may

well be profitable to apply the techniques de-roots and the full-forms that carry them flies

directly in the face of the basic cohort princi- veloped here crosslinguistically, choosing lan-

guages with known differences in morphologyple. The current results suggesting an im-

portant role for semantic transparency in de- and affix structure. Such a test could illustrate

how the perceptual process gets instantiatedcompositional processing is, however, consis-

tent with the conclusion of Marslen-Wilson et differentially in response to different language

environments.al. (1994). Their model may be able to accom-

7/28/2019 Wurm (1997)

20/24

457AUDITORY PROCESSING OF PREFIXED WORDS

APPENDIX A

Judged Prefixedness, Semantic Transparency, and Root Morpheme Frequency

Judged Semantic Root morpheme

Full-form (root) prefixednessa transparencya frequencyb

ablaze (blaze) 6 6 17

abreast (breast) 6 2 83

abridge (bridge) 4 2 71

acquire (choir) 1 1 8

aloft (loft) 5 2 11

antibody (body)c 8 2 705

antisocial (social)c 8 8 537

archbishop (bishop) 6 7 40

ascend (send) 6 2 72aspire (spire) 3 2 20

atingle (tingle) 6 3.5 3

beget (get) 6 2 2379

behold (hold) 5 4 527

belong (long) 3 1 1508

beside (side) 5 5 1236

circumscribe (scribe)c 5 2 2

concave (cave) 6 2 42

condense (dense) 5 4.5 33copilot (pilot) 8 8 25

counterplot (plot)c 7 7 32

default (fault) 7 3 55

defrost (frost) 8 7 16

discredit (credit) 8 7 58

disfigure (figure) 8 4 229

disgust (gust) 4 1 5

distaste (taste) 8 7 93

embattle (battle) 6 4 92embody (body) 6 3 705

engage (gauge) 5 2 0

engulf (gulf) 7 1.5 22

entitle (title) 7 3 46

forecast (cast) 6 2 127

foreshadow (shadow) 7 2 72

forewarn (warn) 7 7.5 50

hyperspace (space)c 7 5.5 154

increase (crease) 3 1 8infringe (fringe) 7 3 20

insecure (secure) 8 8 184

intercourse (course)c 7 2 749

interstate (state)c 7 6.5 497

midyear (year) 7 7 1420

misfit (fit) 7 3 145

mislead (lead) 8 7.5 557

7/28/2019 Wurm (1997)

21/24

458 LEE H. WURM

APPENDIX AContinued

Judged Semantic Root morpheme

Full-form (root) prefixednessa transparencya frequencyb

mistrial (trial) 8 7 66

monorail (rail)c

7 6 89perfume (fume) 1 4 13

persuade (suede) 5 1 3

preamble (amble) 8 3 12

prefix (fix) 8 1 52

preheat (heat) 8 7 143

proclaim (claim) 6 7 177

rebuild (build) 8 7 277

recite (cite) 7 2 26

rephrase (phrase) 8 7 50reprint (print) 8 7 95

reword (word) 8 6.5 502

subsample (sample) 8 7 21

supersede (seed)c 7 2 3

surcharge (charge) 7 7 162

surname (name) 6 7 421

transcend (send) 5 3.5 72

twilight (light) 3 6 530

twinight (night) 5.5 5 625ultrasound (sound)c 7 6 295

unbend (bend) 8 8 77

unborn (born) 8 8 113

unbuckle (buckle) 8 8 7

unbutton (button) 8 8 34

underscore (score)c 7 6 57

unplug (plug) 8 7.5 17

unscramble (scramble) 8 8 19

unusual (usual) 8 8 311

a Values can range from 18.b Per million tokens.c These items were used only in Experiment 1.

7/28/2019 Wurm (1997)

22/24

459AUDITORY PROCESSING OF PREFIXED WORDS

APPENDIX B dle, eclipse, frontier, gratify, grocer, hin-

drance, hockey, image, inert, infant, inspiring,

legal, linguist, minus, mission, mystique, op-Neighborhood Size, Prefix Likelihood, and

Prefix Frequency erate, referee, subsequent, teeter, torpedo, ve-

lour.Neighborhood Prefix Prefix Practice-effect words (these items were heard

Prefix size likelihood frequencya

by 44 of the 110 participants): acumen, audi-tion, brutal, catapult, cognizant, compel, cut-a- 16,803 .090 2454

anti- 72 .028 28 lets, despair, detachment, dinnerware, eyeball,arch- 18 .722 20 fluttering, frankly, heavenly, hesitance, in-be- 6562 .304 204 stinct, iron, mazurka, mermaid, mistletoe,circum- 109 .018 1

murmur, nourish, partial, privilege, pseud-con- 4828 .012 228

onym, ranking, reinforce, shrubbery, slumber,co- 1690 .010 18counter- 22 .545 26 spaghetti, splatter, stainless, subpoena, sweet-de- 5733 .008 104 heart, thinning, towel, twitter, vernacular,

dis- 2416 .092 766 worthy, wristwatch.em- 644 .003 78en- 2017 .092 371

fore- 372 .212 129 REFERENCEShyper- 7 .000b 1

Anderson, S. R. (1988). Morphology as a parsing prob-in- 9444 .094 21963lem. Linguistics, 26, 521544.inter- 1101 .212 256

Anshen, F., & Aronoff, M. (1981). Morphological pro-mid- 163 .264 60ductivity and phonological transparency. Canadianmis- 243 .181 79Journal of Linguistics, 26, 6372.mono- 48 .083 22

Aronoff, M. (1976). Word formation in generative gram-per- 2153 .005 484

mar. Cambridge, MA: MIT Press.pre- 2214 .013 56Baayen, R. H. (1994). Productivity in language produc-pro- 3832 .001 152

tion. Language and Cognitive Processes, 9, 447re- 8561 .067 1881469.sub- 608 .061 649

Baayen, R. H., Piepenbrock, R., & van Rijn, H. (1993).super- 109 .284 49The CELEX lexical database. [CD-ROM]sur- 556 .009 654

Bergman, M. W., Hudson, P. T. W., & Eling, P. A. T. M.trans- 346 .092 47(1988). How simple complex words can be: Morpho-twi- 4 1.000 9logical processing and word representations. Quar-ultra- 9 .889 2terly Journal of Experimental Psychology, 40A, 41un- 3271 .283 107272.under- 543 .232 234

Burnage, G. (1990). CELEX: A guide for users. Nijmegen,

Netherlands: CELEX.a Per million tokens.Bybee, J. L. (1988). Morphology as lexical organization.b The value of .000 for hyper- seems odd, given

In M. Hammond & M. Noonan (Eds.), Theoreticalwords such as hyperactive. In fact, there were 24morphology: Approaches in modern linguistics (pp.words beginning with hyper- in Websters Third New119141). New York: Academic Press.International Dictionary of the English Language (1993)

Caramazza, A., Laudanna, A., & Romani, C. (1988). Lex-that qualified as truly prefixed, but none of them was listedical access and inflectional morphology. Cognition,in the frequency count of Francis and Kucera (1982).28, 297332.

Cohen, J., & Cohen, P. (1983). Applied multiple regres-

sion/correlation analysis for the behavioral sciences(2nd ed.). Hillsdale, NJ: Erlbaum.APPENDIX CCotton, S., & Grosjean, F. (1984). The gating paradigm: A

Words Used in Experiment 2 comparison of successive and individual presentationformats. Perception & Psychophysics, 35, 4148.Critical items: Items from Appendix A not

Cutler, A. (1976). Phoneme monitoring reaction time asdenoted by a superscript c. a function of preceding intonation contour. Percep-Fillers: alphabet, anthem, bedazzle, bridal, tion & Psychophysics, 20, 5560.

Cutler, A., Hawkins, J. A., & Gilligan, G. (1985). Thebugle, curbside, deduction, dependent, dwin-

7/28/2019 Wurm (1997)

23/24

460 LEE H. WURM

suffixing preference: A processing explanation. Lin- Marslen-Wilson, W. D. (1984). Function and process in

spoken word recognition. In H. Bouma & D. G.guistics, 23, 723758.

Bouwhuis (Eds.), Attention and Performance: Vol.Cutler, A., & Norris, D. G. (1979). Monitoring sentence

X. Control of Language Processes (pp. 125150).comprehension. In W. E. Cooper & E. C. T. Walker

Hillsdale, NJ: Erlbaum.(Eds.), Sentence processing: Psycholinguistic studies

presented to Merrill Garrett (pp. 113 134). Hills- Marslen-Wilson, W. D., Tyler, L. K., Waksler, R., &

Older, L. (1994). Morphology and meaning in thedale, NJ: Erlbaum.

English mental lexicon. Psychological Review, 101,Cutler, A., & Norris, D. (1988). The role of strong sylla-333.bles in segmentation for lexical access. Journal of

Experimental Psychology: Human Perception and Marslen-Wilson, W. & Welsh, A. (1978). Processing in-

teractions and lexical access during word recognitionPerformance, 14, 113121.

in continuous speech. Cognitive Psychology, 10, 29Emmorey, K. D. (1989). Auditory morphological priming

63.in the lexicon. Language and Cognitive Processes,

4, 7392. McClelland, J. L., & Elman, J. L. (1986). The TRACE

model of speech perception. Cognitive Psychology,Francis, W. N., & Kucera, H. (1982). Frequency analysis

18, 186.of English usage: Lexicon and grammar. Boston:

Houghton Mifflin. Moby Pronunciator (Version 1.01) (1989). Monterey, CA:

Illumind. [Computer software]Frauenfelder, U. H., & Schreuder, R. (1992). Constraining

psycholinguistics models of morphological pro- Morton, J. (1969). The interaction of information in word

recognition. Psychological Review, 76, 165178.cessing and representation: The role of productivity.

In G. E. Booij & J. Van Marle (Eds.), Yearbook of Morton, J. (1979). Facilitation in word recognition: Ex-

Morphology 1991 (pp. 165 183). Dordrecht, Nether- periments causing change in the logogen model. In

lands: Kluwer. P. A. Kolers, M. E. Wrolstad, & H. Bouma (Eds.),

Processing of visible language (Vol. 1, pp. 259Goldinger, S. D., Luce, P. A., & Pisoni, D. B. (1989).

Priming lexical neighbors of spoken words: Effects 268). New York: Plenum.

of competition and inhibition. Journal of Memory Norris, D. (1994). Shortlist: A connectionist model of

and Language, 28, 501518. continuous speech recognition. Cognition, 52, 189

234.Grosjean, F., & Gee, J. (1987). Prosodic structure and

spoken word recognition. Cognition, 25, 135155. Rubin, G. S., Becker, C. A., & Freeman, R. H. (1979).Morphological structure and its effects on visualHenderson, L. (1985). Toward a psychology of mor-

phemes. In A. W. Ellis (Ed.), Progress in the psy- word recognition. Journal of Verbal Learning and

Verbal Behavior, 18, 399412.chology of language (Vol. 1, pp. 1572). Hillsdale,

NJ: Erlbaum. Salasoo, A., & Pisoni, D. B. (1985). Interaction of knowl-

edge sources in spoken word recognition. Journal ofHenderson, L., Wallis, J., & Knight, K. (1984). Morphe-

mic structure and lexical access. In H. Bouma & Memory and Language, 24, 210231.

D. G. Bouwhuis (Eds.), Attention and performance: Sandra, D. (1994). The morphology of the mental lexicon:

Vol. X. Control of language processes (pp. 211 Internal word structure viewed from a psycholinguis-

226). Hillsdale, NJ: Erlbaum. tic perspective. Language and Cognitive Processes,

9, 227269.Jarvella, R. J., & Meijers, G. (1983). Recognizing mor-

phemes in spoken words: Some evidence for a stem- Schreuder, R., & Baayen, R. H. (1994). Prefix strippingorganised mental lexicon. In G. B. Flores dArcais & re-revisited. Journal of Memory and Language, 33,R. J. Jarvella (Eds.), The process of language under- 357375.

standing (pp. 81112). London: Wiley. Schriefers, H., Friederici, A., & Graetz, P. (1992). Inflec-

Laudanna, A., Burani, C., & Cermele, A. (1994). Prefixes tional and derivational morphology in the mental lex-

as processing units. Language and Cognitive Pro- icon: Symmetries and asymmetries in repetition

cesses, 9, 295316. priming. Quarterly Journal of Experimental Psychol-

ogy, 44A, 373390.Luce, P. A. (1987). Neighborhoods of words in the mental

lexicon. Dissertation Abstracts International, 47, Schriefers, H., Zwitserlood, P., & Roelofs, A. (1991). The

5078. identification of morphologically complex spoken

words: Continuous processing or decomposition?Luce, P. A., Pisoni, D. B., & Goldinger, S. D. (1990). Journal of Memory and Language, 30, 2647.Similarity neighborhoods of spoken words. In

G. T. M. Altmann (Ed.), Cognitive models of speech Tabachnick, B. G., & Fidell, L. S. (1989). Using multivar-

iate statistics (2nd ed.). New York: Harper Collins.processing: Psycholinguistic and computational per-

spectives (pp. 122 147). Cambridge, MA: MIT Taft, M. (1979a). Lexical access via an orthographic code:

Press. The Basic Orthographic Syllable Structure (BOSS).

Journal of Verbal Learning and Verbal Behavior,MacKay, D. G. (1978). Derivational rules and the internal

18, 2139.lexicon. Journal of Verbal Learning and Verbal Be-

havior, 17, 6171. Taft, M. (1979b). Recognition of affixed words and the

7/28/2019 Wurm (1997)

24/24

461AUDITORY PROCESSING OF PREFIXED WORDS

word frequency effect. Memory and Cognition, 7, Tyler, L. K., & Marslen-Wilson, W. D. (1986). The ef-

fects of context on the recognition of polymorphemic263272.

words. Journal of Memory and Language, 25, 741Taft, M. (1981). Prefix stripping revisited. Journal of Ver-752.bal Learning and Verbal Behavior, 20, 289297.

Tyler, L. K., Marslen-Wilson, W. D., Rentoul, J., & Han-Taft, M. (1985). The decoding of words in lexical access:ney, P. (1988). Continuous and discontinuous accessA review of the morphographic approach. In D. Bes-in spoken word-recognition: The role of derivationalner, T. G. Waller, & G. E. Mackinnon (Eds.), Read-

prefixes.Journal of Memory and Language,27,

368ing research: Vol. 5. Advances in theory and practice 381.(pp. 83123). New York: Academic Press.Tyler, L. K., & Wessels, J. (1983). Quantifying contextualTaft, M. (1988). A morphological decompositional

contributions to word-recognition processes. Percep-model of lexical representation. Linguistics, 26, 657

tion & Psychophysics, 34, 409420.667.

Tyler, L. K., & Wessels, J. (1985). Is gating an on-lineTaft, M. (1994). Interactive-activation as a framework for

task? Evidence from naming latency data