Morphosyntax, prosody, and linking elements: The auditory processing of German nominal compounds

Morphosyntax, Prosody, and Linking Elements: TheAuditory Processing of German Nominal Compounds

Dirk Koester1, Th. C. Gunter1, S. Wagner2, and A. D. Friederici1

Abstract

& The morphosyntactic decomposition of German compoundwords and a proposed function of linking elements wereexamined during auditory processing using event-related brainpotentials. In Experiment 1, the syntactic gender agreementwas manipulated between a determiner and the initialcompound constituent (the ‘‘nonhead’’ constituent), andbetween a determiner and the last constituent (‘‘head’’).Although only the head is (morpho)syntactically relevant inGerman, both constituents elicited a left-anterior negativity ifits gender was incongruent. This strongly suggests that com-pounds are morphosyntactically decomposed. Experiment 2tested the function of those linking elements which arehomophonous to plural morphemes. It has been previouslysuggested that these indicate the number of nonhead con-

stituents. The number agreement was manipulated for bothconstituents analogous to Experiment 1. Number-incongruentheads, but not nonhead constituents, elicited an N400 and asubsequent broad negativity, suggesting that linking elementsare not processed as plural morphemes. Experiment 3 showedthat prosodic cues (duration and fundamental frequency) areemployed to differentiate between compounds and singlenouns and, thereby, betwen linking elements and plural mor-phemes. Number-incongruent words elicited a broad neg-ativity if they were produced with a single noun prosody;the same words elicited no event-related potential effect ifproduced with a compound prosody. A dual-route model canaccount for the influence of prosody on morphosyntacticprocessing. &

INTRODUCTION

Compounding, the concatenation of words in order tocreate new words, is one of the major word formationprocesses in most languages (Fabb, 2001). It is veryproductive in German and often used in place of phrases(Fleischer & Barz, 1995; Meyer, 1993). Words (i.e., freemorphemes) from different word classes may be com-bined to new words, whereby the last constituent de-termines all (morpho)syntactic features, for instance,word class, gender, number, and so forth. With fewexceptions, the last constituent, the so-called head(Olsen, 1990; Selkirk, 1982; Williams, 1981), also hassemantic priority. For example, ‘‘Hochhaus’’ (‘‘highhouse’’; multistory building) consists of an adjective(‘‘hoch’’) and a noun (‘‘Haus’’), but the entire com-pound is a noun. ‘‘Steinhaus’’ (stonemasc houseneut) iscomposed of a masculine and a neuter noun, but thecompound’s grammatical gender is neuter, and it refersto a house and not a kind of stone. Such compounds areclassified as ‘‘transparent’’ because the compoundmeaning is related to the meaning of their constituents.Opaque compounds, in contrast, lack such a semanticrelation to their constituents (e.g., ‘‘blackmail’’).

It is possible that entire compound forms are storedin the mental lexicon (‘‘full-listing’’ hypothesis; Butter-worth, 1983). A number of models incorporate this ideaand assume one underlying associative storage mecha-nism that acquires different word forms depending ontheir frequencies of occurrence (Bybee, 1995; Elman,1993; McClelland, Rumelhart, & the PDP research group,1986). This storage strategy is plausible and efficient forhigh-frequency compounds and is logically necessary foropaque compounds. However, others assume that com-pounds have no unique lexical entry in the mentallexicon, but rather that a compound meaning is com-puted from its constituents (‘‘full-parsing’’ approach;Libben, Derwing, & de Almeida, 1999; Taft & Forster,1976). Both theories have their weaknesses. The full-listing perspective cannot satisfactorily account for theprocessing of novel compounds, whereas full-parsingapproaches cannot explain the processing of opaquecompounds, given that their meaning is not related totheir constituents. Dual-route models take an interme-diate position in which the storage of compoundsdepends on compound properties such as frequencyof use, or semantic status (transparent vs. opaque; Isel,Gunter, & Friederici, 2003; Schreuder, Neijt, Van derWeide, & Baayen, 1998; Baayen, Dijkstra, & Schreuder,1997; Zwitserlood, 1994; Sandra, 1990). High-frequencycompounds and semantically opaque compounds are

1Max Planck Institute for Human Cognitive and Brain Sciences,Leipzig, Germany, 2Martin-Luther-Universitat Halle WittenbergHalle, Germany

D 2004 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 16:9, pp. 1647–1668

suggested to have their own lexical entries, whereastransparent low-frequency compounds are not storedin the lexicon. Difficult for dual-route models is thespecification of how a given route is selected.

Decomposition refers to the segmentation of theperceptual input (visual or acoustic) according to lin-guistic units, namely, morphemes (here, constituents).Earlier studies investigated mostly the comprehension oftwo-constituent compounds in the visual modality. Zwit-serlood (1994) and Sandra (1990) showed in repetitionand semantic priming experiments that the languageprocessing system (‘‘parser’’) is not only sensitive toword internal morphological structure (cf. Andrews,1986), but also that opaque compounds do not activatethe meaning of their constituents whereas transparentcompounds do. This suggests that the parser disposes oftwo processing routes: a direct and a decompositionalroute. The direct route accesses lexical entries via awhole word representation and opaque compoundsshould be processed successfully by this route. Thedecompositional route extracts the constituents andactivates their respective lexical entries. Thereby, itshould process transparent compounds successfully, ifthey have no lexical entry. These two routes mayoperate in parallel in a race fashion, or in a cascadingmode. In the latter, the second route is only activatedafter the first has failed (Baayen et al., 1997) or thesecond route is only employed under specific condi-tions. The abovementioned studies do not allow firmconclusions regarding the order of constituent process-ing because constituents are presented simultaneously.Thus, it is unclear when head and nonhead constituentsare processed, or whether both are processed at a time.

Few studies have investigated compound compre-hension in the auditory modality; they have focusedon semantic aspects and their findings are heteroge-neous. Pratarelli (1995) found electrophysiological(event-related potentials [ERPs]), but no reaction time(RT) evidence for the immediate semantic activation ofconstituents in a cross-modal priming study. Subjectswere presented with pictures as primes (e.g., of adogbone). Subsequently, they heard compounds thatserved as targets for a semantic relatedness decision.Unrelated compounds (e.g., ‘‘mailbox’’) elicited a morenegative ERP corresponding approximately to the com-pound duration compared to the ERP in response tocompletely related compounds (e.g., ‘‘dogbone’’). How-ever, if only one constituent was unrelated to the picture(e.g., ‘‘wishbone’’), the differential negativity was re-duced roughly for the related constituent, but not forthe unrelated one (displaced by about the averagelength of constituents). Although the related, but notthe unrelated constituents were repeated, it is suggestedthat the ERP effect reflects decomposition (called ‘‘wordcomplexity effect’’). However, no effect of semanticrelatedness was observed in RT measures. Wagner(2003) probed the semantic activation of nonhead Ger-

man compound constituents at the constituent bound-ary in a cross-modal priming experiment (auditoryprimes and visual target words). She used novel com-pounds, which contained ambiguous nonhead constitu-ents as primes, and obtained reliable N400 effects inresponse to semantically related targets compared tounrelated targets. The N400 effects were significant forthe dominant meaning of the ambiguous constituents.In addition, behavioral priming effects were obtained inthe lexical decision for the dominant and the subordi-nate meanings. These results argue for an immediatesemantic activation of nonhead constituents of novelcompounds.

In contrast, Isel et al. (2003) reported no semanticactivation of nonhead constituents in a behavioral cross-modal priming paradigm in which compounds served asprimes and visually presented words as targets for a lexi-cal decision. The authors used German two-constituentcompounds, which consisted of all combinations oftransparent or opaque constituents. The results showthat the semantic status of head constituents alone wasrelevant for the semantic activation of nonhead constit-uents; they were only activated if the head was trans-parent and only at the offset of the compound. Nonheadconstituents were not semantically activated at theconstituent boundary. The authors interpret their find-ings in favor of a cascading dual-route model, theprosody-assisted head-driven model (PAHD; Isel et al.,2003), in which acoustic input is processed continuouslyand the output of the acoustic phonetic analysis ismapped onto lexical entries via a direct route. Theprosodic structure of the nonhead morpheme is ana-lyzed in parallel to the acoustic phonetic analysis, and ifthe prosodic structure indicates a compound, the de-compositional route is called up. The duration of thefirst constituent (and onset embedded words) wasshown to be a valid parameter of morphological struc-ture (Isel et al., 2003; Davis, Marslen-Wilson, & Gaskell,2002; Cutler, Dahan, & van Donselaar, 1997). Accordingto the PAHD model, the decompositional route works inaddition to the direct route and extracts the headconstituent. Isel et al. (2003) suggest that subsequentlya morphological unit for the first constituent is used asan access code to the respective lexical entry. As soon asone route is found to be appropriate, the other route isdisregarded. The crucial assumption of the PAHD modelis that for each constituent a morphological representa-tion must be available in order to serve as an access codeif constituents need to be accessed (if the head istransparent). In the present experiments, we specificallytested whether a morphological representation is ac-cessed, that is, whether compounds are morphosyntacti-cally decomposed in analogy to the morphologicaldecomposition found in the visual modality (Zwitser-lood, 1994), and whether the morphological represen-tation is specified for number by means of linkingelements.

1648 Journal of Cognitive Neuroscience Volume 16, Number 9

EXPERIMENT 1

Experiment 1 investigates the morphosyntactic access ofcompound constituents by testing whether or not syn-tactic gender is available during comprehension. Genderis a grammatical feature in many languages and Germannouns belong to one of three syntactic gender classes(Comrie, 1999; Corbett, 1991). This gender feature isassumed to be stored in the noun’s lexical entry (Schiller& Caramazza, 2002; Caramazza, Miozzo, Costa, Schiller,& Alario, 2001) and the gender agreement establisheslocal or global coherence within or even across senten-ces. A noun’s gender surfaces, for instance, in therequired form of the definite determiner. The Germanword ‘‘Tisch’’ (table) is grammatically masculine andrequires the determiner ‘‘der’’ (themasc), whereas ‘‘Bett’’(bed) is neuter and requires the determiner ‘‘das’’(theneut). Gender is not marked on the noun itself withthe exception of most derivational suffixes, but these arenot investigated here. Although there is a correlation foranimate objects between the syntactic gender of a nounand the sex of the denoted entity, sex cannot be the solebasis for gender assignment because exceptions existand inanimate nouns have also unequivocally a syntacticgender. There are descriptive probabilities recruitingsemantic, morphological, and complex phonologicalfeatures for the assignment of gender in German(Salmons, 1993; Kopcke, 1982). These assign gendercorrectly for about 80% of German nouns. The explana-tion rate appears to be modest given that more than 40rules are postulated (Kopcke, 1982). In contrast, thegender of 95% of Spanish nouns can be determinedcorrectly based on the terminal phoneme (Teschner &Russell, 1984 as quoted from Wicha, Bates, Moreno, &Kutas, 2003). In German, semantic information (biolog-ical sex) was reported to influence syntactic genderdecision in reading in behavioral and ERP measures(lateralized readiness potentials; Schiller, Munte, Hore-mans, & Jansma, 2003). In the same study, however, aninfluence of phonological rules was only found in be-havioral, but not in ERP, measures. In summary, theabovementioned probabilistic rules may provide helpfulinformation about a noun’s syntactic gender, but ulti-mately the lexical entry has to be consulted to access anoun’s gender (Schiller et al., 2003).

Under the assumption that gender is an inherent partof lexical entries, gender information of nonhead con-stituents should become available during processing ifcompounds are decomposed morphosyntactically, al-though these constituents are syntactically irrelevant.Furthermore, it is assumed that the checking of genderagreement is an automatic process, that is, it is notunder strategic control (Schiller & Caramazza, 2003;Gunter, Friederici, & Schriefers, 2000). In order to testthe availability of the gender information, the ERP will beanalyzed for gender-congruent and -incongruent con-stituents. The comparison for the head constituent will

serve as a benchmark and show what effect is to beexpected for the nonhead constituents. Gender viola-tions have been reported to elicit a left-anterior nega-tivity (LAN) and a P600 in the visual and auditorymodality (Wedel & Hahne, 2002; Deutsch & Bentin,2001; Gunter et al., 2000, but see Schmitt, Lamers, &Munte, 2002; Hagoort & Brown, 1999). LAN effects havebeen found for a range of morphosyntactic violations(Coulson, King, & Kutas, 1998; Penke et al., 1997;Weyerts, Penke, Dohrn, Clahsen, & Munte, 1997) andare interpreted as to reflect the detection of a morpho-syntactic mismatch. A functionally alternative interpreta-tion of the LAN suggests that it reflects increasedworking memory load (Coulson et al. 1998; Kluender& Kutas, 1993). Kluender and Kutas (1993) associate theLAN with storage and retrieval of moved words insentences. Because our stimuli are minimal phrases(determiner + compound), working memory shouldnot play an important role. From these considerations,it is proposed that the presence of a LAN in response togender violations would reliably indicate the availabilityof gender information.

P600 effects were associated with syntactic processing(Hagoort, Brown, & Groothusen, 1993; Osterhout &Holcomb, 1992) and with reanalysis and repair of syn-tactic violations, but are also elicited by unpreferredsyntactic structures (Friederici, 2002; Coulson et al.1998; Friederici, Hahne, & Mecklinger, 1996; Osterhout,McKinnon, Bersick, & Corey, 1996; Neville, Nicol, Barss,Forster, & Garrett, 1991). Note, however, that the P600is more susceptible to controlled processes than earliercomponents (Hahne & Friederici, 1999).

In order to determine whether or not the genderinformation of nonhead constituents becomes available,subjects were presented acoustically with gender-marked definite determiners followed by two-constitu-ent compounds. Because semantic status is a crucialfactor in comprehension (Isel et al., 2003; Baayen et al.,1997; Zwitserlood, 1994; Sandra, 1990), we used bothtransparent and opaque compounds. Opaque, but nottransparent compounds must be stored by definition inthe lexicon and may, therefore, not be decomposed.However, we expect that both transparent and opaquecompounds are decomposed morphosyntactically, be-cause the semantic status depends on both constituentsand cannot be determined by the nonhead constituentalone. For example, upon perceiving /vint/ (‘‘Wind,’’wind), the compound may still turn out to be opaque(e.g., ‘‘Windbeutel,’’ ‘‘wind bag’’; cream puff) or trans-parent (e.g., ‘‘Windboe’’; gust of wind). As it is stillunknown whether both types of compounds are decom-posed morphosyntactically in the auditory domain, weincluded both transparent and opaque compounds inExperiment 1. The gender agreement between deter-miner and each constituent was varied independently ofthe other constituent and of semantic status of thecompound (Table 1). Subjects’ task, indicated after

Koester et al. 1649

presentation, is to judge either the gender agreement orto compare the item to another, visually presented wordon a semantic basis. These tasks ensured that subjectslistened closely and attended to both constituents aswell as to the determiner.

If compounds are not decomposed, gender incon-gruities of nonhead constituents should not result in aLAN, because nonhead constituents are not accessedseparately. If, however, compounds are decomposedmorphosyntactically, as found in the visual domain(Zwitserlood, 1994), a LAN is expected for gender-incongruent nonhead constituents. In both cases,gender-incongruent head constituents are expectedto elicit a LAN as the head is syntactically relevant.

Results

Behavioral Data

Subjects reported no problems and performed well inboth tasks as indicated by the high accuracy rates.However, the grammaticality judgments were moreaccurate (96% correct responses) than semantic judg-ments [91%; t(22) = 3.69; p < .01].1

ERP Data: Head Constituent

The mean ERP amplitude was calculated for each con-dition and each region of interest (ROI) in the timewindow 400–500 msec time-locked to the onset of thehead constituent (Figure 1). An ANOVA was performedwith the factors ROI (4: AL, AR, PL, PR), gender agree-ment of head constituents (C2; 2), gender agreement ofnonhead constituents (C1; 2), and semantic status (2).No interaction of C2 with semantic status, or with C1was found [F(1,22) = 0.00; F(1,22) = 0.79; both ns,

respectively]. The two-way interaction of C2 and ROIwas significant [F(3,66) = 6.58; p < .01; > = .72]. Inorder to determine the scalp distribution of the C2effect, each ROI was tested separately. The main effectof C2 was only significant in the left-anterior ROI[F(1,22) = 10.98; p < .01]. None of the other ROIsshowed a significant effect of C2 [AR: F(1,22) = 0.26; PL:F(1,22) = 2.15; PR: F(1,22) = 0.66; all ns]. Hence, thenegativity for gender-incongruent head constituents wasclassified as a LAN.

Nonhead Constituent

Again, an ANOVA with the factors ROI (4), genderagreement of non-head constituents (C1; 2), and se-mantic status (2) was performed in the time window400–500 msec after compound onset (Figure 2). The in-teraction of C1 and semantic status did not reach sig-nificance [F(1,22) = 0.78; ns]. However, the two-wayinteraction of ROI and C1 was significant [F(3,66) =4.17; p < .05; > = .75], and subsequently the effect of C1was tested separately for each ROI. Only the left-anteriorROI showed a significant main effect of C1 [F(1,22) =4.35; p < .05], whereas the other ROIs were nonsignif-icant [AR: F(1,22) = 0.25; PL: F(1,22) = 0.04; PR: F(1,22)= 0.25; all ns]. Thus, the negativity for gender viola-tions of nonhead constituents is interpreted as a LAN.

Discussion

The results support the hypothesis of morphosyntacticdecomposition of compounds. Gender-incongruentconstituents elicited a LAN that was independent ofsemantic status. The LAN in response to head constitu-

Table 1. Stimulus Examples Used in Experiment 1

TT OO

Condition Determiner C1 C2 C1 C2

AA der Regen – tag Schnee – besen

themasc rainmasc – daymasc snowmasc – broommasc, egg whisk

AV *der Reis – feld Wal – ross

*themasc ricemasc – fieldneut whalemasc – steedneut, walrus

VA das Presse – amt Luft – schloss

theneut pressfem – officeneut airfem – castleneut, daydream

VV *das Nuss – baum Draht – esel

*theneut nutfem – treemasc wiremasc – donkeymasc, bicycle

The four conditions are determined by the agreement (A) or violation (V) of nonhead (C1) and head (C2) constituents. Note that C2 alonedetermines the compound’s gender, and thus the correctness of the assigned determiner. Each item was presented only once to each subject; theitems were presented in one condition to half of the subjects and in another condition (i.e., with another determiner to the other half of thesubjects). The asterisks denote incorrect determiners.


ents was also independent of the gender agreement ofnonhead constituents. As we are concerned with mor-phosyntactic features, we will not discuss any effects ofsemantic status, in particular, because semantic statusdid not interact with the gender manipulations.

Gender-incongruent head constituents, which deter-mine the compound’s gender, elicited a LAN. Thiseffect does not only replicate previous findings forgender manipulations (Deutsch & Bentin, 2001; Gunteret al., 2000), but also serves as a solid benchmark forevaluating the nonhead constituents. The comparisonof the agreement and the violation condition was awithin-item comparison across subjects. Hence, theLAN cannot be explained by differences between itemgroups. The LAN was also independent of the semanticstatus which confirms the previous finding that syntac-tic gender and semantic information is processed inde-

pendently (Gunter et al., 2000) during early processingstages.

The morphosyntactic features of nonhead constitu-ents are irrelevant in German compounds. Yet, nonheadconstituents elicited a LAN if gender was incongruent.We therefore conclude that the gender information ofnonhead constituents is processed (i.e., a morphosyn-tactic representation is accessed). Note that prosodicinformation, namely, duration of nonhead constituents,discloses the constituent as part of a compound word(Isel et al., 2003). Given the prosodic cue, the questionarises as to why nonhead gender information is pro-cessed at all, given that it is not relevant in German. It issuggested that the gender information becomes avail-able as an intrinsic part of the morphosyntactic repre-sentation of the constituent (cf. Caramazza et al., 2001).This morphosyntactic representation might serve as an

Figure 1. The ERPs for gender-congruent (solid line) and -incongruent (dashed line) head constituents of low-frequency compounds. Gender-incongruent violations elicit a larger negativity at left-anterior electrodes (400–500 msec) which was independent of semantic status. Top of the

figure shows frontal electrodes, left (right) hemisphere electrodes are denoted by odd (even) numbers. Constituent onset is at time 0 msec, the

horizontal arrow indicates the average constituent length, and negative is plotted up in this and all subsequent ERP plots.

Koester et al. 1651

access code for a later retrieval of the constituent’ssemantics, for example, if the compound has no lexicalentry. The nonhead LAN was observed for both trans-parent and opaque compounds which parallels themorphological decomposition of both types of com-pounds, found in the visual domain (Zwitserlood,1994). The morphosyntactic decomposition representsstrong evidence against the full-listing hypothesis withrespect to morphosyntactic processing. The data arecompatible with full-parsing models (i.e., with a fullmorphosyntactic specification of nonhead constituents)and with dual-route models.

What function does the nonhead LAN signify? As it wassuggested, it may reflect the automatic checking of agender agreement between two linguistic elements. Thismight help to determine that a constituent is not the lastof a compound, that is, a further constituent has to be pro-cessed. Such a mechanism cannot fully suffice becausenonhead constituents may also agree in gender with a

determiner. Thus, the function reflected by the non-head LAN cannot be fully clarified with the present data.

Subsequent to the LAN for the head, but not forthe nonhead constituent, a larger positivity is present(Figure 3). Although a P600 may be expected, as it wasreported for gender violations within sentences (Gunteret al., 2000; Hagoort & Brown, 1999), the effect can-not be interpreted easily as the task cue is presented200 msec after stimulus offset and processes not directlyrelated to the stimulus might take place in this timerange. When the task cue is presented, subjects have toread it and decide which task to perform, then they haveto decide which answer is required, and finally prepareand execute the response. Thus, the positivity might, forexample, be a P300 in response to the task-relevant cue(Picton, 1992). P300 has been shown to be sensitive tofactors as task difficulty or stimulus salience (Perrin,Garcıa-Larrea, Mauguiere, & Bastuji, 1999; Pritchard,1981). Although it is yet unknown whether a gender

Figure 2. The ERPs for gender-congruent (solid line) and -incongruent (dashed line) nonhead constituents of low-frequency compounds. Genderviolations elicit a larger negativity at left-anterior electrodes (400–500 msec) which was independent of semantic status.


violation is easier to detect or more salient than gendercongruence, it is possible that such a difference causedthe larger positivity for the gender-incongruent condi-tion. Thus, the multiple co-occurring processes precludea valid interpretation of the positivity. If it is not a P600,the absence might be explained by the fact that the min-imal structures could not be corrected by a structuralreanalysis, and that the P600 is susceptible to controlledprocesses (Hahne & Friederici, 1999). Subjects were notinstructed to correct morphosyntactic errors and maynot have done so.

EXPERIMENT 2

Morphosyntactic decomposition is a prerequisite to testthe function of linking elements. Linking elements arephonemes/graphemes between the free morphemes ofcompounds (cf. Table 2), and about 35% of Germannominal compounds contain such linking elements(Krott, 2001, chap. 6). The most common and productiveGerman linking elements are -s-, -n-, -en-, -er-, and -e-.Often they are identical in form to the plural morphemeof the preceding constituent, for instance, ‘‘Trauben-kerne’’ (grape[s] seeds) and therefore one proposedfunction of linking elements is to indicate the numberof the preceding constituent (Clahsen, 1999; Schreuderet al., 1998; Wiese, 1996; Clahsen, Rothweiler, & Woest,1992). In the case of grape seeds, such seeds may indeedstem from several grapes. However, the number function

does not apply systematically. For example, ‘‘Fischge-schaft’’ (fish shop) refers certainly to a shop containingmore than one fish; however, ‘‘Fisch’’ is singular. More-over, the plural reading of the nonhead constituent mayadditionally depend on the number of the head constit-uent as one ‘‘Traubenkern’’ (grape[s] seed) cannotpossibly stem from more than one grape. In such cases,the number of nonhead constituents is not determinedsufficiently by the linking element, thus rendering itsfunction questionable.

Number is a morphosyntactic feature that marks asemantic/conceptual aspect (Schiller & Caramazza, 2003;Corbett, 2000). In German, nouns are marked for pluralby one of five suffixes (-er, -e, -(e)n, -Ø, and -s), some-times combined with a vowel change (umlaut; Wegener,1999; Clahsen et al., 1992). Vowel changes are conceivedof as phonological or morphological variants of theaforementioned plural suffixes (Bartke, 1998; Wegener,1992). Plural morphemes are classified as either irregularor regular (Clahsen, 1999; Weyerts et al. 1997; Marcus,Brinkmann, Clahsen, Wiese, & Pinker, 1995). The suffix-s is assumed to be the regular (default) plural suffix,because it is phonologically not restricted in its applica-tion (Wiese, 1996). That is, it is used for foreign words,

Figure 3. The positivity following the LAN for gender-incongruent

head constituents (700–1200 msec) is reliable at posterior electrodes

[ROI (4) � C2 (2): F(3,66) = 9.53; p < .001; > = 0.72; PL: F(1,22) =14.72; p < .001; PR: F(1,22) = 8.46; p < .01; AL: F(1,22) = 3.81; p < .1;

AR: F(1,22) = 0.33; ns]. However, the positivity cannot be interpreted

as a genuine P600 as several processes co-occurred in this time range;

see text.


Condition Numeral C1 C2

AA einsg Tiersg – versuchsg

an/one animal – experiment

zweipl Liederpl – abendepl

two songs – nights

AV *einsg Feldsg – wegepl

*a/one field – paths

*zweipl Ohrenpl – zeugesg

*two ears – witness

VA zweipl Feldsg – wegepl

two field – paths

einsg Ohrenpl – zeugesg

an/one ears – witness

VV *zweipl Tiersg – versuchsg

*two animal – experiment

*einsg Liederpl – abendepl

*a/one songs – nights

The four conditions are determined by the agreement (A) or violation(V) of the nonhead (C1) and the head (C2) constituent. Note that C2alone determines the compound’s number and thus, the correctnessof the assigned numeral. Linking elements are underlined. Each itemwas presented only once to each subject; in a second session, the samestimuli were presented, but in another condition.

sg = singular; pl = plural.

Koester et al. 1653

abbreviations, proper names, and new word coinages; itis appended to a noun stem according to an abstractrule (Wiese, 1996; Marcus et al., 1995). Irregularly in-flected nouns, taking one of the remaining suffixes forplural formation, are suggested to have both an unin-flected and an inflected (plural) entry in the mentallexicon. Note that this classification has been challengedby theoretical (Wegener, 1992, 1999) and empiricalinvestigations (Penke & Krause, 2002; Sonnenstuhl &Huth, 2002).

Another argument for the classification of -s as theregular plural morpheme comes from the so-called level-ordering approach of word formation (Wiese, 1996;Kiparsky, 1982). Here it is proposed that the formationof words is dependent upon processing on three differ-ent serially ordered levels. On the first of these (level 1)irregular inflection takes places. This is followed bylevel 2, where compounding is supported, and finallyon level 3 regular inflections are carried out. Followingthis serial ordering, -s- as a regular plural morphemecannot appear within compounds because regular inflec-tion takes place after compounding, and the compoundinternal structure is not available anymore (Kiparsky,1982). Only the irregular plural morphemes can ap-pear as linking elements because they are attached onlevel 1 before compounding. Indeed, the linking -s- doesnot appear within German compounds as a possibleplural morpheme (e.g., *‘‘Autosmarkt,’’ *‘‘cars market,’’car sale; Clahsen, 1999; Wiese, 1996; Clahsen et al.,1992).2 Crucial to this argumentation for the regularityof -s is the assumption that linking elements function-ally serve as plural morphemes. However, this has notyet been shown. On the contrary, linking elements havealso been suggested to serve other functions, for exam-ple, prosodic (Jarema, Libben, Dressler, & Kehayia, 2002;Gawlitzek-Maiwald, 1994) or morphological (Aronoff &Fuhrhop, 2002; Fuhrhop, 2000, see also Libben, Jarema,Dressler, Stark, & Pons, 2002). In Experiment 2, weinvestigate whether or not linking elements indeed in-dex number.

Previous ERP studies reported N400 effects for theincorrect application of irregular plural morphemes insentence processing (Wolf, 2004; Luck, Hahne, & Clah-sen, 2001; Weyerts et al., 1997). Because irregular pluralforms are assumed to be stored as unique entries in themental lexicon (Clahsen, 1999), the incorrect applicationof irregular plural morphemes has been suggested toactivate a lexical search (Weyerts et al., 1997). Theincreased difficulty in lexical–semantic search or seman-tic integration is reflected in N400 effects (Kutas &Federmeier, 2000). Hence, the N400 effects for incorrectirregular plural morphemes can be explained by a lexicalsearch, although number is generally seen as a morpho-syntactic feature.

In order to test the proposed function of linkingelements to index number, we presented subjects withspoken numerals3 followed by two-constituent com-

pounds. All constituents require an irregular plural suffixand no feminine nouns were used, as these would havenecessarily been preceded by a saliently inflected nu-meral (‘‘eine’’, a(n)/one). We used only the numeral‘‘ein’’ because this agrees with masculine and neuternouns. We collected semantically transparent, existingGerman compounds that allowed us to vary the numberagreement between the numeral and each compoundconstituent independently (Table 2). Opaque com-pounds cannot be manipulated in such a way (theyemerge diachronically) and were, thus, not included.Subjects’ task was to judge either the correctness of thenumeral agreement or to compare the items semanti-cally to another visually presented word.

Experiment 1 showed that compounds are morpho-syntactically decomposed. If linking elements are pluralmorphemes, that is, if nonhead constituents are speci-fied for number, an N400 effect is predicted for number-incongruent nonhead constituents. If no ERP effect isfound for nonhead constituents, this suggests that thefeature number is not processed for these constituents.However, in both cases, an N400 effect is predicted forincongruent head constituents, because these are syn-tactically relevant.

Results

Behavioral Data

Subjects reported no problems and performed well inboth tasks. Accuracy was high in both the grammaticality(95%) and the semantic task [95%; t(23) = 0.65; ns].

ERP Data: Head Constituent

The mean amplitude was calculated for all conditions forthe time window 300–500 msec time-locked to the onsetof the head constituent. Number-incongruent constitu-ents elicited a more negative ERP than number-congru-ent constituents (Figure 4). An ANOVA was performedwith the factors ROI (4), number agreement of the headconstituent (C2; 2), and number agreement of the non-head constituent (C1; 2). The three-way interaction ROI� C2 � C1 was significant [F(3,69) = 3.61; p < .05; > =.68]. Neither the main effect of C1 [F(1,23) = 0.11; ns]nor any other interaction including C1 or C2 (all F < 1)reached significance. Subsequent ANOVAs were per-formed separately for each ROI with the factors C2and C1. There was a main effect of C2 in the left-anteriorROI [F(1,23) = 4.45; p < .05], and in both posteriorROIs [PL: F(1,23) = 6.79; p < .05; PR: F(1,23) = 12.17;p < .01]. In the right-anterior ROI there was only amarginal effect of C2 [F(1,23) = 3.6; p < .10]. Giventhe posterior distribution of the negativity with a cen-troparietal maximum, we classify it as an N400 effect.No further effect involving C1 could be confirmed (allF < 0.74; all ns).


In addition, a later negativity was observed for num-ber-incongruent head constituents extending from 500to 700 msec after head constituent onset. In order totest its reliability, an ANOVA with the same factors (ROI,C2, C1) was performed on the mean amplitude values inthis time window. A significant main effect of C2 wasobtained [F(1,23) = 16.67; p < .001]. Neither the maineffect of C1 nor any interaction involving C1 reachedsignificance (all F < 2.7; all ns).

It is possible that both the N400 and the subsequentnegativity reflect in fact one component. As it appearsfrom Figure 4, the ‘‘number agreement effect’’ is ratherextended in time, ranging from possibly 50 msec toabout 700 msec. In order to determine more preciselythe onset of the ‘‘number agreement effect’’, we per-formed further ANOVAs with the factors ROI (4) andC2 (2) in subsequent 50 msec time windows from 50 to700 msec. These more fine-grained analyses suggestthat the ERPs in response to number-(in)congruenthead constituents begin to diverge consistently about

200 msec post head onset. The interaction of C2 withthe factor ROI was not significant anymore.

Nonhead Constituent

The ANOVA with the factors ROI (4), and numberagreement (C1; 2) in the time window 300–500 msecyielded neither an interaction [F(3,69) = 0.56; ns] nor amain effect of C1 [F(1,23) = 0.00; ns]. SubsequentANOVAs in consecutive 50 msec time windows from200 to 600 msec after compound onset could not detectany effect of C1. That is, the ERP was not affected bynumber-incongruent nonhead constituents (Figure 5).

Discussion

The results of Experiment 2 do not support theinterpretation of linking elements as plural mor-phemes. Rather, the results are compatible with inter-

Figure 4. ERPs for number-congruent (solid line) and -incongruent (dashed line) head constituents of low-frequency transparent compounds. The

number-incongruent head constituents elicit an N400 effect (300–500 msec, with the ERPs beginning to differ significantly about 200 msec afteronset) followed by a later negativity (500–700 msec).

Koester et al. 1655

pretations that propose a distinct function for linkingelements. Number-incongruent head constituents eli-cited an N400 followed by a broad negativity, whereasno effect at all was observed for number-incongruentnonhead constituents.

The null effect of Experiment 2 is apparently at oddswith the gender effect of nonhead constituents foundin Experiment 1. Note, however, that number differsfrom gender in some respects. Number, but not gen-der, is a variable feature (singular vs. plural), that is, itneeds specification. Number entails semantic/concep-tual consequences, and it is (mostly) marked on thesurface (cf. Schiller & Caramazza, 2002, 2003). The mor-phosyntactic specification of number and its semantic/conceptual consequences constitutes workload that maybe spared because it is syntactically irrelevant for non-head constituents.

An N400 effect in response to number-incongruenthead constituents is in accordance with previous find-ings (Wolf, 2004; Luck et al., 2001; Clahsen, 1999;Weyerts et al., 1997). However, it is possible that morethan one cognitive process is reflected in the observed

N400 effect, especially given its temporal extent. Num-ber incongruities entail, in contrast to gender, concep-tual consequences, in addition to the morphosyntacticdisagreement. At present, these processes cannot bedisentangled for the processing of number. The rele-vance of the present N400 effect lies in its benchmarkfunction for evaluating nonhead constituents.

The N400 effect for head constituents had an earlyonset at about 200 msec after the constituent boundary.The effect reflects the processing of the number incon-gruity, however, it was observed paradoxically, beforethe plural suffix could be perceived.4 However, the suffixis not the only source of information regarding thecompound’s number. Prosodic information may alsocontain cues to the number feature. The precedingnonhead constituent and the suffixation of the headconstituent entail coarticulation and assimilation effects(e.g., terminal devoicing), and may also change thestress pattern (Wiese, 1996; Kohler, 1995; Pompino-Marschall, 1995). Such prosodic effects make it possiblethat the number information is processed before thesuffix is perceived.

Figure 5. ERPs for number-congruent (solid line) and -incongruent (dashed line) nonhead constituents of low-frequency transparent compounds.

The manipulation of ‘‘number agreement’’ via linking elements that are identical in form with plural morphemes did not yield an ERP effect.


The number violation effect was extended up to700 msec. However, the effect did not differ topograph-ically among ROIs between 500 and 700 msec. N400effects with an extended duration and/or an early onsethave been reported previously (Van Petten, Coulson,Rubin, Plante, & Parks, 1999; Holcomb & Neville, 1990;McCallum, Farmer, & Pocock, 1984) but these studieswere concerned with semantic processing. A similareffect was not reported in studies in which plural forma-tion was investigated. The extended negativity may berelated to the suffix processing or reflect a (intentional)postlexical checking of ‘‘number agreement’’ due to theexplicit task, but this remains speculation.

Number-incongruent nonhead constituents shouldelicit an N400 effect, possibly with an extended durationif linking elements are processed as plural morphemes.However, the number violation yielded neither an N400effect nor any other ERP effect. This null effect showsthat linking elements of noun–noun compounds do notindicate by their presence or absence the number of thepreceding constituent. That is, numerals and nonheadconstituents were not processed as incongruent innumber. According to this view, number does not playa role for the processing of nonhead constituents, whichargues against a complete morphological analysis ofnonhead constituents as suggested by full-parsing mod-els (Libben, Derwing, et al., 1999; Taft & Forster, 1976).Hence, the results are compatible with theoretical viewsthat do not ascribe a number function to linking ele-ments (cf. Aronoff & Fuhrhop, 2002; Jarema et al., 2002;Libben, Jarema, et al., 2002; Krott, 2001; Fuhrhop, 1998,2000; Gawlitzek-Maiwald, 1994).

Experiment 1 showed that compounds are decom-posed morphosyntactically, and in Experiment 2, thesyntactically relevant head constituents resulted in anumber agreement effect that was absent for nonheadconstituents. Even if the morphosyntactic features num-ber and gender differ, there must be a factor thatprevents the interpretation of linking elements as pluralmorphemes. Recently, Isel et al. (2003) showed thatprosody is crucial for the semantic processing of com-pounds in the auditory modality. In order to test whetherprosody controls also the morphosyntactic compoundprocessing, a third experiment was conducted.

EXPERIMENT 3

Linking elements and plural morphemes are often ho-mophonous and, therefore, can create an ambiguityregarding the morphosyntactic specification of number.To investigate the involvement of prosody in disambig-uating linking elements and plural morphemes, wescrutinized the number agreement between (i) singlenouns and numerals and (ii) nonhead constituents andnumerals. That is, the same words were presented eitherwith a single noun prosody or with a compound pros-

ody, preceded by a (in)congruent numeral (Table 3). Incontrast to Experiment 2, linking elements are taskrelevant in Experiment 3, that is, the agreement judg-ment is based on these elements.

The design and tasks are equivalent to Experiment 2,but the stimuli differ. If subjects have to judge thenumber agreement between a numeral and a nonheadconstituent, the subsequent constituents would bestrongly distracting. Therefore, after having a naivespeaker produce the same words alone and in properGerman compounds, we extracted the head constitu-ents of the compounds electronically and filled theirintensity envelope with white noise. Thus, the phonemiccontent was removed while preserving the durationaland intensity parameters. In a last step, these ‘‘whitenoise constituents’’ were spliced onto both the singlenoun and the nonhead constituent from which it orig-inated. This procedure gave us words followed by whitenoise that resembled an acceptable head constituentprosodically (in duration and intensity contour). Thecritical items were presented once with a single nounprosody and once with a compound prosody (cf.Table 3). An equal number of unmanipulated com-pounds and single nouns were additionally presented,all items in a randomized order. Hence, there was nobias in the experimental setup towards either interpre-tation of the critical items.

If indeed prosody disambiguates linking elements andplural morphemes, number agreement should affectonly the ERP elicited by single nouns, but not bycompound constituents. Thus, we should find an N400effect for number-incongruent single nouns, and com-pound constituents should result in a null effect repli-cating Experiment 2.

Results

Prosody Data

In order to confirm that the stimuli differ prosodically,we analyzed the duration and fundamental frequency(pitch) of the critical stimuli. The comparisons arewithin-word comparisons, thus, excluding lexical differ-ences. The same words were shorter when they wereproduced as the nonhead constituent of a compound(505 msec on average) than as single nouns [652 msec;t(175) = 27.4; p < .001]. Single nouns and compoundconstituents were tested for pitch differences in 25-msectime slots. Pitch values were registered electronically(Boersma & Weenink, 1992–2003) for each 25-msec timewindow of each critical item and were then averaged foreach time slot across all single nouns and nonhead con-stituents, respectively. The difference in pitch was mar-ginally significant in the third time window (50–75 msec)and significant in all subsequent time windows, namely,from 75–100 msec onwards (Figure 6; Table 4). Hence,duration and pitch are physical parameters that reliably

Koester et al. 1657

indicate whether or not a perceived word is a singlenoun or a compound constituent. So far, only theparameter duration was shown to be a reliable indicatorfor morphological complexity (Isel et al., 2003; Daviset al., 2002; Cutler et al., 1997).

Behavioral Data

Subjects reported no problems and performed with highaccuracy in the grammaticality task (GRA; 95%) and thesemantic task (SEM; 94%). Accuracy did not differ be-tween the tasks [F(1,27) = 1.46; ns], but there was a maineffect of prosody. Accuracy was higher for single nouns(SIN; 96%) than for compound constituents [COM; 93%;F(1,27) = 15.44; p < .001]. There was no interaction oftask and prosody [F(1,27) = 0.18; ns]. RT data are similar.Most importantly, there is also a main effect of prosody[SIN 664 msec; COM 688 msec; F(1,27) = 5.96; p < .05]besides a main effect of task [GRA 570 msec; SEM782 msec; F(1,27) = 214.99; p < .001]. In addition, therewas a marginally significant interaction of task and pros-ody [F(1,27) = 4.03; p = .055], which essentially meansthat the main effect of prosody is restricted to the

grammaticality task [SIN 550 msec; COM 589 msec;t(27) = 2.41; p < .05; semantic task: SIN 778 msec;COM 787 msec; t(27) = 1.43; ns]. That is, the behavioralresults show that prosody affects at least the number


Condition Numeral 1st Session 2nd Session

COM-A einsg Zeltsg, com–[mast]noise Konzertsg, com–[zyklus]noise

a/one tentcom–[pole]noise concertcom–[cycle]noise

zweipl Motorenpl, com–[gerausch]noise Bilderpl, com–[album]noise

two enginescom–[sound]noise picturescom–[album]noise

COM-V *einsg Bilderpl, com–[album]noise Motorenpl, com–[gerausch]noise

*a/one picturescom–[album]noise enginescom–[sound]noise

*zweipl Konzertsg, com–[zyklus]noise Zeltsg, com–[mast]noise

*two concertcom–[cycle]noise tentcom–[pole]noise

SIN-A einsg Rekordsg, sin–[wert]noise Pfeilsg, sin–[gifte]noise

a/one recordsin–[value]noise arrowsin–[poisons]noise

zweipl Lowenpl, sin–[kafig]noise Kleiderpl, sin–[depot]noise

two lionssin–[cage]noise dressessin–[repository]noise

SIN-V *einsg Kleiderpl, sin–[depot]noise Lowenpl, sin–[kafig]noise

*a/one dressessin–[repository]noise lionssin–[cage]noise

*zweipl Pfeilsg, sin–[gifte]noise Rekordsg, sin–[wert]noise

*two arrowsin–[poisons]noise recordsin–[value]noise

The four conditions are determined by the prosody of the items (SIN = single noun, and COM = compound prosody) and the numberagreement (A) or violation (V) between the numeral and the word. Note that the second constituents (in brackets) were not intelligible.They were replaced with white noise that preserved their duration and intensity envelope. A second version was identical, but the prosodywas reversed for all items. That is, words with a compound prosody in version 1 were presented with a single noun prosody in version 2.For instance, instead of presenting ‘‘Zelt,’’ ‘‘Motoren,’’ ‘‘Bilder,’’ and ‘‘Konzert’’ with a compound prosody, ‘‘Rekord,’’ ‘‘Lowen,’’ ‘‘Kleider,’’and ‘‘Pfeil’’ were presented with a compound prosody in the second version, and vice versa. Linking elements are underlined.

sg = singular; pl = plural.

Figure 6. The average pitch contour for the same words produced as

single nouns and as compound constituents. Each time slotcorresponds to 25 msec. Pitch is significantly lower for single nouns

from the 3rd time slot onwards (75–100 msec). The difference is also

marginally significant in the 2nd time slot (50–75 msec).


judgment of single nouns and nonhead constituentsdifferently.

ERP Data: Single Nouns

From Figure 7, it is apparent that there is a broad and,compared to Experiment 2, later negativity for number-incongruent single nouns. In order to test this negativ-ity, the mean ERP amplitude was calculated for thetime window 600–900 msec after noun onset. TheANOVA with the factors anterior–posterior (AP; 2),hemisphere (2), and number agreement (2) yielded amain effect of number agreement [F(1,27) = 6.46; p <.05] and a marginally significant three-way interaction

[F(1,27) = 3.28; p = .08]. Subsequent ANOVAs wereperformed for the resulting ROIs separately to deter-mine the scalp distribution of the effect. The effect ofnumber agreement was significant in the left posterior[F(1,27) = 9.04; p < .001], left anterior [F(1,27) =7.81; p < .001], and was marginally significant in theright anterior ROI [F(1,27) = 3.24; p = .08]. The effectwas not reliable in the right posterior ROI [F(1,27) =2.31; ns].

ERP Data: Compound Constituents

The same ANOVA with the factors AP (2), hemisphere(2), and number agreement (2) was performed for com-pound constituents. In the time window 600–900 msecthere was neither a main effect of number agree-ment [F(1,27) = 0.71; ns] nor an interaction of numberagreement with AP [F(1,27) = 0.22; ns], hemisphere[F(1,27) = 0.05; ns], or both [F(1,27) = 0.75; ns; cf.Figure 8]. In additional 50 msec analyses between 200and 900 msec, there was also no consistent effect ofnumber agreement.

Discussion

Compound constituents and single nouns differ in du-ration and pitch (contour). Compound constituentswere about 150 msec shorter, and pitch was higher afterabout 75–100 msec compared to single nouns5 (fordifference values, cf. Table 4). Duration was expectedto be shorter (Isel et al., 2003), but to our knowledge,pitch differences are reported for the first time. Theaverage pitch contour of compound constituents isroughly level with an early minimum, whereas the pitchcontour of single nouns shows also an early minimumbut then falls (Figure 6). Isel et al. (2003) did not finddifferences in their measure of pitch, which may be dueto a different method of analysis.6

The behavioral as well as the ERP results do not sup-port the concept that linking elements are plural mor-phemes. Compound constituents elicited more errorsand required more time for the agreement judgmentthan single nouns. These effects reflect a functionaldifference in processing because the same words wereused in both prosodic conditions. They suggest that thecompound prosody makes it harder to evaluate thelinking elements as plural morphemes (and their ab-sence as indicating singular). Hence, we conclude thatlinking elements are not processed primarily as pluralmorphemes.

The number disagreement elicited a broad negativitybetween 600 and 900 msec after noun onset for singlenouns, but not for compound constituents. With regardto the left hemispheric preponderance and the late ap-pearance of the N400-like effect, we can only speculate.Generally, N400 effects in the auditory modality were

Table 4. T tests for Pitch Differences between Single Nouns(SIN) and Compound Constituents (COM) for Successive25 msec Time Windows (Time Slots)

Time SlotsPitch—SIN(25 msec)

Pitch—COM(25 msec) t value df p value

1 250 249 �0.45 78 .66

2 244 245 0.46 96 .65

3 243 246 1.88 99 .06

4 247 251 2.51 123 .01

5 251 256 3.16 132 <.01

6 253 258 2.74 135 <.01

7 253 258 2.27 135 .03

8 253 258 2.44 126 .02

9 251 260 4.25 116 <.01

10 249 261 5.30 114 <.01

11 247 260 5.94 117 <.01

12 244 261 7.04 108 <.01

13 243 262 7.56 106 <.01

14 239 259 7.50 102 <.01

15 233 257 9.53 90 <.01

16 234 258 7.57 78 <.01

17 238 258 4.47 73 <.01

18 232 254 6.16 62 <.01

19 232 257 6.92 50 <.01

20 232 261 8.91 46 <.01

21 232 261 7.78 42 <.01

22 232 263 7.51 36 <.01

23 231 264 7.29 34 <.01

24 227 264 8.09 31 <.01

Any word is included in both prosodic conditions. Pitch values are inHz, p values are two-tailed. Pitch differences were only tested if morethan 30 items contributed to the pitch average (all time slots up to 24).

Koester et al. 1659

reported to have a longer duration (Van Petten et al.,1999; Holcomb & Neville, 1990; McCallum et al., 1984),and, more specifically, N400 effects in response toincorrect plural word forms also showed a later ap-pearance (500–700 msec, Luck et al., 2001; 550–700 msec,Wolf, 2004, Experiment V, control group). Althoughthese findings resemble the present result, they are notexactly parallel because in those studies only pluralforms were investigated and only within sentences.Moreover, they did not contain transitions from speechto noise. Although the negativity in response to number-incongruent single nouns is visible at most electrodes, itdoes not reach significance over the right hemisphere.The left hemispheric preponderance might be relatedto the ‘‘noise constituents,’’ which may be distractingand may disturb the comprehension process. In orderto compensate for such a disturbance, left hemisphericcomprehension processes (Springer & Deutsch, 1997)might be strengthened strategically throughout the ex-

perimental session. Still, this tentative explanation needssupport from further research.

Taken together, the N400-like negativity shows that anumber mismatch between single nouns and numeralsis detected after 600 msec. The negativity may reflectthe detection of the violation as well as the processingof their conceptual consequences, but these processescannot be disentangled with the present data.

Compound constituents elicited no ERP effect in thesame time range (600–900 msec). This null effect iscrucial because linking elements were task relevant andthe behavioral data indicate that subjects employed thisinformation. The null effect shows that nonhead con-stituents were distinguished from single nouns beforetheir (proposed) number information became available,and this can only be due to the prosodic informationcontained therein. Here, it is suggested that prosodiccues are used to reconfigure the parser in order tomake processing more efficient. The differential ERPs

Figure 7. The ERPs in response to number-congruent (solid line) and -incongruent (dashed line) single nouns. Number-incongruent nouns elicit a

broadly distributed, larger negativity between 600 and 900 msec that is significant over the left hemisphere.


show that phonologically identical morphemes servedifferent functions in single noun and compound pro-cessing. That is, linking elements are not processed asplural morphemes.

GENERAL DISCUSSION

The present article investigated whether compoundconstituents are morphosyntactically decomposed andwhether linking elements are processed as plural mor-phemes. Experiment 1 showed that syntactic gender in-formation of nonhead constituents is available fortransparent and opaque compounds during auditorycomprehension. Experiment 2, however, showed thatthe number of nonhead constituents is not processed(i.e., linking elements are not plural morphemes). Ex-periment 3 showed that this suppression of numberprocessing is controlled by prosodic cues.

The manipulations of the head constituents served asa benchmark in two respects. First, they showed which

effects should be found for nonhead constituents. Ifnonhead constituents were processed like head constit-uents (regarding access of gender and number spec-ification), they should show the same ERP effects.Second, the results of the head constituents allowedus to relate the present results to previous investiga-tions. The syntactically relevant head constituents eli-cited a LAN if gender was incongruent, as previouslyreported (Deutsch & Bentin, 2001; Gunter et al., 2000),and possibly a P600 (see Discussion Experiment 1).Number incorrect head constituents which take so-called irregular plural morphemes elicited an N400 effect(cf. Clahsen, 1999; Weyerts et al., 1997) and a subse-quent broad negativity. Thus, the results of the headconstituent manipulations are in accordance with pre-vious findings.

The findings for the nonhead constituents are com-patible with dual-route models (Isel et al., 2003;Schreuder et al., 1998; Baayen et al., 1997; Zwitserlood,1994; Sandra, 1990), but not with full-listing (Bybee,

Figure 8. The ERPs in response to ‘‘number’’-congruent (solid line) and -incongruent (dashed line) compound constituents. Number agreement

was determined by the presence or absence of linking elements. The manipulation yielded no significant ERP effect, suggesting that nonhead

constituents were not specified for number.

Koester et al. 1661

1995; Elman, 1993; McClelland et al., 1986; Butterworth,1983) or full-parsing approaches (Libben, Derwing, et al.,1999; Taft & Forster, 1976). The availability of non-heads’ gender information suggests that the parseris sensitive to the internal morphological structure ofcompounds during auditory comprehension. This re-sult extends the morphological decomposition of trans-parent and opaque compounds, found in the visualmodality (Zwitserlood, 1994), to the auditory modal-ity. It contradicts the full-listing idea because opaquecompounds which require a lexical entry are morpho-syntactically decomposed (Experiment 1). The nulleffects for nonhead constituents (Experiments 2 and3) are at odds with full-parsing approaches becausenonhead constituents are apparently not processedregarding number as may be expected if all consti-tuents are fully parsed. These null effects strongly sug-gest that linking elements are processed online assyntactically irrelevant for number. Experiment 3 alsoshows that prosody is used to disambiguate phono-logically identical linking elements and plural mor-phemes. A null effect (nonhead constituents) does notprove the absence of an effect, but under a differentprosodic condition (single nouns) an effect of num-ber violation was observed. Hence, we can reasonablyassume that there was no effect of number (dis)agree-ment for nonhead constituents. Psycholinguistically, itis plausible that the effort to specify a variable fea-ture (number) is spared, whereas gender seems to beaccessed as an intrinsic part of the morphosyntacticnoun representation (Schiller & Caramazza, 2002, 2003;Caramazza et al., 2001).

The present findings are explained by a cascadingdual-route model. The incoming phonetic signal is con-tinuously matched to lexical representations via a directroute which is required in order to access opaquecompounds (Isel et al., 2003; Zwitserlood, 1994; San-dra, 1990). As it was suggested by Isel et al. (2003), weassume that prosodic cues are analyzed in parallel tothe segmental input. The detection of a compoundprosody changes the configuration of the parser andcalls up an additional decompositional route. The ac-cess of syntactic gender suggests that a morphosyn-tactic representation is extracted immediately via thedecompositional route, but number does not seem toplay a role for the extracted morphosyntactic repre-sentations. Such a representation may serve as anaccess code to semantic information of constituentsif no single lexical representation matches the incom-ing signal (via the direct route). Semantic constituentinformation is necessary for (transparent) compoundsthat have no lexical entry. If indeed nonhead con-stituents are only activated semantically at the endof the compound (Isel et al., 2003, but see Wagner,2003), the morphosyntactic representation has to bestored in working memory until the head is identified.Morphosyntactic representations are underspecified

and may, therefore, be less demanding for storage. Ifa morphosyntactic representation is not needed (e.g.,for opaque compounds), it may be discarded. Althoughit is suggested that a morphosyntactic representation isused as an access code, there is another alternative,namely, a phonological constituent representation. Thiswould essentially require a rehearsal process in addi-tion to the ongoing acoustic–phonetic processingwhich we, therefore, assume unlikely.

The reported data do not support the classificationof regular (-s) versus irregular plural morphemes (-er,-e, -(e)n, -Ø). The proposed serial constraint in com-pounding (Wiese, 1996; Kiparsky, 1982) presumes so-called irregular plural morphemes to appear withincompounds, but not regular ones. Because the pres-ent findings show, at least for comprehension, thatlinking elements are functionally distinct from pluralmorphemes, the presupposition is not warranted.Hence, the kind of linking elements that appear withincompounds cannot support the classification of pluralmorphemes.

Another question concerns the interpretation of theN400 effect in response to number violations. It wassuggested that it reflects the increased effort of lexicalsearch for stored (irregular) plural word forms (Clahsen,1999; Weyerts et al., 1997). To date, it is unknown, butwell possible, that such N400 effects consist of morethan one overlapping component. Number violationscomprise a morphosyntactic and a conceptual aspect,and both have to be processed. At present, theseprocesses are not completely understood. It might turnout, that the -s serves as the default plural morpheme(e.g., for novel or foreign words), and that there areother instances of regular pluralization (Penke & Krause,2002; Wegener, 1999) besides the storage of unpredict-able (i.e., irregular forms).

In summary, the present experiments show thattransparent and opaque compounds are decomposedmorphosyntactically during auditory processing. Linkingelements that are homophonous with plural mor-phemes are not processed as plural morphemes. Pro-sodic cues differentiate single nouns and compounds,and are used to disambiguate linking elements andplural morphemes. Further research is needed to deter-mine the functional significance of linking elements.

METHODS

Experiment 1

Subjects

Twenty-three (12 men) right-handed (lateralization co-efficient 95; Oldfield, 1971) volunteers participated witha mean age of 24.7 years (range 19–31). In all three ex-periments, subjects had normal or corrected-to-normalvisual and auditory acuity and were paid for theirparticipation.


Stimuli

We collected 80 two-constituent compounds that weresemantically of transparent–transparent (TT) status andanother 80 compounds that were semantically ofopaque–opaque (OO) status. The latter cannot be con-structed as they emerge historically and, thus, thegender combination of their constituents cannot bemanipulated. However, the gender distribution acrossthe items was fairly matched (34% feminine, 38% mas-culine, and 28% neuter). Both types of stimuli werematched with respect to frequency (TT 17.75; OO 16.89;Quasthoff, 2002)7 and number of syllables (both 3.2).Twenty-five percent of the items (for each group, OOand TT) were assigned a definite determiner that agreedin gender with both constituents, the agreement–agree-ment (AA) condition (Table 1), and for 25% the deter-miner did not agree with the nonhead constituent butwith the second constituent (violation–agreement; VA).In the same way, 25% of the items were each assigned tothe AV and the VV conditions. A total of 120 transparentthree-constituent compounds and 280 single words(containing at most one root morpheme) were addedas fillers. Two lists were constructed whereby the criticalitems were assigned a different determiner in the secondlist. If an item belonged to condition AV on the first list,it belonged to the condition VA on the second. Items incondition AA appeared in condition VV on the secondlist, and vice versa.

A female professional speaker read each item fol-lowing a (phonologically legal) pseudodeterminer forrecording purposes (digitization with 44,700 Hz). Pseu-dodeterminers were used to avoid changes in pitchcontour of the compounds as a result of a speechproduction without a preceding word. In so doing, weexcluded the possibility that prosodic cues might con-tain any information regarding the noun’s gender agree-ment. In the same recording session, the correctGerman determiners were recorded. All stimuli wereelectronically edited in order to adapt for loudness(Johnston, 2000). For the length of constituents for allthree experiments, see Table 5. This procedure was usedin all three experiments.

Procedure

Subjects were seated in a dimly lit, sound-attenuated,and electrically shielded room. They were instructed tolisten carefully to the acoustically presented words andto judge on indication as fast and correctly as possibleeither the gender agreement, or the semantic similarityto another word presented on the screen. One trial in-cluded a fixation cross presentation of 800 msec beforeauditory stimulation. The fixation cross remained onthe screen throughout the auditory presentation. Thedeterminer and the noun were separated by 60 msecin order to make the presentation sound naturally.

After the acoustic offset, no stimulation took place for200 msec, then a task cue was given in the upper part ofthe screen, either ‘‘Grammatikalitat’’ (grammaticality)or ‘‘Interpretation’’ (interpretation). RTs were measuredfrom the task cue presentation (Experiments 1–3). If sub-jects had to compare stimuli (interpretation), anotherword was presented in the lower part of the screen. Thesame location was captured by a string of hash marksin the grammaticality task. These tasks were used in allthree experiments to ensure that subjects attended tothe determiner/numeral and to all constituents. It was,therefore, the best strategy to listen closely as it was notpredictable which judgment would be required. Feed-back was provided by visual stimuli.

The session lasted approximately 60 min consistingof three blocks. Subjects had 16 training trials that werenot used in the experiment. Half of all items weregender congruent and the other half was genderincongruent. In the semantic task, which was appliedequally to the gender-congruent and -incongruent tri-als, half of the items were semantically similar to theprobe word. The order of presentation was pseudo-random with no more than two successive presenta-tions of any experimental condition. Because weconstructed two lists which were randomly assignedto different subjects, each item was presented in twoconditions across all subjects. No constituent was re-peated within a compound position.

Recordings

The electroencephalogram (EEG) was recorded from34 Ag/AgCl electrodes (electrocap; Figure 9). It wasamplified (PORTI-32/MREFA), filtered (DC to 70 Hzband pass), and digitized online at 250 Hz. In order

Table 5. The Durations of Compounds and TheirConstituents

Length (msec)

C1 C2 Total

Experiment 1 500 (109) 554 (107) 1054 (131)

TT 528 (113) 556 (118) 1084 (145)

OO 472 (98) 553 (95) 1025 (110)

Experiment 2 559 (160) 582 (121) 1142 (203)

Experiment 3

SIN 652 (149) [584 (124)] 1236 (201)

COM 505 (140) [584 (124)] 1090 (182)

C1 = nonhead, and C2 = head. Standard deviations are in parentheses.

C2 in Experiment 3 was white noise with the intensity contour andduration of an acceptable head constituent for each item.

TT = transparent compound; OO = opaque compound; SIN = singlenoun, and COM = compound constituent prosody.

Koester et al. 1663

to control for eye movements, bipolar horizontal andvertical electrooculograms (EOG) were recorded. Elec-trode impedances were kept below 5 k� and the leftmastoid (A1) was used as reference. All signals wererecorded continuously. Except for the number of elec-trodes, all technical parameters are identical in Experi-ments 2 and 3.

Data Analyses (Experiments 1–3)

Average ERPs were calculated across subjects for eachcompound constituent separately. The averaged epochexpanded from 200 msec preconstituent onset to1000 msec postconstituent onset. Four ROIs were con-structed to test the ERP for scalp distributions. The ROIswere defined as anterior-left (AL): F7, F5, F3, FT7, FC5,FC3; anterior-right (AR): F4, F6, F8, FC4, FC6, FT8;posterior-left (PL): TP7, CP5, CP3, P7, P5, P3; andposterior-right (PR): CP4, CP6, TP8, P4, P6, P8.

Contaminated EEG epochs were automatically re-jected (EOG rejection ± 40 AV; EEG ± 25 AV), anddouble checked by visual inspection. Approximately 11%of the trials were excluded due to rejection criteria,movement artifacts, or incorrect responses. The ERPwaveforms were quantified by mean amplitude meas-ures in relation to a 200-msec prestimulus baseline.Within-subject factors were gender agreement of non-head and head constituents (number agreement in Ex-periments 2 and 3), and semantic status (TT vs. OO).The dependent variable was subjected to a repeated-

measure ANOVA and the Greenhouse and Geisser (1959)correction was applied where appropriate. The originaldegrees of freedom and corrected p values are reported.All ERPs were filtered (10 Hz low pass) for presentationonly.

Experiment 2

Subjects

In Experiment 2, 24 (13 men) right-handed (lateraliza-tion coefficient 93; Oldfield, 1971) volunteers participat-ed with a mean age of 24.4 years (range 21–30). None ofthem participated in Experiment 1.

Stimuli

We collected 88 semantically transparent two-constitu-ent compounds. Each 25% of the items were composedof two singular noun forms, a singular form followed bya plural form, a plural form followed by a singular form,and two plural noun forms (Table 2). That is, thecompounds in the two latter groups contained a linkingelement between the constituents that were identical tothe plural marker of the nonhead constituent. All con-stituents take, if they are used as single words, an overtplural marking suffix. None of the constituents under-goes a vowel change (umlaut), and only such constitu-ents were selected for which the plural suffix is notconfounded with their singular genitive case. All con-stituents were of masculine or neuter gender. Becausethe numerals used always agreed in gender with mascu-line and neuter nouns, gender was not confounded withthe number manipulation. All stimuli were existingGerman compounds.

In a next step, we constructed the experimentalconditions from the described stimulus set in order tomanipulate the number agreement of each compoundconstituent with a preceding numeral. Any half of thefour stimuli groups mentioned above was assignedeither a singular [‘‘ein’’, a(n)/one] or plural (‘‘zwei,’’two) numeral. Hence, the numeral might agree innumber either with both constituents, only with thenonhead, only with the second, or with neither of thetwo constituents. The AA and the VV conditions wereeach created from half of the singular–singular groupand half of the plural–plural group, depending on theassigned numeral. In a similar vein, from half of thesingular–plural group and half of the plural–singulargroup, the AV and the VA conditions were constructed.All experimental conditions were matched to one an-other with regard to number of syllables (AA 4.3; AV 4.3;VA 4.1; VV 4) and frequency (AA 18.5; AV 19; VA 18.7; VV18.4; Quasthoff, 2002). A total of 88 transparent three-constituent compounds and 176 single words wereadded as fillers. Two lists were constructed whereby thetwo-constituent compounds were assigned a differentnumeral on the second list in analogy to Experiment 1.

Figure 9. A symbolic representation of the electrode positions assuggested by the American Electroencephalograhic Society (1991). The

EEG was recorded from the filled circles in Experiment 1. The half-

filled circles were additionally used in Experiments 2 and 3. The EEG

was not recorded from empty circle positions.


Procedure

The procedure was equivalent to Experiment 1 exceptthat subjects had to judge the number agreement in thegrammaticality task and that each trial started with afixation cross presentation of 1400 msec before auditorypresentation.

The two lists were presented to subjects in twosessions (one list per session) separated by at leastone week. No constituent was repeated within a com-pound position during an experimental session. Thus,the comparison of congruent and incongruent constitu-ents is a within-item comparison within subjects. Onesession lasted approximately 40 min and included threepauses as well as 29 training trials which were not usedin the experiment. Half of all items were presented witha number-congruent numeral. The semantic task wasapplied equally to number-congruent and -incongruenttrials, and half of the items were semantically similar tothe probe word. The order of presentation was pseudo-random as in Experiment 1.

Recordings

The EEG was recorded from 51 Ag/AgCl electrodes(Figure 9).

Data Analyses

Approximately 17% of the trials were excluded due torejection criteria, movement artifacts, or incorrect re-sponses (see Experiment 1).

Experiment 3

Subjects

In Experiment 3, 28 (13 men) right-handed (lateralizationcoefficient 90; Oldfield, 1971) volunteers participatedwith a mean age of 24.6 years (range 21–29). None ofthem participated in Experiments 1 or 2.

Stimuli

We collected 176 transparent two-constituent com-pounds, including the 88 from Experiment 2. All non-head constituents were closely matched to those fromExperiment 2. Half of the compounds contained a link-ing element that was identical with a plural morpheme,the other half did not. Nonhead constituents had onaverage 2.2 syllables and their frequency of occurrence(word form) was 11.4 (Quasthoff, 2002). Seventy con-stituents have a neuter gender and 106 are masculine. Inaddition to the 176 compounds, all nonhead constitu-ents were produced as single nouns; if they contained alinking element, they were produced as plural nouns,otherwise as singular nouns. As a result, the 176 wordswere recorded once with a single noun prosody and

once with a compound prosody, whereas correspondingitems are phonologically identical. The constituentboundary of the compounds was determined undervisual and auditory control and all head constituentswere extracted electronically (Johnston, 2000; Boersma& Weenink, 1992–2003). Subsequently, the intensityenvelope of the head constituents was extracted andfilled with white noise, before it was spliced again ontothe nonhead constituent at the nearest zero-crossing tothe constituent boundary. The same ‘‘noise constituent’’was also spliced to the corresponding single noun.

For both these critical item groups, numeral assign-ment was done equivalently. Half of the plural andsingular forms were assigned a congruent numeral(Table 3). Two lists were constructed. Each list con-tained 50% singular word forms, 50% items with a sin-gle noun prosody, and 50% items with a congruentnumber assignment; the amount of items according toall possible combinations was equal on each list. A totalof 176 transparent two-constituent compounds and 176single words (each 50% singular), both unmanipulated,were added as fillers. Half of theses filler items werepresented with a congruent numeral. By assigningthese two lists to different subjects, it is possible tocompare the ERPs in response to the same words withdifferent prosody across subjects. Additionally, two mir-ror images of the lists were constructed. One originallist and the equivalent mirror list were presented to thesame subjects in two experimental sessions (one list persession). If a word was number correct on the originallists, it was number incorrect on the mirror image lists,and vice versa. The mirror lists permitted a within-itemcomparison.

Procedure

One session lasted approximately 50 min including threepauses as well as 28 training trials that were not used inthe experiment (see Experiment 2).

Recordings

The EEG was recorded from 51 Ag/AgCl electrodes(Figure 9).

Data Analyses

Approximately 13% of the trials were excluded due torejection criteria, movement artifacts, or incorrect re-sponses (see Experiment 1). Due to the prosodic differ-ences between single nouns and compound constituents,they were analyzed in separate ANOVAs.

Acknowledgments

We thank Shirley-Ann Ruschemeyer for helpful comments onan earlier version of the manuscript, Cornelia Schmidt for the

Koester et al. 1665

data acquisition, and Susanne Piehler and Stefanie Regel fortheir help in preparing the stimuli.

Reprint requests should be sent to Dirk Koester, Departmentof Neuropsychology, Max Planck Institute for Human Cognitiveand Brain Sciences, Stephanstr. 1a, 04103 Leipzig, Germany, orvia e-mail: [email protected].

Notes

1. The difference may arise from the intuitively guidedassignment of probe words to the experimental items. Thesemantic relation cannot affect the ERP because subjects neverknew during the presentation which task would be required.2. Note that the -s- often appears in German compounds, butif so, the preceding constituent does not take the plural -s.Following a linguistic convention, an asterisk denotes anunacceptable word form or an ungrammatical phrase.3. Numerals denote semantically numbers, quantifiablemasses, and measures. Some classify numerals like ‘‘ein’’(one) and ‘‘zwei’’ (two), and indefinite articles like ‘‘ein’’ (a/an)which is homophonous with the numeral in German, asindefinite determiners (Don, Kerstens, & Ruys, 1996–1999;Bubmann, 1990). The point here is that the numerals orindefinite determiners are unambiguously marked for numberand need to agree in number with the corresponding noun/compound. In this sense, their syntactic function is equivalentwith definite determiners in Experiment 1. This work is notmeant to support one classification over another; the term‘‘numeral’’ is used for reasons of simplicity.4. In an extensive English gating study, the first 150 msec ofthe acoustic signal contained 1.92 phonemes on average (VanPetten et al., 1999). For the average constituent length of ourstimuli, see Table 5.5. Note that a significant difference in physical parametersdoes not necessarily entail a perceptual change. Here, it is onlyshown that pitch is a reliable cue to morphological complexityat latest 75–100 msec after stimulus onset.6. Isel et al. (2003) tested only the onset, peak, and offsetvalues of the fundamental frequency.7. The frequency values are logarithmic and indicate howoften a word occurs in relation to the definite determiner‘‘der’’ (themasc). That is, higher frequency values indicate rarerappearance.

REFERENCES

American Electroencephalographic Society. (1991). Guidelinesfor standard electrode position nomenclature. Journal ofClinical Neurophysiology, 8, 200–202.

Andrews, S. (1986). Morphological influences on lexical access:Lexical or nonlexical effects? Journal of Memory andLanguage, 25, 726–740.

Aronoff, M., & Fuhrhop, N. (2002). Restricting suffixcombinations in German and English: Closing suffixesand the monosuffix constraint. Natural Language andLinguistic Theory, 20, 451–490.

Baayen, R., Dijkstra, T., & Schreuder, R. (1997). Singularsand plurals in Dutch: Evidence for a parallel dual-routemodel. Journal of Memory and Language, 37, 94–117.

Bartke, S. (1998). Experimentelle Studien zur Flexionund Wortbildung. Tubingen: Niemeyer.

Boersma, P., & Weenink, D. (1992–2003). Praat (4.0.4.3).http://www.praat.org/ (2003). Amsterdam:University of Amsterdam.

Bubmann, H. (1990). Lexikon der Sprachwissenschaft.Stuttgart: Kroner.

Butterworth, B. (1983). Lexical representation. In B.Butterworth (Ed.), Language production, vol. 2(pp. 257–294). London: Academic Press.

Bybee, J. (1995). Regular morphology and the lexicon.Language and Cognitive Processes, 10, 425–455.

Caramazza, A., Miozzo, M., Costa, A., Schiller, N., & Alario,F. (2001). A crosslinguistic investigation of determinerproduction. In E. Dupoux (Ed.), Language, brain, andcognitive development: Essays in honor of JacquesMehler (pp. 209–226). Cambridge: MIT Press.

Clahsen, H. (1999). Lexical entries and rules of language: Amultidisciplinary study of German inflection. Behavioraland Brain Sciences, 22, 991–1060.

Clahsen, H., Rothweiler, M., Woest, A., & Marcus, G. (1992).Regular and irregular inflection in the aquisition of Germannoun plurals. Cognition, 45, 225–255.

Comrie, B. (1999). Grammatical gender systems: A Linguist’sassessment. Journal of Psycholinguistic Research, 28,457–466.

Corbett, G. (1991). Gender. Cambridge: Cambridge UniversityPress.

Corbett, G. (2000). Number. Cambridge: Cambridge UniversityPress.

Coulson, S., King, J., & Kutas, M. (1998). Expect theunexpected: Event-related brain response tomorphosyntactic violations. Language and CognitiveProcesses, 13, 21–58.

Cutler, A., Dahan, D., & van Donselaar, W. (1997).Prosody in the comprehension of spoken language:A literature review. Language and Speech, 40,141–201.

Davis, M., Marslen-Wilson, W., & Gaskell, M. (2002). Leadingup the lexical garden path: Segmentation and ambiguityin spoken word recognition. Journal of ExperimentalPsychology: General, 28, 218–244.

Deutsch, A., & Bentin, S. (2001). Syntactic and semantic factorsin processing gender agreement in Hebrew: Evidencefrom ERPs and eye movements. Journal of Memory andLanguage, 45, 200–224.

Don, J., Kerstens, J., & Ruys, E. (1996–1999). Lexicon oflinguistics [online database]. http://tristram.let.uu.nl/UiL-OTS/Lexicon/ (2003). The Netherlands: UtrechtInstitute of Linguistics OTS, Utrecht University.

Elman, J. (1993). Learning and development in neuralnetworks: The importance of starting small. Cognition,48, 71–99.

Fabb, N. (2001). Compounding. In A. Spencer, & A. Zwicky(Eds.), The handbook of morphology (pp. 66–83). Oxford:Blackwell.

Fleischer, W., & Barz, I. (1995). Wortbildung der deutschenGegenwartssprache. Tubingen: Niemeyer.

Friederici, A. (2002). Towards a neural basis of auditorysentence processing. Trends in Cognitive Science, 6,78–84.

Friederici, A., Hahne, A., & Mecklinger, A. (1996).Temporal structure of syntactic parsing: Early andlate event-related brain potential effects. Journal ofExperimental Psychology: Learning, Memory, andCognition, 22, 1219–1248.

Fuhrhop, N. (1998). Grenzfalle morphologischer Einheiten.Tubingen: Stauffenburg.

Fuhrhop, N. (2000). Zeigen Fugenelemente dieMorphologisierung von Komposita an? In R. Thieroff,M. Tamrat, N. Fuhrhop, & O. Teuber (Eds.), DeutscheGrammatik in Theorie und Praxis (pp. 201–213).Tubingen: Niemeyer.

Gawlitzek-Maiwald, I. (1994). How do children cope withvariation in the input? The case of German plurals and


compounding. In R. Tracy & E. Lattey (Eds.), Howtolerant is universal grammar?: Essays on the languagelearnability and language variation (pp. 225–266).Tubingen: Niemeyer.

Greenhouse, S., & Geisser, S. (1959). On methods in theanalysis of profile data. Psychometrika, 24, 95–112.

Gunter, T., Friederici, A., & Schriefers, H. (2000). Syntacticgender and semantic expectancy: ERPs reveal earlyautonomy and late interaction. Journal of CognitiveNeuroscience, 12, 556–568.

Hagoort, P., & Brown, C. (1999). Gender electrified: ERPevidence on the syntactic nature of gender processing.Journal of Linguistic Research, 28, 715–728.

Hagoort, P., Brown, C., & Groothusen, J. (1993). Thesyntactic positive shift (SPS) as an ERP measure ofsyntactic processing. Language and Cognitive Processes, 8,439–483.

Hahne, A., & Friederici, A. (1999). Electrophysiologicalevidence for two steps in syntactic analysis: Early automaticand late controlled processes. Journal of CognitiveNeuroscience, 11, 194–205.

Holcomb, P., & Neville, H. (1990). Auditory and visualsemantic priming in lexical decision: A comparison usingevent-related brain potentials. Language and CognitiveProcesses, 5, 281–312.

Isel, F., Gunter, T., & Friederici, A. (2003). Prosody-assistedhead-driven access to spoken German compounds. Journalof Experimental Psychology: Learning, Memory, andCognition, 29, 277–288.

Jarema, G., Libben, G., Dressler, W., & Kehayia, E. (2002). Therole of typological variation in the processing of interfixedcompounds. Brain and Language, 81, 736–747.

Johnston, D. (2000). Cool edit 2000 (1.1). Phoenix, AZ:Syntrillium Software.

Kiparsky, P. (1982). From cyclic phonology to lexicalphonology. In H. van der Hulst (Ed.), The structure ofphonological representation (pp. 131–175). Foris:Dordrecht.

Kluender, R., & Kutas, M. (1993). Bridging the gap: Evidencefrom ERPs on the processing of unbounded dependencies.Journal of Cognitive Neuroscience, 5, 196–214.

Kohler, K. (1995). Einfuhrung in die Phonetik des Deutschen.Berlin: Erich Schmidt.

Kopcke, K. (1982). Untersuchungen zum Genussystem derdeutschen Gegenwartssprache. Tubingen: Niemeyer.

Krott, A. (2001). Analogy in morphology: The selection oflinking elements in Dutch compounds. PhD thesis,Max-Planck-Institut fur Psycholinguistik, Nijmegen.

Kutas, M., & Federmeier, K. (2000). Electrophysiology revealssemantic memory use in language comprehension. Trendsin Cognitive Science, 4, 463–470.

Libben, G., Derwing, B., & de Almeida, R. (1999). Ambiguousnovel compounds and models of morphological parsing.Brain and Language, 68, 378–386.

Libben, G., Jarema, G., Dressler, W., Stark, J., & Pons,C. (2002). Triangulating the effects of interfixation inthe processing of German compounds. Folia Linguistica,36, 23–43.

Luck, M., Hahne, A., & Clahsen, H. (2001). How the processingof German plural nouns develops—Auditory ERP studieswith adults and children. 7th Annual Conference onArchitectures and Mechanisms for Language Processing,Saarbrucken, 20.09.–22.09.2001 (pp. 82–83).

Marcus, G., Brinkmann, U., Clahsen, H., Wiese, R., & Pinker,S. (1995). German inflection: The exception that proves therule. Cognitive Psychology, 29, 189–256.

McCallum, W., Farmer, S., & Pocock, P. (1984). The effectsof physical and semantic incongruities on auditory

event-related potentials. Electroencephalography andClinical Neurophysiology, 59, 477–488.

McClelland, J., Rumelhart, D., & the PDP research group.(1986). Parallel distributed processing: Explorations in themicrostructure of cognition: Vol. 2. Psychological andbiological models. Cambridge: MIT Press.

Meyer, R. (1993). Compound comprehension in isolation andin context: The contribution of conceptual and discourseknowledge to the comprehension of German novelnoun–noun compounds. Tubingen: Niemeyer.

Neville, H., Nicol, J., Barss, A., Forster, K., & Garrett, M. (1991).Syntactically based sentence processing classes: Evidencefrom event-related brain potentials. Journal of CognitiveNeuroscience, 3, 151–165.

Oldfield, R. (1971). The assessment and analysis ofhandedness: The Edinburgh Inventory. Neuropsychologia,9, 97–113.

Olsen, S. (1990). Zum Begriff des morphologischen Heads.Deutsche Sprache, 2, 126–147.

Osterhout, L., & Holcomb, P. (1992). Event-related brainpotentials elicited by syntactic anomaly. Journal of Memoryand Language, 31, 785–806.

Osterhout, L., McKinnon, R., Bersick, M., & Corey, V. (1996).On the language specificity of the brain response to syntacticanomalies: Is the syntactic positive shift a member of theP300 family? Journal of Cognitive Neuroscience, 8,507–526.

Penke, M., & Krause, M. (2002). German noun plurals: Achallenge to the dual-mechanism model. Brain andLanguage, 81, 303–311.

Penke, M., Weyerts, H., Gross, M., Zander, E., Munte, T., &Clahsen, H. (1997). How the brain processes complexwords: An event-related potential study of German verbinflections. Cognitive Brain Research, 6, 37–52.

Perrin, F., Garcıa-Larrea, L., Mauguiere, F., & Bastuji, H. (1999).A differential brain response to the subject’s own namepersists during sleep. Clinical Neurophysiology, 110,2153–2164.

Picton, T. (1992). The P300 wave of the human event-related potential. Journal of Clinical Neurophysiology,9, 456–479.

Pompino-Marschall, B. (1995). Einfuhrung in die Phonetik.Berlin: Walter de Gruyter.

Pratarelli, M. (1995). Modulation of semantic processing usingword length and complexity: An ERP study. InternationalJournal of Psychophysioloy, 19, 233–246.

Pritchard, W. (1981). Psychophysiology of P300. PsychologicalBulletin, 89, 506–540.

Quasthoff, U. (2002). Deutscher Wortschatz im Internet[Online database]. http://www.wortschatz.uni-leizpig.de/(2002). Leipzig, Germany: University of Leipzig.

Salmons, J. (1993). The structure of the lexicon: Evidencefrom German gender assignment. Studies in Language, 17,411–435.

Sandra, D. (1990). On the representation and processingof compound words: Automatic access to constituentmorphemes does not occur. Quarterly Journal ofExperimental Psychology, 42A, 529–567.

Schiller, N., & Caramazza, A. (2002). The selection ofgrammatical features in word production: The case ofplural nouns in German. Brain and Language, 81,342–357.

Schiller, N., & Caramazza, A. (2003). Grammatical featureselection in noun phrase production: Evidence fromGerman and Dutch. Journal of Memory and Language, 48,169–194.

Schiller, N., Munte, T., Horemans, I., & Jansma, B. (2003).The influence of semantic and phonological factors on

Koester et al. 1667

syntactic decisions: An event-related brain potential study.Psychophysiology, 40, 869–877.

Schmitt, B., Lamers, M., & Munte, T. (2002). Electrophysiologicalestimates of biological and syntactic gender violation duringpronoun processing. Cognitive Brain Research, 14, 333–346.

Schreuder, R., Neijt, A., Van der Weide, F., & Baayen, R. (1998).Regular plurals in Dutch compounds: Linking graphemesor morphemes? Language and Cognitive Processes, 13,551–573.

Selkirk, E. (1982). The syntax of words. Cambridge: MIT Press.Sonnenstuhl, I., & Huth, A. (2002). Processing and

representation of German -n plurals: A dual mechanismapproach. Brain and Language, 81, 276–290.

Springer, D., & Deutsch, D. (1997). Left brain; right brain:Perspectives from cognitive neuroscience. Houndsmills:Freeman.

Taft, M., & Forster, K. (1976). Lexical storage and retrieval ofpolymorphemic and polysyllabic words. Journal of VerbalLearning and Verbal Behavior, 15, 607–620.

Van Petten, C., Coulson, S., Rubin, S., Plante, E., & Parks, M.(1999). Time course of word identification and semanticintegration in spoken language. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 25,394–417.

Wagner, S. (2003). Verbales Arbeitsgedachtnis und dieVerarbeitung ambiger Worter in Wort- und Satzkontexten.PhD thesis, Max-Planck-Institut fur neuropsychologischeForschung, Leipzig.

Wedel, A., & Hahne, A. (2002). L1 influences on L2 genderprocessing: An ERP study comparing English and Russian

native speakers’ resonse to gender violation in German.24. Jahrestagung der Deutschen Gesellschaft furSprachwissenschaft, Mannheim, 27.02.–01.03.2002(pp. 100–101).

Wegener, H. (1992). Pluralregeln und mentale Grammatik. In I.Zimmermann (Ed.), Fugungspotenzen: Zum 60. Geburtstagvon Manfred Bierwisch (pp. 225–249). Berlin: AkademieVerlag (studia grammatica 34).

Wegener, H. (1999). Die Pluralbildung im Deutschen—einVersuch im Rahmen der Optimalitatstheorie. LinguistikOnline, 4, 1–55.

Weyerts, H., Penke, M., Dohrn, U., Clahsen, H., & Munte,T. (1997). Brain potentials indicate differences betweenregular and irregular German plurals. NeuroReport, 8,957–962.

Wicha, N., Bates, E., Moreno, E., & Kutas, M. (2003). Potato notPope: Human brain potentials to gender expectation andagreement in Spanish spoken sentences. NeuroscienceLetters, 346, 165–168.

Wiese, R. (1996). The phonology of German. Oxford: OxfordUniversity Press.

Williams, E. (1981). On the notions ‘‘lexically related’’ and‘‘head of word’’. Linguistic Inquiry, 12, 245–274.

Wolf, A. (2004). Sprachverstehen mit Cochlea-Implantat:EKP-Studien mit postlingual ertaubten erwachsenenCI-Tragern. PhD thesis, Max-Planck-Institut furKognitions- und Neurowissenschaften, Leipzig.

Zwitserlood, P. (1994). The role of semantic transparency inthe processing and representation of Dutch compounds.Language and Cognitive Processes, 9, 341–368.


Morphosyntax, prosody, and linking elements: The auditory processing of German nominal compounds

Documents

Transcript of Morphosyntax, prosody, and linking elements: The auditory processing of German nominal compounds