Brain potentials indicate immediate use of prosodic cues in natural speech processing

Transcription

1 grmmticl reltions hs een ccomplished. Psycholinguists hve long deted whether or not the initil syntctic preferences cn e immeditely overridden y non-syntctic influences such s semntic or prgmtic cues 6,7. When non-syntctic influences hve een found, the temporl immedicy of these effects hs een controversil 5. When sentences such s 1 nd 1 re presented uditorily, the initil preference in fvor of 1 is considerly diminished y prosodic oundry fter the first ver 2. This finding suggests tht prosodic informtion does influence decisions out syntctic structure t very erly stges. However, little is known out the exct reltionship etween prosody nd sentence processing, nd in prticulr the time course nd neurl sis of influences such s the one just descried 8,9. Event-relted rin potentils (ERPs) re useful tool for the on-line exmintion of oth norml nd impired lnguge processing 10,11, which my contriute to the development of theoreticl ccounts of prosodic nd syntctic interction. The ERP correltes of initil syntctic misnlyses for oth written nd spoken presenttion show tht the dditionl costs re reflected y lte posterior positivity etween 500 nd 1200 ms (termed the P600 component or syntctic positive shift ) In contrst, difficulties of lexicl nd semntic processing (for exmple, He spred the wrm red with socks ) generlly elicit n erlier centroprietl negtivity etween 300 nd 900 ms (the N400 component) rther thn P600. These two clsses of ERP effects indicte specific rin responses to different linguistic fetures. Here we used ERP mesures for the on-line investigtion of prosodic fetures. We exmined the influence of prosody on humn prsing performnce with oth ehviorl nd neurophysiologicl (ERP) mesures. The stimulus mteril consisted of spoken Germn sentence pirs similr to the English exmples in 1 nd 1. To estlish controlled experimentl conditions comprle to those in reding studies, we needed exhustive coustic nlyses of the speech signls. These nlyses reveled systemtic prosodic difrticles Brin potentils indicte immedite use of prosodic cues in nturl speech processing Krsten Steinhuer, Ki Alter nd Angel D. Friederici Deprtment of Neuropsychology, Mx Plnck Institute of Cognitive Neuroscience, PO Box , D Leipzig, Germny Correspondence should e ddressed to K.S. (steinhu@cns.mpg.de) Spoken lnguge, in contrst to written text, provides prosodic informtion such s rhythm, puses, ccents, mplitude nd pitch vritions. However, little is known out when nd how these fetures re used y the listener to interpret the speech signl. Here we use event-relted rin potentils (ERP) to demonstrte tht intontionl phrsing guides the initil nlysis of sentence structure. Our finding of positive shift in the ERP t intontionl phrse oundries suggests specific on-line rin response to prosodic processing. Additionl ERP components indicte tht flse prosodic oundry is sufficient to misled the listener s sentence processor. Thus, the ppliction of ERP mesures is promising pproch for reveling the time course nd neurl sis of prosodic informtion processing. Humn everydy communiction is spoken rther thn written, ut the overwhelming mjority of psycholinguistic reserch underlying models of sentence processing is still sed on reding rther thn speech-processing dt. One reson for this symmetry is tht written text, s compred to spoken lnguge, cn e much more esily controlled in experimentl design. Nturl speech is lwys contminted with prosodic fetures, which increse word-length vriility nd introduce pitch nd mplitude vritions; these vriles cn e completely voided with written text. Prosodic fetures, however, my e importnt for communiction, nd some studies on spoken lnguge suggest tht prosodic informtion my influence sentence processing to such degree tht preferences oserved during reding my not pply eqully to speech processing 1 3. For exmple, eye-trcking mesures in reding tsk indicte 4 tht sentence 1 is much esier to understnd thn sentence 1: (1) Since Jy lwys jogs mile nd hlf this seems like short distnce to him. (1) Since Jy lwys jogs mile nd hlf seems like very short distnce to him. Initilly, the reder prefers to interpret the noun phrse mile nd hlf s the grmmticl oject of the preceding ver jogs rther thn s the suject of the susequent ver seems, s required in 1. Thus the initil nlysis must e revised in 1, which results in prolonged reding time, phenomenon clled the grden-pth effect. The roustness of such effects in reding studies led to the development of the grden-pth model of sentence processing 4. According to this model, initil preferences re exclusively ttriuted to the inherent processing principles of n encpsulted syntctic processing system (clled syntctic prser), which cnnot e influenced y non-syntctic informtion 5. Semntic interprettion (or mening ) occurs only once the syntctic prsing of nture neuroscience volume 2 no 2 ferury

2 rticles Experiment 1 Experiment 2 Fig. 1. Prosody syntx mismtch. ERPs of conditions B (lue) nd C (red) t nine electrodes from the onset of the infinitive mrker zu of the criticl second ver until two seconds lter, displyed seprtely for ech experiment. Negtive mplitudes re plotted upwrds. () In Experiment 1, the prosody syntx mismtch condition C is chrcterized y n N400 component followed y P600. () In Experiment 2, the sme iphsic N400 P600 pttern s in Experiment 1 ws elicited in condition C. ferences etween the two conditions, suggesting specific prosodic phrsing nd ccentution pttern dependent on the syntctic structure. When the sentences were presented uditorily, the listener s event-relted rin potentils reproducily showed chrcteristic positive-going wveform t prosodic phrse oundries, indicting immedite decoding of this informtion. Furthermore, in third condition, we introduced the initil prosodic fetures of one condition into the other condition, leding to prosody syntx mismtch. Both ehviorl nd ERP dt strongly suggest tht the prosodic fetures determined the initil prsing decisions. The mismtch etween prosody nd syntx ws relily detected y the listeners nd elicited n N400 P600 pttern of ERP components reflecting prosody-induced grdenpth effect. Our hypothesis ws tht these erly prosodic differences of spoken lnguge could e sufficient to prevent the grden-pth effect in 2. If the dditionl IPh oundry indeed chnges the initil interprettion of Ann, then it should even e possile to reverse the grden-pth effect. Tht is, if the erly prosodic cues of 2 were introduced in sentence 2, we expected n initil misnlysis nd grden-pth effect in this normlly esy-toprocess structure. The noun phrse Ann would then e erroneously ttched to the second ver, which is intrnsitive nd cnnot tke direct oject s its rgument. Using cross-splicing technique 23, we merged the coustic signls of the first prt of 2 nd the second prt of 2 etween Ann nd the infinitive mrker zu ( to ) of the second ver in ech of the 48 sentence pirs. This resulted in third condition (2c) with mismtch RESULTS The Germn sentence mteril consisted of 48 sentence pirs such s 2 nd 2, where the rcketing indictes the respective intontionl phrses (IPhs) 20 s descried elow: (2) [Peter verspricht Ann zu reiten] IPh1 [und ds Büro zu putzen] IPh2 Peter promises Ann to work nd to clen the office (2) [Peter verspricht] IPh1 [Ann zu entlsten] IPh2 [und ds Büro zu putzen] IPh3 Peter promises to support Ann nd to clen the office As in exmple 1, the noun phrse Ann is the oject of the first ver verspricht ( promises ) in 2 nd therefore elongs to the first IPh. In contrst to 2 ut similr to 1, in sentence 2 Ann is the oject of the second trnsitive ver entlsten ( support ) nd elongs to the second IPh. Becuse of the Germn word order, the correct interprettion of Ann is syntcticlly dismiguted y the second ver only (tht is, reiten ( work ) in 2 versus entlsten in 2). Comptile with the predictions of certin theories of syntx prosody mpping 21,22, however, the 48 speech signls of conditions 2 nd 2 differ considerly even efore the point of syntctic dismigution, vi different intontionl phrsing nd ccentution (Methods). Fig. 2. Closure positive shift. Grnd-verge ERPs of oth experiments (n = 40) t the PZ electrode. The wveforms of conditions A (ornge) nd B (lue) re superimposed. The word onsets of the sentence exmples re ligned to the time xis. Both conditions evoke closure positive shifts t their respective IPh oundries. Only one shift is oservle in condition A, following the second ver reiten, wheres two such shifts occur in condition B, efore Ann nd fter the second ver entlsten. 192 nture neuroscience volume 2 no 2 ferury 1999

3 rticles Experiment 1 Experiment 2 Fig. 3. Sentence-specific ERPs. ERPs of ll three conditions t nine electrode sites from sentence onset until four seconds lter, plotted seprtely for ech experiment. () In Experiment 1, conditions B (lue) nd C (red) shre the sme pttern of two CPSs s opposed to one CPS in condition A (ornge). By two seconds, tht is, fter onset of the second ver, the prosody syntx mismtch condition C diverges from B nd elicits the N400 P600 pttern. () Experiment 2 generlly replictes the findings of Experiment 1. etween prosodic informtion (Methods) nd syntctic constrints (tht is, the intrnsitivity of the ver reiten ): (2c) * [Peter verspricht] IPh1 [Ann zu reiten] IPh2 [und ds Büro zu putzen] IPh3 Peter promises to work Ann nd to clen the office The prosodic inppropriteness of 2c ecomes ovious only when the intrnsitive ver reiten is encountered. At this point, the sentence should initilly e perceived s Peter promises to work Ann, which is certinly ungrmmticl nd requires revision. According to linguistic convention, ungrmmticl sentences re mrked y n sterisk. Forty-eight sentences of ech of the three conditions were presented to 40 sujects in two ERP experiments vrying in their tsk requirements: comprehension tsk (Experiment 1) nd prosody judgment plus comprehension tsk (Experiment 2). The prosody cceptility judgment reveled tht the prticipnts relily detected the prosody syntx mismtch in condition C. In this condition, only 6% of the trils were rted s cceptle, s opposed to more thn 80% in oth conditions 2 nd 2 (p < ). The error rtes in the comprehension tsk were not significntly incresed in the mismtch condition (C) for either Experiment 1 or 2. We focused on two questions: first, whether nd how ERPs reflect the influence of prosodic informtion on erly syntctic processes, nd second, whether ERPs re sensitive to the processing of prosodic fetures per se. We riefly discuss the effect indicting the interply etween syntx nd prosody nd then turn to newly identified prosody-relted ERP component. As Experiments 1 nd 2 hd very similr results, they re presented jointly. The syntx prosody mismtch effect In reding studies using ERPs, violtions of ver s rgument structure elicit n N400 component followed y P600 component 14,24. As predicted, we found similr iphsic N400 P600 sequence during the intrnsitive second ver for condition C s compred to the grmmticlly correct trnsitive ver in condition B (Fig. 1). Men mplitude nlyses cross 8 consecutive 200-ms time windows (from 200 to 1800 ms) confirmed tht oth effects were restricted to centroprietl electrodes. The N400 effect ws significnt etween 400 nd 1000 ms (p < 0.01) nd the P600 effect etween 1200 nd 1800 ms (p < 0.001). For oth ERP components, we oserved only min effects of the sentence condition (condition B versus C) nd no further interctions with the tsk fctor. Bseline-independent, pek-to-pek mesures including condition A ruled out the possiility tht the effects were simply due to the different vers (trnsitive vers in B versus intrnsitive vers in C). Although condition A contined the sme intrnsitive vers s condition C, the mplitude difference etween the N400 pek nd the P600 pek in condition A ws significntly Fig. 4. ERPs fter puse removl. ERPs of conditions B (lue) nd C (red) in Experiment 3 (n = 16) t the CZ nd PZ electrode sites. () Even fter removl of the puse etween the first ver nd the second noun phrse, oth conditions still disply the CPS t the first IPh oundry. () As in the first two experiments, the intrnsitive ver of the prosody syntx mismtch condition C elicits iphsic N400 P600 pttern. This comprison is more relile thn tht in Fig. 4, s the verge window is ligned to the criticl ver nd more trils entered the verges. B (n = 16) C (n = 15) nture neuroscience volume 2 no 2 ferury

4 rticles Word nd puse durtions Fundmentl frequency (FO) Fig. 5. Prosodic prmeters. Prosodic differences etween the speech signls of conditions A (ornge lines) nd B (lue lines). () Durtion mesures of sentence frgments nd puses (#) nd respective differences etween conditions B nd A. Condition B shows oth lengthening of the first sentence frgment Peter verspricht (p < ) nd susequent puse insertion (#1; p < ). () Fundmentl frequency. Wheres the min pitch ccent in condition A is on the ver reiten, it is ligned to the noun phrse Ann in condition B. Becuse of the lengthening in B, however, oth ccents occur t pproximtely the sme time. smller thn tht in C (p < 0.005), nd it did not differ from B (F < 1). Thus the incresed N400 nd P600 mplitudes in condition C were not due to either the ver or the prosodic pttern per se ut rther to the mismtch etween the two. An ERP component reflecting prosodic processing The prosodic processing tht induced the reversed grden-pth effect ws lso reflected in the ERPs. In oth experiments, we found lrge positive wveform t intontionl phrse oundries. The grnd-verge ERPs cross oth experiments (n = 40; Fig. 2) showed tht condition A evoked single shift t its IPh oundry following the second ver. The ERP for condition B, in contrst, contined two such positive shifts, corresponding to its two IPh oundries, the first one preceding Ann nd the second fter the second ver. This pttern of one versus two positive shifts ws confirmed sttisticlly y oth mplitude comprisons (p < 0.001) nd pek locliztions (p < ) nd did not differ etween Experiment 1 (Fig. 3) nd Experiment 2 (Fig. 3; F < 1). Moreover, the ERP for the mismtch condition C, which contins two IPh oundries, lso hd the two corresponding shifts (Fig. 3). Additionl nlyses reveled tht the positive shift t IPh oundries ws not due to the word preceding ech IPh oundry eing ver or to so-clled exogenous ERP components (such s N100 or P200) reflecting the physicl fetures of the stimulus. As we ssume tht the shift primrily reflects the closure of n intontionl phrse, we term this ERP component the closure positive shift (CPS). An importnt question is whether the CPS ctully reflects prosodic phrsing or whether it is more directly relted to the coustic properties mrking the oundry. Becuse in written sentences terminl words preceding puse re lso ssocited with positive ERP components 25,26, the most likely cndidte for such direct correspondence is the puse insertion fter the first ver in conditions B nd C, tht is, the temporry sence of ny speech input t the oundry. Therefore, we rn third ERP experiment in which we crefully removed the entire puse in oth conditions B nd C, preserving other intontionl cues. Behviorl nd ERP results for these new conditions B nd C confirmed tht even without the puse the prosodic oundry ws still perceived y the listeners (n = 16) nd guided their initil prsing decisions (prosodic cceptility rtes, 73.8 % in B, 10.9 % in C ). We still oserved the CPS t the first oundry nd lso the N400 P600 effect in condition C (Fig. 4). This finding supports the ide tht the CPS reflects the processing of the prosodic oundry rther thn the perception of puse interrupting the speech input. DISCUSSION The present study tested whether prosodic cues in spoken lnguge re immeditely used y the listener to solve syntctic miguities tht systemticlly result in initil misunderstndings during reding, nd to determine whether these prosodic influences cn e monitored on-line y ERP mesures. Not only did we demonstrte tht the prosodic informtion ws sufficient to reverse syntctic prsing preferences, ut we lso identified specific ERP component reflecting the decoding of intontionl phrsing, the closure positive shift. With respect to psycholinguistic modeling, the dt provide strong evidence tht the syntctic prser cn e directly influenced y prosodic informtion. The presence of n erly IPh oundry preceding Ann in conditions B nd C stopped further syntctic integrtion into the current first ver phrse nd insted prepred n initil ttchment of Ann to the second ver. This reversed prsing preference, triggered exclusively y prosodic informtion, successfully induced n initil misnlysis ( grden-pth effect) in the mismtch condition C nd elicited the predicted N400 P600 pttern of ERP components on the incomptile intrnsitive ver. The cognitive processes underlying oth components re still suject to discussion 27. Here we follow recent interprettion 28 ccording to which the N400 effect presumly reflects lexicl re-ccess necessry to confirm the outright violtion of the intrnsitive ver rgument structure in condition C 29. The P600, on the other hnd, seems to indicte the susequent structurl revision concerning the ttchment site of Ann. The rpidity of this on-line revision would lso explin why we did not find ny incresed error rtes in the comprehension tsk. When the question ws presented one second lter, the structure hd lredy een repired. As the repir 194 nture neuroscience volume 2 no 2 ferury 1999

5 rticles process is very likely to involve suvocl corrections of the intontionl phrsing, we ssume tht the P600 component my reflect the costs of oth syntctic nd prosodic revisions. We lso found tht prosodic oundries re ssocited with lrge positive-going wveform, which we leled the closure positive shift (CPS). This component did not depend on the presence or sence of puse cousticlly interrupting the strem of speech input. Rther, whenever the oundry ws perceived nd used to guide prsing, the CPS ws found in the corresponding time intervl. The CPS my e ssocited with processes tht serve to structure the mentl representtion of the speech signl nd to prepre the further nlysis of susequent input. It seems tht, t lest for the sentences used in the present study, the CPS enles monitoring of prosodiclly driven prsing decisions long efore the syntcticlly dismiguting element (the second ver) is encountered. Becuse of the confound etween prosodic nd syntctic units in nturl lnguge, the present dt leve open whether the CPS is predominntly relted to prosodic structuring per se or to its consequences on syntctic processing. Our recent ERP study using uditorily presented rtificil lnguge (unpulished dt), however, seems to support the view of prosodic processing underlying the CPS. On reinspection of this dt to replicte our CPS finding, CPS ws oserved even in group of nive sujects who hd not yet cquired the syntx rules of the rtificil lnguge. Syntctic ERP effects, in contrst, were found only in prticipnts who were lredy fmilir with the syntx rules (E. Pfeifer & A.D.R., J. Cogn. Neurosci. Suppl., p. 26, 1998). Wheres the question of which prosodic cues re necessry or sufficient to elicit CPS requires future reserch, ville dt indicte tht the component is tightly linked to the cognitive process of structuring the incoming speech signl. If this finding holds, the component my serve s vlule tool for systemticlly exploring the reltionship etween prosodic prsing nd syntctic prsing, which hs received little ttention until recently Our dt provide strong evidence tht mny grden-pth effects consistently oserved during reding simply my not occur in spoken lnguge processing. This finding would e comptile with so-clled syntx-first models such s the grden-pth model only if prosodic informtion is tken to directly trnsmit syntctic informtion. In ny cse, it underlines the necessity to tke prosodic processing into ccount more explicitly. On-line mesures such s the CPS my help to ly the empiricl foundtions for n dequte theory of nturl speech processing. The present study is only preliminry step towrd understnding the rin functions underlying prosodic processing. However, our dt strongly suggest tht ERP mesures used in controlled experimentl design re promising on-line pproch to shed new light on the role of prosody with regrd to oth norml nd impired speech processing 33. METHODS Sujects. Twenty students prticipted in ech of the first two ERP experiments, nd sixteen in the third experiment. All 56 sujects were righthnded 34, nd without hering or neurologicl disorders. All three experiments complied with Germn legl requirements. Speech signls. Forty-eight sentence pirs such s 2 nd 2 were produced y femle ntive speker of stndrd Germn nd recorded in soundproof chmer. The digitized speech signls (44.1 khz/16 it smpling rte) of ech sentence were mesured with respect to word nd puse durtions, fundmentl frequency (pitch contour) nd loudness (mplitude squres), nd the differences were sttisticlly nlyzed in pired t-tests or with ANOVAs. There were highly significnt durtionl differences etween conditions A nd B (Fig. 5). The dditionl IPh oundry in condition B ws signified prosodiclly y puse insertion efore Ann (p < ), s well s y significnt lengthening of the first constituent, Peter verspricht (p < ). Wheres mjor ccent occurred on the ver reiten in condition A, ccentution ws shifted to the noun phrse Ann in condition B. These differences in ccent positions were confirmed oth y loclly rising pitch contour in the fundmentl frequency (p < ; Fig. 5) nd y corresponding loudness mximum (p < 0.01; not shown). Condition C ws derived y cross-splicing the first prt of B nd the second prt of A in the silent phse of the ffricte /ts/ of the infinitivl mrker zu ( to ). This procedure plus n mplitude normliztion protected ginst detectility of the signl mnipultion t the splicing point. Conditions B nd C of the third experiment were otined y removing the puse efore the second noun phrse, without ffecting the signls of djcent words. Procedure. The 144 experimentl sentences were intermixed with 144 filler sentences nd presented uditorily in pseudo-rndomized order in 8 locks of 36 trils, distriuted over 2 sessions. Block order ws counterlnced cross sujects. EEG ws continuously recorded (250 Hz/12 it smpling rte; Neuroscn DC mplifier) from 17 cp-mounted tin electrodes while sujects listened to the sentences in electromgneticlly shielded chmer. All electrodes were referenced ginst left mstoid. Impednces were kept elow 5 kw. In Experiment 1, the tsk ws to nswer yes or no to comprehension questions such s Does Ann promise to clen the office? in 20% of the trils. In Experiments 2 nd 3, prticipnts were sked to judge the prosodic dequcy of ech sentence immeditely fter presenttion in ddition to the comprehension tsk. Dt nlyses. EEG epochs contining eyelinks or movement rtifcts were rejected nd did not enter the ERP verges. Averges were computed oth cross the whole sentence (Figs.2,3nd 4) nd for the criticl second ver (Figs. 1 nd 4) using 200-ms prestimulus seline. The men mplitudes of 8 susequent 200-ms time windows were computed from 200 ms fter the onset of ver 2 until 1800 ms therefter to mesure the N400 nd P600 components in conditions B nd C. Comprisons including condition A required dditionl seline-independent pek-to-pek mesures 35 to lnce the diverging CPS ptterns. The closure positive shift ws quntified y two different pproches. First, we compred men mplitudes cross eight susequent 500-ms time windows covering the whole sentence length. Second, we evluted the onset nd offset ltencies of lrge positive shifts t midline electrodes seprtely for ech suject in ech condition using pek-to-pek mesures. Both ehviorl nd ERP dt were sttisticlly tested y ANOVAs. ERP nlyses were generlly done seprtely for midline nd lterl electrodes. For the midline, glol three-wy ANOVA with fctors conditions (3) electrode (3) nd the etween-suject fctor tsk (2) ws used. For the lterl electrodes, regions of interest were defined. The resulting ANOVA design included the fctors conditions (3) hemisphere (2) position (3) tsk (2). Single comprisons were only computed if the glol ANOVA reveled significnt min effect of condition with df > 1 or n interction with the fctor condition. Where pproprite, Huynh nd Feldt corrections 36 nd modified Bonferroni correction procedure 37 were pplied. For illustrtive purposes only, the grnd-verge ERPs were smoothed off-line using 5-Hz lowpss filter. ACKNOWLEDGEMENTS We pprecite the comments of Lee Osterhout nd Gerry Altmn on n erlier version of the mnuscript. We lso thnk Ktj Kühn, Annett Schirmer nd Frnzisk Kopp for their ssistnce in generting the stimulus mterils nd in nlyzing the signls. This work ws supported y the Deutsche Forschungsgemeinschft (FR 519/17-1). RECEIVED 10 AUGUST; ACCEPTED 16 DECEMBER Mrslen-Wilson, W. S. et l. Prosodic effects in miniml ttchment, Q. J. Exp. Psychol. 45A, (1992). nture neuroscience volume 2 no 2 ferury

6 rticles 2. Wrren, P., Gre, E. & Noln, F. Prosody, phonology, nd prsing in closure miguities. Lng. Cogn. Processes 10, (1995). 3. Stirling, L. & Wles, R. Does prosody support or direct sentence processing? Lng. Cogn. Processes 11, (1996). 4. Frzier, L. & Ryner, K. Mking nd correcting errors during sentence comprehension: Eye movements in the nlysis of structurlly miguous sentences. Cogn. Psychol. 14, (1982). 5. Frzier, L. in Attention nd Performnce XII: The Psychology of Reding (ed. Colthert, M.) (Erlum, London, 1987). 6. Trn, R. & McClellnd, J. R. Constituent ttchment nd themtic role ssignment in sentence processing: Influence of context sed expecttions. J. Mem. Lng. 27, (1988). 7. Altmn, G. T. M., Grnhm, A. & Dennis, Y. Avoiding the grden-pth: Eye movements in context. J. Mem. Lng. 31, (1992). 8. Cutler, A., Dhn, D. & vn Donselr, W. Prosody in the comprehension of spoken lnguge: A literture review. Lng. Speech 40, (1997). 9. Blumstein, S. E. in The Cognitive Neurosciences (ed. Gzznig, M. S.) (MIT Press, Cmridge, MA, 1995). 10. Hgoort, P. & Kuts, M. in Hndook of Neuropsychology Vol.10 (eds. Boller, F. & Grfmn, J.) (Elsevier, Amsterdm, 1995). 11. Sw, T. Y. The Functionl Locus of Comprehension Deficits in Aphsi: An Electrophysiologicl Approch (Ponsen & Looijen, Wgeningen, The Netherlnds, 1996). 12. Osterhout, L. & Holcom, P. J. Event-relted rin potentils nd syntctic nomly: Evidence of nomly detection during the perception of continuous speech. Lng. Cogn. Processes 8, (1993). 13. Osterhout, L., Holcom, P. J. & Swinney, D. A. Brin potentils elicited y grden-pth sentences. Evidence of the ppliction of ver informtion during prsing. J. Exp. Psychol. Lern. Mem. Cogn. 20, (1994). 14. Osterhout, L. & Holcom, P. J. in Electrophysiology of Mind: Event-Relted Potentils nd Cognition (eds. Rugg, M. D. & Coles, M. G. H.) (Oxford Univ. Press, Oxford, 1995). 15. Friederici, A. D., Steinhuer, K., Mecklinger, A. & Meyer, M. Working memory constrints on syntctic miguity resolution s reveled y electricl rin responses. Biol. Psychol. 47, (1998). 16. Kuts, M. & Hillyrd, S. A. Reding senseless sentences: Brin potentils reflect semntic incongruency. Science 207, (1980). 17. Holcom, P. J. & Neville, H. J. Nturl speech processing: An nlysis using event-relted rin potentils. Psychoiology 19, (1991). 18. Kuts, M. & vn Petten, C. in Hndook of Psycholinguistics (ed. Gernscher, M. A.) (Acdemic, Sn Diego, 1994). 19. Friederici, A. D., Steinhuer, K. & Frisch, S. Lexicl integrtion: Sequentil effects of syntctic nd semntic informtion. Mem. Cogn. (in press). 20. Nespor, M. & Vogel, I. Prosodic Phonology (Foris, Dordrecht, 1986). 21. Selkirk, E. O. Phonology nd Syntx: The Reltion etween Sound nd Structure (MIT Press, Cmridge, 1984). 22. Hlle, M. & Vergnud, J. R. An Essy on Stress (MIT Press, Cmridge, 1987). 23. Gernscher, M. A. & Jeschenik, J. D. Ctphoric devices in spoken discourse. Cogn. Psychol. 29, (1995). 24. Rösler, F., Friederici, A. D., Pütz, P. & Hhne, A. Event-relted rin potentils while encountering semntic nd syntctic constrint violtions. J. Cogn. Neurosci. 5, (1993). 25. Friedmn, D., Simson, R., Ritter, W. & Ppin, I. The lte positive component (P300) nd informtion processing in sentences. Electroencephlogr. Clin. Neurophysiol. 38, (1975). 26. Vn Petten, C. & Kuts, M. Influences of semntic nd syntctic context on open- nd closed-clss words. Mem. Cogn. 19, (1991). 27. Osterhout, L. & Hgoort, P. A superficil resemlnce doen t necessrily men tht you re prt of the fmily: Counterrguments to Coulson, King, nd Kuts (1998) in the P600/SPS-P300 dete. Lng. Cogn. Processes (in press). 28. Friederici, A. D. The time course of syntctic ctivtion during lnguge processing: A model sed on neuropsychologicl nd neurophysiologicl dt. Brin Lng. 50, (1995). 29. Hopf, J.-M., Byer, J., Bder, M. & Meng, M. Event-relted rin potentils nd cse informtion in syntctic miguities. J. Cogn. Neurosci. 10, (1998). 30. Bder, M. in Renlysis in Sentence Processing (eds. Ferreir, F. & Fodor, J. D.) (Kluwer, Dordrecht, in press). 31. Beckmn, M. E. The prsing of prosody. Lng. Cogn. Processes 11, (1996). 32. Fodor, J. D. Lerning to prse? J. Psycholinguistic Res. 27, (1998). 33. Ross, E. D. in Behviorl Neurology nd Neuropsychology (eds. Feinerg, T. E. & Frh, M. J.) (McGrw-Hill, New York, 1997). 34. Oldfield, R. C. The ssessment nd nlysis of hndedness: The Edinurgh inventory. Neuropsychology 9, (1975). 35. Regn, D. Humn Brin Electrophysiology: Evoked Potentils nd Evoked Mgnetic Fields in Science nd Medicine (Elsevier, New York, 1989). 36. Huynh, H. & Feldt, L. A. Conditions under which men squre rtios in repeted mesurement designs hve exct F-distriutions. J. Am. Stt. Assoc. 65, (1979). 37. Keppel, G. Design nd Anlysis: A Resercher s Hndook (Prentice-Hll, Englewood Cliffs, New Jersey, 1991). 196 nture neuroscience volume 2 no 2 ferury 1999