Exploring the Revised OLLC

Transcription

1 haper 5 Exploring he Revised OLL 5.1 Inroducion In his chaper, I will give an overview of he sysem ha was developed in he previous chapers by showing how i accouns for he various garden pah effecs we encounered hroughou his disseraion. he model consiss now of wo ingrediens. One is Priche s srucurebuilding mechanism of hea Aachmen: (1) hea Aachmen: he hea crierion aemps o be saisfied a every poin during processing given he maximal hea grid. he oher ingredien is a mechanism of reanalysis. In he previous chaper, I exended Priche s OLL by adding a clause ha saes ha he Human Senence Processor can auomaically perform reanalysis from he firs edge of a phase. he resuling formulaion is he following: (2) Revised On-Line Localiy onsrain (ROLL): he Human Senence Processor can auomaically perform reanalysis, iff: a. he source posiion dominaes he arge posiion; or b. he source posiion governs he arge posiion; or c. i can locae maerial o be reanalyzed in he firs edge of a single phase, and he arge posiion is locaed ouside ha phase Where firs edge is undersood as follows: (3) he firs edge of a phase is is lefmos edge conaining over maerial

2 152 Exploring he Revised OLL Obviously, he addiion of (2c) is a significan addiion o Priche s sysem as described in chaper 2; i allows for a whole new class of reanalyses. I seems o me ha i would herefore be a good idea o ake a sep back and look a he resuling sysem as a whole, o see which kinds of reanalysis i allows, which kinds i excludes, and o observe in deail is mode of operaion. In doing so, I will also pay some more aenion o he inuiions underlying he hree clauses of he ROLL, and poin ou some opics for fuure research. In addiion, I will presen some Duch cases ha have no been discussed so far. he addiion of he edge-clause o he ROLL has he effec ha we can disinguish wo ypes of reanalysis: reanalysis wihin a phase, which is poenially allowed by he governmen-clause or by he dominance-clause, bu never by he edge-clause; and reanalysis o a posiion ouside a phase, which can poenially be allowed by he edge-clause. I will firs discuss he cases of phase-inernal reanalysis we have seen so far. In secion 5.2 and 5.3 I will briefly recapiulae which kinds of reanalysis are permied by he dominance clause and he governmen clause of he (R)OLL, respecively. In his discussion, I will also poin ou some deails of Priche s sysem ha were ignored in chaper 2. In secion 5.4, I will briefly recapiulae he moivaion for he edgeclause. I will hen provide some addiional cases of uncosly reanalysis in Duch ha suppor he edge-clause in secion In secion 5.4.2, I will look a he English cases ha involve reanalysis from a phase o a posiion ouside he phase, and check if he ROLL makes he righ predicions. We will see ha indeed, in almos all he English cases of reanalysis from a phase ha we saw in chaper 2, he source posiion is no locaed in he firs edge, so he reanalysis here is sill correcly ruled ou by he ROLL. In secion 5.5, I will sugges a more elegan formulaion of he ROLL. Finally, in secion 5.6, I will discuss he predicions ha he curren model makes for he processing of relaive clauses. We will see ha he ROLL, conrary o he OLL, allows for reanalysis from an iniial main clause analysis o a reduced relaive clause analysis. We will also see ha he formulaion of he reanalysis consrain ha I propose in secion 5.5, does rule ou his ype of reanalysis, jus like he OLL, which can be viewed as an empirical argumen for his reformulaion. I will show ha i is acually difficul o ake a posiion here because he daa are very murky. I will furhermore show ha hea Aachmen enails many ineresing predicions abou he iniial analysis of reduced relaives. 5.2 he dominance clause of he ROLL he dominance clause of he (R)OLL says ha reanalysis is possible if he source posiion dominaes he arge posiion. he underlying inuiion

3 5.2 he dominance clause of he ROLL 153 seems o be ha reanalysis is possible if he Human Senence Processor coninues o work on he mos recenly aached consiuen. As we saw in chaper 2, he dominance clause of he (R)OLL allows for reanalysis o he specifier posiion of he original aachmen sie. his ype of reanalysis never leads o a garden pah effec, and i can occur recursively: (4) a. I like her b. I like her sudens c. I like her sudens papers d. I like her sudens paper s qualiy onsider for insance he seps from (4b) o (4c), and from (4c) o (4d): (5) a. P b. P like like her sudens papers her sudens papers qualiy A similar case of reanalysis where he source posiion dominaes he arge posiion arises when a nominal direc objec is reanalyzed o he subjec posiion of a verbal complemen, as in (6): (6) John believed Mary came When Mary is encounered, i is hea-aached as he direc objec of believed. he necessary reanalysis when came appears is depiced below:

4 154 Exploring he Revised OLL (7) P John P P believed P Mary P came he dominance clause of he (R)OLL also allows for recursive compounding. onsider (8): (8) We admire heir inelligence agency policy decisions (Priche s (238)) onsider for insance he reanalysis from We admire heir inelligence a- gency policy o We admire heir inelligence agency policy decisions:

5 5.3 he governmen clause of he ROLL 155 (9) P we P admire De heir decisions inelligence agency policy Since he source posiion dominaes he arge posiion, his ype of reanalysis is also freely allowed by he (R)OLL. 5.3 he governmen clause of he ROLL In he discussion of he OLL in chaper 2, we saw ha he governmen clause of he OLL is needed o accoun for only one case of uncosly reanalysis in English: (10) OKhey gave her books o Ron During he parsing of (10), her is iniially aached as he indirec objec of gave; his analysis has o be revised when he acual indirec objec, o Ron is encounered. In chaper 2, we already saw ha his reanalysis is allowed by he OLL because he source posiion governs he arge posiion. he relevan srucure was given in (47), and is repeaed below:

6 156 Exploring he Revised OLL (11) OK P hey P gave De her books PP o Ron A couple of hings mus be noed abou Priche s analysis. Firs, observe ha wo elemens are reanalyzed here: her and books. he reanalysis of her is unproblemaic: is source posiion dominaes is arge posiion. he oher elemen ha is reanalyzed here is books; books is removed from is original hea posiion, and is merged in he posiion conaining her, wih reanalysis of her o he specifier of ha posiion. Second, if he reanalysis process would be a reanalysis of her o he specifier posiion of books, he reanalysis would no be permied by he OLL. o see his, ake a look a he srucure below, which depics reanalysis of her raher han books: (12) P gave De her books PP o Ron he reanalysis in (12) is no allowed by he OLL, because he source posiion does no govern he arge posiion: here is a maximal projecion (he maximal projecion of books) inervening. 1 he reason ha his problem does no arise if we assume he reanalysis in (10) o be reanalysis of books raher han her, as depiced in (11), is 1 However, here is in fac a governmen relaion beween he source posiion of her and is arge posiion if we assume a DP-analysis for her books:

7 5.3 he governmen clause of he ROLL 157 ha he arge posiion here is he head posiion of he newly formed consiuen her books; he source posiion governs he maximal projecion of he arge posiion, hence he reanalysis is allowed according o Priche. As discussed in chaper 2, Priche makes a poin of explaining ha here is no problem wih aking he reanalysis in (10) o be reanalysis of books o a posiion in which a differen hea role was assigned before he reanalysis. 2 One way of looking a wha he governmen clause of he (R)OLL does, is o say ha i disinguishes beween (uncosly) reanalysis of a Pinernal argumen o a differen argumen posiion wihin he same P, and (cosly) reanalysis of a P-inernal argumen o a posiion adjoined inside a P-inernal argumen. he basic conras is he conras beween (13) and (14): (13) OKhey gave her books o Ron (i) P gave DP DP he D PP o Ron D r books his analysis also allows us o mainain he original source posiion; see foonoe 2. 2 oe ha here is anoher consrucion in which Priche does mainain he original hea posiion as he arge posiion. he relevan case is he following: (i) John warned Mary lef In he iniial parse, Mary is aached as an argumen of warned: (ii) P John P warned Mary his analysis has o be revised when came is encounered: Mary mus be reanalyzed as he subjec of he clause expressing he conen of he warning.

8 158 Exploring he Revised OLL (14) I pu he candy in he jar ino my mouh Le us consider again how he reanalysis in (14) is barred under he (R)OLL: (iii) P John P warned P P Mary P came oe ha in he srucure in (iii), he original hea posiion of Mary is mainained as he source posiion (which neiher governs nor dominaes he arge posiion). If he original hea posiion would no be mainained, he source posiion would be he maximal projecion of he complemen clause; and hen, he reanalysis would be permied by he OLL, since he source posiion would hen dominae he arge posiion. I does seem reasonable o mainain he source posiion in (iii), since he embedded P does indeed saisfy a differen hea role han Mary did in is source posiion. However, here is some ension beween reaining he original posiion here, bu no in he double objec cases. A more sophisicaed P-srucure migh shed some ligh on his issue.

9 5.3 he governmen clause of he ROLL 159 (15) P I P PP pu De he PP ino my mouh candy in he jar he crucial facor disinguishing (15) from (11) is ha in (15), he reanalyzed elemen is reanalyzed o an adjoined posiion inside he complemen. he maximal projecion of candy consiues an inervening barrier, and his excludes governmen of he arge posiion by he source posiion. I may seem as if he governmen clause of he (R)OLL is a raher sophisicaed echnical device which excludes cases of reanalysis on he basis of raher unsophisicaed (fla) P-srucures. However, he underlying inuiion seems prey clear: reanalysis from one argumen posiion o anoher argumen posiion inside he P is allowed, and reanalysis from an argumen posiion o an embedded adjoined posiion is no. I should be possible o reconsruc his noion in a synacic framework wih more P-inernal srucure, and wihou he noion of governmen, such as he Minimalis Program. he exac formulaion is a opic for fuure research. Le us for now proceed wih he governmen clause of he (R)OLL as formulaed. he same phenomenon as in (14) is found in Duch: (16) ( )Jan legde he snoep op afel in de kas Jan pu he candy on able in he cupboard he iniial parse is he following:

10 160 Exploring he Revised OLL (17) P Jan P legde P PP he snoep op afel his analysis mus be revised when in de kas in he cupboard is encounered: (18) P Jan P legde P PP De he PP in de kas snoep op afel

11 5.3 he governmen clause of he ROLL 161 his analysis is no allowed by he (R)OLL, because here is no governmen beween he source posiion and he arge posiion. he garden pah effec disappears when (16) is embedded: (19) OKIk denk da Jan he snoep op afel in de kas legde I hink ha Jan he candy on able in he cupboard pu I hink ha Jan pu he candy on he able in he cupboard his is exacly wha we expec under hea Aachmen: aachmen of he PPs will be posponed unil he verb is encounered. Since he verb appears a he end of he clause, here is no misaachmen and no reanalysis. he cases we discussed above all involve PPs ha are locally ambiguous beween an argumen aachmen and aachmen as an adjunc. We can find a similar ype of local ambiguiy wih Ps. Recall (13) from chaper 2, repeaed here as (20): (20) he man convinced he girl ha he lef o smile. he iniial analysis is depiced below: (21) P he man P convinced he girl P ha P he P lef And he necessary reanalysis looks like his:

12 162 Exploring he Revised OLL (22) P he man P P convinced De he girl P OP P PRO o P ha P he smile P lef Wha he cases in (15), (16), and (22) have in common is ha he arge posiion of he reanalysis is an embedded adjoined posiion; his means ha he source posiion does no govern he arge posiion. In chaper 2, we saed Priche s accoun of (20) wihou furher commen. On closer inspecion, however, i urns ou ha hea Aachmen acually predics opionaliy beween aachmen of he firs P as an argumen of convinced, and aachmen as a relaive clause. he reason for his is ha he resuling srucures each have one hea role lef o assign. onsider firs he argumen analysis in (23):

13 5.3 he governmen clause of he ROLL 163 (23) P 1 he man P convinced 2 he girl P 3 ha P he 1 P lef 2 In his analysis, he inernal hea role of lef is no assigned. his hea role is opional, bu i does coun for hea Aachmen because hea Aachmen refers o he maximal hea grid. In he relaive clause analysis (24), we also find one violaion of he hea crierion:

14 164 Exploring he Revised OLL (24) P 1 he man P 2 3 convinced De he P girl OP ha P he 1 P lef 2 Here, he unassigned hea role is he hird hea role of convinced, which is also opional. he iniial analysis in (24) is he correc analysis for he sring in (20); so (20) is acually prediced o give rise o only a 50% garden pah effec. In order o deermine wheher hea Aachmen makes he righ predicion here, more research involving a sufficienly large number of subjecs is necessary. A Duch equivalen is given below: (25) Jan verelde he meisje da de hond bee da hij naar huis ging Jan old he girl ha he dog bi ha he home wen hea Aachmen predics opionaliy here, for he same reason. he way o es wheher his predicion is borne ou, is o presen a sufficien number of subjecs wih senences like hese, and see if he judgmens really fall ino wo groups of roughly he same size, one group reporing no par-

15 5.3 he governmen clause of he ROLL 165 icular difficuly, he oher group reporing a garden pah effec. 3 he same predicion is made for he following case: (26) olgens de meneer die ik aan de elefoon kreeg, beeken According o he miser REL I on he phone go, means de uispraak van de kanonrecher da wij een he verdic of he canonal judge ha we an arbeidsovereenkoms hebben da we in loondiens employmen conrac have, ha we in salaried employmen zijn. are According o he person I alked o on he phone, he canonal judge s verdic ha we have an employmen conrac, means ha we are in salaried employmen. (source: personal , dd ) In his case, hea Aachmen predics opionaliy because he P da wij een arbeidsovereenkoms hebben can be eiher aached as a complemen of uispraak verdic (in which case no reanalysis needs o occur), or as a complemen of beeken means, in which case reanalysis does have o occur. his reanalysis violaes he (R)OLL. In chaper 2, we also saw he following case of phase-inernal reanalysis, excluded by he (R)OLL. For compleeness sake, i is repeaed below. 3 I esed his senence wih six naive speakers of Duch, and indeed received mixed judgmens, bu ha is hardly significan since he group was oo small.

16 166 Exploring he Revised OLL (27) P Karina P gave De he he wine P man who P was eaing P he fudge he reanalysis in (27) is excluded by he (R)OLL because he source posiion of he fudge neiher dominaes nor governs he arge posiion, and because he source posiion does no appear in he firs edge of a phase. 5.4 he edge clause of he ROLL he edge clause of he ROLL was moivaed by he possibiliy of reanalysis in Japanese of elemens originaing in he edge of a phase (see he discussion of (47), (51), (56), (63), (9), (19), (31), (35), (41), and (45) in he previous chaper). We saw ha a similar ype of reanalysis can ake place in Duch ((85b) in chaper 4). In his secion, I will discuss cases of reanalysis from a phase o a posiion ouside he phase in English and Duch. In secion 5.4.1, I will discuss an environmen in Duch ha provides some addiional suppor for supplemening he OLL wih he edge-clause (2c). In secion 5.4.2, we will see ha he edge-clause added o he OLL does no affec he English

17 5.4 he edge clause of he ROLL 167 garden pahs we discussed in chaper 2, which involve reanalysis from a complemen posiion Permied reanalysis from he edge of a phase onsider he following Duch case: (28) Giseren walgde de aio van de Urechse yeserday was disgused he PhD-suden of he Urech leerenfaculei van haar werk. faculy of leers of her work Yeserday, he PhD-suden of he Urech faculy of Ars was disgused wih her work he PP van de Urechse leerenfaculei is locally ambiguous; i can be aached o he subjec, or as an argumen of he verb walgde was disgused. By hea Aachmen, i is iniially aached as an argumen of he main verb: (29) P giseren yeserday P walgde was disgused de aio PhD P PP van de Urechse leerenfaculei of he Urech faculy of leers his analysis mus be revised when van haar werk of her work is encounered:

18 168 Exploring he Revised OLL (30) P giseren yeserday P walgde was disgused De de aio PP van de Urechse leerenfaculei of he Urech faculy of leers PP P van haar werk of her work his reanalysis can be performed wihou conscious processing difficuly. I is permied by he ROLL because i akes place from he firs edge of he P-phase. I is no, however, permied by he OLL, because he source posiion neiher governs nor dominaes he arge posiion; so he lack of processing difficuly in his case is an addiional argumen for he ROLL as opposed o he OLL. oe he conras in accepabiliy wih he reanalysis in (18), which is more difficul o process, and which is barred by he ROLL. he crucial srucural difference beween (18) and (30) is ha in (18), he reanalysis akes place phase-inernally, whereas i does no in (30). In (31), we see an example involving reanalysis from he firs edge of P o a P-adjoined posiion, which is also fine: (31) a. De soldaa dach aan he fron aan zijn meisje he soldier hough a he fron a his girl he soldier hough of his girlfriend a he fron b.... da de soldaa aan he fron aan zijn meisje dach ha he soldier a he fron a his girl hough ha he soldier hough of his girlfriend a he fron

19 5.4 he edge clause of he ROLL 169 he same kind of reanalysis can ake place wih a clausal complemen: (32) a. De gebedsgenezer bad om nie in slaap e vallen om een he miracle healer prayed for no asleep o fall for a wonder miracle he miracle healer prayed for a miracle in order no o fall asleep b.... da ha de gebedsgenezer om nie in slaap e vallen om een he miracle healer o no asleep o fall for a wonder bad miracle prayed ha he miracle healer prayed for a miracle in order no o fall asleep his reanalysis is also performed wihou conscious processing difficuly Reanalysis from complemen posiion he inuiion underlying he edge clause of he ROLL is ha a phase is inaccessible for operaions relaing o posiions ouside he phase, excep for maerial in is edge. his predics ha reanalysis from a complemen posiion o a posiion ouside a phase is always impossible. Indeed, hese are exacly he configuraions ha lead o he mos severe garden pah effecs in English. In his subsecion, I will review hese cases. he firs case of reanalysis from inside a phase o a posiion ouside ha phase ha we will explore is he mos sraighforward case of objecsubjec ambiguiy in English: (34) from chaper 2, repeaed here as (33): (33) Afer Susan drank he waer evaporaed he srucure is he following (relevan phase boxed):

20 170 Exploring he Revised OLL (34) P PP P P P afer Susan P he waer P evaporaed drank learly, he phase-clause of he ROLL does no allow his insance of cosly reanalysis, as he source posiion is no locaed in he firs edge of a phase; i is in he complemen posiion of drank. oe how srucurally similar his case seems o be o he Japanese example (53) we saw in he previous chaper, repeaed here as (35): (35) OK Yumiko-o yobidasia kissaen-ni nagai koo maa-sea Yumiko-A summoned ea room-lo long ime wai-made made Yumiko wai for a long ime a he ea room o which he summoned her (Mazuka and Ioh s (18b)) Is srucure is given below, wih a box indicaing he phase from which reanalysis akes place:

21 5.4 he edge clause of he ROLL 171 (36) P pro P Yumiko-o maa-sea wai-made P OP P kissaen-ni ea room pro P pro yobidasia summoned Boh (36) and (34) involve reanalysis of a direc objec from an embedded clause o a higher clause, forced by he main verb. In boh cases, he original argumen posiion (he direc objec posiion of drank in (34), and he direc objec posiion of yobidasia summoned in (36)) is lef unexpressed phoneically, even hough he hemaic role associaed wih he posiion is presen semanically. In neiher case does reanalysis obey he governmen clause of he OLL. Bu whereas he reanalysis is allowed in Japanese, i

22 172 Exploring he Revised OLL is no in English. oe ha he reanalysis consrain formulaed in (2c) allows for he reanalysis we see in he Japanese case, since he source posiion is locaed in he firs edge, whereas in (35), he source posiion is in complemen posiion. he example in (33) has no counerpar in Duch, since in Duch he direc objec appears before he verb in he embedded clause. However, we can consruc a counerpar involving a preposiional objec, since preposiional objecs can appear afer he head noun of he relaive clause, bu o he righ of he embedded verb. onsider he following example: (37) Ik dach da Jan de man die had gewach op de bus I hough ha Jan he man who had waied for/on he bus naar Zalbommel zee o Zalbommel pu I hough ha Jan pu he man who had been waiing, on he bus o Zalbommel In he iniial parse, he PP op de bus naar Zalbommel is analyzed as an argumen of gewach waied. he srucure of he conaining he relaive clause looks like his: (38) De de he P man OP die who P had P gewach waied PP op de bus naar Zalbommel on he bus o Zalbommel

23 5.4 he edge clause of he ROLL 173 When he verb zee pu is encounered, he PP has o be reanalyzed as an argumen of zee: (39) P Jan P De de he PP op de bus naar Zalbommel on he bus o Zalbommel zee pu man P OP die who P had P gewach waied PP e he reanalysis consrain (2c) correcly rules ou his reanalysis, since he source posiion does no occur in he firs edge of he P. Again, we see ha he srucural posiion of he source is relevan o he availabiliy of reanalysis. Anoher case of reanalysis from a complemen posiion inside a phase o a posiion ouside he phase, is (9) from chaper 2, repeaed here as (40): (40) Below he sairs collapsed In (40), he sairs is iniially aached as an argumen of below, and needs o be reanalyzed as an argumen of collapsed. he reanalysis is depiced

24 174 Exploring he Revised OLL below: (41) P P below P he sairs P collapsed I is excluded by he OLL: he source posiion does no appear in he firs edge of he PP-phase 4, nor does i govern or dominae he arge posiion. o complee our overview of cases of cosly reanalysis from a complemen posiion, recall he following case, which was discussed in chaper 2: (42) Karina gave he man who was eaing he fudge If he fudge is iniially analyzed as he complemen of eaing, a garden pah effec arises. Subsequen reanalysis would resul in (43): 4 However, i does appear inside he firs edge of he P-phase; apparenly, reanalysis is resriced o members (i.e. he highes maximal projecion) of he firs edge; he parser canno look inside he consiuen in he firs edge.

25 5.5 owards a more elegan formulaion of he ROLL 175 (43) P Karina P gave De he he fudge P man who P was eaing P he reanalysis in (43) is no allowed by (2c), since he source posiion is no locaed in he edge of he phase. 5.5 owards a more elegan formulaion of he ROLL In he sysem developed here, here are hree possible configuraions in which he Human Senence Processor can auomaically perform reanalysis: i can build maerial on op of a consiuen (he dominance clause of he ROLL), i can shuffle P-inernal argumens around (he governmen clause), and i can reanalyze maerial from he edge of a phase o a posiion ouside he phase (he edge-clause). While he ROLL as saed works empirically, as we saw in his chaper and he previous one, he way i is formulaed is no paricularly elegan: he governmen and he dominance clause are of a differen kind han he edge-clause. Ideally, i should be possible o sae he ROLL as follows:

26 176 Exploring he Revised OLL (44) wice Revised On-Line Localiy onsrain (ROLL): he Human Senence Processor canno access maerial inside a phase once i is closed, unless he maerial is locaed in he firs edge of ha phase he paerning of he daa under his consrain would be essenially he same as under he ROLL. Reanalysis from complemen posiion would be excluded because he P is a closed phase, and he parser can access maerial inside a closed phase, unless i s in he edge. As we saw, reanalysis from a complemen posiion resuls in he mos severe garden pah effecs: (45) Below he sairs collapsed (46) Afer Susan drank he waer evaporaed (47) Wihou her conribuions failed o come in he cases ha are ruled in by he dominance clause, would in principle follow because maerial is added o a phase (which shows ha i is no ye closed); he same would go for he cases ha are ruled in by he governmen clause; hese cases involve shuffling maerial around inside a phase, namely a P under consrucion. Recall ha here are cases ha also involve shuffling maerial around inside a P which do evoke a garden pah effec: (48) I pu he candy in he jar ino my mouh hese cases should be excluded because he renalysis involves adjuncion ha adds maerial o a phase ha has already been closed (he candy in (48)). Of course, he par of (44) ha remains o be defined is he once i is closed par; we have o find a way o define when exacly a phase is closed. Finding he righ definiion is no rivial. On he up side, i seems ha he ROLL, also in he way i is saed now, opens a possibiliy o accoun for he difference in he severiy of garden pah effecs in Japanese versus English. In he previous chaper, in foonoe 1, I referred o Mazuka and Ioh s observaions ha here is a difference in judgmens beween English garden pah senences and Japanese garden pahs. According o Mazuka and Ioh, he Japanese garden pahs are less difficul: By using he enaive aachmen sraegy, we can also reflec he Japanese speaker s inuiive judgmen ha none of he reanalyses we considered here cause a garden-pah effec as clearly and robusly as some of English garden-pah senences. A he same ime, i is no he case ha he garden-pah effec is missing in Japanese. Raher, he difficulies caused by various

27 5.5 owards a more elegan formulaion of he ROLL 177 ypes of reanalysis seems o be of various degrees. If we aemp o ascribe he garden-pah phenomena o he fac ha cerain reanalyses violae some processing principle (e.g. a relicensing consrain, as in Priche, 1991, or as a parsing breakdown, as in Marcus, 1980), hen he effec should be all or none, and i would be difficul o accoun for he "somewha difficul" senences we have discussed here. (Mazuka and Ioh (1995), p.325) Recall (21) in chaper 4, repeaed here as (49): (49) Yoko-ga kodomo-o koosaen-de mikakea akusii-ni Yoko-OM child-a inersecion-lo saw axi nosea pu-on Yoko pu he child on he axi she saw a he inersecion. (Mazuka and Ioh s (14a)) In (49), wo argumens have o be reanalyzed, boh from an edge posiion. Mazuka and Ioh do label senences like (49) a garden pah, because here is a conscious processing effor associaed wih hem; bu hey repor ha he psychological effec is no he same as in English. I m assuming ha hey have in mind English garden pahs like he following: (50) Wihou her conribuions failed o come in. his ype of garden pah is no found in Japanese, because his ype of complemen-subjec ambiguiy canno be consruced in Japanese. Garden pahs like hese can also no be found in Duch, for he same reason (excep for cases like (37) in his chaper, which indeed really is very difficul, in he same way ha (50) is). We do, however, find inuiions of reanalysis complexiy in Duch, jus as in Japanese; recall he discussion of (85b) in he previous chaper. he environmens in which we find hese inuiions of processing complexiy raher han a more or less persisen impression of ungrammaicaliy (as in reanalysis from complemen posiion), is undersandable if we look a he daa from he perspecive of he ROLL. Reanalysis from edge posiions in principle lies wihin he capabiliies of he Human Senence Processor, albei only one edge a a ime. I is no surprising ha having wo or more of hese operaions o be performed simulaneously, causes a conscious impression of complexiy of reanalysis, even hough hey are permied in principle. his seems o me o be a convincing argumen in favor of he approach in erms of phases ha has been defended in his disseraion.

28 178 Exploring he Revised OLL 5.6 Reduced relaives Reanalysis and he ROLL We have seen ha so far, he ROLL is raher successful in disinguishing cosly from uncosly reanalysis. However, here is one srucure where he edge-clause fails (quie specacularly): i allows for reanalysis from he subjec posiion of an iniially parsed main clause o he subjec posiion of he acual main clause. his is he ype of reanalysis we find in reduced relaive-main clause ambiguiies, and as is well known, his can no ake place cos-free: (51) he horse raced pas he barn fell. In chaper 2, we saw ha Priche s OLL can accoun for he difficuly of reanalysis in (51), since he source posiion (he subjec posiion of raced) neiher governs nor dominaes he arge posiion (he subjec posiion of fell).

29 5.6 Reduced relaives 179 (52) P De he P fell P horse OP P P raced PP pas he barn he ROLL does allow for reanalysis of he horse: he reanalyzed elemen originaes in he firs edge of he embedded P-phase, and he arge posiion is ouside he phase. his clearly is a problem for he ROLL: hese ypes of srucures are generally very difficul o process:

30 180 Exploring he Revised OLL (53) a. he sudens advanced o he nex grade had o sudy very hard b. he clipper sailed o Porugal carried a crew of eigh c. he roops marched across he fields all day resened he general d. he model plane roaed on he meal axis fell off he sand e. he dog walked in he park was having a good ime f. he ship glided pas he harbor guards was laden wih reasure (Sevenson and Merlo s (1997) (3)) On he oher hand, we know ha here are cases in Japanese which are very similar srucurally o (51), and which do no lead o a garden pah effec, such as (51) from chaper 4, repeaed here as (54): (54) P akamura-ga P pro P P oki when pro P syuuri-sie-kurea repaired (for me) yuuko-no pasokon-o second-hand P kaa bough In boh (51) and (54), reanalysis involves he subjec of a clause (an iniial main clause in (51), and an unaached adverbial clause in (54)). In boh cases, he arge posiion is he subjec posiion of he marix verb which riggers he reanalysis. In boh cases, he reanalysis is allowed by he edge clause of he ROLL. Bu (51) is a garden pah, and (54) is no. here is one ineresing srucural difference beween he wo cases; his difference lies in he aachmen sie of he phase from which he sub-

31 5.6 Reduced relaives 181 jec is reanalyzed. In (51), he aachmen of he iniial main clause srucure, is in a posiion adjoined o horse. In (54), he adjuncion sie of he adverbial clause is a projecion of he verb which riggers he reanalysis. Under he ROLL, his laer adjuncion should be allowed, since he P headed by kaa bough in (54) is no ye closed: afer all, is head kaa has jus been encounered. However, i would be enirely reasonable o suspec ha he horse in (51) is closed a he poin when he reduced relaive should be aached o i. his would mean ha adjuncion o horse is no possible under he ROLL (compare (48)), and his could explain why (51) is a garden pah, whereas (54) is no. If his explanaion is on he righ rack, we have also found an empirical way o disinguish he ROLL from he ROLL. However, I am hesian o ake a final posiion here, because i is no clear o me wheher i is no acually desirable o allow for reanalysis from an iniial main clause o a reduced relaive clause, as is prediced o be fine by he ROLL. o see his, le us ake a look a some of he daa. In chaper 4, I already discussed Priche s observaion ha reduced relaive clauses do no lead o a garden pah effec if he verb in he relaive clause is obligaorily ransiive. Recall he conras beween (55) and (56): (55) a. he horse raced pas he barn fell b. (Rex knows) he boy hurried ou he door slipped (56) a. he spaceship desroyed in he bale disinegraed b. he bird bough in he sore flew away c. he children found in he woods were frozen Priche resored o a sysem of rebuffering: he proposed ha he reason why he examples in (56) are processed wihou effor is ha he srucure is broken up when he PP is encounered, because he PP shows ha he main clause analysis is no he righ one. ow, i has been observed in he lieraure ha reduced relaives are no only processed efforlessly when he verb is obligaorily ransiive, bu also when he verb is he ransiive alernae of an unaccusaive raher han an unergaive verb. onsider he examples in (57) and (58): (57) OKhe buer meled in he microwave was lumpy (Sevenson and Merlo s (2)) (58) a. he wich meled in he Wizard of Oz was played by a famous acress b. he genes muaed in he experimen were used in a vaccine c. he oil poured across he road made driving reacherous d. he picure roaed 90 degrees was easy o prin (Sevenson and Merlo s (10))

32 182 Exploring he Revised OLL oe ha Priche s rebuffering sysem predics no rebuffering here (for insance, he buer meled in he microwave is a perfecly accepable main clause), bu he senences in (57) and (58) are sill easy o process, even hough hey violae he OLL. Sevenson and Merlo (1997) argue ha he reason why (51) is bad is no ha he reanalysis ha has o be performed here canno be performed, bu ha here is some synacic problem wih building a reduced relaive clause wih he ransiive alernae of an unergaive verb. o suppor his claim, hey esed (in a quesionnaire) senences wih morphologically unambiguous pariciple forms of verbs of his ype, in conexs which were eiher neural or biased owards he reduced relaive reading. he resul hey go was ha people had grea difficuly even wih hese morphologically unambiguous pariciple forms: (59) a.??he greyhound run around he rack all day was ired b.??he children wihdrawn from he religion class could sudy music insead c.??he albaross flown o Ausralia was very ired (Sevenson and Merlo s (4)) (60)??he bell rung only a Easer rused from disuse (Sevenson and Merlo s (5)) So i does indeed seem as hough he problem wih (51) could be due o a more general problem, and is no a problem wih he required reanalysis per se. I is no easy o ake a posiion here, since he lieraure on he processing of reduced relaive abounds wih conflicing judgmens and consideraions of (discourse and oher) facors ha supposedly faciliae or complicae reanalysis 5. For insance, Sevenson and Merlo s analysis is challenged by Filip (1998), who claims ha he problem wih (51) lies in he rigger for reanalysis (he main verb). She claims ha he fewer agenive properies and he more paien-like properies he main verb assigns o is subjec, he easier he whole senence wih a reduced relaive clause is judged (Filip s (1998), p.5) o subsaniae his claim, Filip gives he following examples wih ransiive alernaes of unaccusaive verbs ha are acually easy o process: (61) a. he vicims rushed o he emergency room died upon arrival b. he prisoners paraded pas he mob was [sic, IM] laer execued 5 Or selecion of he main clause or reduced relaive analysis, for he adherens of parallel processing models.

33 5.6 Reduced relaives 183 c. he ca rolled down he ramp overurned when i hi a bump d. he dog walked in he park was wearing a choke collar e. he debuane walzed across he floor wore a beauiful dress f. he diplomas jeed o Iraq were unable o diffuse he crisis g. he convic moved ino an isolaion cell became depressed h. he conraband floaed down he river was discovered by he border parol i. he soldiers marched across he fields were ambushed by he enemy (Filip s (5)) And she also gives a se of examples wih ransiive alernaives of unaccusaive verbs which are difficul o process: (62) a. he heare darkened for he movie frighened some preschoolers b. he Klingon disinegraed during he bale had launched a rocke c. he soluion crysallized in he oven burned a hole ino he peri dish d. he plaser hardened in he oven cracked wih loud popping sounds e. he bubble burs in he made he principle jump [sic, IM] f. he paper yellowed in he sun was wrinkled (Filip s (6)) She also gives he following consrass: (63) a. he paiens rushed o he emergency room complained o he nurse b. he paiens rushed o he emergency room died (Filip s (7)) (64) a. he Grea Dane walked in he park ugged on he leash b. he Grea Dane walked in he park wore a choke collar (Filip s (8)) (65) a. he heare darkened for he movie frighened some preschoolers b. he heare darkened for he movie smelled like popcorn (Filip s (9)) (66) a. he genes muaed in he experimen aacked heir hos b. he genes muaed in he experimen were used in a new vaccine (Filip s (10))

34 184 Exploring he Revised OLL My main poin here is no o develop a comprehensive analysis of he daa above, bu o show ha here seems o be significan confusion abou he daa. he reduced relaive-main clause ambiguiy is one of he mos widely researched ambiguiies in he field, and probably for good reasons: he reduced relaive clause consrucion is raher frequen in English, so i would be odd if i were as exremely difficul o process as many processing heories predic. In he remainder of his secion, I will focus on he predicions of he model ha has been developed here; wheher hese predicions are righ is a opic for fuure research hea Aachmen and reduced relaive clauses In his secion, we will see ha hea Aachmen enails many ineresing predicions abou he iniial analysis of reduced relaives. In he inroducion o Priche s model in chaper 2, I preended ha hea Aachmen always forces reduced relaives o be analyzed as main clauses iniially, bu as we will see in some deail in his secion, hea Aachmen acually predics opionaliy beween he main clause analysis and he reduced relaive clause analysis. his is ineresing because, as we have seen above, here is a lo of confusion abou he daa, and he opionaliy prediced under hea Aachmen may acually give us a handle o beer undersand his confusion. o see how hea Aachmen predics opionaliy in reduced relaive clauses, reconsider for insance he all-oo-familiar (67): (67) he horse raced pas he barn fell. Le us go over his example sep by sep. A he horse, nohing happens; here is no hea assigner available. When raced is encounered, is maximal hea grid is rerieved; race can assign wo hea roles: (68) race here are in principle wo opions for srucure building: a main clause can be consruced, wih he horse as subjec, or a reduced relaive clause modifying he horse. he wo alernaives are given below:

35 5.6 Reduced relaives 185 (69) a. P he horse P raced b. he P horse OP P P raced Boh analyses saisfy hea Aachmen equally well. In (69a), one hea role from he maximal hea grid of raced is lef unassigned, namely he inernal hea role. In he analysis in (69b), boh hea roles of raced are assigned: since he verb is passivezed, he exernal hea role of raced is absorbed, while he inernal hea role is assigned o he relaive operaor. he consruced headed by horse does need a hea role iself, hough. So, he srucure in (69b) consiss of an wihou a hea role, while he srucure in (69a) conains an unassigned hea role. hey herefore saisfy hea Aachmen (as formulaed in (1)) equally well. his is a problemaic siuaion, since he predicion now is ha he senence in (67) leads o a garden pah only half he imes i is parsed. o see his, le us consider wha happens exacly under each alernaive. Firs, le us consider wha happens if he parser chooses o sar wih

36 186 Exploring he Revised OLL he srucure in (69a). he parse coninues wih he inpu of pas he barn, which is aached as an adjunc 6. When fell is encounered, he iniial main clause analysis mus be revised; horse mus be reanalyzed as he subjec of fell. (70) P De he P fell P horse OP P P raced PP pas he barn As noed in he previous secion, his reanalysis violaes he OLL (and he ROLL). he source posiion (he subjec posiion of raced) does no dominae or govern he arge posiion (he subjec posiion of fell). he parse ha sars wih he srucure in (69a) herefore correcly predics a garden pah effec here. he siuaion is quie differen, however, if he Human Senence Processor sars wih he reduced relaive clause analysis in (69b). Under ha scenario, no reanalysis is needed since he reduced relaive analysis 6 oe ha hea Aachmen does no give any direcive for he processing of adjuncs.

37 5.6 Reduced relaives 187 is pursued from he beginning, and we predic no garden pah effec. So he problem is ha hea Aachmen acually predics only 50% garden pah here, while (67) is always a garden pah. One can hink of a number of soluions for his problem. One soluion is o say ha he Human Senence Processor favors hea role recepion over hea role assignmen, i.e. i is more imporan for s o receive a hea role, han i is for hea assigners o assign heir hea roles. Le us formulae his guideline for he parser as follows: (71) When faced wih a choice beween wo analyses for a paricular sring, one of which leaves a hea role undischarged, and one of which leaves an wihou a hea role, he parser chooses he firs analysis. his guideline is no made explici in Priche s work, bu i seems o be presen implicily in his discussion on pp A sraegy like his seems reasonable, considering ha hea roles may be opionally assigned, while s can never occur wihou a hea role. he guideline in (71) provides an accoun for why he analysis in (69a) is chosen over he analysis in (69b) in he following way. In (69a) one hea role has been assigned and every has a hea role; his is beer han he siuaion in (69b) because here one hea role role has been assigned and one absorbed, while he resuling sill needs a hea role. In oher words, he local hea crierion violaion in (69b) involves an wihou a hea role, while he local hea crierion violaion in (69a) involves an undischarged hea role. herefore, he main clause analysis mus be pursued firs under guideline (71). 7 oe ha he approach aken in (71) predics ha he siuaion changes when he ambiguous region appears embedded under a higher verb: (72) John knows he horse raced pas he barn fell o see he difference, le us look carefully a wha happens during he parsing of his senence. John and knows can be inegraed in he familiar 7 A relaed "soluion", suggesed by Priche on p.90, is ha he synacic analysis of reduced relaive clauses should in fac be differen from he radiional analysis involving a relaive clause. Following Abney (1987) and Mcawley (1988), he suggess ha reduced relaives migh in fac be beer analyzed as APs. he alernaives beween which he parser has o choose hen, are (69), repeaed here as (ia), and somehing like (ib): (i) a. P he horse P raced

38 188 Exploring he Revised OLL way, John being aached as he subjec of knows. When he horse is encounered, i has o be aached as he direc objec of knows in accordance wih hea Aachmen: (73) P John knows P he horse he nex inpu consiss of raced. here are wo ways o fi in raced ino he exising srucure. One is o simply build a reduced relaive clause and adjoin i o he projecion of horse. he resuling srucure is he following: b. he AP raced horse Under he assumpion ha APs don assign hea roles, we see ha we sill need guideline (71) o explain why he Human Senence Processor iniially chooses analysis (ia). In he srucure in (ia), here is one hea role lef unassigned, while he srucure in (ib) consiss of an wihou a hea role. his means ha, again, he number of local violaions of he hea rierion is he same for boh srucures (namely one); and we need an addiional mechanism, like (71), o force iniial analysis as (ia) raher han (ib). Since he AP-analysis of senences like (67) seems o make he same predicions as he reduced relaive clause analysis, I will ignore i here and coninue o use he reduced relaive clause analysis.

39 5.6 Reduced relaives 189 (74) P John knows P De he P horse OP P P PP raced pas he barn John knows he horse (which) raced pas he barn In his srucure, no hea roles are lef unassigned, and no s are lef wihou a hea role; specifically, he horse coninues o receive is hea role from knows. Apparenly, assuming ha hea role absorpion couns as an opional saisfacion of he maximal hea-grid, we expec he reduced relaive analysis o be considered firs, raher han second, in an embedding conex, so ha no garden pah effec should occur. When fell is encounered, he srucure has o be reanalyzed; he he horse raced pas he barn has o be reanalyzed as he subjec of fell, and he embedded senence he horse raced pas he barn fell receives he inernal hea role of knows.

40 190 Exploring he Revised OLL (75) P John P knows P P De he P P fell horse OP P P raced PP pas he barn As can be seen in he srucure above, he source posiion of he horse raced pas he barn dominaes is arge posiion; herefore, his reanalysis is permied by he OLL, he ROLL, and he ROLL. Le us now go one sep back, o he poin where raced was encounered. A his poin, he srucure looked like (73), repeaed here as (76):

41 5.6 Reduced relaives 191 (76) P John knows P he horse In principle, he parser can also choose o reanalyze he srucure o one conaining an embedded clause when i encouners raced; i can consruc he horse raced pas he barn as an embedded senence, assigning i he inernal hea role of knows: (77) a. P John P knows P P he horse P raced PP pas he barn John knows (ha) he horse raced pas he barn However, his analysis leaves open one hea role: he second hea role of he maximal hea grid of raced. ompare his siuaion o he srucure in (74). here, no hea role is lef unassigned. his means ha for hea Aachmen, wih or wihou guideline (71), he analysis in (74) is he only opion. We have already seen ha ha analysis only leads o an uncosly

42 192 Exploring he Revised OLL reanalysis when fell is encounered. Hence, hea Aachmen predics no garden pah effec in (72). Guideline (71), which gave us he righ resul for he choice beween he main clause analysis (as given in (69a)) and he reduced relaive analysis (given in (69b)) in non-embedded conexs, does no have an effec in embedded conexs. oe ha he same predicions are also made for reduced relaives wih an obligaorily ransiive verb. his means ha hea Aachmen predics no garden pah effec in (78a); i also means ha in (78b), reanalysis is prediced under hea Aachmen. (78) a. I saw he gangser jus murdered lay in he sree. b. I saw he gangser jus murdered a kid. o sum up: we saw ha hea Aachmen has some difficulies in accouning for he garden pah effec in main clause-reduced relaive clause ambiguiies. We discussed one possibiliy for an addiional guideline ha forces he main clause analysis in he simple cases, which seems o be implici in Priche s work. his guideline has no effec in embedded conexs. As we have seen, he model makes many ineresing predicions ha should be looked ino. Given ha a crucial aspec of he model is ha in case of opionaliy one analysis is chosen, reading ime sudies do no seem o be a suiable mehod o es he predicions of he curren model, as we would expec some readers o choose one opion, and anoher group of readers o chose he oher opion. I is no clear o me how his could be esed in a reading ime sudy. A beer mehod would be o use quesionnaires, and ask readers for inrospecive judgmens Reduced relaives in Duch o complee he discussion of he predicions made by he curren model for reduced relaive clauses, I would like o poin ou some predicions for he parsing of reduced relaives in Duch. Under he ROLL, we expec ha reanalysis of wo argumens from an iniial main clause analysis, will be much more difficul han reanalysis of one argumen. his canno be esed in English main clause-reduced relaive clause ambiguiies, bu i can be esed in Duch because Duch is verb-final. I mus be noed, however, ha reduced relaives are much less common in Duch han hey are in English (hey are somewha archaic). everheless, I esed he following cases of reanalysis of one argumen: (79) a. Deze afrikaanse volkeren beschreven in de aional hese african peoples described in he aional Geographic hun hopeloze siuaie Geographic heir hopeless siuaion hese African peoples described heir hopeless siuaion in he

43 5.6 Reduced relaives 193 aional Geographic b. OKDeze afrikaanse volkeren beschreven in de aional hese african peoples described in he aional Geographic leven nog in he senen ijdperk Geographic live sill in he sone age hese African peoples described in he aional Geographic are sill living in he sone age (79b) is a reduced relaive clause. If he iniial analysis is a main clause, he necessary reanalysis involves he subjec of his iniial analysis. his is allowed by he edge clause of he ROLL because i involves only one argumen, bu no by he ROLL, nor by he OLL. (80) P De deze hese P leven live P PP afrikaanse African volkeren peoples OP P in he senen ijdperk in he sone age P beschreven described PP in de aional Geographic in he aional Geographic My subjecs generally seemed o have no garden pah effec in (79b). One speaker acually repored a garden pah effec on he simple main clause in (79a).

44 194 Exploring he Revised OLL For compleeness, I also include he oher cases I esed below. (81) a. De bergbeklimmers beschreven in de aional Geographic he mounainiers described in he aional Geographic een moeilijke och vol hindernissen. a difficul journey full obsacles he mounainiers described a difficul journey full of obsacles in he aional Geographic b.( )De bergbeklimmers beschreven in de aional Geographic he mounainiers described in he aional Geographic volooiden een moeilijke och vol hindernissen compleed a difficul journey full obsacles he mounainiers described in he aional Geographic compleed a difficul journey full of obsacles (82) ( )De ondekkers van di merkwaardige verschijnsel beschreven he discoverers of his srange phenomenon described in een invloedrijk arikel hebben de nobelprijs gewonnen. in an influenial aricle have he obel prize won he discoverers of his srange phenomenon described in an influenial aricle won he obel prize (example from Oele Koornwinder) (83) ( )Een slikker is een drugskoerier die bollejes drugs verpak a swallower is a drugs rafficker who balls drugs wrapped in sevig plasic vervoer in de maag. in srong plasic carries in he somach A swallower is a rafficker who carries balls of drugs wrapped in srong plasic in he somach (example heard in he v-show Opsporing erzoch ) he verbs here are all obligaorily ransiive, and he judgmens for he reduced relaives vary from perfecly fine o complex bu no unprocessable. As for cases where wo over argumens mus be reanalyzed, I esed he following examples: (84) De agen he police officer heef has bollejes balls drugs verpak in sevig plasic drugs wrapped in firm plasic gevonden. found he police officer found balls of drugs wrapped in firm plasic