A C T A U N I V E R S I T A T I S L O D Z I E N S I S FO L IA L IN G U IST IC A 36, 1997 P iotr S ta lm a szczyk T H E M A T IC A N D C O N C E P T U A L S E M A N T IC S O F E N G L IS H M ID D L E V ER B S* 1. INTRODUCTION
T h o u g h it is w o rth rem em bering th a t “ th e lexicon is n o t ju s t v erb s” [ P u s t e j o v s k y 1991: 410], recent research in the G o v ern m en t an d B inding th eo ry o f g ra m m a r (and related syntax-oriented fram ew orks) seems to be co n c en trate d m ainly o n verbs and co n stru c tin g a p p ro p ria te lexical rep resen ta tio n s fo r them . A nd the p roblem central to this research is diathesis: the re la tio n betw een the sem antic (i.e. th em atic) roles subcategorized fo r by a verb o r predicate, and the surface expression o f these roles as arg u m en ts. In this p a p e r I discuss different ap p ro ach e s to th em atic roles in generative g ra m m a r an d th eir inadequacies. I also provid e an altern ativ e a p p ro a c h w hich enables a m ore co h eren t in te rp re ta tio n o f syntactic c o n stru c tio n s w ithin the fram ew o rk o f conceptual sem antics. T h e sy ntactic co n stru c tio n u n d e r analysis is the so called middle verbal diathesis.
In the sta n d a rd G o v e rn m en t and B inding m odel o f generative g ra m m a r, lexico-sem antic in fo rm atio n associated w ith a p red icate is p rovided by a theta-grid (th-grid). T h e th-grid o f a verb is a list o f the arg u m e n ts th a t th e p re d ic a to r req u ires. E ac h o f these arg u m e n ts is id entified by th e sem antic re la tio n it bears to the p re d ic ato r (i.e. its th em atic role - th -ro le )1.
* This is an extended and revised version of a paper presented a t the 1993 K azim ierz Conference on Semantics and Lexicography, and published in: H. K o r d e l a and G. P e r s s o n (ed s), New Trends in Semantics and Lexicography. A cta Universitatis Umensis 1995: 151-165.
1 The relation between arguments and roles is governed by the Theta-Critcrion, a biuniqueness condition on th-role assignment, which forces the requirem ents o f the lexicon to be projected into the syntax. The standard form ulation o f the Theta-C riterion relates roles to argum ents [ C h o m s k y 1981: 36]: (i) T heta-Criterion: Each argum ent bears one and only one th-role, and each th-role is assigned to one and only one argum ent.
A typical th-grid is given in (1). In fo rm a tio n included in th is grid specifies th a t p u t is a triad ic verb w hich requires th ree argum ents: a n A g en t, a T hem e, and a L o ca tio n , as in (2):
(1) put: < A g ent, T hem e, L o ca tio n > , (2) Jo h n p u t the b o o k on the shelf.
A g en t T h em e L ocation.
2. T H E M A T IC R O LES
T h em a tic roles used by researchers in the G B fram ew o rk o rig in ate from th e early w ork connected with lexical sem antics, especially F i l l m o r e [1968] an d G r u b e r [1965]. B oth F illm ore and G ru b e r p o stu la te a finite set o f underlying categories w hich serve to unite the sem antic an d syntactic levels.
F o r F i l l m o r e [1968: 20] it is Case which is an “ u nd erlyin g sy n tac tic-sem antic re la tio n sh ip ” . C ases can be identified b o th sem antically and sy n tactica lly . T h e sem an tic id en tificatio n pro ceed s th ro u g h p o in tin g to intuitiv e n a tu ra l classes based on th e way in which we con ceptu alize states an d events, w hereas the syntactic identificatio n is d o n e by show ing covert gram m atical distinctions in th e ways in w hich n om inals behave in th e syntax. F i l l m o r e [1968: 24-25] suggests th a t the follow ing cases exist: agentive, dative, in stru m en tal, factive, locative, objective (sentential and nom inal), benefactive and tem p o ral.
G r u b e r [1965] proposed a set o f thcmatic relations, originally based on verbs o f m o tio n . T h e system was fu rth e r elab o rated by J a c k e n d o f f [1972], and in c o rp o rated into the th eta-th e o ry m o d u le o f the G B fram ew ork [cf. C h o m s k y 1981; S t o w e l l 1981; W i l l i a m s 1981]. T h e follow ing is a list o f th em atic re la tio n s developed by J a c k e n d o f f [1972] to g eth er w ith som e later m o d ificatio n s and additions:
(3) A gent - an N P expressing will tow ard the action,
T h em e - fo r verbs o f m o tio n : th e m o v in g o b ject, fo r v erb s o f location: the th in g w hich is located,
L o ca tio n - the N P (usually w ithin a P P ) expressing lo catio n , Source - the initial position o f the T hem e,
G o a l - the final d estin atio n o f the T hem e,
E xperiencer - the individual w ho feels o r perceives the event, P ercept - an entity w hich is experienced o r perceived,
P atien t - an entity w hich undergoes an action,
In stru m e n t - the object with w hich th e action is p erfo rm ed , B enefactive - the entity fo r w hose benefit the event to o k place.
T h e F illm o re - G ru b e r - Jac k e n d o ff ac co u n t is based o n th e follow ing m ain assum ptions:
(4) a. T h em a tic relations are atom ic labels d raw n from a fixed list, b. T h e labels are ordered in a hierarchy an d linked to th e syntactic
positions,
c. Every arg u m e n t has exactly one th em atic role.
A different ap p ro ach is advocated by A n d r e w s [1985]. H e suggests th a t th e re are p ro b a b ly infinitely m a n y sem an tic roles sig nifican t fo r a g ra m m a r o f a language. T h e ones he m en tio n s belong to tw o m ain groups: the Participatory roles (“ b o rn e by w h at one w ould th in k o f as ac tu al p artic ip a n ts in th e situ atio n im plied by the v erb ” , 1985: 68), and the Circumstantial roles (“ b o rn e by entities th a t d o n o t really p artic ip a te , b u t instead form p a rt o f the setting o f th e ev en t” , 1985: 69).
T h ere are tw o P articip ato ry roles - A gent (“A ” - vaguely ch aracterized as a p a rtic ip a n t w hich the m eaning o f the verb specifies as d o in g o r causing som ething, possibly in tentionally) and P atien t (“ P ” - a p a rtic ip a n t w hich the verb characterizes as having som ething h ap p e n to it, an d as being affected by w hat hap p en s to it). T h e C ircu m stantial roles include D irectio n al (Source - “ S” , and G oal - “ G ” ), E xperiencer (“ E ” ), R ecipient (“ R c ” ), In n e r L ocative (“ I L ” ), T hem e (“T h ” ), C auser (“ C ” ), In stru m e n ta l (“ I ” ), R easo n (“ R ” ), Benefactive (“ B” ), O u ter L ocative (“ O L ” ), C ircu m stan tial C o m itativ e (“ C C ” ), and T em p o ra l (“T ”).
T h e follow ing sentences provide exam ples o f the abo ve m en tio n ed roles [ A n d r e w s 1985: 70]:
(5) a. T iger s n a k e s ^ in h ab it A ustralian., b. G eo rg eA/Th w alked fro m /to the shores/G, c. IE love Lucy,
d. F re d e rik a c annoys m eE,
e. B ruceA handed D a rlen eR a steakTh, f. BillA pro d d ed the sn ak eP w ith a stickj, g. T h e ea rth c a ttra c ts the m o o n xh, h. T h e car-n, is expensive,
i. S usan cau g h t a lizard in the garden0L. j. Bruce barbecued a steak fo r D a rlen eB, k. A lvin shot up a sign fo r funR,
1. Shirley w ent diving w ith a sp earguncc, m . Jac k ate a sausage d u rin g the raceT.
A n d r e w s [1985: 70] stresses the fact th a t “ no presently k n o w n system o f sem antic roles can be com prehensively applied in a convincing m a n n e r” . N evertheless, various researchers have suggested th a t reference to th-ro le labels is involved in the d escription o f num ero u s linguistic p hen om en a: g ra m m a r o f a n a p h o ra [ J a c k e n d o f f 1972], th eo ry o f c o n tro l [ J a c k e n
-d o f f 1972], a-djectival passive fo rm a tio n [ W i l l i a m s 1981], m i-d-d le co ns tru c tio n s [ R o b e r t s 1987], etc.2
3. PROBLEMS WITH ГНЕТА ROLES
T h ere is, how ever, no consensus am o n g linguists o n the im p o rtan ce and co n ten ts o f th-roles, and som e researchers seem to diverge from explicit reference to th-role labels. T his tendency is m o tiv ated by the fact th a t there a p p e a r n o t to exist any clear criteria fo r determ in ing w hat th -ro les given arg u m en ts bear. F o r exam ple, H o e k s t r a [1984: 34] states th a t th e specific c o n te n t o f n o tio n s such as A gent, T hem e, etc., m ay be o f som e relevance fo r th e ultim ate sem antic rep resen tatio n , b u t n o t fo r the pu rp o ses o f sentence gram m ar. J a e g g l i [1986: 588] points to the frequent “indeterm inacy s u rro u n d in g the n a tu re o f the p a rtic u la r them atic role assigned to any p a rtic u la r arg u m e n t” , an d therefore he in tro du ces the follow ing sym bols to cover the th-roles (w ith o u t a ttrib u tin g to them any theoretical significance): (6) th-s - represents the th-ro le assigned to the subject o f a predicate,
th-d - represents the th-role assigned to the direct object o f a predicate, th-1 - represents the tra d itio n a l L o catio n,
th -g - represents the trad itio n al G o al, etc.
in fo rm a tio n provided in such th-grids is very lim ited w hen co m pared to earlier represen tatio n s (cf. (1) above):
(7) put: th-s, th -d, th-1 [J a e g g l i 1986], put: th-1, th-2, th-3 [ H o e k s t r a 1984].
M o re recently, R o z w a d o w s k a [1989] p ro poses a featu re-b ased a p p ro a ch to th-roles. A fte r analyzing derived nom inals, P olish im p erson al co n stru c tio n s, Polish reflexive verbs, and binding o f a n a p h o ra in experiential co n stru c tio n s, she com es to a conclusion th a t instead o f trea tin g th -roles as discrete u n decom posable atom ic w holes, it is m o re a p p ro p ria te to view them as bundles o f features, such as [ + / - sentient], [ + / - cause], and [ + / — change].
W ith features it is possible to acco u n t fo r a great deal o f o v erlap am o n g th-roles, how ever, even a very small set o f features can be com bined in such a way th a t it p ro duces definitions n o t co rresp o n d in g to an y attested roles. T his is also tru e a b o u t the system devised by R ozw ad o w sk a - o f the 9 possible feature com b in atio n s, tw o ( [ + sentient, - c a u s e , - c h a n g e ] and [ —sentient, + cause, + change]) seem no t to define any k no w n relations.
2 F o r a comprehensive discussion o f these issues see L e v i n [1985], R o z w a d o w s k a [1992], S t a l m a s z c z y k [1992].
Still o th er researchers - J a c k e n d o f f [1987], R a p p a p o r t , L a u g - h r e n a nd L e v i n [1988], an d Z u b i z a r r e t a [1987] - p resen t several argum en ts against any usage and reference to th e co n ten ts o f th-roles.
A nd so, J a c k e n d o f f [1987: 378] argues th a t som e concepts, th o u g h o f the sam e form al type as S ource or G o a l, d o n o t have any tra d itio n a l label, such as the role o f the object N P o f the verb pass in (8):
(8) Jo h n passed the house.
A lso the direct object N P s o f ju m p , approach, pierce in sentences (9) have n o stan d ard nam es fo r their th-roles [ J a c k e n d o f f 1987: 378]: (9) a. Jo h n jum p ed the gorge,
b. J o h n ap p ro ach e d H arry, c. T h e arro w pierced the target.
A n o th e r problem arises w ith argum ent N P s having m u ltip le th-ro les, such as the subject o f give which is an agent and a Source at the sam e tim e. T h e subject o f roll down the hill could be an A gent or a T hem e. A lso in sentences w ith verbs such as buy, sell, exchange, trade, tw o actio ns are going a t th e sam e tim e, and therefo re th e subject an d th e (p repo sitio n al) object N P s b ear tw o th-roles each. A s pointed ou t by J a c k e n d o f f [1987: 382] a n o th e r verb w ith m ultiple th-roles on each N P arg u m en t is chase. T his is so because fo r an action o f the form X chase Y to be tru e, at least three co n d itio n s m u st be satisfied (JackendofT s (21), from w hich it follow s th a t X has tw o roles and Y three:
(10) a. Y is in m o tio n ,
b. X m oves to w ard (or in p a th of) Y, c. X intends to go to (or catch) Y.
J a c k e n d o f f [1987: 382-383] also discusses cases w here m u ltip le N P s hold a single th-role, as illustrated below:
(11) a. The b o x has books in it,
b. The list includes m y nam e o n it.
In b o th (11a) and ( l i b ) tw o different N P s in the sam e sentence satisfy the sam e th-role. T o g eth er w ith cases o f argum ents w hich have m u ltiple th-ro les the sentences in (11) co n stitu te counterexam ples to b o th clauses o f the T h eta -C rite rio n as form ulated by C h o m s k y [1981: 36].
J a c k e n d o f P s [1987: 378-379] conclusion is clear: “ th em atic relatio ns are to be reduced to stru ctu ra l co nfigurations in con cep tual stru ctu re ; the nam es fo r them are ju st convenient m nem onics fo r particu larly p ro m in en t co n fig u ratio n s [...] the term s T hem e, A gent, and so on are n o t prim itives o f sem antic th eo ry ” 3.
3 Z u b i z a r r e t a [1987: 12] arrives a t a similar conclusion: “ [...] substantive notions like theme, patient, goal, experiencer have no grammatical im port: rules and principles o f gram m ar are never form ulated in term s o f these notions” .
Several o th er p ap ers clearly d em o n strate th a t in a n u m b er o f m o rp hological processes and syntactic altern atio n s a p p ro p ria te rules (or gen era lizations) are “ th -b lin d ” . L e v i n and R a p p a p o r t [1986] an d R a p - p a p o r t , L a u g h r e n a nd L e v i n [1988] discuss th e adjectival passive fo rm a tio n (A P F ). P revious accounts o f A P F [ex. W i l l i a m s 1981] used th e th -ro le T h em e to single o u t the arg u m e n t o f a verb w hich becom e th e external argum ent o f the related adjectival passive. H ow ever, as dem o n strated by L aughren, Levin and R a p p a p o rt, reference to this role is b o th unnecessary an d u n ten ab le, and it is possible an d desirable to recast the rule o f A P F as a rule w hich results in th e extern alization o f a single a rg u m e n t o f the base verb. T h e in tera ctio n o f various principles o f g ra m m a r ensures th a t the a p p ro p ria te arg u m e n t is externalized, w h a t is im p o rta n t how ever, is th a t the principles discrim inate betw een argum en ts in term s o f manner o f th -ro le assignm ent w ith o u t referring to th eir content. U n d e r this ac co u n t th e th-grid becom es an an n o ta te d list o f arg u m e n ts4:
(12) put: X < y , P z >
Sim ilar conclusions follow from the analysis o f non-agentive -er n o m inals [ L e v i n a nd R a p p a p o r t 1988], relations betw een m o rp h o lo g y an d sy n tax in D u tc h an d E nglish w o rd -fo rm a tio n [ B o o i j 1992], and th e d irectio n o f th -ro le assignm ent [ T r a v i s 1984]. In discussing such issues the term th -ro le is used as a synonym o f the term ‘a rg u m e n t’ a n d p a rtic u la r sem antic c o n te n t o f this arg u m en t is irrelevant fo r th e m o rp h o lo g ical or syn tactic processes.
A d d itio n ally , an analysis o f such p h en o m en a as the locative a lte rn a tio n , m id d le , u n ac cu sativ e an d in ch o ativ e c o n stru c tio n s, p o in ts to w a rd s th e existence o f deep sem antic processes, m o re general th a n tho se described by co n v en tio n al th -ro le s5.
4. THE MIDDLE CONSTRUCTION
A n in -d ep th analysis o f ju st one co n stru c tio n m ay brin g in terestin g results fo r the theory o f g ram m ar. In recent studies related to th e G o v e m - m ent-B inding fram ew ork one o f the constru ctio n s undergo ing com prehensive analyses is the m iddle construction. T h e m o st im p o rta n t studies include K e y s e r a n d R o e p e r [1984], F e l l b a u m [1986], F e l l b a u m a nd Z r i b i - H e r t z [1989], H a l e a nd K e y s e r [1986, 1987, 1988], R o b e r t s
4 Cf. W i l l i a m s [1981] for a discussion of argum ent types, and S t a l m a s z c z y k [1992] for mechanisms o f argum ent identification and various aspects o f lexico-syntactic representations fo r verbs.
[1987], Z u b i z a r r e t a [1987], F a g a n [1988, 1992], A c k e m a a n d S c h o - o r l e m m e r [1993] and H o e k s t r a a nd R o b e r t s [1993].
T h e co n stru c tio n in q u estio n is exem plified by th e follow ing sentences (ex. (a.-d) from F a g a n 1992: 247; (e.-h) from Q u i r k et al. 1985: 744)6: (13) a. G lass recycles.
b. (N a stu rtiu m ) D oes n o t tra n s p la n t well. c. U m brella-style fram e sets u p easily. d. C lear plastic d o o rs lift u p fo r access. e. H er b o o k s tra n sla te well.
f. T h e sentence reads clearly.
g. M y te a p o t pou rs w ith o u t spilling. h. T h e sheets w ashed easily.
I t is w o rth n otin g here, th a t there exists a huge discrepancy betw een the alm o st com plete lack o f interest in this co n stru c tio n in tra d itio n a l and university gram m ars (cf. the very b rief notes in Q u i r k , et al . [1985: 744], o r D o w n i n g and L o c k e [1992: 124]), and con sid erab le in terest w ithin various theoretical fram ew orks (early T ra n sfo rm atio n al G enerative G ra m m a r, G overnm ent-B inding T heory, Lexical F u n ctio n al G ra m m a r, W ord G ra m m a r, D ix o n ’s ‘G ra m m a r o n Sem antic Principles’, etc.).
In this p ap e r I present pro p erties o f this co n stru c tio n an d p ro b lem s connected w ith devising an a p p ro p ria te lexico-sem antic re p resen ta tio n fo r verbs entering the m iddle a lte rn a tio n 7.
5. BASIC PROPERTIES OF THE MIDDLE CONSTRUCTION
T h e M iddle C o n stru ctio n (M C ) is derived from basically tran sitiv e verbs, the verb has active m orphology, and the S-structure subject corresponds to the object o f tran sitiv e co n stru ctio n s (the gap in (14b, d ) represents this position). In this respect m iddles resem ble inchoatives (unaccusative verbs), cf. (15b, d):
(14) a. J o h n read the boo k.
b. T his b o o k reads ___ easily. c. M a ry ironed the clothes. d. T h e clothes iro n well.
6 In this paper I discuss only the English middle; for studies dealing with other languages see: A c k e m a and S c h o o r l e m m e r [1993], H o e k s t r a and R o b e r t s [1993] for D utch; F a g a n [1988, 1992] for G erm an; F e l l b a u m and Z r i b i - H e r t z [1989] for French; and Z u b i z a r r e t a [1987] for Romance.
1 The discussion o f properties (together with m ost examples and judgm ents) is based upon the papers m entioned above, especially the w ork o f Fellbaum and Fagan.
(15) a. H e b ro k e the cup. b. T h e cup b ro k e __ . c. T h e enem y sank the boat. d. T h e b o a t sank ___ .
T ra d itio n al gram m ars trea t this alternatio n as a subtype o f m orphological conversion [ Q u i r k et al . 1985: 1565]; in W ord G ra m m a r th e co n stru c tio n is analysed as an exam ple o f a w o rd -fo rm atio n relatio n [ R o s t a 1992: 327]; and acco rd in g to D i x o n [1991: 327] the m iddle is a m ark ed co n stru c tio n , representing th e process o f pro m o tio n -to -su b ject. O n the o th e r h a n d , early T G studies suggested a tra n sfo rm a tio n a l accou nt o f th e m iddle, w hereas Lexical F u n c tio n a l G ra m m a r [ex. В r e s n a n 1980: 115-116] in tro d u ced rules affecting g ram m atical fu n c tio n s8.
T h e p ro m o tio n o f the D -stru c tu re object (the in ternal, direct arg u m e n t in the sense o f W i l l i a m s 1981) m akes the process o f m id dle fo rm a tio n rem iniscent o f passivization: in b o th cases there is an im plied agent, lacking in the unaccusative co n stru c tio n s (cf. (15b, d)).
H ow ever, in the M C the m issing agent arg u m en t c a n n o t be lexically represented and there is n o possibility o f re-linking it, in co n tra st to the 6j>-phrasc o p tio n available fo r passives:
(16) a. T h is b o o k was read by Jo h n . b. * T h is bo o k read s well by John.
F ro m the above pro p erties it follows th a t a t th e level o f A rg u m en t S tru c tu re (i.e. th e lexico-sy ntactic re p re se n ta tio n ) verbs d isp lay in g th e m id d le a lte rn a tio n have th e follow ing re p re se n ta tio n (w here x is th e extern al, an d y the internal argum ent, cf. W i l l i a m s 1981):
(17) read - a. (transitive): V [x < y > ] , - b. (m iddle): V [ < y > ] .
A closer analysis o f the p ro m o ted direct object reveals th a t th ere exists a co n stra in t on the type o f objects which can becom e subjects in M C , as illu strated by the c o n tra st in (18):
(18) a. T h is poem tran slate s easily.
b. * T h is poem learns by h e a rt easily. c. Old cars sell easily
d. * Old cars buy easily.
T h e class o f transitive verbs w hich m ay a p p e a r in the M C is restricted to th o se re q u irin g an affccted in tern a l arg u m e n t (w ith som e p ossible exceptions, as for ex. the verb read). T he affectedness co n stra in t on M C is fu rth e r confirm ed by the fact th a t only one g ro u p o f th e psych-verbs can a p p e a r in this construction :
' For an overview o f some previous accounts o f the middle construction, cf. S t a l m a s z c z y k [1993].
(19) a. J o a n fears mice. b. * M ice fear easily. c. M ice terrify Jo an . d. J o a n terrifies easily.
U sing tra d itio n a l them atic relatio n s it m ay be said th a t th e M C is a T h em e -o rien ted co n stru c tio n : it states th a t any A g e n t ca n /m a y /w ill p erfo rm th e ac tio n , ta k in g u n d e r ac c o u n t som e in v a ria n t o r in h e re n t pro p erties o f the T h em e9.
6. ANALYSES OF THE MIDDLE CONSTRUCTION
A nalyses w ithin the GB fram ew ork consider the M C fo rm a tio n at tw o levels: syntactic an d pre-syntactic (or lexical)10. In K e y s e r and R o e p e r [1984] it is assum ed th a t m iddles are derived from their transitive co u n terp arts by m eans o f a lexical rule th a t absorbs the objective case an d th e subject th -ro le. In c o n tra st to unaccusatives, m iddles em erge from th e lexicon as transitives, an d a syntactic rule o f m o v e-а (in this case N P -m o v em en t) m oves the D -stru c tu re object in to th e S -structure subject p o sitio n.
T h e M C is derived th ro u g h a lexical rule; in fo rm u latin g this rule K eyser and R o ep er follow the stan d ard ap p ro ach es to R o m an ce languages and establish in the g ra m m a r o f E nglish an ab stra ct, phon olog ically null, reflexive clitic si w hich ab so rb s objective case an d th e subject th-role. H ow ever, as p ointed o u t by J a e g g l i [1986], it does n o t seem reaso n ab le to assum e the existence o f a null clitic on the one h an d , and th e existence o f a phonologically spelled o u t clitic on the o th er, while the po stu lated ab stra c t elem ent lacks relevant p ro p erties o f the R o m an ce reflexive se/si m o rphem e.
A m o re recent analysis is due to R o b e r t s [1987], w ho in tro d u ce s a m iddle fo rm a tio n rule o f the form “ E xternalize (Them e)” , ro u g h ly in th e sense o f W i l l i a m s [1981]. T his rule is an o p eratio n o n th-grid s, w here the A gent role is suppressed and the T hem e is ex tern alized 11:
(20) [Agent, Them e] — ► [(A gent), Theme}.
R ule (20) restricts the class o f verbs underg oin g m idd le fo rm a tio n to tran sitiv e verbs as it can only affect verbs w ith T hem es. R estrictin g the n o tio n o f T hem e, R o b erts can apply this rule to th e relevant class o f
9 In the case o f psych-verbs (ex. terrify) the construction is Experiencer-oriented and the unexpressed argum ent is a Percept.
10 Cf. the discussion in A c k e m a and S c h o o r l e m m e r [1993].
11 The following rule represents one sub-class of psych-verbs, (cf. (19) above): (i) [Percept, Experiencer] —» [(Percept), Experiencer].
tra n sitiv e s, i.e. th e ones w ith a n affected in tern a l a rg u m e n t (Affected Theme). A very im p o rta n t aspect o f rule (20) is th a t it m ak es th e claim th a t th e A g en t th -ro le is n o t elim inated, b u t only p revented fro m being assigned to th e external arg u m e n t (i.e. it is un p ro jected an d unlinked). Still o th e r p ro p e rtie s o f the M C p o in t to w ard s th e im p o rta n c e o f a m ore sem antically based ap p ro ach .
T h e M C receives a non-eventive, generic, h ab itu al o r p o ten tial in te r p re ta tio n , an d acco rd in g to K e y s e r and R o e p e r [1984] because o f this pro p e rty it is incom patible w ith the progressive, perfective p ast o r im perative (th o u g h ju d g m en ts vary, especially in the case o f (21b, c))12:
(21) a. T h is b o o k reads easily.
b. * T h is b o o k re ad s easily at the m o m ent. c. * T his bo o k read easily yesterday. d. * R ead easily, book!
Sentences (21b, c) contrast with the non-deviant unaccusative constructions: (22) a. T h e b o a t is sinking.
b. T h e b o a t sank yesterday.
Sentence (21a) can have the follow ing, generic, in terp re tatio n : (23) It is easy, for everyone, to read this book.
O ne recent a p p ro a c h explicitly dealing w ith the sem antic p ro p e rtie s o f m iddles is ad v o cated by S ara h F ag an . In h er recent w ork [ F a g a n 1988,
1992], she pro p o ses to tre a t m id dle fo rm atio n as an exam ple o f a general process o f genericization. G enericization is a process w hich assigns a gene ric in te rp re ta tio n to a th-ro le th a t is subsequently left unrealized (u n lin ked). U n derlying this conception is the n o tio n o f saturation o f th-roles developed by R i z z i [1986]. S a tu ra tio n is u n d ersto o d as an asso ciatio n o f a th-ro le w ith som e referential co n ten t - “ th a t is, w hen we can u n d ers ta n d ‘w ho does w h a t’ in the situation referred t o ” [ R i z z i 1986: 508]. Typically, the P rojection Principle and the T h -C rite rio n ensure th a t s a tu ra tio n is accom plished in th e syntax. R izzi, how ever, allow s fo r th e possibility th a t the P rojection Principle operates in the lexicon th ro u g h the rule o f arbitrary interpretation. A ccording to R i z z i [1986: 512] a rb it ra ry in te rp re ta tio n should be characterized by a collection o f featu res [ + h u m an , -fg en eric, + / — plural]. Such features are in h ere n t in certain n o m in al elem ents (G e rm an m an, F ren c h on, Ita lia n si, etc.) o r are assig ned th ro u g h (24):
(24) A ssign -f-arb to the d irect th-role.
12 Cf. different judgm ents in K e y s e r and R o e p e r [1984] and R o s t a [1992]; and the discussion in F a g a n [1992]. A ccording to D i x o n [1991: 326] ‘present’ is the m ost comm on choice but p ast tense is also possible.
T h e direct th -ro le is the direct arg u m en t th -role, i.e. the only role directly th -m ark ed by the verb, th-roles w hich are satu ra te d lexically are n o t realized in the syntax, and therefore Rizzi presents a re fo rm u latio n o f the P ro jection Principle [ R i z z i 1986: 509]13:
(25) C atego rial stru ctu re reflects lexically unsaturated th em atic stru ctu re a t all syntactic levels.
T h e P ro jection Principle as stated in (25) asserts th a t on ly u n sa tu ra te d arg u m en ts are accessible to syntactic in terp retatio n .
In su m m ary , th -ro les can be s a tu ra te d in the sy n tax th ro u g h th e sta n d a rd P rojection Principle, o r in th e lexicon - by virtue o f b o th (24) a n d (25). I f a th-ro le is satu ra te d already in the lexicon it never ap p e ars in the syntax, nevertheless it m ay be un d ersto o d because it still belongs to the lexical m ean in g o f the verb. T he tw o different ways o f sa tu ra tin g th-ro les m ay be observed in th e co n tra stin g b eh aviour o f verbs like eat and devour:
(26) a. F ra n k ate an en o rm o u s b u rger in the bar. b. F ra n k ate in the bar.
c. F ra n k devoured an eno rm o u s b urger in th e b ar. d. * F ra n k devoured in the bar.
In (26a, c) the P atien t th -ro le assigned by the verbs to the o bject is sa tu ra te d syntactically and therefore overtly projected in to th e syntactic stru c tu re o f th e sentence. In (26b) the th -ro le is n o t projected in to the syntax, bu t is satu ra te d lexically, pro v in g th a t there exist tw o o p tio n s fo r sa tu ra tio n in the case o f the verb eat; how ever, th e sem antically related verb devour requires syntactic sa tu ra tio n o f the relevant ro le (i.e. the arg u m e n t bearing the role m u st be explicit - cf. (26d)).
F a g a n in co rp o rates R izzi’s o b servation into h er w ork, an d suggests th a t th ere exist tw o rules responsible fo r m iddle fo rm a tio n [ F a g a n 1988: 198]14: (27) A ssign + a rb to the external th-role.
(28) E xternalize the direct th-role.
By rule (27) the external th-ro le o f m iddle verbs - usually, b u t n o t alw ays, an A gen t - is no longer associated w ith a stru ctu rally projected p o sitio n th o u g h it is still un d ersto o d (generically). R ule (28) acco u n ts for the fact th a t th e direct th-ro le o f the transitive verbs becom es the extern al a rg u m e n t of the detransitivized m iddle verb. T his rule bears som e sim ilarity to R o b e rts’ rule (20), it is, how ever, m o re ad e q u ate as it deals w ith
13 I he standard form ulation o f the (Extended) Principle is provided by C h o m s k y [1981: 29] and [1982: 10]: Projection Principle: (i) Representations at each syntactic level (i.e. LF, and D- and S- structure) are projected from the lexicon, in that they observe the subcategorization properties o f lexical items. // (ii) Every IP (S) m ust have a subject.
14 l a g a n [1992: 160-170] further elaborates these rules in order to provide an account for French and G erm an middles.
po sitio n s o f arg um ents in the A rg u m e n t S tru ctu re o f th e verb and n o t w ith the specific co n ten t o f the roles.
F a g a n assum es th a t by assigning the index arb to a role, it becom es lexically saturated, and therefore it will n o t be realized syntactically. U n der this ac co u n t m iddle fo rm a tio n is treated as an o p eratio n on th e A rg u m e n t S tructure o f a verb. A possible, negative, consequence o f such an ap p ro ach is the existence o f tw o external roles (argum ents): the original external argum ent, an d the externalized T hem e (i.e. the internal argum ent). O ne way o f avoiding this pro blem is to assum e th a t lexical sa tu ra tio n deletes th e p rim ary external arg u m e n t a t the lexico-syntactic level o f re p resen ta tio n (for ease o f p re se n ta tio n , in the rules below the assignm ent o f < + a r b > replaces the external argum ent). Schem atically, the deriv atio n o f a M C m ay be presented as the follow ing o p eratio n o n A rg u m en t S tructure (where: * - ex tern al arg u m e n t, y - direct argum ent, y x - externalized direct argum ent):
(29) 1. U nderived AS: V [x < y > ] ,
2. A ssign < + arb > to x: V [ < + a r b > < y > ] , 3. E xternalize y: V [y j,
4. D erived AS: V [у].
H ow ever explicit the above solution m ight seem, it does n o t ta k e u n d er a c co u n t all im p o rta n t (sem antic) p roperties o f the M C .
The subject o f the M C m u st have certain in h eren t q u alities th a t trigger the process d en oted by the v erb 15:
(30) a. T hese figures add u p to 1000. b. W hich apples bak e best? c. O ranges peel easily.
K eeping this property in m ind, we m ight further p arap h rase sentence (21a): (31) It is easy, fo r everyone, to read this b o o k because o f its certain
pro p erties (such as large p rin t, o r clear style, etc).
T h e M C requires the presence o f a m od ifier - ad verbial, n eg atio n , c o n trastiv e stress, em p h atic do, reflexive, e tc .16:
(32) a. T his novel reads quite well.
b. M o d ern fem inist lite ratu re sim ply d o e sn ’t read. c. G B p ap ers read like detective stories.
This property seems more explicit in languages which require reflexives in middle constructions, сГ. the G erm an and Polish equivalents o f (21a): (i) D as Buch liest sich leicht. / / (ii) T a książka czyta się łatwo.
16 Discussing this property, D i x o n [1991: 325] suggests th at “ prom otion to subject is possible when there is some m arker o f the success of the activity” . L a k o f f [1977: 251-252] discussing sentences in (i) argues that (i.a) is an instance o f an agent-focused construction, as opposed to patient-focused middles: (i) a. Rollce Royces drive themselves. // b. * Rollce Royces drive themselves easily.
However, the impossibility o f (i.a) follows rather from a general constraint on multi-adverbial m odification: (ii) a. * This book reads easily well. // b. * John reads easily well.
d. B u reau crats B R IB E . e. B ureau crats do bribe!
f. G o o d cars drive them selves.
T h e m o d ifier denotes th e q u ality o f th e process an d em phasizes th e generic, h ab itu al o r p otential in terp re tatio n . T he in h eren t p ro p erties o f th e M C subject are o ften stressed by the use o f th e verb will, especially in the negative form :
(33) a. T h e figures will n o t add. b. T his b o o k w o n ’t sell.
c. T h e suit-case w ould n o t lock.
A s o ften n oted [cf. F e l l b a u m 1986; F a g a n 1988] the n a tu re o f the m o d ificatio n , o r even its very presence, is connected w ith p ragm atically given in fo rm atio n , as attested by (34):
(34) a. T h is um brella folds up. b. T his dress b u tto n s. c. T h a t dress zips up. d. G lass recycles.
T h e above sentences d em o n strate th a t in som e cases it is n o t th e generic in te rp re ta tio n („people in general” ) w hich is m o st ch aracteristic o f the M C , b u t ra th e r th e specific q u alities o f the subject, o ften in te rp re te d an d prop erly u n d ersto o d because o f som e extralinguistic factors.
All o f the above m entio n ed appro ach es to th e M C ta k e fo r g ranted th e existence o f the A gent th-role (or external arg u m en t) in th e u n derlying rep resen ta tio n o f the m iddle verbs. In R o b e rts ’ ac co u n t this th -ro le is still p resen t afte r the ap p lica tio n o f rule (20), th o u g h it is u n p ro jected an d u nlinked. F a g a n , on the o th e r h an d , argues fo r a process o f genericization w hich leaves the syntactically unexpressed arg u m e n t u n d ersto o d in a generic sense.
7. M ID D L E C O N ST R U C TIO N S AND CO N C EPTU A L STR U C TU R E
Below I suggest a different approach, nam ely th a t there is no A gent th-role in the M C a t any level o f representation. T his is the p osition tak en by H a 1 e an d K e y s e r [1987] which I ad o p t and im plem ent w ith ideas stem m ing from the above discussed rules o f sa tu ra tio n an d genericization. W ithin th e fram e w o rk o f Conceptual Semantics, as developed by J a c k e n d о f f [1987, 1990], th e follow ing lexical re p resen tatio n s fo r the verb break can by pro vided : (35) J o h n b ro k e the cup,
AS: V [x < y > ] ,
(36) T h e cu p broke, AS: V [y],
LCS: [Y B E C O M E “ B R O K E N ”].
In C o n c e p tu a l S em antics th e lexical sem an tic re p re se n ta tio n w hich encodes the m eaning o f a verb is called the Lexical Conceptual Structure (LC S). LCS is a re p resen ta tio n o f the co ncept nam ed by the verb an d the p a rtic ip a n ts in the ac tio n (represented by variables); m ean in g is encoded th ro u g h pred icate decom position. Lexical C on ceptual S tru ctu re is com posed o u t o f elem ents fro m a u niversal set o f p rim itiv e fu n c tio n s a n d th e b ac k g ro u n d assu m p tio n is th a t a t som e level o f rep resen ta tio n the m eaning s o f verbs are n o t u n analyzab le entities. A ccording to P u s t e j o v s k y [1991] lexical d ecom position is possible if it is perform ed generatively, i.e. if generative devices c o n stru c t sem antic expressions.
C A U S E an d B E C O M E a re the prim itiv e fu n c tio n s o f C o n c e p tu a l S tru ctu re, X an d Y the argum ents, an d “ B R O K E N ” is an ab b rev iatio n fo r a m o re articu lated expression17. T h e a p p ro p ria te re p resen ta tio n fo r the m id d le v a ria n t o f break is identical w ith the one fo r th e un accu sative v ariant:
(37) C h in a cups b reak easily, AS: V [y],
LCS: [Y B E C O M E “ B R O K E N ”].
T h e o p tio n in (35) - a causative p red icate w ith an ag en t p a rtic ip a n t - is the only one w ith the external arg u m en t p resent a t th e deepest level o f re p resen tatio n . I claim here th a t no agent (external arg u m en t) is p resen t in the CS re p resen ta tio n fo r unaccusatives an d m iddles, an d th a t the re la tio n betw een dyad ic (35) and m o n ad ic (36) an d (37) is governed by a causative rule [cf. H a l e an d K e y s e r 1986: 19] w hich em beds the m o n a d ic CS as a com plem ent o f the general causative function:
(38) [X C A U S E (Y...)]
w here (Y ...) can be in terp reted as “ Y u n d erg o ch a n g e” . T h is CS (“ u n d erg o c h a n g e”) defines the crucial p ro p e rty o f verbs w hich allow th e m idd le v a ria n t an d p o in ts directly to the affectedness o f th e object. T h e rule responsible fo r m iddle fo rm a tio n (40) is an instance o f a m o re general ru le (39):
(39) [X P R E D IC A T E (Y...)] — ► [Y...],
(40) [X C A U S E (Y “ und erg o ch an g e” )] — ► [Y “ u n d erg o c h a n g e” ]. H a l e and K e y s e r [1987: 20] call this rule th e Ergative-Middle Alter nation, an d state th at: “ o n this view a m id dle does n o t differ in any interestin g sense from th e unaccusative m em ber in an erg ative a lte rn a tio n ” .
17 On the status o f primitive functions, cf. H a l e and K e y s e r [1987, 1988] and Z u b i z a r r e t a [1987].
C o m p a riso n o f th e re p re se n ta tio n s in (35)-{37), an d especially th e existence o f the co m m o n , em bedded, elem ent in all th ree re p resen ta tio n s, suggests a possible “ d ee p er” level o f re p resen ta tio n , em b o dy in g crucial aspects o f transitives, unaccusatives and m iddles. Such a deeper representation h as in fact been p roposed by G u e r s s e l [1986] an d ad o p ted in H a l e a nd K e y s e r [1987, 1988].
G u e r s s e l [1986: 69] states th a t “ a m o re basic level o f co n cep tu al stru ctu re , to be referred to as the Prim itive C o n cep tu al S tru ctu re (PC S), m u st be posited. PC S is intended to be m o re basic th an the LC S in its expression o f the m eaning o f a p redicate in th a t it sim ply expresses the n o tio n conveyed by a p redicate, and d oes n o t involve the re p resen ta tio n o f th e p a rtic ip a n ts in term s o f v a ria b le s” . G u erssel suggests th a t th e prim itive hu m an classification o f processes recognizes a class w hose realization is alw ays the result o f som e external force o r co n dition .
H a l e a nd K e y s e r [1987, 1988] follow G uerssePs ideas in assum ing th a t th e PC S (o r in their term inology, ad o p ted here, the Elementary Lexical Conceptual Structure - ELC S) is th e m o st elem ental re p re se n ta tio n o f L C S, a “ preling uistic” level, an d it is devoid o f linguistic elem ents o f L C S (such as the a rg u m e n t variables which project into syntax an d th e event p o sitio n , in th e sense o f H i g g i n b o t h a m 1985). L a te r on, how ever, th ey d o n o t discuss the “ p u re ” E L C S b u t ra th e r equip it w ith tw o d ifferent sets o f variables: unrestricted an d restricted. U nrestricted variables (o f th e fo rm X , Y , Z) represent the argum ents o f a predicator which are associated obligatorily w ith g ra m m a tic a l fu n c tio n s (G F s, as defined in C h o m s k y 1981) in Lexical S tru ctu re (i.e. A rg u m e n t S tructure). R estricted variables o n the o th e r h a n d , are only o p tionally projected as G F s in syntax. T h e re is one re stric ted v a ria b le o f im p o rta n c e fo r this d iscu ssio n - th e co n c e p tu a l category - Circumstance (“ C ” ).
A p ro p o se d E L C S fo r verbs w ith a transitive, u naccu sative an d m idd le v aria n t is given in (41a), w ith reading as in (41b):
(41) a. С — ► [Y “ undergo chan g e”],
b. Some circum stance “ C ” results in Y ’s change.
In th e system originally proposed by G uerssel th e PC S fo r a verb like break (and its n om inalizations) is sim ply (42):
(42) PCS: [B R EA K ].
A s a PCS it is intended to represent the n o tio n o f b re ak in g an d does n o t involve a form al re p resen ta tio n o f the p articip an ts. H ow ever, it seems possible and justified to introduce an abstract, unspecified variable (represented as “ Q ” ), realized at the lexico-sem antic level as an a rg u m e n t (variable), c o n sta n t (cf. th e case o f eat ), o r unprojected variab le (cf. th e case o f wash). T h e idea behind this n o tio n m ay be explained w ith a b rie f discussion o f the a p p ro p ria te sem antic re p resen tatio n s fo r verbs behaving like eat and
wash. A s has already been noted in the discussion o f th e co n tra st exem pli fied in (26), th e re exist tw o v a ria n ts o f th e verb eat', tra n sitiv e and in tra n sitiv e 18. I assum e here th a t they have different lexico-sem antic re p re sen tatio n s, w ith the intransitive v arian t co n tain in g a c o n sta n t arg u m e n t “ F O O D ” :
(43) a. eat: [X E A T Y],
b. eat: [X E A T “F O O D ”].
T h is co n stan t, unlike variables, an d co n stan ts in idiom s, is n o t projected from LC S o n to o th er levels o f re p resen tatio n and therefo re th e respective a rg u m e n t stru ctu re grids (i.e. m odified and restricted th-grids) h ave th e follow ing form :
(44) a', eat: V [x < y > ] , b'. eat: V [x].
T h e tw o v ariants o f eat have different LCS re p resen tatio n s (43a) and (43b), how ever, it seems reasonable to p o stu late the existence o f a m o re p rim itive level o f lexico-sem antic re p resen tatio n from w hich b o th LCS form s are derived (cf. G uerssel’s PCS). F o r this p u rp o se I p ro p o se the follow ing prim itive LC S fo r the verb eat:
(45) eat: [X E A T Q].
In (45) “ Q ” is an ab stra ct, unspecified variable, realized as “ Y ” in the tran sitiv e v aria n t, o r as “ F O O D ” in the intran sitiv e one. “ Y ” is fu rth e r projected o n to the syntactic rep resen tatio n (and ap p e ars as th e N P object o f the verb), “ F O O D ” on th e o th er h a n d , is n o t projected [cf. Z u b i z a r r e t a 1987: 10]:
(46) cat: [X E A T Q] — > a. [X E A T Y], AS: V [x < y > ], b. PC E A T “F O O D ”], AS: V [x].
T h e in tro d u c tio n o f co n stan ts and ab stra c t variables a t a p p ro p ria te levels o f lexical re p resen ta tio n o f verbs allow s fo r an elegant an d co m p rehensive tre a tm e n t o f verbal diathesis, as illu strated by th e follow ing lexico-sem antic re p resen ta tio n fo r verbs like dress, shave, wash, etc. V erbs o f this type a p p e a r in tran sitiv e constru ctio n s, co n stru c tio n s w ith a reflexive object, an d intran sitiv e constructions:
(47) a. M a ry w ashed the baby, b M ary w ashed herself. c. M ary w ashed.
18 As observed already by G r u b e r [1965], intransitive eat has a more restrictive m eaning than transitive eat : sentence (i) cannot be interpreted as (ii), i.e. it can only m ean th a t the baby ale food: (i) The baby ate. / / (ii) T he baby ate a piece o f chalk. Cf. also the discussion in Z u b i z a r r e t a [1987: 10].
T h e three v aria n ts have th e respective LCSs: (48) a. w ash : [X W A SH Y],
b. w ash : [X W A SH X], c. w ash : [X W A SH X].
In LC S (48b) there are tw o identical variables, the second being spelled o u t in th e syntax as an a p p ro p ria te reflexive. In (48c) th e variab le is u n projected, and so it does n o t a p p e ar in arg u m en t stru c tu re (and syntax). A unified entry fo r wash has th u s the follow ing form :
(49) w ash : [X W A SH Q] — > a. [X W A SH Y] AS: V [x < y > ], b. [X W A SH X] AS: V [x; < У ;> ], c. [X W A SH X] AS: V [x].
T h e ab stra c t variable is realized as a transitive object (47a), reflexive object (47b), o r as an unprojected variable, yielding the in tran sitiv e co n s tru c tio n (47c).
It is also possible to use the a b stra c t variable in co n stru c tin g the prim itive, elem ental CS fo r th e verb break:
(50) break: [Q C A U S E ( B E C O M E “ B R O K E N ” )].
I f th e variable “ Q ” is projected o n to the CS it is realized as an external a rg u m e n t o f C A U S E and yields the tran sitive co n fig u ratio n (35); next, this arg u m e n t is satu ra te d in syntax, in accordance w ith the P ro jectio n Principle. O n th e o th er h a n d , the variable “ Q ” m ay be sa tu ra te d alread y a t the deepest sem antic level (and th us unpro jected), yielding the u n accusative (36) an d m id dle (37) constru ctio n s. In o rd e r to explain the difference betw een these tw o co n stru c tio n s I rein tro d u ce th e featu re [ + / — arb], w here [ + a r b ] ab b rev iates all relevant features o f the M C : inh erent p ro p e rty , non-eventive, generic an d p o ten tial in terp re tatio n , etc. (cf. th e discussion in F a g a n 1988, 1992)19.
N o w th e d eriv a tio n o f transitive, m iddle and unaccu sativ e v aria n ts o f b re ak proceeds as below: (51) Transitive: E LC S: [Q C A U S E (Y B E C O M E “ B R O K E N ” )], LCS: [X C A U S E (Y B E C O M E “ B R O K E N ” )], AS: V [x (y)]. (S a tu ra tio n in syntax).
lg The m otivation for retaining this feature is similar to the one discussed in the context o f the them atic role agent in Z u b i z a r r e t a [1987: 12], where it is pointed out th a t this notion is “perhaps a rem nant o f a prim ary semantic category in early stages o f language acquisition” .
(52) Middle: ELC S: [Q C A U S E (Y B E C O M E “ B R O K E N ” )] — ► [X — < + arb > C A U S E (Y B E C O M E “ B R O K E N ” )], LCS: [Y B E C O M E “ B R O K E N ”], (S a tu ra tio n in LCS), AS: V [у]. (53) Unaccusative: E LC S: [Q C A U S E (Y B E C O M E “ B R O K E N ” )] — ► [X — < — arb > C A U S E (Y B E C O M E “ B R O K E N ” )], LCS: [Y B E C O M E “ B R O K E N ”], (S a tu ra tio n in LCS), AS: V
[y]-F o r a verb to en ter the m iddle altern atio n , the follow ing sem antic co n d itio n s m u st be fulfilled:
(54) 1. T h e E L C S has the form : [Q C A U S E (Y...)].
2. (Y ...) is interp reted as “Y u ndergo ch a n g e” ; this in te rp re ta tio n a cco u n ts fo r the affectedncss o f the object.
3. T h e ab stra c t variab le “ Q ” is realized as [ + a r b ] and it is sa tu ra te d in the LC S, yielding the m o n ad ic LC S [Y...].
T h e oblig ato ry presence o f th e C A U S E co m p o n en t in the un derlying re p resen ta tio n fo r the M C is confirm ed by the follow ing p aradigm :
(55) a. J o a n wiped the dishes. b. * T hese dishes w ipe easily. c. Jo a n wiped the dishes dry. d. T hese dishes wipe dry easily.
(55b) is ill-form ed because the a p p ro p ria te LCS fo r the tran sitiv e stru ctu re lacks the n o tio n C A U S E :
(56) [X “ W IP E ” Y],
H ow ev er, the c h a ra c te ristic fe a tu re o f re su ltativ e fo rm a tio n is th e em b ed d in g o f th e sim ple stru ctu re (56) into a C A U S E function:
(57) [X C A U S E (Y B E C O M E Z ) BY (X “ W IP E ” Y)].
S tru ctu re (57) com plies w ith the req uirem ents fo r m id dle fo rm a tio n and th ere fo re the resultative m iddle (55d) is w ell-form ed.
O ne m o re aspect o f the M C rem ains so far unexplained: the presence o f an ad v erb , or o th e r m odifying elem ent. Sentences (58) d e m o n stra te th a t the ad v erb in the M C is process-oriented, in c o n tra s t to the ag ent-o riented ad v erb in tran sitiv e sentences (59):
(58) a. T h is oven cleans easily. b. New cars sell easily.
c. T his bo o k reads effortlessly. (59) a. J o a n cleaned the oven easily.
b. D ealers sell new cars easily. c. J o h n reads b ooks effortlessly.
In sentences (58) the pro p erties o f the derived subject are responsible fo r the ease o f cleaning, selling, reading. It is therefo re possible to assum e th a t th e presence o f the m o d ifier is triggered by the feature [4 -arb ]. T h is m ay be an effect o f som e sem antic “ incom pleteness” forced by the featu re [ + a r b ] , Z u b i z a r r e t a [1987: 148] suggests a possibility o f ex plaining the oblig ato ry presence o f a m odifier in term s o f an in tera ctio n o f focus and A rg u m e n t S tru ctu re. W hen the v ariab le “ Q ” is realized as “X ” (i.e. a variable projected o n to the external arg u m e n t po sitio n ) th e presence o f a m o d ifier is o p tio n al, how ever, w hen it is realized as X — < + a r b > it requires a con cludin g elem ent: adverb, negation, co n trastiv e stress, etc. T h is co n clu d in g elem ent ap p ears in the CS after the “ Q ” variable is realized as X — < + arb > ; its presence m ay be captured by adding a m an n er co m p on ent to the CS:
(60) ELC S: [Q C A U S E (Y...)] — [ X - < + a r b > C A U S E (Y...)], LCS: [(Y...) (IN M A N N E R Z)].
A bove, I have provided the follow ing generic in te rp re ta tio n for sentence (21a) (repeated here as (61)):
(61) T his b o o k reads easily,
(62) It is easy, fo r everyone, to read this book.
T h e form alism s introduced above enable now p ro viding a m o re ac cu rate in te rp re ta tio n o f sentence (61):
(63) F o r X — < + a r b > , it is possible th a t X R E A D Y in m a n n e r Z (because o f inh eren t P R O P E R T Y o f Y).
In this LC S, the variable X is b ound by the o p e ra to r X — < + a r b > and th erefo re only the variable Y can be projected on the level o f AS. In ad d itio n , the LC S also specifies the obligatory presence o f a m o difying elem ent. T echnically, this requ irem en t is realized by creatin g a place in the A S to be sa tu ra te d by an adjunct. If the m odifying elem ent is an adverbial it ap p ears as an oblig ato ry ad ju n ct in the AS, an d in syntax it is C h o m s- k y-adjoin cd to projections o f V(erb). T h e ac co u n t o f n eg atio n m ig h t be sim ilar. C o n trastiv e stress is m o re p ro b lem atic as we have to tak e u n d er c o n sid e ra tio n th e level o f P h o n etic F o rm . H ow ever, in th e m o d e l o f g ram m ar presented by J a c k e n d о f f [1987, 1990], there exist correspondence rules which link phonological, syntactic, and con cep tu al levels.
T h e co ncluding elem ent o f the M C m odifies the process described by the m iddle verb, and it seems plausible to suggest th a t th e sem antic (o r in som e cases even p ra g m a tic ) in te ra c tio n betw een th e m o d ifier an d th e predicate results in the sense o f agency felt to be presen t in the M C and attrib u te d in the previous accounts to an im plicit A g en t role.
REFERENCES
A c k e r n a, P. and Schoorlemmer, M . (1993) “The Middle Construction and the Syntax-Semantics Interface” , О Т S Working Papers, U trecht, June 1993.
A n d r e w s , A. (1985) ‘T h e m ajor functions of the noun phrase” . In T. S h o p e n , (ed.) Language Typology and Syntactic Description, vol. 1, Cambridge: Cam bridge University Press, 62-154.
B o o i j , G . (1992) “ M orphology, Semantics and A rgum ent Structure” . In I. R o c a , (ed.) Thematic Structure. Its Role in Grammar. Berlin and New Y ork: Foris Publications, 45-62. B r e s n a n , J. (1980) “ Polyadicity: P art I o f a T heory of Lexical Rules and R epresentations” .
In T. H o e k s t r a , H. v a n d e r H u l s t , and M. M o o r t g a r t , (eds) Lexical Grammar, D ordrecht: Foris, 97-121.
C h o m s k y , N. (1981) Lectures on Government and Binding. D ordrecht: Foris.
C h o m s k y , N . (1982) Som e Concepts and Consequences o f the Theory o f Government and Binding. Cambridge, M ass.: M IT Press.
D i x o n , R. M . W. (1991) A New Approach to English Grammar, On Semantic Principles. O xford: Oxford University Press.
D o w n i n g , A. and L o c k e , P. (1992) A University Course in English Grammar. New York: Prentice Hall International.
F a g a n , S. (1988) ‘T h e English M iddle” . Linguistic Inquiry 19: 181-203.
F a g a n , S. (1992) The Syn ta x and Semantics o f Middle Constructions. Cambridge: Cam bridge University Press.
F e l l b a u m , C. (1986) On the Middle Construction in English. Bloom ington: IU L C .
F e l l b a u m , C. and Z r i b i - H e r t z , A. (1989) The Middle Construction in French and English: A Comparative Study o f its Syntax and Semantics. Bloomington: Indiana University Linguistics Club Publications.
F i l l m o r e , C. J. (1968) “ The Case for C ase". In E. B a c h , and R. H a r m s , (eds) Universals in Linguistic Theory. New Y ork: H olt, R inehart & W inston, 1-88.
G r u b e r , J. (1965/1976) Lexical Structures in Syn ta x and Semantics. A m sterdam : N o rth H olland (published version o f M IT doctoral dissertation, 1965).
G u e r s s e l , M . (1986) “ On Berber Verbs o f Change: A Study o f Transitivity A lternations” . Lexicon Project Working Papers 9, Cambridge, M ass.: Center for Cognitive Science, M IT Press.
H a l e , K. and K e y s e r , J. (1986) “ Some Transitivity A lternations in English” . Lexicon Project Working Papers 7, Cambridge, M ass.: Center for Cognitive Science, M IT Press. H a l e , K. and K e y s e r , J. (1987) “ A View from the M iddle” . Lexicon Project Working
Papers 10, Cambridge, Mass.: Center for Cognitive Studies, M IT Press.
H a l e , K. and K e y s e r , J. (1988) “Explaining and Constraining the English M iddle” . In C. T e n n y , (ed.) Studies in Generative Approaches to Aspect, Lexicon Project Working Papers 24. Cambridge, Mass.: Center for Cognitive Studies, M IT Press, 41-57.
H i g g i n b o t h a m , J. (1985) “O n Semantics” . Linguistic Inquiry 16: 547-593.
H o e k s t r a , T . (1984) Transitivity. Grammatical Relations in Government-Binding Theory. D ordrecht: Foris.
H o e k s t r a , T. and R o b e r t s , I. (1993) “ M iddle Constructions in D utch and English” . In E. R e u 1 a n d and W. A b r a h a m (eds) Knowledge and Language. Volume II: Lexical and Conceptual Structure. D ordrecht: Kluwer Academic, 183-221.
J a c k e n d o f f , R. (1972) Semantic Interpretation in Generative Grammar. Cam bridge, M ass.: M IT Press.
J a c k e n d o f f , R. (1987) “T he Status of Them atic Relations in Linguistic T heory” . Linguistic Inquiry 18: 369-411.
J a c k e n d o f f , R. (1990) Semantic Structures. Cambridge, M ass.: M IT Press. J a e g g l i , O. (1986) “ Passive” . Linguistic Inquiry 17: 587-622.
K e y s e r , J. and R o e p e r , T. (1984) “ O n the M iddle and Ergative C onstructions in English” . Linguistic Inquiry 15: 381-416.
L a k o f f , G . (1977) “Linguistic Gestalts” . Papers from the Chicago Linguistic Society 13: 225-285. L e v i n , B. (1985) “ Lexical Semantics in Review: An Introduction” . In B. L e v i n (ed.) Lexical
Semantics in Review, Lexicon Project Working Papers 1. Cambridge, M ass.: Center for Cognitive Science, M IT Press, 1-62.
L e v i n , B. and R a p p a p o r t , M. (1986) “The Form ation o f Adjectival Passives” . Linguistic Inquiry 17, 623-661.
L e v i n , B. and R a p p a p o r t , M . (1988) “ N on-event -er Nominals: A Probe into A rgum ent Structure” Linguistics 26: 1067-1083.
P u s t e j o v s k y , J. (1991) “The Generative Lexicon". Computational Linguistics 17: 409-441. Q u i r k , R. et al . (1985) A Comprehensive Grammar o f the English Language. London and
New Y ork: Longman.
R a p p a p o r t , M. , L e v i n , B. and L a u g h r e n , M . (1988) “ Niveux de représentation lexicale” . Lexique 7: 13-32.
R i z z i , L. (1986) “ Null Objects in Italian and the Theory of pro". Linguistic Inquiry 17: 501-557. R o b e r t s , I. (1987) The Representation o f Implicit and Dethematized Subjects. D ordrecht: Foris. R o s t a , A. (1992) “ English M ediopassives” . Working Papers in Linguistics 4, London:
University College, 327-351.
R o z w a d o w s k a , В. (1989) “ A re Them atic Relations Discrete?” In R. C o r r i g a n , F. E c k m a n and M. N o o n a n (eds) Linguistic Categorization. A m sterdam : John Benjamins, 115-130.
R o z w a d o w s k a , В. (1992) Thematic Constraints on Selected Constructions in English and Polish, Wroclaw: W ydawnictwo Uniwersytetu W rocławskiego.
S t a l m a s z c z y k , P. (1992) The Structure o f Thematic Relations in English, Łódź: E K O R N O . S l a l m a s z c z y k , P. (1993) “The English M iddle C onstruction and Lexical Sem antics” ,
Papers and Studies in Contrastive Linguistics 27: 133-147.
S t о w e 11, T. (1981) Origins o f Phrase Structure. Cambridge, M ass.: M IT Press (doctoral dissertation).
I r a v i s , L. (1984) Parameters and Effects o f W ord Order Variation. Cam bridge, M ass.: M IT Press (doctoral dissertation).
W i l l i a m s , E. (1981) “Argument Structure and M orphology” . The Linguistic Review 1: 81-114. Z u b i z a r r e t a , M .-L. (1987) Levels o f Representation in the Lexicon and in the Syntax.
D ordrecht: Foris.
P iotr S ta lm a szczyk
T E M A T Y C Z N A I K O N C E PT U A L N A SEM A N TY K A A N G IE LSK IC H CZA SO W N IK Ó W M E D IA L N Y C H
Celem au to ra jest omówienie wzajemnych związków semantyki i składni we współczesnej gram atyce generatywnej (zwłaszcza modelu Chom sky’ego). Pierwsza część artykułu zawiera analizę różnych teorii ról tematycznych i problemów związanych z wyszczególnieniem wszystkich typów ról w teoriach wypracowanych przez Fillm ore’a, G rubera i Jackendoffa, a stosowanych przez gram atyków gencratywnych.
D ruga część artykułu poświęcona jest analizie jednej konstrukcji - angielskiej stronie medialnej (np. This book reads easily - Ta książka czyta się łatwo) i czasowników w niej występujących, tzw. middle verbs. Po przedstawieniu właściwości składniowych i sem antycznych tej konstrukcji konieczne staje się zaproponow anie odpow iednich poziom ów rcpezentacji leksykalnych dla middle verbs.
Analiza zaproponow ana w artykule zakłada, że na żadnym poziom ie reprezentacji w k o n strukcjach m edialnych nie występuje rola tem atyczna agensa. Właściwe rozw iązanie polega na zaproponow aniu elementarnej reprezentacji leksykalno-semantycznej, z której m ożna derywować poszczególne konstrukcje, w tym również stronę medialną. Sem antyka konceptualna leżąca u podstaw przyjętej w artykule analizy w znacznym stopniu nawiązuje do propozycji R aya JackendolTa.