Event paths, conflation, argument structure and VP shells 1 Andrew McIntyre I discuss the semantics and syntax of a phenomenon often called 'lexical subordination' and here called 'conflation', in which a VP with a single verb expresses both an activity and a result predication, although only the former is licensed by the verb's permanent lexical entry. Alongside standard cases such as resultatives and particle verbs, I discuss what I call ‘event path structures’ in English and German. In these, an activity is argued to conflate with a predication expressing a (sometimes metaphoric) ‘path’ of the activity. These little-known data challenge many argument structure theories because the path expressions sometimes disallow the linking of the verb’s normal object. They require a theory where the lexical verb does not contribute arguments in conflation structures, a conclusion motivated empirically. I present a theory of syntax-semantics mapping which relies on VP shells with meaningful light verbs which constrain the interpretations of their specifiers and complements. Conflation is treated as compounding of a verb root to one of the light verbs. This accounts simply for the argument-structural patterns of resultatives and event path structures. None of the argumentation appeals to operations at a lexical-semantic level between conceptual structure and syntax. 1 Introduction This study is partly concerned with a type of VP seen in (1) which, I argue, contains a predication expressing an EVENT PATH, the temporal or spatial 'path' of a situation. Discussion of this is overdue because most theories of the syntax-semantics interface are ill-equipped to accommodate cases like (1b-d) which are ATRANSITIVE in the sense that direct objects normally selected by the verb cannot be linked. (1) a. I read through the book ('I read and the reading went through the book') b. I rang (*the number) through to her ('I rang the number and the call went through to her') c. I saw (*them) into the window ('I saw them and the seeing extended into the window') d. I played (*the guitar) on ('I played the guitar and the playing went on') These structures will be analysed as instances of a parameterised phenomenon often called 'lexical subordination' (Levin & Rapoport 1988) but here called 'conflation', roughly following Talmy (1985). In this phenomenon, a VP with a single verb stem expresses both an activity named by the verb, and a subevent, usually represented by a secondary predicate, expressing a change of state or position. One instance of this is a resultative construction like eat onself sick, which conflates an eating event with a becoming-sick event. We will be concerned as much with conflation in general as with event paths, since I believe that neither phenomenon can be understood in isolation from the other. We proceed as follows. Section 2 introduces the constructions standardly assumed to involve conflation and assumptions about their semantics. I assume two types of subevent conjunction, &cause (for causal relationships) and &contemp (for two events perceived as a single event). I illustrate the relevance of the distinction for unaccusative and transitive resultatives (dance into the house vs. dance oneself to death). Section 3 gives a semantic analysis of English and German data involving event paths, arguing for the paraphrases in terms of conflation in (1). I record most of the relevant data I have found, in a bid to demonstrate the pervasiveness of this little-known type of VP and to aid further research on it. Section 4 gives an account of the argument structure of conflation VPs and of other facets of their syntax-semantics mapping. The atransitivity in (1b-d) can be explained if we assume that these structures involve conflation and that verbs cannot link their arguments in conflation VPs. Section 4.1 supports the latter claim empirically, mainly with little-known data where unaccusative and obligatorily transitive verbs fail to link their arguments in resultative and particle constructions. Sections 4.2-4.4 propose a theory which explains the absence of verbal argument linking in conflation structures. Section 4.2 introduces two types of light verbs which are part of the explanation, INIT, which has a causative function, and CHANGE, which mediates predication of a path or result predicate over a DP. Section 4.3 proposes that the conflation VPs are licensed by a mechanism called mconflation, in which a verb root is compounded with a light verb. M-conflation is constrained such that the light verb and verb root must refer to the same event. This is shown to give attractive accounts for problems in syntax-semantics mapping in resultatives, e.g. the presence/absence of reflexives in dance oneself to death and dance into the house, the appearance of unselected arguments in cases like the metal edge tore off the blind and the ambiguity in cases like Ann walked Jo out, where either Ann or Jo could walk. Sections 4.4 and 4.5 apply the theory to atransitivity and discuss potential alternative accounts. Section 5 gives some conclusions and addresses problems for the analysis.
2 The semantics of conflation In the constructions in (2), seen by Levin & Rapoport (1988) as instances of lexical subordination (in my terms: conflation), a causing event is represented by the agent and the verb stem, while the result event is represented by other material in the VP, including result predicates and direct objects, notably objects which are UNSELECTED, i.e. not subcategorised by the verb, compare work a debt off with *work a debt.) (3) gives examples for the semantic representations (SRs) I assume for these constructions. (2) a. Resultatives: shout someone deaf, wrestle people to the floor, knock a vase flying b. (Some) particle verbs: work a debt off, vote a party in, edit out a passage c. Effected object verbs: dig/drill a hole, shoot/carve/stare holes in the wall d. The way-construction: lie one's way into the building, fight one's way out (3) a. Dave ate himself sick: DO(DAVE,EAT) &cause BECOME(SICK(DAVE)) b. Dave scratched a sticker off: DO(DAVE,SCRATCH) &cause GO(STICKER,TO(NOT(ON([Thing Ø]))) c. Dave drilled a hole: DO(DAVE,DRILL) &cause BECOME(EXIST(HOLE)) d. Dave shot his way to Texas: DO (DAVE,SHOOT) &cause GO(DAVE'S WAY, TO TEXAS) The decompositions in (3) ignore matters irrelevant here like quantification and tense. Causation (notated as &cause here) is not seen as an agent-event relation (e.g. Jackendoff 1990) but as a relation between two events (e.g. Dowty 1979, Wunderlich 1997b). Other notational choices follow Jackendoff (1990). I use his GO function, which relates a moving object to successive parts of a path. Some would reject GO in favour of BECOME. However, BECOME is problematic for PPs like around the city, along the wall, southwards where the theme reaches no goal. We need GO for at least these expressions, and I will treat all spatial transitions uniformally. See also Jackendoff’s (1990:93f) defense of GO. Like Dowty (1979) and unlike Jackendoff, I use the standard BECOME operator with adjectival state change, e.g. (3a). Some constructions in (3) need comment. Firstly, in particle verbs (also called 'phrasal verbs', 'separable verbs', 'verb-particle combinations') like (2b) and (3b), the particle differs from its related transitive preposition (cf. scratch the sticker off the wall) in that the ground2 is defocussed, remains implicit and must be identified via recourse to the context or encyclopaedic knowledge. The symbol Ø indicates that the entity is defocussed. I leave open whether this purely conceptual notion suffices to predict the entity's syntactic non-realisation. It would suffice if particles have silent pronominal complements (Svenonius 1996). If not, we need extra stipulations, not formulated here, about possible argument structures of prepositions. The way construction in (2d), studied in Goldberg (1995), Israel (1996), Jackendoff (1992), Marantz (1992), Mateu (2000) is provisionally formalised in (3d). I treat the x's way part of the construction as iconic of a conceptual theme argument, as in the analyses of Goldberg, Marantz and Mateu. I do not discuss the nature of the righthand conjunct here, since I only appeal to the assumption that the subject's action is seen as causing the motion. (3d), beside the causal reading where Dave reached Texas because he shot those who tried to stop him, has a marginal non-causal reading in which shooting was a salient concomitant of the motion. The latter reading may be a jocular use of a causal structure in a non-causal context. If not, then &cause should give way to a second type of conjunct, &contemp, which expresses that the conjoined subevents are contemporaneous and spatially indistinct, and are perceived as the same event. The conjunctions used here are inspired by Kaufmann (1995a,b,c) and Wunderlich (1997a,b), although these writers leave the difference between &contemp and &cause up to inference. Single-DP resultatives like (4), which I call 'unaccusative resultatives', are often seen as causal (e.g. Levin/Rapoport 1988, van Valin 1990:224). On this view, (4c) means 'Ethel got into the studio by dancing'. However, I follow e.g. Rappaport Hovav & Levin (2001), Kaufmann (1995a,c) and Kaufmann & Wunderlich (1998) in rejecting this causal analysis in favour of one in terms of contemporaneous conjunction. The sound emissions in (b) accompany the motion rather than causing it, so an &cause analysis makes far less sense than an &contemp analysis in which the subevents are indistinguishable because they occur at the same place and time. (4) a. I strolled into the house, the box broke open, the toast burned to a crisp, the tires wore thin b. a motorbike roared down the highway, a bee buzzed past c. Ethel danced into the studio: DO (ETHEL,DANCE) &contemp GO (ETHEL, TO IN STUDIO) Let us explicitly compare &cause and &contemp. The unaccusative structure in (4c) portrays the dancing and the traversal of a path into the studio as the same happening. If this condition is not met, we require transitive structures like (5). Here the entry into the record books or the trance can take place
1
2
after the dancing, so that there can be no contemporaneous conjunction. Rappaport Hovav & Levin (2001) offer similar examples. I explain why the &cause/&contemp distinction correlates with the unaccusative/transitive distinction in resultatives in section 4.3. (5) a. Ethel danced her way into the record books: DO (ETHEL,DANCE) &cause GO(ETHEL’S WAY,TO IN RB) b. Ethel danced herself into a trance: DO (ETHEL,DANCE) &cause GO (ETHEL, TO IN TRANCE) Despite what my notational format may imply, I understand my SRs as simplified models of pure conceptual structure, not as belonging to a lexical-semantic level of grammar (sometimes called Semantic Form or Lexical Conceptual Structure) which interfaces between conceptual and syntactic structure and encodes grammatically relevant aspects of meaning. While the extra level simplifies the resolution of mismatches between conceptual structure and syntax, I will explore a more parsimonious and restrictive theory which dispenses with this level. For instance, conflation is often taken to be the addition of result subevents to verb meanings in a lexical-semantic operation, be it lexical subordination (Levin/Rapoport 1988, Legendre 1997:84, Spencer/Zaretskaya 1998), event composition (Pustejovsky 1991), template augmentation (Rappaport Hovav & Levin 1998) or lexical adjunction (Wunderlich 1997a,b). Languages (e.g. Romance) lacking conflation phenomena lack the relevant lexical operation in such theories. Rejecting a lexical-semantic level leaves me unable to appeal to this reasoning. In section 4.3, I pursue an alternative in which the ability of a language like English to express in a VP with a single verb stem the (by hypothesis universally available) complex event structures such as those in the SRs in (3-5), and the impossibility of this in Romance, reduces to some properties of the syntax of the languages. 3 Event paths This section argues that conflation can license structures like (6), event path structures. A VP with a single verb stem can express a situation s in which a subevent s' represented by a verb and its external argument is conjoined with a subevent s'' specifying the ‘path’ followed by s'. The theme of s'' is an item coindexed with s'. Direct objects are only allowed if they are part s'', a fact explained in section 4. (6) [EVENT ]i &contemp [EVENT GO ([EVENT]i,[PATH ])] 3.1 Atransitive particle uses The particle uses in (7) are atransitive, i.e. are incompatible with the direct objects normally selected by the simplex verb. My (2001) study gives more data and an analysis superseded by the present one. (From (7) onwards, all non-English examples are German, unless otherwise marked.) (7) a. read (*notes) ON; fight (*one’s battles/enemies) ON b. sing (*a song) ALONG; read (*the text) ALONG c. play (*a silly game) AROUND; hammer (*the metal) AROUND d. (*ein Lied) LOSsingen; (*Fußball) LOSsspielen (a song) off-sing; (football) off-play ‘start singing a song'; ‘start playing football’ e. kick OFF *(with) a new football; hit OFF *(with) a new golfball f. (*spekulaas) DOOReten; (*de brief) doorschrijven [Dutch] (gingerbread) through-eat; (the letter) through-write 'continue eating gingerbread'; 'continue writing the letter' Atransitive particles are mainly durative or ingressive, leading one to ask if atransitivity is connected to Aktionsart. Stiebels (1996:64f) and van Hout (2000) pursue this course. The idea is that the telic interpretation of verbs with bounded (quantised) incremental theme objects would clash with the durative or ingressive semantics of the particle. This is refuted by cases like (8) where the particles block objects which do not yield telic events. Moreover, sections 3.2 and 3.4 show that atransitivity occurs with PPs in non-durative VPs, e.g. shoot (*a bird) into a tree. Thus, I bypass accounts of atransitivity which connect it to the particles’ effects on Aktionsart. (8) a. She played (*her guitar) on/around/along. (cf. She played her guitar for an hour.) b. They fought (*their enemies) on. (cf. They fought their enemies for a day.) My account of atransitivity in section 4.4 will use the idea that the particles express a metaphorical path followed by an event through time. As this analysis is not intuitively obvious, I discuss the particles in some detail, in each case describing firstly the spatial sense of the particle which is claimed to be the metaphoric basis for the atransitive particle use. Beside the atransitive, so-called 'aspectual' use of on in (7a) and (9b), there is a spatial use seen in (9a), where a theme follows a path characterisable as EXTENDED AND UNINTERRUPTED. This spatial
use must be distinguished from the 'aspectual' use. The spatial use allows transitivity, cf. (9a), as is normal with paths predicating over DPs. An aspectual semantics for on is implausible in stative structures like Further on down the road was a church or The path goes on into the woods. Moreover, each sense has its own synonym: onwards matches only the spatial sense and on and on only the aspectual sense. (9) a. We prodded/moved/passed them on; We marched/walked on b. She worked/talked/typed/danced on unperturbed; It continued/droned on c. The shouting went on all night The atransitive use of on in (7a) and (9b) has a durative effect, characterising the event as continuing for a longish time. The description EXTENDED AND UNINTERRUPTED which applies to the relevant spatial reading of on also applies here, except what is extended and uninterrupted is the course of the event rather than the path of an entity. I assume that the intuitive notion of an event’s course is conceptualisable as a path followed by the event through time. The proposed conceptualisation gains support from (9c), where an event nominal is the theme in a metaphorical motion event. In my analysis of atransitive on, formalised below, a sentence like I read on is taken to be a conflation structure whose interpretation is ‘I read, and this reading went on’. The analysis of on may carry over to Dutch door in (7f) (cf. footnote 3), and to German weiter, literally 'further', on which see section 5. German los (cognate with loose, cf. losbrechen 'break loose') has a spatial sense roughly translatable with 'off, away' (losfahren 'drive off, depart', losstiefeln 'walk off'). Stiebels (1996:93) subsumes this sense under the so-called 'inchoative' sense in (7d), but there are cases where los is clearly a path predicated of an entity, not an inchoative marker. For instance, Briefe losschicken means ‘send off letters', not 'start sending letters', and losschicken is transitive, unlike in the inchoative use in (7d). Also, losschwimmen 'swim off' is unaccusative, forming its perfect with sein 'be', patterning with other cases where schwimmen combines with a directional expression, cf. (10b,c). Without a directional PP, schwimmen can take the haben 'have' perfect when expressing an activity rather than directed motion, cf. (10a). If los lacked a spatial sense and were always an inchoative marker, we would not expect losschwimmen to disallow the have perfect. (On the ability of path expressions to license be perfect with manner of motion verbs, see e.g. Abraham 1993, Kaufmann 1995b, Keller & Sorace 2003, van Valin 1990.)3 Finally, los has a privilege otherwise unique to directional expressions in allowing the elliptical motion verb construction in (10d), on which see van Riemsdijk (2002). (10) a. wir haben geschwommen we have swum b. wir {sind/*haben} losgeschwommen we {are/have} off.swum c. wir {sind/*haben} zum Ufer geschwommen we {are/have} to.the bank swum d. ich muß {los/runter/weg/zum Arzt} I must {off/down/away/to.the doctor} 'I must go {off/down/away/to the doctor}' e. das Geheule ging los the sobbing went off 'the sobbing started’ I conclude that there is a purely spatial use of los. I define it as A PATH WHICH IS THE INITIAL BOUND OF A LARGER PATH. This definition captures the intuition that gives initial plausibility to an analysis of los with motion verbs as an inchoative marker. We now turn to the non-spatial, atransitive inchoative use of los in lostippen 'start typing', loswählen 'start dialing' and (7d). My claim is that the atransitive use still expresses A PATH WHICH IS THE INITIAL BOUND OF A LARGER PATH, but the path in atransitive uses is metaphorical, proceeds through time rather than space, and has an event instead of an entity as theme. The ‘inchoative’ effect of the particle follows from the fact that the path of the event (in common parlance: its ‘course’) is the initial part of a larger path (to wit, the course of a larger event). Analogous to go on in (9c), losgehen 'go off, start' allows event nominal themes, cf. (10e). These points lead to the claim that a construction like wir arbeiten los ‘we start working’ conjoins with &contemp the situations expressed by wir arbeiten ‘we work’ and die Arbeit geht los ‘the work starts’ (lit. ‘goes off’). A benefit of this localistic approach is that the differences between the spatial and inchoative senses are captured without having to assume homonymy. The punctuality of los verbs (Stiebels 1996:93, fn. 15) is also explained because the agentive subevent cannot otherwise be in an &contemp relationship with the event's metaphorical movement along the minimal path expressed by los.
3
4
The analysis of los applies to the off examples in (7e). The relevant spatial meaning is seen in walk off, carry the goods off. The atransitive use of the particle is unproductive, however. The only other examples I found are the golf term tee off 'start playing golf' and the discussion started off, where the particle arguably has the same semantics as in hit off but predicates over an overt event nominal. One spatial reading of around expresses a path of entities which is characterised as lacking an intended goal (cf. walk around, and transitives like push cars around). I suggest that the atransitive use of around (cf. (7c) and fiddle/joke/experiment around) indicates that the course of an event metaphorically lacks a goal ('gets nowhere', so to speak), whence the intuition that around is a verb diminutive which portrays an event as aimless, unplanned, ineffectual, etc. The connection between the spatial and atransitive senses is clear from the fact that we find both senses in the non-cognate items about (push cars about vs. play about, bash about on a piano) and German (he)rum, which far outmeasures around in productivity. Beside spatial cases like rumschieben 'push around', we find atransitives like rumflicken 'do minor repairs', rumkritisieren 'criticise pettily', rumdiskutieren 'discuss (in place of action)', ruminstallieren ‘install useless programmes’, rumstipulieren 'stipulate rather than explain', rumerklären 'explain away unconvincingly', rumformulieren ‘do minor formulating’ (from an obligatorily transitive verb). The fact that the historically unrelated items around, about and (he)rum have all developed both senses must be explained. We can achieve this if we accept that the goalless path denoted by around/about/-herum has come to allow metaphoric construal, licensing events as well as entities as theme.4 More evidence for the metaphor posited for atransitive around comes from (11). With motion verbs, hin und her means 'to and fro, back and forth'. In (11), it suggests that the active role passes from one participant to another, and disallows objects. Duden (1996:709) observes that hin und her may connote that the event does not reach a conclusion or result. This 'going-nowhere' aspect is also suggested by around/herum; indeed, herum can often replace hin und her with little difference in meaning. Under localistic assumptions, the parallels are unsurprising inasmuch as the spatial uses of hin und her and around both express a goalless path. Note also the overt event nominal subject predication in (11c), cf. (9c), (10e). (11) a. sie diskutierten/redeten/debatiertierten hin und her they discussed/talked/debated to and fro b. solange hin und her gemordet wird... [Spiegel 8.10.01, p. 56] as.long.as to and fro murdered gets 'as long as there is all this senseless killing from both sides' c. die Diskussion ging hin und her the discussion went to and fro 'the discussion passed from one participant to another or went nowhere, in circles' Even along in (7b) can be analysed localistically. The PP in walk along the wall is a path parallel to an elongated object. Particle uses like I brought the book along or she came along describe a path which is parallel to another path (in bring along, the latter path is that of the 'bringer', in come along, it is the path of other people moving to the same goal; see my 2001 study on such uses). In the atransitive use in (7b), along also expresses parallelness, but it subsists between two events and is temporal, not spatial. As we see from parallel lectures, temporal parallelness is contemporaneousness. Thus, Ann played guitar and Fred sang along means simply that Fred's singing was parallel to (contemporaneous with) a contextually present event, viz. the guitar playing. This localistic view explains why along apparently has a comitative meaning ('with others'). An analysis denying the spatial basis of this particle use forces one to posit homonymous alongs and cannot explain why the German translational equivalent, mit 'with', a genuinely comitative item, is not atransitive (cf. einige Lieder mitsingen 'sing (*some songs) along'), while the path analysis of along will later give us an explanation. I now propose more explicit SRs for atransitive particle verbs. (12) summarises my approach. The claim is that these structures involve conflation. An activity (represented by a verb stem and its agent) conflates with a predication of a path over an element coindexed with the activity. (12) Atransitive pv's: [Event ]i &contemp [Event GO ([Event]i , [])] All we need do now is fill in the activity and the path positions. SRs for the particles discussed are suggested in (13a-d). They only aim to capture the main aspects shared by both the spatial and the atransitive readings of the particles.5 Using GO ahead of BECOME complicates the statement of selection restrictions that particles impose on themes. I resort to the licensing condition in (13e) as a way stating that these particle uses can take either an entity or an event as theme. I suggested that the
paths expressed by along and los are to be defined in terms of another path or event. Recall from section 2 that Ø is for syntactically unexpressed, defocussed entities identified by pragmatic construal. (13) a. around: [Path -INTENDED GOAL] b. along: [Path PARALLEL TO ([event/path Ø])] c. on: [Path EXTENDED] d. los/off: [Path INITIAL PART OF ([path Ø])] e. for the above paths, [Path...] is licensed in the context [Event GO ([THING/EVENT],[Path...]) A concrete SR is (14). The SR glosses as ‘Fred talked and this talking went on (i.e. continued)’. (14) Fred talked on [Event DO(FRED,TALK)]i &contemp [Event GO([EVENT ]i, [Path EXTENDED])] A reviewer queried the legitimacy of positing two coindexed contemporaneous subevents in (12) and (14). But (15) shows that an event e can conjoin with another event giving the course or direction of e. I do not see why it should be problematic to conjoin SUBevents in the same way. (15) a. We rehearsed, and the rehearsal went on into the early hours b. Wir diskutierten, und die Diskussion ging hin und her We discussed, and the discussion went back and forth (cf. (11)). The reviewer also claimed that it is 'not at all clear that subevent decomposition is a legitimate way to encode aspectual operators' and that the use of a GO predication (crucial in explaing atransitivity in section 4.4) is stipulative because there is 'no real motion'. While my evidence for the subevents and metaphoric motion is perhaps less compelling than suggestive, readers sceptical toward my analysis (whether or not as part of a general aversion to the localism of e.g. Jackendoff and Cognitive Grammar) now have the burden of proof. My analysis explains the existence of the particle senses in (7) simply: it is due to widening the set of possible themes of the particle and metaphoric construal, hardly implausible notions6. Denying this in favour of the notion 'aspectual operator' (a notion applied in no detailed study of Germanic particles, which does scant justice to along and around and which problematically equates particles with progressive markers) forces one to treat the particles' atransitive and spatial senses as accidental homonyms, which ought to arouse suspicion given that the same 'accident' occurs with several particles in several languages. Also, sections 3.2 and 3.4 discuss cases of atransitivity where GO is less controversial. For instance, 3.4 notes cases where the theme is not an event but an implicit entity. Thus, in shoot (*a bird) into a tree, it is a bullet which is the theme of GO. My account will identify a common symptom in VPs suffering from atransitivity (namely that they involve a conflation-licensed GO predication with an implicit theme) which will later allow us to explain atransitivity, while it is unclear if the aspectual operator analysis can offer any explanation beyond the account refuted by (8). 3.2 Atransitive prepositional phrases There are also full PPs which express event paths and are incompatible with objects. To understand the data, we must firstly note two stative uses of directional PPs. One is (16), where PPs express imaginary paths projected from the intrinsic frontal region of entities, formalised in Jackendoff (1983:172f; 1990:92) with the ORIENT function as in (16a). (16) a. the window faces onto the street: [STATE ORIENT([WINDOW],[PATH TO ON STREET])] b. the arrow points to the centre, the gun is aimed at me c. Das Fenster geht nach Norden raus the window goes to north out 'the window faces northwards' In (17), the path is coextensive with a stationary, elongated theme to which directedness is imputed. Such cases are described in Matsumoto (1996) and Jackendoff (1983:173f; 1990:44, 92-94), whom I follow in using the EXT (=EXTEND) function as in (17a). (17) a. the road goes to Rome: [STATE EXT([ROAD],[PATH TO ROME])] b. the road extends/leads/cuts/runs through the forest, the bridge crosses the river, the beam sticks/juts out of the wall, the rope dangles into the water Now consider a directed perception verb like sehen 'see'. (18a) shows the complementary distribution of direct objects and PPs giving the direction of the subject's gaze. I see two possible semantic integrations of the PP. (18b) says that the PP is an event path (more exactly: a state path), in which case see into the window means that the seeing event is oriented into the window. (c) uses EXT, the gloss being 'Anne saw, and her line of vision extended into the window'. (c) involves not an event path but the path of an abstract entity. Either analysis will suffice for my explanation in section 4 for the atransitivity. (18) a. Anne sah (*einen Mann) ins Fenster Anne saw (*a man) into.the window
5
6
b. [STATESEE (ANNE,MAN)]i &contemp [STATE ORIENT ([STATE]i, TO IN WINDOW))] c. [STATESEE (ANNE,MAN)] &contemp [STATE EXT ([LINE_OF_VISION], TO IN WINDOW))] d. Anne sah Gesichter in die Wolke Anne saw faces into.the cloud Sebastian Löbner (p.c.) pointed out (18d), where a directional PP is compatible with a direct object. However, this is a resultative construction where Anne is said to cause faces to enter the cloud by (deliberately) distorted perception. That speakers can assert result predications they know to be illusory is confirmed by (19). (19) a. He read arcane messages into the text that simply weren't there. b. In the kitchen, the dope heads smoked their way to some far-flung corner of the cosmos. c. The system does not 'hallucinate' arbitrary meanings into an expression. (Levin & Rapoport 1988:281) Atransitivity is also found with PPs with 'telephone verbs', cf. (20-21). Direct objects and path PPs cooccur only if the path is a result predicate on the object (as in (20d), where the details go through to the secretary due to the phone call) but not when the PP gives the direction of the phone call, as in (20c) and (21). (20) a. Stan rang {the secretary/the number}. b. Stan rang through (with the details). c. Stan rang {*the secretary/*the number} through. d. Stan rang the details through (to the secretary). (21) a. You can't call (*people) out on this phone: it's for internal calls. b. I phoned/faxed/dialed (*the number) through/in. I formalise the resultative pattern in (22d), and the atransitive structure in (22a-c). (b) glosses as 'Stan rang, and the call went through'. DOPUNCTUAL marks ring as a punctual agentive verb (the 'semelfactives' of Smith 1991, e.g. cough, knock). To me, he rang for ten minutes is not good, showing that ring is not durative. For speakers who disagree, the correct SR would be (c), glossing as 'Stan rang and the call extended through'. (22) a. Stan rang (*the secretary) through b. [EVENT DOPUNCTUAL (STAN,RING(SECRETARY))]i &contemp [EVENT GO ([EVENT]i, THROUGH)] c. [EVENT DO (STAN,RING(SECRETARY))]i &contemp [STATE EXT ([EVENT]i, THROUGH)] d. Stan rang the details through [EVENT DOPUNCTUAL (STAN,RING)] &cause [EVENT GO (DETAILS, [Path THROUGH])] Another verb where objects and directional PPs are in complementary distribution is German greifen 'grasp, try to gain manual control of something', cf. (23). (23d,e) are two suggestions on what the directional PP predicates of. (d) says that the grasping event moves in some direction, and (e) that the theme of the PP is the agent’s hand, which is syntactically unexpressed. Again, either analysis will do for my purposes. (23) a. Er griff einen Akkord. He gripped a chord 'He played a chord (on a piano, guitar)' b. Er griff in die Klaviertasten He gripped into the piano keys He played the piano c. *Er griff einen Akkord in die Klaviertasten. Intended: ‘he played a chord on the piano’ d. greifen + PP: [EVENT DO (X,GRASP(Y))]i &contemp [EVENT GO ([EVENT]i, [Path ])] e. greifen + PP: [EVENT DO (X,GRASP(Y))] &contemp [EVENT GO ([HAND_OF_X], [PATH])] Similar is reach in (24). In (c), the path does not predicate over the object, and unacceptabilty results. (d) is a resultative where the object is a theme over which the path predicates. (24) a. I can't reach up to the top shelf. b. I can't reach [DPthe bottle on the top shelf]. c. I can't reach (*the bottle) up to the top shelf. d. I reached my hand up to the top shelf. (25) lists further cases of complementary distribution of subcategorised objects and directional PPs. (a) and (b) are metaphorical EXT paths giving the extent of the government/ binding relationships. (c) seems to be a temporal EXT path: the remembering extends back in time. (d) and (e) seem to be a
7
different type of structure (here irrelevant) where the object is impossible because the structures are unaccusative resultatives where the PP expresses Mary's entry into the family/company. (25) a. Verbs govern (*noun phrases) into prepositional phrases after P-incorporation. b. NPs in subordinate clauses don't bind (*anaphors) into matrix clauses. c. I remember (*John) back to my youth. d. Mary married (*her second husband) into a happy family. e. Mary bought (*a lot of shares) into a big company. Finally, consider the cicumposition an x vorbei 'past x' in (26a). In (26b-e) we see yet another instance of an event path reading of a directional PP. The sense of the literal gloss 'past us' is that the activity ignores our interests. The same metaphor is seen in the discussion bypassed us. As usual, direct objects normally associated with the verbs are disallowed in the event path reading, cf. (26b-d). (e) is an event nominal theme use parallel to (9c), (10e) and (11c). (26) a. Sie schob die Karre an uns vorbei 'she pushed the cart past us' b. Sie reden (*Unsinn/*Englisch) an uns vorbei 'they talk (nonsense/English) past us' c. Sie planen (*die Aktivitäten) an uns vorbei 'they plan (the activities) past us' d. Sie diskutieren (*die Probleme) an uns vorbei 'they discuss (the problems) past us' e. Die Diskussion ging an uns vorbei 'the discussion went past us' 3.3 Event path particles where objects are conceptual grounds of particles The event path VPs seen thus far were atransitive. I also argue that there are transitive event path VPs. In these, the conceptual ground argument of a particle is somehow 'promoted' to object, a phenomenon not investigated here, but independently attested in (27). See McIntyre (2003) for descriptive remarks on these little-known constructions, which remind one of locative alternations and may involve a passive-like operation acting on a preposition. (27) wipe the table off (=wipe dust off the table), pour the bucket out (=pour water out (of the bucket), squeeze the orange out, pump the cellar out, run someone through (with a sword) Stiebels (1996:162-165) discusses the German particle verbs in (28). The complex verbs are obligatorily transitive, but most simplexes are intransitive. Comparing these with their English glosses suggests that the objects of the an verbs are conceptual grounds of the particle which are promoted to object, instead of being Case-marked complements inside a PP as in the English glosses. (28) anlügen ‘lie to’, anmotzen ‘whinge to’, anschweigen ‘be silent to’, anschreien 'scream at', anstarren ‘stare at’, anbellen ‘bark at’, anzweifeln ‘have doubts about’, anhusten ‘cough in the face of’, anstaunen 'marvel at', anfauchen 'hiss at', anstrahlen 'shine on' Stiebels proposes the SR in (29a), where the situation argument is asserted to have a direction. This is one of the few previous analyses which is in the spirit of the event path concept. My analysis is (29b). (29) assumes that an event is the theme of the orientation predication in (28). Evidence that such a conceptualisation is possible comes from her laughter/gaze/cough was directed at him. (29) a. anlachen ‘laugh at’: λu λx λs [LAUGH(x)(s) & DIRECTED_TOWARDS(s,u)] b. anlachen/laugh at: [EVENT DO(X,LAUGH)]i &contemp [STATE ORIENT([EVENT]i,[PATH TO(U)])] I also assume an event path analysis for the so-called 'aspectual' use of through in (30). I briefly sketch the analysis here, saving a fuller exposition for another study. (30)
read a book through, play a song through, think a problem through
(31) is my analysis. The SR glosses as 'Joe thought and this thinking metaphorically went through the issue'. That through implies that the event encompasses the whole of the object is epiphenomenal of so-called holistic readings which apply to direct objects (e.g. Anderson 1971, Tenny 1994). (31) a. Joe thought the issue through [EVENT DO(JOE,THINK)]i &contemp [EVENT GO ([EVENT ]i,[PATH THROUGH (ISSUE)])] b. [PATH THROUGH (X)] denotes a path inside X. The assumption that the direct object of 'aspectual' through verbs is the conceptual ground of the particle which has somehow been promoted to direct object allows us to capture semantic parallels with the prepositional verbs in (32a). These show the same atransitivity effects as other event path 8
expressions, cf. (32b). I suggest that the particle verbs and the prepositional verbs both have the type of SR illustrated in (31). However, the holistic effect applies only to the particle verbs, since these have the ground in direct object position. Consequently, we find telicity differences like those in (32c). Additionally, we capture the intuition that reading through a book is less thorough than reading a book through. Although the former could exhibit the bounded reading of through in the sense that the reading encompasses the beginning and end of the book, it is compatible with skim-reading or leaving out some sections because there is no holistic effect to ensure that the whole of the book is involved. (32) a. read/look/glance/work quickly [PP through the book], think/talk [PP through the matter] b. She read (*the important passages) through the book. c. I thought through the issue {for/in} an hour. vs. I thought the issue through {in/*for} an hour. 3.4 Event paths or paths of entities? The idea that there are path expressions predicating over items coindexed with an agentive event will now be compared with an alternative. Concerning (33a)=(14), J. Mateu (p.c.) asked why the path expressed by atransitive particles could not be a predication over an agent, as in (33b). Likewise, (34) contrasts my analysis of read through with an analysis implied by Jackendoff's (1996:332f) suggestion that expressions like read a book are conceptualised as paths followed by an agent through an entity. (33) Fred talked on a. [Event DO(FRED,TALK)]i &contemp [Event GO([EVENT]i, [Path EXTENDED])] 'Fred talked and the talking went on' b. [Event DO(FRED,TALK)] &contemp [Event GO([FRED], [Path EXTENDED])] 'Fred went on talking' (34) Fred read through the book a. [Event DO(FRED,READ)]i &contemp [Event GO([EVENT ]i, [Path THROUGH BOOK])] ‘Fred read, and this reading went through the book’ b. [Event DO(FRED, READ)] &contemp [Event GO([FRED], [Path THROUGH BOOK])] ‘Fred read and he went through the book’ Mateu's and Jackendoff's suggestions merit consideration, for constructions like I got up to the last chapter and the gloss in (34b) (and that in (33b), unless go on is a raising verb here) suggest that an agent's progress can be conceptualised as his/her position. However, the agent predication analyses in (33b) and (34b), being parallel to spatial transit expressions like (4c), predict unaccusativity, but none of the German structures analysed as event paths in section 3 shows the be auxiliary found with unaccusative structures where a path predicates over a unique DP argument. Also, we find impersonal passivisation peculiar to unergatives with event paths, but not paths of entities: (35) a. hier wird nur rumgespielt here is only around.played 'people are just playing around here' b. ??hier wird nur rumgelaufen here is only around.run 'people are just running around here' There are constructions superficially similar to event path structures, but whose grammar suggests that they involve agent-oriented path predication, cf. (36a). Dispensing with event paths in favour of agent paths fails to capture the difference between (36a) and (36b). (36) a. ich arbeite mich durch das Buch durch I work myself through the book through 'I work my way through the book' b. ich arbeite das Buch durch I work the book through 'I work through the book’ Thus, the Mateu-Jackendoff hypothesis cannot replace the event path metaphor. This said, I do not claim that all unergative directional PP structures involve event paths. The PPs in (37) presumably express paths of the implicit entities given in brackets. (18c) and (23e) also assume paths of implicit entities. However, other cases studied above (e.g. play around, eat on, work through a book, look at someone...) are better analysable in terms of event paths than paths of implicit entities. (37) a. He shot (*a bird) into the tree. [a bullet] b. I sawed (*the wood) through the door. [a saw]. c. She spoke/breathed into the microphone. [words/air]
d. I was waving around. [arms] The explanations given later for the complementary distribution between directional expressions and direct objects will apply irrespective of whether an event or an implicit entity is theme of the path. What is important in the analysis below is that these PPs/particles express paths and that direct objects normally associated with the verb are unlinkable because they are not the themes of these paths. 3.5 Summary of the argument-structural effects of event paths I argued that there are conflation structures where a situation s conflates with a predication expressing either the path or direction of s (section 3.1-3.3) or the path of an implicit entity (3.4). The data in 3.1 and 3.3 included (but were not confined to) a subset of the particles sometimes pretheoretically called 'aspectual', as in play on and read books through. These are not accidental homonyms of other senses of the particle, but arise from metaphoric extension of a spatial sense of a particle so that events as well as entities are possible themes. Let us summarise the argument-structural behaviour of the constructions. Particles/PPs expressing event paths (or paths of implicit entities) disallow the linking of direct objects associated with the verb unless such DPs are conceptually the ground of a particle which is promoted to direct object (section 3.3). If a prepositional element has no overt ground, or if the ground is realised as a complement of P, the structure is atransitive. Table 1 clarifies this. Subject Verb Direct Object Particle/PP Ground I ate * on I saw * into the house =complement of P I read the book through =direct object Table 1. Event paths and argument linking 4 Argument structure and conflation To explain the argument-structural facts just summarised, I will propose a general theory of conflation, proceeding as follows. Section 4.1 argues that direct arguments in transitive resultative and particle constructions are not arguments of the verb, and more generally that conflation bans the linking of a verb's arguments. This generalisation, which predicts atransitivity, is itself explained in the theory of the syntax of conflation VPs presented in sections 4.2-4.3. The theory is applied to atransitive VPs in section 4.4. 4.1 Against inheritance of verbal arguments in conflation structures Many writers (Carrier/Randall 1992, Haider 1997, Neelman/Weerman 1993, Winkler 1997) maintain that, in resultative constructions and particle verbs, direct objects conforming to the verb's object selection (e.g. nails in (38)) are arguments of the verb, not just of AP/PP. I term this the hypothesis of INHERITANCE of verbal arguments. (38) hammer nails flat, hammer nails into the wall, hammer in nails If inheritance is impossible in conflation VPs, we would have the makings of an explanation for atransitivity. We would expect conflation VPs to be transitive exactly when the secondary predicate has an external DP argument. We would expect objects in (38) because the AP/PP has a theme argument, and in read the book through because the particle's ground is not realised as complement of through, and must leave the PP in search of case. We would predict atransitivity with read (*notes) on and see (*people) into the window because the prepositions license either no DP argument or a single DP argument realised PP-internally. To pursue this account for atransitivity, we need an independent demonstration -undertaken now- that the inheritance hypothesis is wrong. One argument for the inheritance hypothesis is based on the claim that unselected objects (defined in section 2) do not occur in resultative constructions with obligatorily transitive verbs, suggesting that the verbs' object selection properties are preserved in resultative constructions; unselected objects are assumed to occur only with intransitive (variants of) verbs (I ate vs. I ate myself sick). Carrier & Randall (1992:187; henceforth C&R) support this with data like (39). (a) shows that frighten is obligatorily transitive, and (b) and (c) suggest that result predications respect the verb's object selection, which is taken to show that the object is indeed an argument of the verb. The open question is whether the three verbs with which C&R support their claim are representative. If they are, then the inheritance hypothesis would be convincing. If not, it could be dismissed with little further ado.7 (39) a. The bears frightened *(the hikers). b. The bears frightened the hikers speechless. c. *The bears frightened the campground empty.
9
10
The resultatives and particle verbs in (40) (and more German examples in Kaufman & Wunderlich 1998:20) show that C&R's data do not generalise. (b) and (c) refute their analysis of frighten. To save space, indications of the verbs' obligatory transitivity and the unselected status of the object are omitted from (b) onwards; they can be constructed analogous to the bracketed material in (a). (40) a. Adrian lit/rolled up (a cigarette) vs. Adrian lit/rolled *(a cigarette). b. They frightened/scared/bored the hell out of me c. He did not draw music out of his players; he frightened it out of them. [from a description of an intimidatory conductor, found by web search under www.google.com] d. I shut/locked him in the cellar; I beat the crumbs out of the blanket; I wrenched the photograph out of his hand; I poured the bucket out; I ripped the machine out of its packaging e. Ich stellte die Garage voll I stood the garage full 'I stood things in the garage, filling it up’ Furthermore, C&R's empirical claim, if extended to particle verbs, cannot make sense of optionally transitive particle verbs based on obligatorily transitive verbs: (41) a. Adrian lit/rolled up (a cigarette) vs. Adrian lit/rolled *(a cigarette). b. Sally locked/tidied up (the house) vs. Sally locked/tidied *(the house). The correlation of obligatory transitivity and resistance to unselected objects is undermined by wayconstructions based on obligatorily transitive verbs, easily attestable by web search, cf. (42). Clearly, the unselected object x’s way ignores verbs' selection restrictions. (42) a. Cooper frightened his way into the hearts of defiant adolescents [www.worldguitarist.com/rock/nl6.html] b. They {bribed/beat/ground/defined} their way out of the situation The DP in an intransitive structure like I ran is an obligatory argument. As with objects in transitive resultatives, one could argue either that the DP argument in unaccusative resultatives like I ran into the house is shared by the verb and PP (e.g. Kaufman & Wunderlich 1998) or that it is solely an argument of the PP (e.g. Hoekstra & Mulder 1990:4). But (43) shows that even unaccusative resultative structures can exhibit unselected arguments, contra Kaufmann (1995a:144). Note incidentally that (43f,g) and (40g,i,j) are examples of the type seen in (27) where a particle’s ground is realised at the expense of the theme. (43) a. the door blew shut [*the door blew] b. he broke into the house [*he broke] c. the hammer head broke off [the hammer, not the hammer head broke] d. the water emptied out of the bathtub [*the water emptied] e. das Fenster wächst zu [*das Fenster wächst] the window grows to/obscured 'the window is becoming overgrown' f. die Wanne fließt schlecht ab [*die Wanne fließt] the bathtub flows badly away 'the bathtub empties badly' g. die Tasse schwappt über [*die Tasse schwappt] the cup splashes over 'the cup overflows, liquid splashes out of the cup' There is also a stative parallel of unaccusative resultatives where a position verb conflates with a locational predication in such a way that the position verb's argument structure is violated: (44) a. die Wand hängt voller Bilder [*die Wand hängt] the wall hangs full.of pictures [STATE HANG(PICTURES)] &contemp [STATE BE(WALL, FULL OF PICTURES)] b. der Keller steht halbvoll [*der Keller steht] the cellar stands half.full 'the cellar is half full (of furniture, large objects)' c. der Fußboden lag zu mit Papier [*der Fußboden lag] the floor lay full with paper 'there was paper lying all over the floor' The conclusion from the data in this section is (45a). From it follow the corollaries (b) and (c). (45) a. Obligatory direct arguments (i.e. direct objects, unaccusative subjects) normally selected by verbs are not obligatory when result predicates and particles are present.
b. A verb's internal argument selection requirements are inactive when conflation takes place. c. Direct arguments in conflation constructions are not arguments of the verb, but are arguments of a predicate in the subevent introduced by conflation. The theses in (45b,c) have precedents in the literature, including all studies which see direct arguments in resultatives and particle verbs as being initially generated in a small clause or projection of a particle/ result predicate (e.g. den Dikken 1995, Hoekstra 1988, Kayne 1985, Zeller 2001a). I anticipate two objections to (45b,c). The first is that it does not predict cases where V DP XP entails V DP, e.g. that (46a) suggests hammering the metal. Kayne (1985:122) refuted this objection by noting that (46b) also entails hammering the metal. In (b), this is an inference based on world knowledge about the connection between hammering and metal becoming flat. There is no reason why this reasoning should not apply to (a) as well. Hoekstra (1988:117f) gives related arguments. An example showing the inferential strategies at work is (47), from a real conversation about retrieving a ball stuck in a tree. Here the speaker became aware of the default interpretation, heavily primed by the stereotype of a ball as a thrown object, and provided extra context favouring the correct interpretation ahead of the default one. These examples speak for the claim of Kaufmann & Wunderlich (1998:19) that 'Formally, the direct object of the construction is the argument introduced by resultative formation, which is pragmatically identified with the argument of the verb'. (46) a. They hammered the metal flat. b. A little more hammering should get the metal flat. (47) Ich hab den Ball runtergeworfen, also mit nem Stock runtergeworfen I have the ball down.thrown, to.be.exact with a stick down.thrown 'I got the ball down by throwing a stick at it' A second challenge to (45b,c) was raised by a reviewer, who claimed that the status of the particles in (48a) as predicates over DPs is refuted by (48b). This reasoning -call it the COPULA ARGUMENT- is often levelled against small clause theories of particle verbs (e.g. Booij 2002, Dehé 2002, Jackendoff 2002, Stiebels 1996). The copula argument presupposes that compatibility with a copula is a necessary condition for secondary predicates, a premise falsified by the following considerations. (48) a. I drank up the beer; She read the article over b. *The beer is up; *The article is over Firstly, most directional PPs and continuous state change comparatives (i.e. inherently eventive PPs/APs) are incompatible with copulas, cf. (49), although they uncontroversially predicate over DPs. The copula argument is thus not compelling with respect to (48a) without a demonstration that up and over are not inherently eventive like the PPs/AP in (49a). (49) a. I walked to the station/ around (the house); I got colder and colder b. *I am {to the station/ around (the house)/ colder and colder} Secondly, consider the metaphoric resultatives in (50a,c). The object is an argument of the PP/AP (not the verb: *drink someone/the cellar), but the PP/AP cannot be a copula predicate (cf. (b,d), where * applies to readings parallel to (a,c)). (50) a. Fritz could drink/talk anyone under the table. (i.e. 'outdrink/outtalk') b. *Don't try to outdrink/outtalk Fritz: you'll be under the table in no time. c. We drank the cellar dry; I drained the boiler dry; Rapacious imperialists bleed us dry. d. *The cellar/boiler is dry; *We are totally dry due to rapacious imperialists. The senses of dry and under the table in (50a,c) ('empty of liquid', 'no longer able to compete') are only found in conflation structures and only with certain verbs. This applies in equal measure to many particles. Stiebels (1996), Zeller (2001b) and McIntyre (2002) note that the appearance of semiproductivity found with many readings of particles is epiphenomenal to the fact that they only appear in verb-particle combinations and that the senses of a particle usually constrain the (classes of) verbs which may appear in the construction. Given that a copula has little in common with other verbs, it is unsurprising that it often fails to appear with verb particles, including those in (48a) The copula+particle constructions we do find are highly irregular, cf. (51) (see also McIntyre 2002:101f). The copula argument makes the wierd prediction that away and down are secondary predicates in (51a,c) but not in (b,d). The copula argument is also undermined in that different copulas make different predictions. (b) and (d) allow stay. German Anne ist weg can be used of (51b). (51) a. Anne went away (on holidays) = Anne is away b. Anne walked/ran away ≠ Anne is away c. The problem brings/drags Mary down = Mary is down (about the problem) d. I bolted the food down; The food went down ≠ The food is down
11
12
In sum, the copula is untrustworthy as a test for secondary predicates. The copula argument thus fails to support the claim that the objects in (48a) are arguments of verbs rather than of particles. Indeed, the particles in (48a) sometimes openly flaunt their status as secondary predicates. The relevant uses of up (the so-called 'aspectual' uses, or, to use a more accurate notion available for free, its uses as an underspecified result predicate, see Spencer & Zaretskaya 1998:6, McIntyre 2003) license unselected objects, cf. (52a), and optional intransitivity impossible with the simplex verb, cf. (41). Similarly, repetitive over (Jackendoff 2002:78f) forms obligatorily transitive structures even with intransitive or optionally transitive verbs, cf. (52b). Clearly, the verb is not selecting the objects in (52), in accord with (45b,c). (52) a. chat someone up, soak water up [cf. *chat someone, *soak water] b. read *(books) over, think *(the matter) over [cf. read (books), *think the matter] Beside up and over, the so-called 'inchoative' use of German an (anlesen 'read partly', anknabbern 'nibble partly') is another apparent threat to (45b,c), for it co-occurs with direct objects, unlike the at first sight synonymous los in (7d). However, the objects are obligatory, whatever the verb's subcategorisation, cf. andenken 'think about partly' from intransitive denken. See Zeller (2001a) and McIntyre (2001) for data and analyses of an as a secondary predicate expressing partial affectedness of direct objects. Thus, we can maintain the empirical claim that direct objects in conflation VPs, including resultatives and particle verbs, are not arguments of the verb. As noted earlier, this predicts atransitivity when the PP/particle licenses no external DP argument, as in eat (*cakes) on and see (*her) into the window. Thus, atransitivity reduces to an independently motivated fact. The ensuing sections try to explain this fact. 4.2 VP shells, INIT and CHANGE My explanation for atransitivity will assume with recent generative studies that the traditional VP can consist of more than one embedded maximal V-projection ('shell') where at least one verbal head is a null light verb. In contrast to Larson's (1988) original proposal, most writers using shells assume that the head of the upper shell has a (e.g. causative, agentive) meaning (e.g. Baker 1997, Bowers 1993, Hale/Keyser 1993, 1997, Harley 1995, Kratzer 1996, Pylkkänen 2002, Stechow 1996). I call this head INIT. Its specifier is an initiator, an entity with which an event originates. The complement of INIT expresses an initiated event. I assume a broad semantics for INIT, covering eventive notions like CAUSE and DO, but also stative relations like 'is the initiator/source of'. Something like the initiator is found in the very different theory of Croft (1998), who sees subjects as initial points in a causal chain. My INITP is like 'FP-initiation' in Ritter & Rosen (1998), which also licenses an initiator in its specifier, but I reject their (redundant?) upper v-shell and their idea that FP-initiation is not projected if there is no delimitation of the event, falsified by causativisations of unergatives like I work them hard, I run/operate the machine, which involve stacked INITPs in my terminology. A second type of light verb used here is called CHANGE, which mediates a predication between a PP or AP predicate in its complement and its specifier. A DP specifier is an instance of a direct argument, i.e. an unaccusative subject or direct object. (53) gives a schema for a VP with both light verbs. Readers endorsing agreement projections can add them between or above the shells without detriment to my argumentation. Note that the full structure is not always present. An unaccusative structure omits INITP. (53) [InitP [Init’ INIT [ChangeP [Change' CHANGE ]]]] If we think of CHANGE as a verb, the overall structure in (53) parallels one common analysis for resultative and particle constructions in generative theories, viz. one in which the secondary predicate and lower verbal head form a V’, often called a complex predicate (e.g. Winkler 1997, Haider 1997, Zeller 2001b). This contrasts with analyses where the verb takes a small clause complement consisting of the DP and secondary predicate (den Dikken 1995; Hoekstra 1988; 1992; Hoekstra & Mulder 1990; Kayne 1985). My proposal is a compromise. CHANGEP can be likened to a small clause with a functional head (Abraham 1993; Bowers 1993; Hoekstra 1992; Svenonius 1994; 1996), with the difference that CHANGE expresses the eventive content expressed by GO and BECOME in my SRs, while small clause theories usually assume that a verbal element above the small clause contributes this information. If my approach is altered such that CHANGE takes a small clause complement, it will make the same predictions, provided that CHANGE cannot introduce another argument in its specifier. I called CHANGE a 'type of light verb' because I intend it to be understood as a cover term for distinct light verbs sharing the same basic predication properties. The subvariants of CHANGE relevant here are VBECOME and VGO, which correspond to BECOME and GO in SR and take respectively AP and PP
complements. Support for seeing VBECOME and VGO as distinct morphemes comes from elliptical modal+path constructions of the type I want out and German structures like (10d). Van Riemsdijk (2002) argues that their syntax must contain a null GO (i.e. VGO). Dutch, but not German, has AP constructions of the type Jan wil dood 'Jan wants to die', (lit. 'Jan wants dead'). This suggests that Dutch allows an empty VBECOME in the complement of a modal, while German allows VGO but not VBECOME here. This contrast can be captured only if VGO and VBECOME are distinct morphemes. (A further argument for this view appears in section 4.3.) In the following, references to CHANGE are to be read as shorthand for VGO and/or VBECOME. INIT and CHANGE are not seen as categories unto themselves, but as two different ways in which the category V can relate a specifier to a complement. Put differently: if V projects a specifier and complement, it can either relate an entity to an event, in which case V is called INIT, or an entity to a property or path, where V is called CHANGE. I leave open whether the grammar stipulates the existence of the INIT/CHANGE distinction, or whether precisely these manifestations of V are predictable given some general theory of verbhood. (Hale & Keyser 1993:68-74 offer relevant speculations.) The concepts introduced so far are not claimed to be able to capture all types of VPs. Since my theory of conflation will not require any particular theory of syntax-semantics mapping in nonconflation VPs, I do not propose such a theory here. I have discussed only INIT or CHANGE, since they will be used in describing conflation. They are not claimed to head every VP shell. English may have other light verbs, e.g. Pylkkänen's (2002) beneficiary-licensing applicative head and others noted in section 4.3. I do not claim that the complement of INIT is always a CHANGEP as in (53). For instance, unergative verbs might originate as (perhaps category-neutral) roots which incorporate into INIT, starting from a structure like (54a) (e.g. Hale & Keyser 1993; 1997; Chomsky 1995:315f; Harley 1995; Uriagereka 1998:410). Some of these roots may themselves project arguments appearing as direct object, as in (54b). (54) a. Fred spoke: [InitP [DP Fred] [Init’ INIT [X(P) speak]]] b. Martha played the piano: [InitP [DP Martha] [Init’ INIT [XP/VP play [DP the piano]]]] 4.3 M-Conflation and resultatives Theories of resultatives and particles which use shells normally assume that a lexical verb heads the lower shell, as in (55a) (cf. Dehé 2002; Haider 1997; Winkler 1997; Zeller 2001b). I will put forth a different approach where lexical verbs can merge directly with INIT, cf. (55b), giving a structure where each shell is headed by a light verb. I firstly describe the theory, and then support it by showing that it acquits itself well in handling various problems of resultative VPs. (55) a. [VP Jana drinki [VP the bar ti empty]] b. [InitP Jana [Init drink+ INIT] [ChangeP the bar CHANGE empty]] I illustrate my approach with the transitive resultative in (56). Ignoring the material under the higher INIT, (56b) means that Ethel is the initiator of an event in which she becomes tired. INIT merges with the root dance in the morphology, yielding a compound verb whose nonhead (the root) does not affect INIT 's relationship to its specifier and complement. The lexical verb licenses no arguments, commensurate with the anti-inheritance findings in (45). The morphologically formed complex dance+INIT is interpreted such that the initiation is identified as dancing. (57) proposes a condition for the interpretation of the morphological structures.8 Note the terminology: I speak of 'conflation' when abstracting away from the mechanism which yields it, and of ‘m-conflation’ when speaking of the mechanism in (57) (as opposed to the lexical operations mentioned in section 2).
13
14
(56)
a.
Ethel danced herself sore: &cause BECOME (SORE(ETHEL))
DO (ETHEL,DANCE)
b.
INITP INIT '
DP INIT ROOT
Ethel
CHANGEP
M-Conflation INIT
dance
DP
CHANGE’ CHANGE
herself
VBECOME
AP sore
(57)
M(ORPHOLOGICAL)-CONFLATION: Compound a root R with INIT or CHANGE if R names an event which is identical to the initiation or change expressed by those heads.
(56) shows a syntax-semantics mismatch. The SRs assumed here treat causation as relating two events rather than an entity and an event, but the syntax (INIT) requires an initiating entity. (The same mismatch holds between the levels of conceptual structure and semantic form in Wunderlich 1997b:34-38 for verbs with underspecified causing events, e.g. empty a glass.) The only way a binary branching syntax can translate the symetrically conjoined SR in (56a) is to represent the entity in the causing (DO) event as an initiator. This can be derived from the SR because the agent is the initiator of the causing event, and therefore indirectly the initiator of the result event. One form of recompense for embracing the unconventional and stipulated procedure in (57) comes from the simple account it gives us for the different syntax-semantics mappings in transitive and unaccusative resultatives. Unaccusatives are seen in (58) and (59). The GO subevents map onto a CHANGEP, as usual in this theory. The &contemp relation defined in section 2 is the same as the identity between the events expressed by the root and the light verb insisted on in (57). Thus, the roots expressing the events in the respective lefthand conjuncts (dance/tear) can compound with CHANGE. (dance cannot compound with CHANGE in (56), for the dancing and becoming sore are distinct: the soreness may set in after the dancing ceases.) The arguments in (58) and (59) are licensed not by the lexical verb but by CHANGE. This seems odd until we recall the finding in (43) that unaccusative resultatives need not show the same argument selection as the verb without the result predicate. (59) is an unaccusative resultative with an unselected argument. (the metal edge can be seen as an argument of tear only if we forego all chance of verifiability and explanatoriness by stipulating that tear has a second sense 'move as a result of some other object's tearing' which is possible only when a directional PP is present.) (58) a. Ethel danced into the theatre DO (ETHEL,DANCE) &contemp GO(ETHEL, [PathTO IN THEATRE]) b. [ChangeP [DP Ethel] [Change’ [Change dance+ VGO] [PP into the theatre]]] (59) a. (The blind became rotten.) The metal edge tore off. b. *The metal edge tore. c. *BECOME(TORN(METAL EDGE)) &contemp GO (METAL EDGE,OFF) d. BECOME(TORN(BLIND)) &contemp GO (METAL EDGE,OFF) e. [ChangeP [DP the metal edge] [Change’ [Change tear+ VGO] [PP off]]] The theory accounts simply for why (56) needs a reflexive and (58) does not. A BECOME or GO event maps onto a CHANGEP. If a verb root names a situation seen as identical to this event, then the root mconflates with CHANGE. There can only be one DP argument in this case. If an activity A is distinct from the event expressed in CHANGEP, and causes it, A, including its entity argument, maps onto an INITP. There will thus be two DP arguments. If the DP arguments of INIT and CHANGE are coreferential, the grammar need only ensure that the lower DP is realised as a reflexive and not as a DP phonologically identical with the subject DP, a problem equally difficult for all theories of resultatives. Another account of why (56) needs a reflexive and (58) does not is Rappaport Hovav & Levin (2001). They argue that the reflexive is needed because of a condition that distinct (in my terms: noncontemporaneous) subevents in SR must each be represented by at least one argument phrase (cf. Rappaport Hovav & Levin 1998). This constraint looks ad hoc unless we interpret it as a type of recoverability condition serving to identify the event structure of a VP. Even so, there are problems. The argument-per-subevent-condition is formulated in terms of XPs, but seems to be applied solely to NPs, witness the claim that 'event structures with two subevents must give rise to sentences with both a subject and an object, while simple event structures would give rise to sentences that require only a subject' (p.779). Apparently, a result predicate does not count as an 'XP argument' for the purposes of representing a subevent. The questions begged are: Why is the identification of subevents solely the prerogative of NPs? Why does a directional PP like that in she worked *(herself) to death not suffice for this purpose? Why do subevents which are in my terminology linked with &cause require surface identification when those linked with &contemp do not? Principled answers to these questions look unlikely enough to tempt one to look elsewhere for an explanation for the empirical content behind the argument-per-subevent condition. My account buys us freedom from these conceptual problems, the price being the stipulation of the existence of m-conflation. Given that conflation is parameterised and thus stipulated in a given language, the cost of a stipulation is incurred in any theory of conflation.
Another approach to reflexives like that in (56b) assumes that result predication can only reach (underlying) direct objects (the 'Direct Object Restriction', DOR). Unlike unaccusatives, unergatives lack direct objects, so reflexive objects are inserted to mediate the result predication to the subject (Levin & Rappaport Hovav 1995; Li 1999; Simpson 1983; Spencer/Zaretskaya 1998; Winkler 1997). A problem concerns unaccusative resultatives from sound verbs (rustle/buzz/roar in). For these, DOR accounts must posit an unaccusative variant of the verb that preexists resultative formation. This approaches circularity, for the unaccusative variant is only evident in the verbs' behaviour in resultative constructions. It remains unclear in existing DOR theories why this unaccusative variant (be it generated by lexical rule or lexically listed) should exist at all. My account avoids such problems: rustle in does not involve an unaccusative variant of rustle but the use of rustle in an unaccusative structure. My theory of conflation is what may be called a 'plug-in theory' since argument structure and much of the VP's meaning come not from the verb but from the structure it is 'plugged into'. In this I agree with Construction Grammar (see section 4.5), but differ from that theory in not giving up hope that constructions can be dissected into independently needed elements. Instances of plugging-in are also found in Hale & Keyser (1993:90f) and Harley (2002), who posit devices (Hale & Keyser: 'tags'; Harley: 'manner incorporation') which name the problem of conflation without asking how it works. Another plug-in theory is Mateu (2001), whose proposal inspired my theory of m-conflation. Mateu assumes a cross-sentential generalised transformation which replaces a causative light verb with a verb formed by the Hale-Keyser incorporation mechanism in (54a). The differences between my proposal and Mateu's are that I assume that Mateu's separate derivational workspaces correspond to the more widely assumed distinction between morphology and syntax, and that conflation is constrained by the semantic condition in (57). A virtually identical form of plugging-in to that posited here is used in Embick (2003), which came to my attention just before the submission of the final version of this article. Embick argues for reasons independent of those given here that resultatives are formed by 'direct merge' of a root to a light verb, a complex-head-forming process also said to be responsible for the formation of certain types of compounds. In lieu of my morphological approach to conflation, one could obtain the same results using Mateu's (2001) generalised transformation, provided it is constrained semantically in a way similar to (57). Linguists doubting the existence of non-syntactic morphological operations could also devise an account where verb roots in conflation structures are introduced by a mechanism akin to adverbial modification in theories like Cinque (1999), except that the roots are heads rather than specifiers and originate in projections immediately above INIT or CHANGE. A reason to prefer the morphological view of conflation is Snyder's (2001) finding that languages with what I call conflation all have productive compounding, and that children acquire compounding and conflation at the same time. Snyder assumes that English complex predicates involve morphological compounding at LF because this is the only way that two syntactically independent expressions can characterize the event-type of a single event-argument (2001:328,336). The compounding of which Snyder speaks is not m-conflation but an LF union between a verb and secondary predicate, but Snyder's parallel between conflation and compounding fits equally well with my theory of conflation, provided it is understood as a morphological operation. In fact, given that effected object constructions (I drilled a hole) involve no overt secondary predicate but are among the conflation constructions (Levin & Rapoport 1988), my account may be preferable to Synder's abstract compounding as an account for his observations. Most other theories of resultatives cannot capture Snyder’s findings. In the present theory, (e.g. Romance) languages lacking conflation phenomena simply lack the type of compounding in (57). I assume that in such languages, SRs like (56a) and (58a) are possible. However, there is no way to introduce the root dance into the syntax. To express these SRs syntactically, such languages must use constructions closer to Ethel went dancing into the theatre or Ethel made herself sore dancing. To demonstrate the theory's coverage, I briefly note more applications of m-conflation. As well as cases of m-conflation with INIT like (56b), the theory predicts the existence of transitive VPs where a verb m-conflates with CHANGE. The ambiguity in (60) confirms this. In the reading in (a), the walking is identical to the causation, so walk m-conflates with INIT, while in (b) the walking is identical to the object's motion, so walk m-conflates with CHANGE. Conflation with CHANGE in transitive VPs is also needed for causativisations of unaccusative resultatives like (61), where the burning is identical to the toast's becoming black, not Bill's action (e.g. turning on a grill). With AP predicates, cases like (61)
15
16
and parallel unaccusatives (the toast burnt black) are ill-formed in German (Kaufmann & Wunderlich 1998). In my theory, this is expressed by stating that German lacks productive compounding with VBECOME. (This gives a second theory-internal argument that VBECOME and VGO are distinct morphemes, cf. section 4.2.) (60) Ann walked Jo up. a. (Ann walks, pushing Jo in a pram): [Ann [[walk+INIT] [Jo VGO up]]] b. (Ann, stationary, makes Jo, a puppet, walk): [Ann [INIT [Jo [walk+ VGO] up]]] (61) Bill burnt the toast black [Bill [INIT [the toast [burn+ VBECOME] black]]]
Various factors predict the unlinkability of the verbal object. Firstly, m-conflation as defined in (57) says that in (63) (as in (56), (58) and (59)) the lexical verb is a nonhead of a compound, and thus unable to license arguments outside the compound, cf. the standard observation that 'the arguments of the nonhead are not part of the argument structure of the compound' (Di Sciullo & Williams 1987:30). Thus, the factors blocking *scrubwomen of floors, *a crybaby of bitter tears and *a bakehouse of cakes are one source of ungrammaticality in cases like *scrub floors on, *cry bitter tears on and *bake cakes on. 9
Even in the unlikely event that the lexical verb in (63) could license an object in the lower specifier position in lieu of X, the semantics of CHANGE (VGO) would force on to predicate over the object, ruling out the event path reading of the particle and forcing a resultative interpretation, even if this is pragmatically deviant. This is one of the first intuitions one has about atransitivity violations: *scrub floors on and *eat cakes on, if interpretable, sound as if the objects are somehow caused to ‘go on’. #bash the piano about is construed parallel to push the piano about, although the event path reading 'bash about on the piano' is far more likely. Thus, the nature of m-conflation and the relational semantics of CHANGE forbid the linking of verbal objects in (63). It remains to clarify certain aspects of the proposed syntax. (63) shows a syntax-semantics mismatch similar to (but more complex than) that in (56). That Fred initiates the GO event in (63) is not stated in the SR in (62), but is inferred from it. The initiating event is identified as scrubbing by m-conflation although it is not represented explicitly as the cause of the GO subevent in the SRs, where the GO event is linked by &contemp (not &cause, unlike in (56)). However, since the GO subevent presupposes the scrubbing, the scrubbing can be inferred to be the initiating event. &contemp allows verb roots to m-conflate with CHANGE in unaccusatives like (58), so we must ask why &contemp in SR in (62) does not allow an unaccusative structure (i.e. a bare CHANGEP). (64) illustrates the possibilities. (a) is blocked similarly to *scrub floors on: the semantics of CHANGE would wrongly force predication of on over Fred. The theory can arguably generate (b), but this cannot be part of a well-formed sentence due to the requirement that English sentences have overt subjects. That this problem is not remedied by inserting an expletive subject in (64c) is either a problem for my theory or is due to properties of expletives. The latter view seems tenable. Bowers (2002:194-199) notes that it is inappropriate with unaccusatives. There is usable with unaccusatives, but many such verbs reject there: *there died/disappeared many people, *there descended many into the pit, *there opened a great cavity in the ground. It is thus legitimate to ask whether there is insertable with sufficient freedom for (64c) to be possible in a grammar which can generate (b). (64) a. *[ChangeP [DP Fred] [Change’ [Change scrub+CHANGE] [P(P) on]]] b. *[ChangeP X [Change’ [Change scrub+CHANGE] [P(P) on]]] c. *There scrubbed on, *It scrubbed on ‘scrubbing went on (continued)’ The above analysis also applies to PPs like those in section 3.2 (see (*them) into the house), but a full PP rather than a particle is complement of CHANGE (VGO). The question arises of whether the stative GO EXT EXT and ORIENT functions used in section 3.2 map onto V or whether we need extra light verbs V and VORIENT. I leave this open, for want of a precise theory of how EXT and ORIENT conceptually relate to (and differ from) GO. The data in section 3.3 (including non-event-path cases like (27)), where the conceptual ground of the particle appears as direct object, must also be accorded a separate study. 4.5 Alternative approaches to atransitivity Previous explanations for atransitivity are less than adequate. The approaches in Stiebels (1996) and van Hout (2001) were criticised in section 3. My (2001) study settled for the observation that particles must predicate over any direct objects present without trying to explain it. Toivonen (2001:144) stipulated that Swedish på 'on' must combine with intransitive (uses of) verbs. This is unexplanatory and empirically questionable given the observation in section 4.4 that atransitivity is attested with obligatorily transitive verbs. It is more accurate to say that the particle is incompatible with the direct object than that it only combines with intransitive (variants of) verbs. Zeller's (2001a) study, which partly inspired my theory, requires more discussion. He catalogues the argument-structural effects of German verb particles, concluding that adding particles leads to the loss of the verb's linking information, and that internal arguments of particle verbs are arguments of (and licensed inside the projection of) the particle. This predicts the range of possible argument structures of particle verbs. The theory could be extended to include directional PPs and result predicates. Zeller suggests that argument blockages of the type eat (*cakes) on occur because the ParticleP and object compete for the same position, namely the complement of V in a single-shell structure. To prevent the verbal object from appearing in the specifier of V, as in (65), Zeller needs a mysterious stipulation, akin to designation or underlining of external arguments in older GB theories, which gives 'external arguments' sole rights to the verb's specifier position. By contrast, thematic restrictions on particular positions in VP follow naturally in theories like mine which use syntactic lexical decomposition (cf. Baker 1997). Thus, cakes in (65) is in no wise an initiator, and thus not a possible specifier of INIT, and is not predicated over by on, and thus cannot be the specifier of CHANGEP. (65) *Cakes ate on ('someone ate on, eating cakes')
17
18
Extending the inventory of light verbs yields more possibilities. Constructions like (44) (the wall hangs full of pictures) m-conflate a position verb with a silent copula BE. More radical possibilities include m-conflation with a silent MAKE (drill a hole) and CONTAIN (the room sleeps/sits ten people). The idea of argument-structurally inert verb roots central to m-conflation reminds one of serial verblike constructions such as go see, go find it, go give her the book, where go does not affect the argument structure. These seem to present the going and the event occuring at the destination as a single event, an intuition found with serial verbs in many languages (Durie 1997:291). Perhaps go enters the structure by m-conflation as in resultatives, except that go sanctions overt V+V compounding, unlike other verbs. These structures resist inflection (*she go(es) get(s) it), perhaps because the English Tense head cannot check features with two verb stems. Deciding whether there are any serial verb(-like) contructions which should be generated by m-conflation is too complex an issue to enter into here, however. To conclude, it appears that the approach recommended here is capable of a neat explanation for the syntax-semantics mapping in resultative constructions. Assessing the consequences of this nonstandard proposal requires more work, but I leave the discussion now, hoping that the theory is worth considering. 4.4 The syntax of event path structures We now see how the theory fares with event path structures. An account for atransitivity must work with SRs that encode illicit direct objects, e.g. because *Fred ate cakes on has an acceptable literal German translation in Fred aß Kuchen weiter (cf. section 5) and because atransitivity can befall verbs like hit (*hit a golfball off), whose otherwise obligatory transitivity suggests that the action cannot be conceptualised without the object. I use (62) as a sample SR. The choice of scrub as a sample verb and the representation for the verb's meaning in the lefthand conjunct is arbitrary, as the explanation will cover any verb appearing in an atransitive VP. (62) Fred scrubbed (*floors) on [DO(FRED,SCRUB(FLOORS))]i &contemp GO([EVENT ]i, ON) (63) is the proposed syntax. As usual in my theory, the GO subevent in the SR maps onto a CHANGEP. The specifier of CHANGEP in (63) is an empty element (notated as X) which represents the theme of the GO event in SR (i.e. the coindexed Event constituent). X is not essential for my purposes, but economising on it would reduce the clarity of the mapping to the semantics. X occupies the position for the direct object. I will not rely on this in explaining atransitivity, however. (63) Fred scrubbed (*floors) on INITP
DP Fred
INIT ' CHANGEP
INIT
scrub
INIT
X
CHANGE' CHANGE
V
GO
P(P) on
I know no general theory of argument realisation which predicts atransitivity. Given that the meaningful light verbs crucial to my account are not part of the architecture of most such theories (e.g. Croft 1998; Dowty 1991; Grimshaw 1990; Jackendoff 1990; Larson 1988; Levin & Rappaport Hovav 1995; Pinker 1989; Tenny 1994; van Valin & LaPolla 1997; Wunderlich 1997a,b), linguists subscribing to such theories must either devise an alternative account of atransitivity or change their theories. I know no theory of resultatives which predicts atransitivity. As noted in section 4.1, any theory that allows verbs to project objects in transitive resultatives wrongly predicts that the verb can project an object in cases where a path phrase has no external argument. Theories which do not posit linking of verbal objects in resultatives are the small clause theories of Hoekstra (1988) and others cited in section 4.2. Such theories could try to capture atransitivity by assuming that an atransitive PP/particle forms a small clause with an empty theme, as in my account. However, all such theories assume that the small clause is complement to a lexical verb. As in Zeller (2001a), it is unclear what prevents the verb from projecting its object in its specifier position. Consider next Construction Grammar (Goldberg 1995; Jackendoff 1997). Booij (2002:31f) applies this theory to atransitive particles as in (66a) (which replaces Booij's Dutch example with an English one). There is no slot for an object in the construction's lexical entry, so the theory can describe atransitivity. But it cannot PREDICT it, for constructions are seen as idioms, and a non-occurring configuration like (66b) must therefore be a lexical gap. A challenge for this approach is to ensure that the cross-linguistic presence of the 'lexical gap' in English, German, Dutch and Swedish does not force Construction Grammarians to regard atransitivity as a relic acquired via negative evidence by generation after generation ever since Proto-Germanic times. It may be replied that the archiconstruction (Jackendoff 1997:555) which underlies all conflation constructions is endowed with a generalisation that any objects present must be predicated over by PPs/APs. I see no non-stipulative account for this generalisation in this theory. A second answer might be that (66b) is bad because it blends the transitive construction (Goldberg 1995) with the durative on construction, and hybrid constructions are not acquired by language learners without positive evidence for their existence. This predicts that hybrid constructions could not come into existence except as mistakes which somehow gain currency. Now English has several hybrid particle constructions, including the double object+particle construction (print her out a copy), the double object+particle+PP construction (send people out leaflets to their homes) and the particle+NP predicate construction (make him out a liar). These constructions (not peculiar to English) are too infrequent for it to be certain that direct evidence alerts children to their existence (let alone to constraints on their syntax noted in den Dikken 1995). The existence of such hybrid constructions makes it hard to believe that (66b) is really unexpected in such a theory, in contrast to a theory where the general makeup of the syntax of VP leads language learners to expect atransitivity with event path particles. (66) a. Particle verbs with durative on (e.g. eat on): [[x]V[on]P]V' 'go on x-ing' b. Pseudo-English particle verbs with durative on (e.g. eat cakes on): [[x]V [y]NP [on]P]VP 'go on x-ing y' Another atransitivity account was sketched by a reviewer, who wrote that much atransitivity data is blocked similarly to (67a), which is claimed to succumb to a constraint against multiple independent state changes (including location changes) in a single event (cf. Goldberg 1995:81-89, and Tenny's 1994 ban on multiple measures). This does not predict e.g. see (*it) into a window, ring (*her) through, remember (*him) back to one's youth and the particle data in section 3.1. Anyway, a blanket ban on multiple state changes wrongly rules out (67c) parallel to (b). My theory predicts (c), for verb roots coding state changes can m-conflate into VPs with separate result predicates. Similar examples appear in (d) and Goldberg (1995:171f). (a) and (b) are ruled out trivially in theories like mine which assume single complements. (67) a. *They kicked him black and blue out of the room; *They kicked him out of the room to death b. *The plaster fell to pieces to the floor. c. The plaster crumbled to the floor BECOME(CRUMBLED(PLASTER)) &contemp GO(PLASTER, TO FLOOR) d. I broke the bread into the bowl, the water emptied out of the tank, I bent the pipe down
5 Conclusions and potential problems This article presented a theory of 'conflation' (also called 'lexical subordination'), the parameterised phenomenon in which an activity combines with a path or result predication in such a way that only the activity is represented by a verb root. I argued that conflation generates not only well-studied phenomena like resultative constructions, but also event path structures, VPs where a directional expression was argued to express the spatial or metaphorical path of a situation. Section 3 presented many instances of PPs and particles which can be argued to be event paths. The expressions analysed as event paths included, but were not limited to, a subclass of the so-called 'aspectual particles'. The event path analysis connects the spatial and 'aspectual' use of the particle, unlike non-localistic analyses. Except for systematic exceptions in section 3.3, event paths distribute complementarily with direct objects (ring (*Ann) through, read (*books) on). This atransitivity can be predicted if (a) event path VPs involve conflation and (b) verbs cannot link their arguments in conflation structures. Section 4.1 independently motivated (b). Thus, direct objects in resultative and particle constructions are NOT arguments of the verb, even if the object fits the verb's s-selection, as in hammer nails in. Sections 4.24.3 presented a theory which explains (b). The theory treats conflation as compounding ('mconflation') of a verb root with one of two light verbs, INIT (with causal or agentive meaning) and CHANGE (which mediates the predication of result and path expressions). Section 4.3 showed that mconflation yields natural explanations for the syntactic manifestations of event-structural differences between unaccusative and transitive resultatives and for some less familiar conflation patterns. Section 4.4 applied the account to atransitive VPs. The verb cannnot introduce arguments because it is nonhead of a compound. Also, we cannot insert a DP in the object position, the specifier of the lower shell, because the lower shell is headed by a light verb whose semantics forces a predication relationship between the particle/PP and object, yielding a resultative semantics rather than the event path reading. The chief empirical claims to be evaluated are that (a) the event path analysis is valid, (b) event path VPs can involve conflation, (c) conflation blocks the linking of verbal arguments, and (d) conflation is the compounding of a verb root to a meaningful light verb. If accepted, (d) supports e.g. Baker (1997), Hale & Keyser (1993; 1997) and Pylkkänen (2002) in their claim that argument structure is at least in part a projection of syntactically represented lexical decomposition predicates. This position is incompatible with theories which see all arguments in the VP as being projected from the verb itself and which derive alternative argument linking patterns from lexical-semantic operations performed on the verb (Jackendoff 1990; Pinker 1989; Levin & Rappaport Hovav 1995 and many others). It remains to address potential problems for my theory. Firstly, a reviewer noted cases like (68a,b), rightly observing that the complement of CHANGE cannot contain both an aspectual particle and a PP. But the particles discussed in section 3.1 all have spatial path senses as well as 'aspectual' ones. I claim that on and los in (68a,b) are being used in the spatial, non-aspectual senses seen in it is further on down the road and Briefe losschicken 'send off letters'. I see no evidence against this claim, and it receives support from cases like (c), involving particles whose atransitive and spatial senses are easier to tell apart than those of on and los. In (68c), along and around cannot have their atransitive senses ('with others', 'aimlessly', cf. (7b,c)), but receive unambiguously spatial readings (corresponding to their transitive prepositions: 'along the street', 'around the corner'). (68) a. They danced ON out of the barn past the well into the yard b. Sie rannten LOS auf die Straße they ran off onto the street c. I walked {along/around} to the church d. [LOS auf die Straße] sind sie gerannt. (=(b)) e. Jo is down in the parlour by the fire; With Jo down in the parlour by the fire... Since my theory eschews multiple complements of any kind, we must ask why (68a,b) are good even if on and los express spatial paths. The answer is that the PPs form a complex PP, iconic of a complex path in SR. Complex PPs are standardly assumed in studies on PP syntax (e.g. Haftka 1997; Koopman 2000; Olsen 1999; van Riemsdijk 1990; Wunderlich & Herweg 1991). Topicalisations like (d) are one argument for complex PPs. Stative contexts like (68e) render dubious an imaginable alternative where motion verbs have three complements (source, passage and goal).10 Incidentally, I do not claim that complex PP formation is unconstrained. For instance, the spatial and a metaphoric PPs in (67a,b) cannot form a complex PP because the paths are in different domains ('landscapes', cf. Goldberg 1995:81-89) and thus not conceptualised as part of the same path.
19
20
The final potential problem is a set of rare cases where directional PPs coexist with direct objects over which they do not predicate. If we assumed that all VP-internal directional expressions are complements of CHANGE, my theory would wrongly predict predication over the object. However, it is mostly possible to show that these PPs are not complements of CHANGE. We start with some cases that turn out to be adjuncts.11 Talmy (1985:66) suggests that the non-bracketed parts of (69) involve conflation of a path with a complex consisting of the verb and object. However, the bracketed continuations in (69) coordinate the directional expressions with place adjuncts. Such coordinations are impossible with standard conflation VPs (*throw the ball into the basket and at the park), because the coordinated PPs form a larger PP/coordination phrase which can be inserted either in the complement or the adjunct position, but not in both positions. The coordinations in (69) can be explained if both PPs are adjuncts. Specifically, I claim that the directional PPs are being used as adjuncts of location. This is clear in in (69b), for all the way (to x) is independently capable of a locational reading (there are trees all the way to the shore). A locational analysis for to the party in in (69a) is surprising until we note that the sentence does not actually assert motion. If motion were asserted, we would expect the to PP to delimit the event, as in walk to the party in an hour. But (a) is atelic: *wear a dress to the party in an hour. Hence, Talmy's gloss of (a) as 'she went to the party, wearing a green dress' must be rejected for something like 'she wore a green dress on the way to the party'. (69) a. She wore a green dress TO THE PARTY (and at the meeting the next day). b. I read comics ALL THE WAY TO NEW YORK (and on the subway as well). (70) a. I studied the problem (-indeed I did so) TO THE POINT OF UTTER FRUSTRATION b. They drink beer (every night, often doing so) {INTO THE EARLY HOURS/ TO EXCESS} c. A strange man was opening his overcoat TO PASSERS-BY (and then he did so TO A POLICEMAN) d. Kindly smoke that filthy cigar OUT THE WINDOW rather than (doing so) INTO MY FACE. The PPs in (70) look like event paths: in (b) and perhaps (a), they seem to map the temporal course of an event onto a path with a state or time as goal. (70c,d) seem to place the flashing/smoking events on ORIENT paths (section 3.2). No direct object in (70) is theme of a PP. Thus, event paths are not intrinsically in complementary distribution with direct objects. I assume that direct objects are disallowed when an event path is projected in the complement of CHANGE. However, the PPs in (70), like (69), are adjuncts. The bracketed material applies the standard do so test, confirming this. Applying this test to the particles and PPs treated in earlier sections yields severe deviance (*I ate, and indeed did so on). Fu et al. (2001) are among those who note that do so replaces a whole VP. This predicts that it will replace all complements, but not necessarily adjuncts. If the PPs in (69) and (70) are adjuncts, then they are not complements, and do not refute my claim that a complement of CHANGE must always predicate over any direct object. My theory entails no analysis of directional PP adjuncts, beyond the claim that they exist, but the novelty of this claim calls for some brief remarks. Firstly, an analysis in which they predicate over the event seems correct in most cases; this is the most plausible analysis in (70b-d). A case like (70a) raises problems analogous to those raised by subject-oriented adverbs (I left willingly), where one can debate about whether the adverb is predicated over the subject, the event or both. Secondly, like the marginal French resultative adjuncts noted in Legendre (1997), the use of English directional PPs as adjuncts is idiosyncratic. *We drank (beer) into the afternoon is bad despite (70b) and the drinking lasted/extended into the afternoon, where the PP is an obligatory complement and the verb lexicalises a temporal use of EXT (section 3.2). Secondly, (71) contrasts to the point of with to. The former is an idiom (perhaps calqued on French jusqu’au point de) whose lexical entry stipulates its use as an adjunct. The to PPs only appear in object-predicating resultative structures.12 Thus, adjunct directional PPs which co-occur with direct objects are sporadic and lexically stipulated, unlike the paths in section 3, which block objects with clockwork regularity. (71) a. she worked/ran *(herself) to exhaustion b. she worked/ran (*herself) to the point of exhaustion c. I studied the problem to *(the point of) exhaustion Another problem is German weiter 'further', which, like on, can express either a continuing path (ich schickte es weiter 'I sent it on') or a continuing event path (wir saufen weiter 'we drink on'; die Diskussion ging weiter 'the discussion went on'), but does not resist objects when expressing an event path (ich trank mein Bier weiter 'I drank (*my beer) on'). Showing that weiter is an adjunct would involve issues in German syntax and prosody too complex to discuss here. Weiter is a potential
problem for my theory, but every account of atransitivity will face this problem given the semantic similarity between on and weiter. Rappaport Hovav & Levin (2001:770; RHL) note rare cases like (72) where result predication seemingly targets subjects, bypassing objects. These VPs would destroy my theory of the semantic structure of VP if they had the same syntax as normal transitive resultatives, but I doubt that they do. (73) shows that if we use out as a result predicate, we find that RHL’s ‘subject-oriented resultatives’ resist particle shift13, unlike the standard object-oriented predication in the primed variants. (The (c)(c’) contrast makes sense because, unlike follow, chase can be used when the subject induces motion on the part of the object, witness the instrument role of the stick.) The sensitivity of word order to different semantic integrations of the object and particle would be inexplicable unless these semantic integrations resulted in different initial syntactic configurations. Thus, the particle and/or object are not in the same position as in standard resultative particle constructions. A hypothesis to test is that subject-oriented predicates are adjuncts. (72c,d) remind one of the adjuncts in (69) in their failure to yield telicity. Pace Dowty (1991:571) and RHL, there are also grounds for asking whether the subject and object follow are not derived from lower positions. Conservative English allows unaccusativetypical there insertion even if follow has two DP arguments (e.g. there followed him a great company of people in Luke 23:27 in the Revised Standard Version of the Bible). German folgen is unlike normal transitive verbs in taking the be perfect and in that its object receives inherent Dative Case. Thus, RHL’s data do not automatically bring down my theory. However, my theory requires a fuller analysis of these data in future work. (72) a. John danced mazurkas across the room b. I rode the breeze clear of the rocks c. I took the subway to the city (*in ten minutes) d. I followed/accompanied them to their home (*in ten minutes) (73) a. We flew {*out} American Airlines {out} a’. We flew {out} the airplane {out} b. I took {*out} the subway {out} b’. I took {out} the books {out} c. I followed {*out} the dog {out} (with a telescope/*with a stick) c’. I chased {out} the dog {out} (with a stick/*with a telescope) Another challenge for the theory comes from the 'possessor ascension' structures in (74). (Levin 1993:71f) lists data and literature.) (74) a. She touched/kissed/hit/punched him on the nose; She stabbed/kicked him in the leg b. Sie {schlug/schnitt/biß/trat} {ihn/ihm} ins Bein she {hit/cut/bit/kicked} himaccusative/dative into.the leg If it can be shown that these have the same syntax as resultatives, i.e. a structure where CHANGE has a DP and PP argument, more work is needed to clarify how CHANGE relates PPs to DPs. Clearly, the PP with touch does not indicate a location of the object, and it is unclear if the relational character of CHANGE should be extended to include possessive relationships. However, English possessor ascension VPs probably do not have the same structure as resultatives. (75a-b) show that the possessive constructions, but not resultatives, allow an instrumental phrase between object and PP. (cd) show that the PPs are not in complementary distribution to result predicates, whereas normal result predicates are. Finally, the English PPs give no formal indication that they are directional (cf. *punch him into the face). The English facts are thus all consistent with an analysis of the PPs as adjuncts rather than one where they occupy the same position as a result predicate. I must, however, concede that the parallel German structures like (74b) are not covered by these arguments, and must be taken up in future work. (75) a. She {slammed/cracked/hit/touched} him with a golfstick on the head. b. *She slammed it with a stick off the table. c. He was shot dead in the back of the head. [attested severalfold in the internet] d. *He was shot dead {to a pulp/full of holes}. We have seen some structures where a directional PP can coexist with a direct object over which it apparently does not predicate. Some of these do not threaten my theory because the PPs are adjuncts. The German accusative possessors, weiter and the structures noted by RHL are problems needing further work. Only time will tell whether these problems can be solved without abandoning the theory in section 4. In the meantime, my work will have been to some avail if it raises awareness of event paths and brings some new perspectives into the discussion of conflation and argument structure.
21
22
Notes 1. Parts of this study were presented at the 2001 Sinn und Bedeutung conference at Osnabrück, the 2002 Scandinavian Conference in Linguistics in Troms∅ and the 2002 workshop Complex Predicates, Particles and Subevents in Konstanz. The audiences at these venues are thanked for useful discussion. I also thank Jaume Mateu, Jochen Zeller and the journal's anonymous reviewers for detailed and useful comments on earlier versions. The foregoing does not acknowledge my way out of responsibility for errors. Author's address: Universität Leipzig, Beethovenstr. 15, D04107 Leipzig, Germany. E-mail:
[email protected] 2. The internal argument of a preposition (the one which receives case from it in a full PP and serves to locate the external argument) answers to the terms 'ground', 'reference object', 'landmark', 'location', and the external argument to 'figure', 'theme', 'trajector', 'locatum', ‘located object’. I use 'ground' and 'theme' or 'figure' here. 3. I cannot say why unergative uses of los+motion activity, e.g. losschwimmen in the sense 'start swimming', are bad. They do not seem to be excluded by general principles, witness den Dikken's (1995:32f) observation that Dutch doorlopen (literally 'walk through') is unergative in the meaning 'keep up a pace' and unaccusative in the meaning 'walk on, continue walking'. The glosses are den Dikken's; I suspect that the particle conveys an extended spatial path in the unaccusative variant and that, in the unergative variant, it has the same (at first sight durative) meaning as the particles in (7a,f) or Dutch doorwerken 'work on, continue working'. 4. While there is a connection between the spatial and atransitive senses, I assume that the metaphor is semiproductive, meaning that the lexical entry for the particle must sanction the atransitive use. I base this on the fact that some varieties like mine do not allow the contracted form round in the atransitive sense, although its spatial use is identical to around (they walked (a)round but they played *(a)round). 5. (13) only aspires to capture the spatial senses to which the atransitive senses are directly related, not e.g. on in the contact/support use in (i), and around in the sense of motion via an object's exterior, as in (ii). A reviewer finds it problematic that I ignore what s/he sees as the basic senses of these particles. But since the spatial sense of on in push the cart on is not related to the sense in (i), neither sense can be more basic than the other. The sense in (ii) is not the basic sense of the PARTICLE around, for around does not mean 'via the exterior of something' in run around, push the car around and analogous VPs with many other motion verbs. The question of how the entire set of senses of on and around (both as particles and with complements) are related is of interest to polysemy theory, but cannot be studied here. It suffices for my purposes to find the spatial sense which is the point of departure for the atransitive sense, whether or not this is the basic sense of the particle. (i) I put the kettle on (the stove) (ii) I ran around the lake/ the corner 6. The metaphor 'events qua moving entities' of which I propose a subinstance surely exists. For instance, many verbs meaning 'happen' derive from motion verbs, e.g. French arriver 'happen, arrive', German passieren 'happen' (formerly 'pass', cf. come to pass), vor sich gehen, vorgehen 'occur' (
