Goal and Source: Asymmetry in their Syntax and Semantics

Seungho Nam
(Seoul National University) Presented at the workshop on Event Structures March 17-19, 2004. Leipzig, Germany
This paper focuses on the syntactic and semantic asymmetry between Goal (e.g., into the store) and Source locatives (e.g., from the store). Based on their syntactic and semantic asymmetry, the paper argues that they have distinct underlying base positions in extended VP-structure and further that they have different semantic scope/contribution in event structure. Thus, we claim (i) Goal PPs are generated under the lower VP2, and they semantically compose a core event (result state: E2) denoted by the lower VP2. And (ii) Source PPs are generated under the higher VP1, and semantically modify the process sub-event (E1). Source locatives do not compose a core event. This paper identifies an interface principle between syntax and semantics of Goal/Source locatives, and further argues that the interface principle should account for the non-directional readings of Goal/Source phrases in natural language. Key Words: goal, source, locative, event structure, argument, directional, non-directional, incorpration, pseudo passive, adverbial modification, aspectual composition, eventuality

1. Goals
This paper focuses on the syntactic and semantic differences between two types of directional PPs (i) Goal locatives (e.g., into the store) and (ii) Source locatives (e.g., from the store). The contrast between their syntactic behavior is identified in various constructions, and we account for their asymmetry by assigning them two distinct underlying base positions. Further, we argue that their systematic semantic differences are predicted by their different semantic scope in event structure. Directional locatives can be divided into the following three types: (i) Goal locatives, (ii) Source locatives, and (iii) symmetric Path locatives. 'Goal' and 'Source' have been well defined in the

- 1 -

literature: If an event involves a movement of an object and its trajectory, Goal designates the final/end point of the trajectory and Source the initial/starting point of the trajectory. Thus we use the terms 'Goal locatives' and 'Source locatives' to refer to the phrases designating the initial point and the final point of a trajectory, respectively. Each language has a way of expressing Goal and Source: e.g., prepositional/postpositional phrases or affixes. Jackendoff (1983, 1990), contrary to the treatment of Relational Grammar and Lexical Functional Grammar, does not take thematic roles like Source and Goal as grammatical primitives, but the notions are defined in his lexical conceptual structure. Thus in Jackendoff (1990), Goal and Source are defined as an argument of Path-functions, [Path TO ([Place ])] and [Path FROM ([Place ])], respectively. (1) shows a conceptual structure of a sentence like John came to/from the office. (1) [Event GO [Thing JOHN], [Path TO/FROM([Place OFFICE])]] Prepositions like to, into and onto typically take a Goal argument, and from, from under, from behind, and off can take a Source argument. Some other prepositions in English can bear either a directional or a non-directional sense, so the sentences in (2) are all ambiguous. (2a) means either (i) 'the suspect walked, staying inside of the store,' or (ii) 'the suspect walked into the store from outside.' (2) a. b. c. d. The suspect walked in the store. Harry swam under the bridge. Chris drew the box behind the curtain. The boys jumped on the bed.

Notice that each of the PPs in (2) only gives a Goal reading when the sentence denotes a directional movement. That is, the PPs never refer to a source location, thus (2b) does not mean that 'Harry swam from under the bridge.' The prepositions in (2) do not carry a morpheme denoting a (goal) directional sense, but they can give a goal directional reading of a motion event. This is not true of many

- 2 -

other languages. There is another set of prepositions that behave differently from Source or Goal. This includes through, over, across, past, and around. They are categorized as 'symmetric prepositions' by Nam (1995), since they denote a symmetric relation between the source and goal regions of a movement. For instance, John threw the ball over the fence refers to a symmetric spatial relation such that 'the goal location of the ball is on the other side of the fence from the source location, and vice versa.' Thus, it does not matter from which side of the fence the ball moved. Jackendoff (1990) uses various path functions for the symmetric prepositions, e.g., OVER, ACROSS, THROUGH (= VIA-IN), and VIA-NEAR, etc.1 The paper is organized as follows: Section 2 illustrates the Goal-Source asymmetry in syntax, specifically in the structures of Preposition Incorporation, Prepositional/Pseudo Passives, PP-dislocation and locative alternations. Section 3 shows the semantic contrast between Goal and Source PPs with respect to adverbial modification and aspectual composition. Section 4 proposes two distinct underlying base positions of Goal and Source locatives, and accounts for the syntactic and semantic contrast in terms of more fine-grained event structure. Further, we propose a set of mapping rules which link the locative PPs in event structure to their syntactic positions.

2. Goal-Source Asymmetry in Syntax 2.1 Directional vs. Non-directional PPs

There has been much work on the dichotomy directional vs. non-directional PPs. Recently, Koopman (1997), and Travis (2000) among others. Koopman argues for the need of two locative functional heads in Dutch, i.e., 'Path' for directional PPs and 'Place'
1 Jackendoff (1990:47) says the direct object of transitive pass is understood as the argument of this Path-function [VIA-NEAR]; it is neither Source nor Goal nor Theme in the usual sense. Further, he notes that the lexical verbs like pierce and jump lexically incorporate THROUGH (or VIA-IN) and OVER (or ACROSS), respectively. (i) The arrow pierced the target. (ii) John jumped the gorge.

- 3 -

for non-directional ones. She claims that prepositional PPs in Dutch contain a functional category Place, but postpositional phrases consist of a functional projection of Path which embeds a Place Phrase, as shown in (3). (3) Het vliegtuig is [PathP [PlaceP vlak onder de brug] door] gevlogen The airplane is right under the bridge] through] flown 'The airplane flew right under the bridge.' She argues for this structural configuration, illustrating various syntactic phenomena such as PP-Movement (pied-piping and PP-over-V movement), P-stranding and P-incorporation. Postpositional PPs in Dutch only have a directional reading, but prepositional PPs may have either a non-directional or a directional reading. And their syntactic behavior clearly shows that postpositional PPs are more integrated with the verb than prepositional PPs are. In other words, PPs with a directional reading are more integrated with the predicate than those with a non-directional reading are.2 Maienborn (2001), dealing with only non-directional locatives in German, proposes three syntactic base positions for the locative PPs. (4) a. In Argentina, Eva still is very popular. b. Eva signed the contract in Argentina. c. Eva signed the contract on the last page. The PP in Argentina in (4a) is classified as 'frame-setting modifier,' which often carries temporal reading, and the same PP in (4b) is called 'external modifier,' and on the last page in (4c) is called 'internal modifier.' Maienborn (2001) claims that the three types of locatives are generated in different base positions: The frame-setting modifier of (4a) is generated in periphery of TopP(topic phrase), and modifies the whole event of the sentence. The external locative
2 Kracht (2002) also identifies directional and 'Modalizer' and 'Localizer,' respectively. Thus below as a Modalizer Phrase (MP) containing a (i) The cat appeared [MP from [LP under non-directional senses in terms of he analyses the whole PP in (i) Localizer Phrase (LP). [DP the table]]]

- 4 -

of (4b) is generated in periphery of VP, and modifies the eventuality of the VP. Finally, the internal locative of (4c) is generated in periphery of V, and modifies the eventuality of the V.

2.2 Preposition Incorporation

Preposition incorporation reveals the Goal-Source asymmetry as well as the contrast between directional vs. non-directional locatives. Baker (1988) illustrates that the PPs of Dative and Goal are most common in preposition incorporation, and claims that the arguments associated with the applicatives are theta-marked ones, i.e., inner locatives in the sense of Hornstein and Weinberg (1981). (5-6) of Kinyarwanda are quoted from Kimenyi (1980), and (7) of Chichewa from Baker (1988). (5) a. Umukoobwa a-ra-som-a igitabo girl SP-PRES-read-ASP book 'The girl is reading the book.' b. Umukoobwa a-ra-som-er-a umuhuungu igitabo girl SP-PRES-read-for-ASP boy book 'The girl is reading the book for the boy.' (6) a. Abaana b-iica-ye ku meeza children SP-sit-ASP on table b. Abaana b-iica-ye-ho ameeza children SP-sit-ASP-on table 'The children are sitting on the table.' (7) a. Ndi-na-tumiz-a chipanda cha mowa kwa mfumu 1sS-PAST-send-ASP calabash of beer to chief 'I sent a calabash of beer to the chief.' b. Ndi-na-tumiz-ir-a mfumu chipanda cha mowa 1sS-PAST-send-to-ASP chief calabash of beer 'I sent the chief a calabash of beer.' The verbal complex of (5b) contains an applicative suffix -er denoting Benefactive role. The verbal complex in (6b) also contains an applicative suffix -ho 'on' instead of the lexical preposition ku 'on' in (6a). (7a) has a lexical preposition kwa 'to', but in (7b) the applicative suffix -ir 'to' (Goal) is incorporated into the verbal

- 5 -

complex. The following data from Kimenyi (1980) show that PI is impossible from the outer locatives. (8) a. Abaana b-iica-ye ku musozi children SP-sit-ASP on mountain b. *Abaana b-iica-ye-ho umusozi children SP-sit-ASP-on mountain 'The children are sitting on the mountain.' Baker (1988) and Kimenyi (1980) illustrate few source locatives. The following is the sole example of PI with a Source argument in Baker (1988:240). (9) Kambuku a-na-b-er-a mkango njinga. leopard SP-PAST-steal-APPL-ASP lion bicycle 'The leopard stole the bicycle from the lion.' Baker reports that the sentence also has a Benefactive reading: That is, 'The leopard stole the bicycle for the lion.' But the Source reading is possible because the Source argument is a true argument of the verb 'to steal.' That is, steal does not take a Goal argument, so *The man stole the book into his bag, but takes a Source as a true argument. Koopman (1997) shows that Dutch postpositions and particles can incorporate to V deriving a directional interpretation whereas prepositions cannot due to their non-directional reading. A prepositional PP may derive a goal directional reading, if it is selected by a motion verb. Thus (10a) has an incorporated (directional) preposition in between the auxiliary verb is and the main verb geklommen 'climb', and (10b) has a directional postposition door 'through' incorporated in gelopen after the auxiliary verb is 'be.' (10) a. omdat zij de boom is in geklommen because she the tree is in climbed 'because they climbed into the tree' b. omdat zij het bos is door gelopen because she the forest is through walked 'because she walked through the forest'

- 6 -

Further, unlike Goal directional PPs, Dutch Source directional PPs do not allow P-incorporation. Thus, the goal type particle heen may be incorporated to V as in (11a), but the source type particle vandaan in (11b) may not. (11) a. dat zij de jas over de stoel hebben heen gelegd that they the coat over the chair have prt put 'that they laid the coat over the chair' b. dat dit book (van) onder het bed is (?*vandaan) gekomen that this book from under the bed is from come. 'that this book came from under the bed' Notice that van 'from' in (11b) shows up as a preposition. The source PPs with van, however, can be dislocated by pied-piping and PP-over-V movement, while goal type PPs cannot. The following data are quoted from Koopman (1997, and pc). (12) Movement of non-directional PPs: boven in welke la heb jij de sokken gelegd up in which drawer have you the socks lay 'Up in which drawer did you lay the socks?' (13) Movement of directional PPs: a. *Onder welke brug door is het vliegtuig gevlogen Under which bridge through is the airplaine flown 'Under which bridge did the airplane fly?' b. *Welk bos in ben jij gelopen Which forest in are you walked 'Into which forest did you walk?' c. *Zij zijn gelopen het bos door [PP-over-V movement] they are walked the forest through 'They walked through the forest.' (14) Movement of Source directional PPs with van: a. van welke brug ben jij gelopen from which city are you walked 'From which city did you walk?' b. (?)Zij zijn gelopen van Amsterdam they are walked from Amsterdam 'They walked from Amsterdam.'

- 7 -

Munro (2000) illustrates that some verbs in Choctaw and Chickasaw can occur with more than one applicative prefixes. She claims there is a strong constraint against verbs with a total of more than four arguments. (15) show combinations of multiple applicatives in Choctaw: i.e., comitative and dative in (a), and comitative and benefactive in (b).3 The Chickasaw sentences in (16) also show combinations of applicatives among comitative, goal, source, and stative locatives. (15) a. Charles-at [sa-baa]-[chi]-taloow-aachi. Charles-nom [lsII-com]-[2sIII:dat]-sing-irr 'Charles will sing to you with me' b. Lynn-at kaah [chi-baa]-[ami]-chopa-tok. Lynn-nom car [2sII-com]-[1sIII:ben]-buy-pt 'Lynn bought a car for me with you' (16) a. [ibaa-in]-taloowa [with-to]-sing 'sing to with' (directional in- 'to') b. [aa-imaa]-chompa [in-from]-buy 'buy from in' (non-directional aa- 'in') c. ibaa-okaa-malli [with-into]-jump 'jump into with' (directional okaa- 'into') Munro notes that the order of the prefixes in the combinations, specifically in Chickasaw, is subject to the following constraint: Non-directional prefixes precede Source ones which precede Goal-directional ones. Thus the applicatives in (16), even though their combinations are somehow restricted, are prefixed to the verb in the following order: [aa-](non-directional locative) + [imaa-](source directional) + [okaa-](goal directional) + [in-](dative) + verb-root. This applicative prefix ordering suggests that the different locative argument/adjuncts occupy different syntactic positions, and further they modify different semantic domains.

3 The underlined vowels in (15a,b) are nasalized ones, which are phonemically contrastive to unmarked non-nasalized ones.

- 8 -

2.3 Prepositional (Peudo) Passives

Prepositional passives show a similar contrast between Goal and Source locatives. (17-19) illustrate active-passive pairs, where the passive sentences have a stranded preposition, and its object NP is promoted to the subject position. When a PP denotes a benefactive (17), goal (18), or comitative (19), its object NP is allowed to be the passive subject. (The data are quoted from Couper-Kuhlen 1979). (17) a. b. (18) a. b. (19) a. b. His surviving brother provided for John's widow. [benefactive] John's widow was provided for by his surviving brother. Several magistrates spoke to him. [goal] He was spoken to by several magistrates. Anyone cannot room with Martha. [comitative] Martha can't be roomed with by anyone.

When the PP denotes a circumstantial location or source, however, its NP is hardly found in the passive subject position. Thus, the (a)-sentences with a goal locative in (20-22) are acceptable, whereas the (b)-sentences with a source locative sound bad. (20) a. b. (21) a. b. (22) a. b. The store can be run to in a matter of minutes. [goal] *The store can be run from in a matter of minutes. [source] If the boat is jumped into it may capsize. [goal] *If the boat is jumped from it may capsize. [source] The house was moved into three weeks ago. [goal] ?*The house was moved from three weeks ago. [source]

Now consider the following minimal pairs, where the same PP may give a directional reading or a non-directional reading. (23) a. b. (24) a. b. The road could be driven across only at great risk *The road could be played across only at great risk The gate mustn't be gone beyond *The gate mustn't be played beyond

Couper-Kuhlen (1979:54) says "On the other hand, if some of these same examples are manipulated in order to express, say, motion +

- 9 -

direction (or resultative position) [e.g.., (23b) and (24b)] rather than locomotion + direction [e.g., (23a) and (24a)], then the notion in object position becomes one of location and a passive is no longer possible." Her distinction between motion and locomotion, however, does not seem to be responsible for the contrast here, since the sentence (25a) is ambiguous with the same motion verb: (i) 'the boy was running crossing the road,' and (ii) 'the boy was running on the other side of the road.' (25) a. The boy was running across the street. b. The road can be run across only at great risk. (25b) contains the same verb and preposition in passive, but the sentence is not ambiguous. It has a directional reading only, i.e., 'the road can be crossed by running only at great risk.' This contrast shows that non-directional PPs do not undergo passivization, stranding the preposition. There are, however, some apparent counterexamples where non-directional PPs allow passivization. (26) a. This house cannot be lived in any longer. b. This cart must not be sat in by more than two people at once. Kouper-Kuhlen (1979:64) says that (26a-b) are acceptable since the PPs carry an instrumental sense, i.e., the locations are used to serve a certain purpose. But the acceptability does not seem to depend on their instrumental reading but what matters here is their argumenthood. The PPs in (26) are semantically selected by the verbs live and sit, and their syntactic status is different from that of the PPs in the following. (27) a. John slept in New York yesterday. b. *New York was slept in by John yesterday. That is, the PPs in (26) are true arguments of the verbs live and sit, but the PP in (27) is not a true argument but an adjunct modifying the whole event 'John's sleeping yesterday.' Following Baker's (1988) distinction, the PPs in (26) are theta-marked by the

- 10 -

verb, while that of (27) is not.4

2.4 Movement and Ordering

We have another syntactic evidence revealing the contrast between Source and Goal PPs: That is, Source PPs can be easily dislocated, while Goal PPs cannot. The Source PP from Los Angeles in (28) can move to the front by Topicalization, whereas the Goal PP to Chicago resists to move. This suggests that the Goal PP behaves more like a true complement of the verb send than the Source PP does. (28) a. From Los Angeles John sent the letter to Chicago. b. ??To Chicage John sent the letter from Los Angeles. Now we note that a Source PP is more ready to scramble with a temporal/aspectual PP. Thus, the PP from the library in (29) can move over the durative adverbial for ten minutes, but the Goal PP to the library in (30) is not allowed to move over the time-frame adverbial in ten minutes. (29) a. b. (30) a. b. He ran from the library for ten minutes. He ran for ten minutes from the library. He ran to the library in ten minutes. ??He ran in ten minutes to the library.

Goal PPs, combining with a transitive verb, always specify the location or the movement of Theme argument, i.e., the argument in direct object position. So (31a) entails that 'Mary was in the garden' and (31b) 'the hay moved onto the truck.' (31) a. John saw Mary in the garden.
4 Let us note here another type of locative PP which is not easy to be passivized. The following contains an orientational locative which does not refer to a goal or a source but refers to an orientation of trajectory. Thus its passive counterpart (ii) sounds bad. (i) John advanced towards the house. (ii) ??The house was advanced towards by John.

- 11 -

b. John loaded the hay onto the truck. If the verbs take a Source PP, however, it may denote either the location of the subject argument or the location of the object argument. That is, from the rooftop refers to the location of John (the subject), and from the ground in (32b) refers to the source location of the hay. In other words, we can say that the Source PP from the rooftop in (32a) is 'subject-oriented' and the other PPs in (32a-b) are 'object-oriented.' (32) a. John saw Mary in the garden from the rooftop. b. John loaded the hay onto the truck from the ground. We note here that the PPs are not free in ordering: That is, if an object is followed by an object-oriented PP and a subject-oriented PP in English, the former always precedes the latter. Therefore, the two PPs in (32a) cannot scramble as in (33a), but those in (32b) can scaramble as in (33b). (33a) may have a reading where 'the rooftop was in the garden,' which is not the intended reading of (32a), though. (33) a. *John saw Mary from the rooftop in the garden. b. (?)John loaded the hay from the ground onto the truck. We have seen that Goal PPs are always oriented to object/theme argument but Source PPs may be oriented to subject argument. The data also show that two locative PPs oriented to the same argument (e.g., Theme) can change their positions, but those oriented to different arguments cannot.

2.5 Locative Alternations

English and many other languages allow locative alternations like the following: (34) a. Bees are swarming in the garden. b. The garden swarms with bees. (35) a. John sprayed paint on the wall.

- 12 -

b. John sprayed the wall with paint. (36) a. The woman embroidered flowers on the jacket. b. The woman embroidered the jacket with flowers. These alternation patterns have been attested and well described in many languages. In (34a), the intransitive verb swarm takes a locative PP in the garden, but the same location the garden shows up as a subject in (34b). (35) and (36) illustrate two more alternation patterns between two transitive structures: Locative arguments is not a direct object in (35a, 36a), but they show up as a direct object in the other structures (35b, 36b).5 Locative PPs involved in such alternations are mostly Goal-type locatives, i.e., the nouns of the PPs denote a goal/result location of the relevant argument, typically a Theme. Thus, (34-36) entail 'bees are in the garden,' 'paint ends up being on the wall,' and 'flowers come to exist on the jacket,' respectively. Syntactically, the locative arguments are promoted to subject in (34) or to direct object in (35-36), and we claim that the promotion should be subject to a syntactic constraint: That is, only V'-internal/inner locatives can be promoted by locative alternation. In a more general context, we claim that V-modifiers like Goal-type PPs allow locative alternation while VP-adjuncts like Source and Path-type PPs hardly do. Further, non-directional PPs - higher VP-adjuncts - do not participate in locative alternation, either.

3. Directional PPs in Event Structure 3.1 Lexical Complex Event Structure

Since Vendler (1967) characterized four classes of Aktionsarten, there have been many proposals on the complex lexical event structure: Dowty (1979), Jackendoff (1990), Grimshaw (1990), Parsons
5 There have been many proposals to account for these locative alternations. Pustejovsky (1991, 1995) accounts for the transitive alternations like the following in terms of HEAD underspecification of event structure. (i) The enemy sank the boat. (ii) The boat sank. Further, Lee et al. (1998) and Alsina (1999) extend the underspecification method to locative and causative alternations.

- 13 -

(1990), Pustejovsky (1991, 1995) among others. Most of them identify the structure of an event with the organization of the arguments in the clause, and extends the lexical event structure to compositional structures. The following illustrate some event structures of lexical verbs proposed in the literature. Since McCawley (1968) used CAUSE and BECOME in lexical decomposition of kill, i.e., 'CAUSE-BECOME-NOT-alive', they are accepted as primitives in many proposals. Thus, Dowty (1979) and Jackendoff (1993, 1990) take them as major primitive concepts in event/proposition structure. (37) Dowty 1979: open: [[DO(...)] CAUSE [BECOME[...]]] (38) Jackendoff 1990: open: [Event1 CAUSE ([Thing x], [Event2 GO ([Thing y], [Path TO [Property OPEN]] (39) Parsons 1990: close: (e)[Cul(e) & Agent(e, x) & (e')[Cul(e') & Theme(e', y) & CAUSE(e,e') & (s)[Being-closed(s) & Theme(s, y) & Hold(s) & Become(e',s)]]] (40) Pustejovsky 1991: John closed the door E:Transition = e1:Process = [act(j,the-door) & not-closed(the-door)] + e2:State = [closed(the-door)] Notice that the event structures of Jackendoff and Parsons are embedding/recursive ones: In Jackendoff's lexical conceptual structure of open, the outer event (Event1) embeds an inner event (Event2), and the outer event (e) of Parsons' representation (39) embeds an inner event (e'), which again contains a state (s). In addition to such recursive event structure, Pustejovsky proposes a parallel event structure: The whole event named 'transition' contains two sub-events e1:Process and e2:State which are conjoined in parallel, i.e., none of them embed the other. In section 4, we will adopt and extend the conjoined event structures in order to account for the event composition of locative PPs and predicates.

- 14 -

3.2 Adverbial Modification and Locative PPs in Event Structure

We argue that the various modes of locative PP semantics require a more fine-grained event structure. Alsina (1999), Tenny (2000), and Travis (2000) identify outer (causing) event and inner (caused/core) event. Pustejovsky (1995), Eckardt (1998) and Ernst (1998) account for (scopally) ambiguous adverbial modification in terms of event structure. Let us consider a few adverbs which can be interpreted ambiguously as in the following: (41) a. Harry departed the room rudely. (i) 'Harry's way of departing the room was rude.' (ii) 'the event of Harry's departing the room was rude.' (42) a. The police quickly arrested John. b. John clumsily spilled the beans. The three sentences above contains a manner adverb, thus (41.i) shows the manner reading of rudely, where Harry might have interrupted others by banging the door. We can easily get such manner readings in (42a-b), too. Further, the sentences have another reading where the adverbs modify the whole event, i.e., they are predicated of the whole event. Thus (41.ii) shows that rudely is predicated of the whole event. In (42a) quickly means 'without delay or hesitation' when it modifies the whole event, and clumsily in (42b) means that the whole event was clumsy. Pustejovsky (1991) represents the ambiguous readings in his event structure as the following. (43) Pustejovsky (1991): Harry departed the room rudely. a. E0:Transition / \ E1:Process E2:State / \ | MOD E1:Process [harry is-not-in the-room] | | [rudely] [harry departed]

- 15 -

b. / MOD | [rudely]

E0:Transition \ E0: Transition / \ E1:Process E2:State | | [harry departed] [harry is-not-in the-room]

(43a) represents the reading (41.i), i.e., the manner reading of rudely, and (43b) represents the reading of (41.ii) where the modifier (MOD) scopes over the whole event E0. Let us now see another adverb again, so called repetitive adverb, which can be interpreted ambiguously in the following sentences. They are quoted from Dowty (1979). (44) a. John closed the door again. b. All the king's horses and all the king's men couldn't put Humpty Dumpty together again. c. John fell asleep during the lecture, but Mary quickly shook him awake again. d. The book had fallen down, but John put it on the shelf again. According to Dowty (1979), (44a) above is ambiguous: (i) the event of John's closing the door is assumed to have occurred previously, (ii) the state of the door being closed is assumed to have existed previously, i.e., not necessarily as a result of John's action. Tenny (2000) calls the first reading 'repetitive', and the second 'restitutive' reading. In other words, the first reading indicates that the whole event is repeated, while the second reading means that only the result state of being closed is repeated. We have rather clear ambiguity in (44b-d), so (44b) implies either (i) the whole event denoted by the sentence (interpreted positively) previously occurred, or (ii) 'Humpty Dumpty is assumed to have previously been together in one piece, but not as the result of anyone's doing. This type of ambiguity in adverbial modification rudely, quickly, again naturally suggests that each of the adverbs should be generated in two distinct base positions, and the base positions can

- 16 -

fit into the extended VP structures: i.e., VP-internal subject structure of Koopman & Sportiche (1991); VP-shell structure of Larson (1988); and Hale and Keyser's (1993) division of L-syntax and S-syntax, among others. Further, we may find the positions of adverbials in Cinque's (1999) universal hierarchy of functional head projections illustrated as follows: (45) Cinque's universal hierarchy of clausal functional projections: [Mood-speech act [Mood-evaluative [Mood-evidential [Mod-epistemic [T(Past) [T(future) ... [Mod-necessity [Mod-possibility [Asp-habitual ... (i) [Asp-repetitive(I) [Asp-frequentative(I)... [Asp-celerative(I) [T(Anterior) [Asp-terminative ... [Asp-perfect(?) [Asp-retrospective [Asp-proximative [Asp-durative [Asp-generic/progressive ... [Asp-SgCompletive [Asp-PlCompletive [Voice (ii) [Asp-celerative(II) [Asp-repetitive(II) [Asp-frequentative(II) Cinque places the adverb again under two projections i.e., Asp-repetitive(I) on the line (i) above and Asp-repetitive(II) on the line (ii). He also identifies the positions of quickly at the two projections, i.e., Asp-celerative(I) and Asp-celerative(II). Now let us consider how Goal and Source PPs interact with the adverb again. Their syntactic behavior we discussed in section 2 suggests that Goal PPs should be syntactically much closer to the verb than Source PPs. Now considering their interaction with again, we claim that Goal PPs constitute a core event (i.e., result state) whereas source PPs do not. Therefore, again does not allow restitutive (narrow scope) reading with a Source PP. (46) a. b. (47) a. b. John drove to New York again. [ambiguous] John drove from New York again. [repetitive reading only] John sent the book to NewYork again. [ambiguous] John sent the book from NewYork again. [repetitive reading only] 'the event restitutive York is that John

Again in (46a) gives two readings: (i) repetitive reading of John's driving to New York is repeated,' and (ii) reading - 'the state of John's being at New resumed/restituted.' The second reading does not imply

- 17 -

drove to New York previously. (46b) however, only gives a repetitive reading, and does not give a restitutive reading, since the sentence lacks an expression that may denote a result state. The same constrast holds for (47a) and (47b). We will see shortly that the semantic contrast between Source and Goal PPs can be accounted for by assigning them two independent semantic scopes in the event structure.

3.3 Aspectual Division

Cinque (1999) and Travis (2000) identify (at least) two aspectual domains in syntax, which Tenny (2000) labels higher/viewpoint aspect and middle/situation aspect. We argue that Source-type PPs scope over the whole situation aspect, so they, unlike Goal PPs, do not shift the aspectual character (situation aspect) of the inner event denoted by the lower VP. (48) a. Mary ran (for ten minutes/*in ten minutes). b. Mary ran to the store (in ten minutes/*for ten minutes). c. He ran from the library (for ten minutes/*in ten minutes). (48a) denotes an atelic activity, which does not normally go with a time-frame adverbial like in ten minutes. When a Goal PP combines with the verb, however, it changes the aspectual character of the verb, so the sentence denotes a telic event of accomplishment. Thus, (48b) is fine with a time-frame adverbial but it is bad with a durative adverbial for ten minutes. Unlike Goal PPs, the Source PP from the library does not change the aspectual character of the verb, so (48c) behaves in the same way as (48a) does. This contrast between Goal and Source PP on aspectual shift suggests that the Goal PPs can be treated just like an internal argument which participate in aspectual composition. Tenny (1994), Verkuyl (1993), and Krifka (1995) report that an internal argument like (incremental) theme determines the aspectual character of the VP. Thus we have the following aspectual alternation: The quantized NPs like a house in (49a) and the whole tank of beer in (50a) make the VP denote a telic event, but the unquantized (bare plural or mass) NPs like houses in (49b) and beer in (50b) compose an atelic

- 18 -

event. (49) a. a'. b. b'. (50) a. a'. b. b'. Mary built a house in a year. ??Mary built a house for a year. Mary built houses for a year. *Mary built houses in a year. The men drank the whole tank of beer in a couple of hours. ??The men drank the whole tank of beer for a couple of hours. The men drank beer for a couple of hours. *The men drank beer in a couple of hours.

Let us assume that, in a fine-grained VP internal structure, Asp-head separates the lower and the higher VPs. Then, we propose that Goal PPs are generated under the AspP, while Source PPs are generated in a position higher than the AspP. We will implement the asymmetry in the extended VP structure and event structure proposed shortly in section 4.

3.4 Non-locative Source PPs

Source PPs headed by from in English often give a non-locative readings, thus the from-PPs below refer to Cause in (51-52) and Agent in (53). (51) a. Harry died from AIDS complications. [cause] b. We have reports of death from AID complication. (52) a. No damage was caused from the shooting. [cause] b. Preliminary estimates indicate damage from the freeze totals $385 million. (53) a. We have been promised from the top in Moscow that we will receive... [agent] b. challenge/support/donation/testimony/help/approval/threat from the company The Source PPs in (51) and (52) both denote a state or an event which caused a result state. In (51a) Harry's state of being under AIDS complications caused his death, and in (52a) the shooting event caused no damage. We also find in (51b) and (52b) the same

- 19 -

semantic relation between a predicative noun (reports or damage) and a Source PP. Thus the sentences take these PPs as composing a causing sub-event of their event structures. Further, (53a) shows that a Source PP can denote an Agent of passive sentences, and such Agent reading can be obtained in nominal constructions in (53b). If an Agent argument shows up in a complex event denoted by a transitive verb, it does not paly a role in its result state. Instead the Agent role is essential in its causing sub-event. We will see that these non-locative readings of Source PPs can be properly represented in the event structures proposed in 4.2.6

4. Proposal: Base Positions of Locative PPs and their Semantic Scope

Here we adopt the extended VP structure of Hale and Keyser (1993), and propose three base positions where Goal PPs, Source PPs, and non-directional PPs are generated. Further, extending Pustejovsky's (1991, 1995) event structure, we represent their semantic scope in event structure. A complex event contains at least two conjoined sub-events: One is normally a Process which denotes a causing sub-event, and the other is a State which denotes a result state.7

6 Source phrases may be used as an obligatory adjunct in the passives (i-ii) below. Grimshaw & Vikner (1993:143) claim that obligatory adjuncts in passive identify causing sub-event of the whole sentence as shown in (iii). (i) The homemade stove was built *(from the rims of coal truck tires). (ii) The five horses were brought *(from New Delhi). (iii) This house is built/designed/constructed *(by a French architect/yesterday/in ten days). 7 A typical complex event of causation contains a causing process and a caused (result) state, but this is not a structural constraint on complex event structures. The causing sub-event may be a state, or the caused sub-event may be a process. The former case will show up in 4.3 when we represent the event structure of (i) below, and some causative verbs like walk and jump give a complex event structure with a result process. (i) Harry died from AIDS complications. (ii) John walked the baby.

- 20 -

4.1. Goal PPs: Internal Locatives

Let us first consider Goal PPs like those in (54): the Goal PPs (PPG) are generated under the lower VP, where a Goal PP combines with V2 to form V2' as shown in (55). V2' may contain an internal argument (Theme). (54) a. John swam to the boat. b. Marta loaded the hay onto the truck. (55) [VP1 DP1 [V1' V1 [VP2 [V2' (DP2) V2 PPG]]]] As we have seen in section 3, Goal PPs are interpreted as composing a result state. A Goal PP generated under the lower VP specifies the final location of Theme argument, which shows up as a direct object of a transitive verb or as a subject of an intransitive verb. Then the result state will be composed of the Theme and the Goal. Thus, we represent the event structure of (54a) as (56) below. The lexical verb swim does not denote a complex event by itself, but the Goal PP extends the simplex event (E1) to a complex one with a result state (E2). (54a) entails the result state (E2) 'John was at the boat,' and the event structure of (54b) should entail 'the hay was on the truck.' (56) John swam to the boat. E0:Transition / \ E1:Process E2:State | | [john SWIM] [john BE-AT the-boat] We take the event structure (56) as a semantic structure which can be mapped to its syntactic VP structure of (55). Thus, we characterize the mapping as follows: (57) Mapping-1: PPs constituting a result state are generated in the lower VP. Notice that the Goal PP is then treated exactly like a resultative

- 21 -

phrase, which evidently forms a result state in the following sentences. (58) a. The potter baked the clay hard. b. She cooked the food brown. c. The dog barked the neighbors awake. That is, each of the resultative phrases in (58) denotes a predicate of a result state, and they are generated under the lower VP. As we discussed in 3.2, again modifies a result state to give a restitutive reading, then due to (57) Mapping-1 the adverb is also generated under the lower VP.

4.2. Source PPs: Intermediate Locatives

Now let us consider the Source PPs in (59) below: We claim that the Source PP adjuncts (PPS) should be generated under the higher VP, so they scope over V1' containing the lower VP2 as shown in (60). (59) a. John swam to the boat from the beach. b. Marta sent the book from Chicago. (60) [VP1 DP1 PPS [V1' V1 [VP2 [V2' (DP2) V2 ]]]] As we have seen in 3, the Source PPs do not compose the result state, so do not affect the aspectual character of the verb. The Source PP in (59a) indicates the initial point of John's movement, so it is represented as a modifier of the causing event in (61) below.8 (62) states the mapping relation between the semantic structure of (61) and the syntactic configuration of (60). (61) John swam to the boat from the beach.

8 Again, the meaning of the verb determines what entity is located by the source PP: (59a) implies John's change of location, and (59b) implies the book's change of location.

- 22 -

E0:Transition / \ E1:Process E2:State / \ | MOD E1 [john BE-AT the-boat] | | [from the beach] [john SWIM] (62) Mapping-2: PPs modifying a causing event are generated under the higher VP. As we have seen in 3, there are other adverbials that modify the causing sub-event (Process). For instance, subject-oriented adverbials like reluctantly in (63a), temporal (frame) adverbials like in an hour in (63b), and manner adverbials like clumsily in (63c). We claim that the adverbials, just like Source PPs, are also generated under the higher VP. (63) a. The man reluctantly sold the car to me. b. John painted a picture in an hour. c. The boy clumsily spilled the beans over the floor. Some verbs of removing like empty, remove, clear and wipe take a Source PP as a core argument as in (64). Then the PP should be generated in the lower V2' just like Goal PPs. (64) a. They emptied water from the tank. b. He wiped crumbs off the table. Source PPs may have a non-locative reading, so we noted in 3.4 that the PP from AIDS complications in (65) repeated below denotes the cause of Harry's death. That is, Harry's state of being under AIDS complications caused his death. Die lexically denotes an achievement event, which may involve an external cause (i.e., killing action) and an Agent. (65) does not express an external cause like 'killing' process, but has an internal cause, i.e., Harry's state of being under AIDS complications.9 The event structure of (65) is
9 (65) does not involve an active Agent role in the event, but an NP like death

- 23 -

represented as (66) below, where the Source PP composes the causing sub-event, i.e., E1:Sate = [harry BE-UNDER AIDS complications]. (65) (66) Harry died from AIDS complications. [cause] E0:Transition / \ E1:State E2:State | | [harry BE-UNDER AIDS compl] [harry BE dead]

4.3. Non-directional PPs: External Locatives

(67a,b) below contain a non-directional PP. The PP in the lake in (67a) locates the event of John's swimming to the boat, and at the meeting locates the event of Marta's meeting with Mary.10 We claim that the non-directional PPs (PPND) are generated as a VP1-adjunct as shown in (68). (67) a. John swam to the boat in the lake. b. Marta met Mary at the meeting. (68) [VP1 PPND [VP1 DP1 [V1' V1 [VP2 ... ]]]] Non-directional PPs denote the location of the whole event that VP1 denotes, thus (69) represents the event structure of (67a). (69) John swam to the boat in the lake. E0 / \ MOD E0:Transition | / \ [in the lake] E1:Process E2:State | | [john SWIM-ACT] [john BE-AT the-boat]

from the shooting overtly expresses an active causing event (Process) with a Source PP. 10 (67a) has another reading where in the lake modifies the boat within the same DP. This adnominal reading of the PP is not of our concern here, but Maienborn (2001) analyses this use of locative PPs as one of the three types of locative modifiers in German.

- 24 -

(69) implies that 'John was in the lake' and 'the boat was in the lake,' since the PP in the lake locates the whole event of the sentence. Now we have the following mapping rule for the non-directional locatives: (70) Mapping-3: PPs modifying the whole event are generated adjoined to the higher VP. In 3.2, we noted that again and quickly may be ambiguous with respect to their semantic scope. Thus the adverbs in (71) repeated below can modify either a sub-event (result state or process) or the whole event. When they modify the whole event, the PPs should be generated adjoined to the higher VP due to (70). Thus we can represent the ambiguity of again as in the following: (71) a. John drove to New York again. b. The police quickly arrested John. (72) a. John drove to New York again. [again in restitutive reading] E0:Transition / \ E1:Process E2:State | / \ [john DRIVE-ACT] MOD E2:State | | [again] [john BE-AT New York] b. John drove to New York again. [again in repetitive reading] E0 / \ MOD E0:Transition | / \ [again] E1:Process E2:State | | [john DRIVE-ACT] [john BE-AT New York] Let us note that locative PPs in a sentence initial position hardly give a directional reading. Thus (73a) and (74a) are ambiguous: they

- 25 -

have both a directional and a non-directional reading. But (73b) and (74b) lack a directional reading, so the PPs do not carry a goal directional reading. (73) a. b. (74) a. b. John slipped in the bathtub. In the bathtub John slipped. John jogged across the street. Across the street John jogged.

As illustrated in (75), the sentence initial position also hosts a sentence-level or a discourse-level adverbial (such as perspectival, intensional, speech act oriented, or frame-setting adverbials). (75a,b) are from Maienborn (2001). We are not dealing with these adverbials, since they are independent of the event structure proposed here. (75) a. In Argentina, Eva still is very popular. b. In Italy, Lothar bought his suits in France. c. Hopefully/Certainly, the war will end soon.

5. Concluding Remarks
Based on the extended VP-structure of Hale and Keyser (1993) and the event structure of Pustejovsky (1991, 1995), we proposed a more explicit mapping between syntax and semantics of directional PPs particularly Goal and Source PPs. We distinguished three syntactic base positions for locatives, and represented their semantic scopes in event structure: (i) goal locative PPs are generated under the lower VP and compose a result state sub-event, (ii) source locative PPs are generated under the higher VP and modify a process sub-event, and (iii) non-directional locative PPs are generated adjoined to the higher VP and scope over the whole event. The syntactic behavior of Source and Goal PPs discussed in section 2 suggests in general that Goal PPs have more integrity with the verb than Source PPs do. We illustrated their contrast in terms of Preposition Incorporation (2.2), Pseudo-passives (2.3), Movement (2.4), and Locative Alternation (2.5). The semantics of

- 26 -

Source and Goal locatives is characterized largely in terms of scope: That is, their scope properties in event structures are supported by the ambiguous readings of various adverbs (like again, quickly, and rudely) (3.2), and further by the clear contrast in their contribution to aspectual interpretation (3.3). In order to support the proposed account, we need further explore their syntactic and semantic characteristics in relation to wider range of PPs and adverbials. We have neither dealt with intensional locative PPs (e.g., frame-setting, perspectival, and speech act oriented locatives) nor with Path-type PPs (e.g., through the tunnel and over the bridge). But these PPs should be included in further research on locative modification.

References
Alsina, Alex (1999) 'On the Representation of Event Structure,' in Tara Mohanan and Lionel Wee (eds.) Grammatical Semantics Evidence for Structure in Meaning. CSLI, Stanford. pp. 77-122. Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing. Chicago University Press. Cinque, G. (1999) Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press. Oxford. Couper-Kuhlen, E. (1979) The Prepositional Passive in English. Max Niemeyer, Tuebingen. Davidson, D. (1967) 'The Logical Form of Action Sentences,' in N. Rescher (ed.) The Logic of Decision and Action. University of Pittsburg Press, Pittsburg. Reprinted in D. Davidson (1980) Essay on Actions and Events. Clarendon Press, Oxford. Dowty, D. (1979) Word Meaning and Montague Grammar. Reidel, Dordrecht. Eckardt, R. (1998) Adverbs, Events, and Other Things: Issues in the Semantics of Manner Adverbs. Tuebingen: Max Niemeyer Verlag. Ernst, T. (1998) 'Scope Based Adjunct Licensing,' in NELS 28. Amherst, GLSA. Ernst, T. (2000) 'Manners and Events' in J. Pustejovsky and C. Tenny (eds.) (2000): 335-358. Grimshaw, J. (1990) Argument Structure. MIT Press. Cambridge, MA. Grimshaw, J. & S. Vikner. (1993) 'Obligatory Adjuncts and the Structure of Events,' in E. Reuland and W. Abraham (eds.) Knowledge and Language II: Lexical and Conceptual Structure: 143-55. Kluwer Publishers. Hale, K. and S. J. Keyser (1993) 'On Argument Structure and the Lexical Expression of Syntactic Relations,' in K. Hale and S. J. Keyser (eds.) The

- 27 -

View from Building 20. MIT Press. Cambridge, MA. Hale, K. and S. J. Keyser (1993) Prolegomenon to a Theory of Argument Structure. Linguistic Inquiry Monograph 39. MIT Press. Cambridge, MA. Hornstein, N. and A. Weinberg (1981) 'Case Theory and Preposition Stranding' Linguistic Inquiry 12: 55-92. Jackendoff, R. (1983) Semantics and Cognition. MIT Press. Cambridge, MA. Jackendoff, R. (1990) Semantic Structures. MIT Press. Cambridge, MA. Kimenyi, A. (1980) A Relational Grammar of Kinyarwanda. University of California Press, Berkeley. Koopman, H. (1997) 'Prepositions, Postpositions, Circumpositions and Particles: The Structure of Dutch PPs' ms. UCLA. Koopman, H. & D. Sportiche (1991) 'The Position of Subjects,' Lingua 85: 211-258. Kracht, M. (2002) 'On the Semantics of Locatives' Linguistics and Philosophy 25: 157-232. Krifka, Manfred (1992) 'Thematic Relations as Links between Nominal Reference and Temporal Constitution,' in I. Sag and A. Szabolcsi (eds.) Lexical Matters. CSLI Publications. Stanford, California. Larson, R. (1988) 'On the Double Object Construction,' Linguistic Inquiry 19: 335-392. Lee, C., S. Nam, and B. Kang (1998) 'Lexical Semantic Structure for Predicates in Korean,' in J. Bos and P. Buitelaar (eds.) Lexical Semantics in Context. European Summer School in Logic, Language and Information. U. of Saarbruecken, Germany. Maienborn, C. (2001) 'On the Position and Interpretation of Locative Modifiers,' Natural Language Semantics 9: 191-240. McCawley, J. (1968) 'Lexical Insertion in a Transformational Grammar without Deep Structure,' in B. Darden, C.N. Bailey, and A. Davidson (eds.) Fourth Regional Meeting of the Chicago Linguistics Society: 71-80. University of Chicago. Munro, P. (2000) 'The Leaky Grammar of the Chickasaw Applicatives,' in Arika Okrent and John P. Boyle (eds.) The Proceedings from the Main Session of the Chicago Linguistic Society's Thirty-sixth Meeting. Volume 36-1: 285-310. Chicago Linguistic Society. Chicago. Nam, S. (1995) Semantics of Locative Prepositional Phrases in English, Doctoral dissertation. UCLA. Parsons, T. (1990) Events in the Semantics of English: A Study in Subatomic Semantics, Number 19 in Current Studies in Linguistics, MIT Press. Cambridge, MA. Pustejovsky, J. (1991) 'The Syntax of Event Structure,' Cognition 41: 47-81. Pustejovsky, J. (1995) The Generative Lexicon. MIT Press. Cambridge, MA. Pustejovsky, J. and C. Tenny (eds.) (2000) Events as Grammatical Objects. CSLI Publications. Stanford, California.

- 28 -

Ritter, E. & S. Rosen (2000) 'Event Structure and Ergativity,' in J. Pustejovsky and C. Tenny (eds.) (2000): 187-238. Smith, C. (1991) Parameter of Aspect. Kluwer Academic Press. Dordrecht. Tenny, C. (1994) Aspectual Roles and the Syntax-Semantics Interface. Kluwer Academic Publishers. Dordrecht. Tenny, C. (2000) 'Core Events and Adverbial Modification,' in J. Pustejovsky & C. Tenny (eds.) (2000): 285-334. Travis, L. (2000) 'Event Structure in Syntax' in J. Pustejovsky & C. Tenny (eds.) (2000): 145-185. Vendler, Z. (1967) Linguistics in Philosophy. Cornell University Press. Ithaca, New York. Verkuyl, H. (1993) A Theory of Aspectuality. Cambridge University Press.

Department of Linguistics, Seoul National University Sillim-dong San 56-1, Kwanak-ku, Seoul, 151-742, KOREA E-mail: nam@snu.ac.kr webpage: http://cl.snu.ac.kr/nam

- 29 -