The Event Structure of English Prepositional Resultatives

Edward Kenschaft
LING879 Seminar in Semantics, Fall 2005

This is a work in progress, containing sections which were not in my original class paper.

• Abstract
• Background
• Semantic Interpretation
• Event Complexity
• Simple Causatives
• Complex Causatives
• Path Resultatives
• Typology of Resultatives
• Prepositional Resultatives
• Transitive, unselected, object-control
• Transitive, selected, object-control
• Intransitive, object-control
• Transitive, selected, subject-control
• Intransitive, subject-control
• Transitive, unselected, subject-control
• Conclusions
• Future Work
• References

Abstract

A complex causative such as (01) can be analyzed (Williams 2005) as a complex predicate (means + result), with a single complex event e containing (at least) two subevents M and R, and a causative relationship between them.

(01)	Al pounded the cutlet flat.

In this example, pound is the means predicate M, flat is the result predicate R, and the complex event e can be roughly described as pound-flat.

Technically, a distinction should be made between, for instance, the result event e2 and the result predicate R with which is it associated.  I ignore this distinction for simplicity of presentation.

Likewise, one should distinguish between the result event flatten and the result state flat, but I will not concern myself with this distinction.

Building on (Pietrosky 2005), Williams argues that arguments – in this case, agent al and patient the cutlet – attach to the complex event e, not to the subevents.  For the purpose of discussion, I will refer to an analysis involving a predicate such as pound-flat as compound-complex, and an analysis with a complex predicate but no such compound event as analytic-complex.

Could a compound-complex analysis apply to complex predicates with other types of resultative phrases, particularly (English) prepositional phrases, as in (02)?

(02)	a.	John hit the ball (out of the stadium) (with an aluminum bat) (at 100 mph).
	b.	Sue tickled her daughter (into a frenzy) (with a feather).
	c.	Gretchen swam across the river.
	d.	Al pounded the cutlet into a pancake.
	e.	Al pushed the perambulator across the park.
	f.	Al (inadvertently) pounded the bottle off the table.
	g.	Al (accidentally) knocked Bill over the rail.
	h.	Harry coughed himself into insensibility.
	i.	Bill sneezed out the window.
	j.	John danced mazurkas across the room.
	k.	The car honked down the road.

Evidence will show that an analytic-complex event structure is never the best analysis.  Some cases will strongly favor a compound-complex event structure, while others leave open the possibility of a simple event structure.

All data judgments are my own.

Background

Semantic Interpretation

I assume a generally neo-Davidsonian, conjunctivist framework (e.g. Carlson 1984, Landman 2000, Pietrosky 2005).  I will not elaborate further on this.

Event Complexity

Various arguments can be used to determine whether a given clause has a simple or complex event structure.  Some are more compelling than others.

Impoverished Lexicon

Lacking evidence to the contrary, we would like to assume that the lexicon is as simple as possible.  In particular, we propose that a verb instantiates the same lexical item regardless of the number of syntactic or semantic arguments, e.g. pound in (03).

(03)	a.	Al pounded the cutlet.
	b.	Al pounded the cutlet with a hammer.
	c.	Al pounded the cutlet flat.
	d.	Al pounded the cutlet flat with a hammer.
	e.	Al pounded the cutlet flat for Stella on the table with a hammer.

This principle is widely disputed (e.g. Bresnan 1982; Klein & Sag 1985; Kim, Srinivas & Trueswell 2000; Joshi 2004).  I will not accept it as gospel truth, but I will appeal to it as possible evidence.

Thematic Uniqueness

A popular tradition (e.g. Carlson 1984, 1998; Krifka 1998; Landman 2000) suggests a principle of thematic uniqueness.  This principle is stated with various formalisms, but indicates roughly that for any simple event e and thematic role r, there can be no more than one participant in role r of event e.  This principle can be used to differentiate subevents; e.g. if two participants p1 and p2 are both agents of the (complex) event, there must be at least two subevents.

For instance, (04) can be analyzed as a simple event, where john is the agent, a hot dog is the theme, and at zab's eatery is a locative (or whatever you want to call it).

(04)	John ate a hot dog at Zab's eatery.

On the other hand, (05) must describe two distinct events, since there are two agents and two themes.

(05)	John ate the hot dog and Betty ate the apple.

Note that it would be perfectly reasonable to describe this as a single complex event with two subevents, say, if john and betty are having a snack together.

At first glance, (06) might appear to be a single event, with an agent sam, a theme car, and a source (or whatever) harry.

(06)	Sam bought a car from Harry.

Upon reflection, however, it must be acknowledged that both sam and harry are agents, volitional participants.  By the thematic uniqueness principle, this requires (at least) two subevents, one for each agent.  These subevents might be conceptualized as a buying event and a selling event, differing only in perspective.

Note that this analysis contradicts (Dowty 1989), who claims that the description of an event cannot depend on perspective. 

Carlson (1998) begins to build a theory of plural entities, further developed by Landman (2000).  This allows single-event interpretations of sentences such as (07).

(07)	a.	Bob and Sally went to the store.
	b.	Jacob and Esau met on the plain.

Pending evidence to the contrary, I will take thematic uniqueness (or lack thereof) as a comparatively reliable indicator of thematic structure.

?? Argument per Subevent

Levin & Rappaport Hovav (1999) claim that all complex events are causative, and all causatives are transitive.  Any clause with only a single lexical argument, therefore, must indicate a simple event structure, as in (08).

(08)	The clothes steamed dry.

Enough evidence (e.g. Williams 2005) casts doubt on this hypothesis to prevent my relying upon it; however, I will consider it as a possible indicator.

Pervasive Theme

Levin & Rappaport Hovav (1999) suggest that for a resultative to be considered a single event, the result must be predicated upon an entity which is a participant in both subevents.  Thus, (09a) could be a single event, since the cutlet participates in both pound and flat, but (09b) would have to be considered a complex event, since no entity – particularly, not the theme us – participates in both talk and into a stupor.

(09)	a.	Al pounded the cutlet flat.
	b.	The professor talked us into a stupor.

I see no reason to dispute this.  The inverse claim – that a pervasive theme necessarily implies a simple event structure – is unclear, and is not stated by L&RH.

?? Paraphrase

Levin & Rappaport Hovav (1999), following (Parsons 1990), suggest a paraphrase test for event complexity, as in (10).

(10)	a.		The clothes steamed dry.
	b.		The clothes became dry by steaming.
	c.		The pond froze solid.
	d.	#	The pond became solid by freezing.

According to L&RH, (10b) is a perfectly natural paraphrase of (10a), while (10d) is a totally unacceptable paraphrase of (10c).  Their conclusion is that (10c) consists of only one event, while (10a) is composed of multiple subevents.

  Using a principle known as event coidentification, they go on explain that (10a), although composed of multiple subevents, yet has a simple event structure.  They use the same principle to suggest, for example, that both sentences in (11) refer, in fact, to a single event.

(11)	a.	Robin danced out of the room.
	b.	Robin went out of the room dancing.

I find their evidence for coidentification unconvincing.  I don't see how multiple subevents could combine into anything other than a complex event.  Furthermore, the judgment that (10d) is drastically worse than (10b) is not obvious to me.  My inclination is to take the Paraphrase condition with a grain of salt.

?? Lexical Entailment

Levin & Rappaport Hovav (1999) suggest a similarity between the two sentences in (12), where the result predicate specifies an endpoint on a scale already lexicalized in the verb.

(12)	a.	The bag burst open.
	b.	The pond froze solid.

They argue that this property indicates a simple event structure, and that lack of such a relationship indicates a complex event structure.  They also claim that this property correlates with the Paraphrase condition.

The claim that such a predicate lexically entails an endpoint that can be readily dropped, as in (13), is difficult to confirm or deny.

(13)	a.	The bag burst.
	b.	The pond froze.

Furthermore, L&RH go on to explain (via coidentification) why the complex structure describes only a single event, which I find difficult to accept.  All told, I consider this condition as L&RH employ it to be highly suspect.

On the other hand, the inverse principle seems more reasonable – that a result predicate which introduces a participant not entailed by the means predicate necessarily introduces a new subevent.  This reasoning is related to that of the Impoverished Lexicon, but is not entirely redundant with it, as we will see in some of our examples.

? Temporal Dependence

Levin & Rappaport Hovav (1999) note that in an intransitive resultative, temporal adverbials must apply to the entire complex event (or to both subevents), rather than to just one of its subevents, as in (14).

(14)	a.		Lewis ran quickly to the theater.
	b.	#	Lewis ran quickly to the theater, but it took him a long time to get there since he took a circuitous route.
	c.	#	Lewis ran quickly to the theater, even though he ran slowly, because the theater was only a half-block away.

L&RH did not include (14c).

This is in contrast to reflexive resultatives, as in (15).

(15)	a.	Peter quickly read himself into an inferiority complex.
	b.	Peter quickly read himself into an inferiority complex, by slow, deliberate reading of his classmates' theses.

L&RH's actual example was, "Peter quickly read himself into an inferiority complex, after a few slow deliberate readings of his classmates' theses."  I think (15b) is a better parallel for (14c), and hence better supports their point.

From this and other observations they conclude (16).

(16)	For intransitive resultatives, but not for reflexive resultatives:
	a.	The two subevents are temporally dependent, i.e.:
		i.	They occur over the same timespan.
		ii.	They unfold at the same rate.
	b.	The event structure is simple, i.e. noncausative.

However, Williams (2005) demonstrates the same temporal dependence for so-called complex causatives, as in (17), which to the best of my knowledge are uncontroversially recognized as both complex and causative.

(17)	a.		Al quickly pounded the cutlet flat.
	b.	#	Al quickly pounded the cutlet flat, although it took a long time for the cutlet to become flat.

This casts doubt on the claim that temporal dependence is truly an indicator of simple event structure. 

Furthermore, it is not even clear that the putative subevents in (14a) are temporally dependent, given that (18) is perfectly acceptable.

(18)	Lewis ran quickly to the theater, through the main hall, and out the back door.

It's not as if the run event necessarily culminates in the to the theater result.

It is possible that the inverse principle still holds:  If two alleged subevents are not temporally dependent, this implies a complex event structure.  If so, this might suggest that (14a) is, in fact, complex, contrary to L&RH.

Simple Causatives

Pietrosky (2005) argues that a morphologically simple causative clause such as (19) is best analyzed as a complex event which terminates in the result event – in this case, in the boiling of the soup.

(19)	Pat boiled the soup.

The meaning of (19) can then be roughly described as (20).

(20)

∃e1 ∃e2 [ terminater(e1,e2) & boiling(e2) & agent(e1, pat) & theme(e1, the soup) ]

compound- complex

Under this compound-complex analysis, both participants are arguments of the larger event e1.  The corollary fact theme(e2, the soup) follows by entailment.

Contrast this with an earlier analytic-complex approach (e.g. Parsons 1990) which would have analyzed (19) as something like (21).

(21)

∃e1 ∃e2 [ cause(e1,e2) & boiling(e2) & agent(e1, pat) & theme(e2, the soup) ]

analytic- complex

Here there is no complex event e with two subevents.  Rather, there are two independent events e1 and e2, with a causal relationship between them.

Pietrosky offers the following arguments to distinguish between (20) and (21).

Coparticipants

There is an intuitive sense in (19) that pat and the soup are coparticipants in a single event.  This intuition is reflected in (20), but not in (21).

Immediacy of Causation

It can be readily seen that (19) is not equivalent to (22).

(22)	Pat did something that caused the soup to boil.

For instance, take scenario (23).

(23)	Pat was walking through the kitchen, carrying a load of laundry. Pat accidentally bumped the stove without realizing it, and turned on the switch for the left rear burner. After Pat left the room, a pot of soup on the left rear burner came to a boil.
	a.		Pat did something that caused the soup to boil.
	b.	#	Pat boiled the soup.

Here, (23a) is clearly true, whereas most people would say (23b) is false, or at least bizarre.

There are various ways to exclude this scenario under (20).  For instance, one could argue that there is no single event e1 which has agent(e1, pat) & theme(e1, the soup).  Or, one could claim that there is an event e1 with participants pat and the soup, but pat isn't really the agent, because the effect is not volitional.  (Relaxing the agentivity requirement would, in fact, explain why some people might rate (23b) as true.)

The only way to exclude this scenario under (21) is to define cause as 'cause in just the right way'.  Even if we allow for the possibility of making this definition appropriately precise, it is then difficult to see how a true analysis for (23a) could be articulated.

Temporal Modification

Pietrosky notes that temporal modification must apply to the event as a whole, not to its individual subevents.

For instance, take scenario (24).

(24)	Before Pat goes to bed Monday night, he sets the timer on his stove to turn on the next morning. At 6:00 Tuesday morning, the timer goes off, and the soup comes to a boil.
	a.		Pat did something on Monday that caused the soup to boil on Tuesday.
	b.	#	Pat boiled the soup on Monday.
	c.	#	Pat boiled the soup on Tuesday.

The temporal modifier only makes sense when it applies to the entire event of causation.

This behavior follows if there is one atomic event e1 which can take modifiers, as in (20).  It is difficult to explain with the analysis in (21), where two independent events e1 and e2 could presumably take modifiers independently.

Indeed, other adverbial modifiers would appear to apply to only one subevent, as in (25).

(25)	Pat carefully boiled the soup.

Intuitively, it was not the boiling that was careful, but rather Pat's (unspecified) initiating action.  This observation supports the claim that this is a complex event, not a simple event.

Complex Causatives

Williams (2005) extends the compound-complex analysis to complex causatives such as (01), repeated here as (26), with roughly the meaning in (27a).  Earlier analytic-complex analyses (e.g. Parsons 1990) would have represented the meaning as something like (27b).

(26)		Al pounded the cutlet flat.
(27)	a.	∃e ∃M ∃R [ cause(e,M,R) & pounding(M) & flat(R) & agent(e, al) & theme(e, the cutlet) ]	compound- complex
	b.	∃M ∃R [ cause(M,R) & pounding(M) & flat(R) & agent(M, al) & theme(M, the cutlet) & theme(R, the cutlet) ]	analytic- complex

The same arguments apply as for simple causatives.

Coparticipants

There is an intuitive sense in (26) that al and the cutlet are coparticipants in a single event.  This intuition is reflected in (27a), but not in (27b).

Immediacy of Causation

Although not as straightforward as the simple case, it can be shown that (26) is not equivalent to (28).

(28)	Al pounded the cutlet, and this caused the cutlet to become flat.

Take scenario (29).

(29)	Al works at a fish market with several subordinates.  He finds it easier to direct his subordinates with hand gestures rather than words. Al wants a cutlet to be flattened.  He indicates this to Sam by pounding his fist into the cutlet. Following Al's direction, Sam picks up a huge block and drops it on the cutlet, thus flattening it.
	a.		Al pounded the cutlet, and this caused the cutlet to become flat.
	b.	#	Al pounded the cutlet flat.

This dichotomy is reflected naturally under analysis (27a), which would be false because there is no single event e of which al is the agent, which involves a pounding and a flattening of the cutlet.

In contrast, (27b) would appear to be true, reflecting the meaning of (29a).

Temporal Modification

Williams notes that it is impossible to attach a temporal modifier to the M event alone, without modifying the entire event e.  For instance, (30a) can only mean (30b), even though the meaning expressed in (30c) is perfectly reasonable.

(30)	a.		Al quickly pounded the cutlet flat.
	b.		Al pounded the cutlet flat, and the cutlet was flattened quickly.
	c.	#	Al pounded the cutlet quickly, although it took a long time for the cutlet to become flat.

This interpretation follows naturally given a compound-complex analysis, but is difficult to explain given an analytic-complex analysis.

Telicity

Another argument uses a common test for telicity, as in (31).

(31)	a.	Al will pound the cutlet (for two hours, #in two hours). [repetitive reading]
	b.	The cutlet will be flat (for two hours, #in two hours).
	c.	The cutlet will become flat (#for two hours, in two hours).
	d.	Al will flatten the cutlet (#for two hours, in two hours).
	e.	Al will pound the cutlet flat (?#for two hours, in two hours).

(31e) demonstrates that the complex predicate pound-flat is telic.  This is straightforward with analysis (27a), but is difficult to explain with analysis (27b).  What predicate does in two hours attach to?  It can't be M-pound, given (31a).  It can't be R-flat, given (31b).  But there is no other event to attach to.  The only apparent explanation is an implicit event flatten as in (31c,d), but that would suggest a highly complex hidden derivation.

Path Resultatives

Beavers (2002) builds on (Krifka 1998) in suggesting that a motion verb such as walk in (32) has a semantic path argument p.  The verb also entails a movement event e which is homomorphic to p.  Since p always has at least two sub-parts, source and goal, e must likewise have at least two subevents.

(32)	Bill walked to the store.

Beavers continues, in keeping with long tradition, to suggest that PP resultatives and other change-of-state predicates behave as predicates of abstract motion.  The context is a discussion of to and into with non-motion verbs indicating abstract motion, as in (33).

(33)	a.	Tatyana Tarasova coached skaters to 31 gold medals.
	b.	I was startled into indiscretion.

These abstract uses of prepositional resultatives are also analyzed in terms of an abstract path p.  As with motion verbs, p has at least two sub-parts, indicating that the event e also has at least two subevents.

Beavers (2004) extends the argument to all change-of-state predicates, as in (34).

(34)	a.	John ate the apple.
	b.	John wiped the table clean.
	c.	John hiked the Barton Springs trail.
	d.	John pounded the metal flat.

Every change-of-state predicate, he argues, entails an implied scale of change s.  In (34a), s measures the consumption of the apple; in (34b), s measures the cleanliness of the table; and so on.  This scale of change s behaves just as the abstract path p in (Beavers 2002), entailing a homomorphism with event e.  Again, since s has at least two sub-parts corresponding to the start and the end, e also has at least two subevents.

Typology of Resultatives

Goldberg & Jackendoff (2004) build on (Rappaport Hovav & Levin 2001) to identify the following (mostly) independent dimensions of variation in resultative sentences.

(38)	a.	resultative phrase type: AP vs. PP
	b.	configuration type: property vs. spatial
	c.	intransitive vs. transitive
		i.	within transitive: selected vs. unselected
			1.	within unselected: normal vs. fake reflexive
	d.	control
		i.	within transitive: subject vs. object
		ii.	within intransitive: explicit vs. implicit

Some of their examples are in (39).

(39)	a.		The pond froze solid	AP, property, intransitive, explicit-control
	b.		Bill rolled out of the room	PP, spatial, intransitive, explicit-control
	c.		Bill sneezed out the window.	PP, spatial, intransitive, implicit-control
			The toilet leaked through the floor to the kitchen below.
			Bill ate off the floor.
	d.		Bill broke the bathtub into pieces.	PP, property, transitive, selected, object-control
			Bill broke the bathtub.
	e.		The professor talked us into a stupor.	PP, property, transitive, unselected, object-control
		*	The professor talked us.
	f.		Harry coughed himself into insensibility.	PP, property, transitive, unselected, fake reflexive, object-control
		*	Harry coughed himself.
		*	Harry coughed us into insensibility.
	g.		Bill followed the road into the forest.	PP, spatial, transitive, selected, subject-control
			We drove Highway 5 from SD to SF.
			John danced mazurkas across the room.
	h.		The car honked its way down the road.	PP, spatial, transitive, unselected, subject-control
		*	The car honked its way.
			Bill whistled his way past the house.
		*	Bill whistled his way.

I added (39h) to complete the paradigm in (42).  The 'way' construction appears to be the only licenser of this configuration in English.

G&J's analysis for virtually (not quite) all such constructions can be summarized in (40).

(40)	a.	A resultative contains two separable subevents, one (M) introduced by the verb, the other (R) introduced by the resultative construction.
	b.	The relationship between M and R is either noncausative (transitive subject-control, intransitive explicit-control) or causative (transitive object-control, intransitive implicit-control).
	c.	M is either the (noncausative) means by which R takes place, or the (causative) cause of R taking place.
	d.	The (a)telicity of R determines the (a)telicity of the overall event (c.f. Beavers 2004).

There is considerably more to G&J's analysis, but it is not directly relevant to this discussion.

My analysis is only interested in PP constructions.  Following Beavers (2002, 2004), I will consider property resultatives to indicate abstract motion, thus eliminating the configuration distinction.  I will also ignore the normal vs. fake reflexive distinction, since it has no bearing on this discussion.

My hypothesis is that fake reflexives will behave precisely the same as other unselected transitives, but I will not explore this here.

As seen in (40b), the various types of control can be lumped into noncausative vs. causative; however, I will instead call these subject-control vs. object-control, since I think these terms better describe the obvious facts.  Thus, we are left with the distinctions in (41), exemplified in (42).

(41)	a.	intransitive vs. transitive
		i.	within transitive: selected vs. unselected
	b.	subject-control vs. object-control

(42)	a.	Bill rolled out of the room	intransitive, subject-control
	b.	Bill sneezed out the window.	intransitive, object-control
		The toilet leaked through the floor to the kitchen below.
		Bill ate off the floor.
	c.	Bill broke the bathtub into pieces.	transitive, selected, object-control
	d.	The professor talked us into a stupor.	transitive, unselected, object-control
		Harry coughed himself into insensibility.
	e.	Bill followed the road into the forest.	transitive, selected, subject-control
		We drove Highway 5 from SD to SF.
		John danced mazurkas across the room.
	f.	The car honked its way down the road.	transitive, unselected, subject-control
		Bill whistled his way past the house.

Prepositional Resultatives

For each configuration based on (41), we will examine evidence to determine:

• Event Complexity: Is the event simple or complex?
• Event Structure: If complex, is the event compound-complex or analytic-complex?

For reasons which should become apparent, we will address the cases in a different order than they are listed in (42).

Transitive, unselected, object-control

(43)	a.	The professor talked us into a stupor.
	b.	Harry coughed himself into insensibility.
	c.	Al pounded the bottle off the table.

This is perhaps the most straightforward case to analyze, because the means and result have no thematic participants in common.

Let's look at (43a).  Possible rough interpretations are in (44).

(44)	a.	∃e [ action(e, talk) & goal(e, into a stupor) & agent(e, the professor) & theme(e, us) ]	simple
	b.	∃M ∃R [ cause(M,R) & action(M, talk) & goal(R, into a stupor) & agent(M, the professor) & theme(R, us) ]	analytic- complex
	c.	∃e ∃M ∃R [ cause(e,M,R) & action(M, talk) & goal(R, into a stupor) & agent(e, the professor) & theme(e, us) ]	compound- complex

Event Complexity

Impoverished Lexicon

talk does not typically take a theme, and cannot take a theme without a goal, as illustrated in (45).

(45)	a.		The professor talked.
	b.	*	The professor talked us.

To avoid positing a distinct homomorphic lexeme talk<agent,theme,goal>, we posit another subevent R other than talk, which introduces the theme and goal in (43a).  This therefore suggests a complex event structure.

Thematic Uniqueness

Not relevant: There is only one agent, one theme, one goal.

?? Argument per Subevent

Not relevant: Multiple arguments allow, but do not entail, multiple subevents.

Pervasive Theme

The theme of R is us.  This is not a participant of talk (assuming an impoverished lexicon).  Hence, the event is complex.

?? Paraphrase

One might paraphrase (43a) as (46).

(46)	We went into a stupor because of the professor talking.

To the extent that one is convinced by this principle, this suggests a complex event structure.

?? Lexical Entailment

talk does not lexically entail a theme and goal.  Hence these participants must belong to a subevent distinct from that introduced by talk; hence the event is complex.

? Temporal Dependence

The two putative subevents appear to be temporally dependent, at least in the same sense as (14a), as shown in (47).

(47)	a.		The professor talked us into a stupor.
	b.	?#	The professor quickly talked us into a stupor by talking slowly.
	c.	#	The professor quickly talked us into a stupor, although it took a while.
	d.		The professor talked us into a stupor, and then a deep sleep, and even after we woke up and left.

If (47b,c) are bad, this demonstrates that temporal adverbials attach to the complex event, not the individual subevents.  (47d), however, suggests that the two subevents do not need to be temporally dependent.  I would say this condition is inconclusive.

Complexity Summary

Altogether, the evidence is quite strong that the event in (43a) is complex.

Event Structure

Having determined that the event structure is complex, we now must decide whether it is analytic-complex (44b) or compound-complex (44c).

Coparticipants

There is an intuitive sense in which the professor and us are involved in the same event.  This suggests a compound-complex structure such as (44c).

Immediacy of Causation

The question is whether (43a) is equivalent to (48).

(48)	The professor talked, and this caused us to go into a stupor.

Picture scenario (49).

(49)	The professor speaks a command to the orderly. As a result, the orderly injects us with sodium pentathol. As a result, we go into a stupor.
	a.		The professor talked, and this caused us to go into a stupor.
	b.	#	The professor talked us into a stupor.

(43a) apparently requires a more immediate cause than (48).  This favors a compound-complex analysis.

Temporal Modification

As noted in (47c), it does not appear to be possible to attach a temporal modifier to the M event alone.  This supports the compound-complex analysis.

Event Structure Summary

Altogether, the evidence is quite strong that the event in (43a) is compound-complex, with a meaning as in (44c).

Transitive, selected, object-control

(50)	a.	Bill broke the bathtub into pieces.
	b.	John hit the ball out of the stadium.
	c.	Sue tickled her daughter into a frenzy.
	d.	Al pounded the cutlet into a pancake.
	e.	Al pushed the perambulator across the park.
	f.	Al knocked Bill over the rail.

The sentences in (50) form a natural class only if one accepts the methodology of Beavers (2002) in treating PP change-of-state resultatives as abstract predicates of motion.  Even then, they may need to be differentiated based on their aspectual qualities, as we will see below.

The main quality that distinguishes (50) from (43) is that the means and result share a common thematic argument.

Let's look at (50c).  Possible rough interpretations are in (51).

(51)	a.	∃e [ action(e, tickle) & goal(e, into a frenzy) & agent(e, sue) & theme(e, her daughter) ]	simple
	b.	∃M ∃R [ cause(M,R) & action(M, tickle) & goal(R, into a frenzy) & agent(M, sue) & theme(R, her daughter) ]	analytic- complex
	c.	∃e ∃M ∃R [ cause(e,M,R) & action(M, tickle) & goal(R, into a frenzy) & agent(e, sue) & theme(e, her daughter) ]	compound- complex

Event Complexity

Impoverished Lexicon

Unlike talk, tickle does typically take a theme for the relevant reading, but not a goal, as in (52).

(52)	a.	#	Sue tickled.
	b.		Sue tickled her daughter.

We would like to avoid positing distinct tickle lexemes for (52b) and (50c).  We can do this either by positing a separate subevent which introduces the goal, or by introducing the goal as an adjunct.  Therefore, this condition is indeterminate.

Thematic Uniqueness

Not relevant: There is only one agent, one theme, one goal.

?? Argument per Subevent

Not relevant: Multiple arguments allow, but do not entail, multiple subevents.

Pervasive Theme

The theme of R-into a frenzy is her daughter, which is also the theme of M-tickle.  A pervasive theme allows, but does not require, a single-event reading.  Thus, this condition is indeterminate.

?? Paraphrase

Judgments may vary, but (50c) might reasonably be paraphrased as (53).

(53)	a.	?	Her daughter went into a frenzy because Sue tickled her.
	b.	?	Sue's daughter went into a frenzy by Sue tickling her.

This provides weak evidence for a complex event structure.

?? Lexical Entailment

tickle does not lexically entail motion, and hence does not entail a goal.  Adding a goal therefore indicates a distinct subevent.  Thus the event is complex (unless we believe in coidentification).

? Temporal Dependence

The two putative subevents appear to be temporally dependent, at least in the same sense as (14a), as shown in (54).

(54)	a.		Sue tickled her daughter into a frenzy.
	b.	?#	Sue quickly tickled her daughter into a frenzy by tickling her slowly.
	c.	#	Sue quickly tickled her daughter into a frenzy, although going into a frenzy took a while.
	d.	?	Sue tickled her daughter into a frenzy, although she'd already stopped tickling her when she went into a frenzy.
	e.	?	Sue tickled her daugher into a frenzy and for another five minutes.

(54b,c), if bad, demonstrate that temporal adverbials attach to the complex event, not the individual subevents.  (54d,e), however, suggest that it is at least possible to get a reading without temporal dependence.

The questionable acceptability of (54d,e) provides weak evidence for a simple event structure.

Note that the aspectual difference between (50c) and (50a) makes the case for temporal dependence in (50a) much stronger, as shown in (55).

(55)	a.		Bill broke the bathtub into pieces.
	b.	#	Bill quickly broke the bathtub into pieces by breaking it slowly.
	c.	#	Bill quickly broke the bathtub into pieces, although going into pieces took a while.
	d.	#	Bill broke the bathtub into pieces, although it didn't fall into pieces until several minutes later.
	e.	?#	Bill broke the bathtub into pieces and even into smaller pieces.

Thus the case for a simple event structure is stronger for (50a).

Complexity Summary

Altogether, the evidence is mixed.

Event Structure

The arguments for compound-complex vs. analytic-complex structure assume a complex structure to begin with.  They can show that the structure, if it is complex, is compound-complex.  They might also shed more light on the question of simple vs. complex.

Coparticipants

There is a strong intuitive sense that sue and her daughter are involved in the same event.  This argues against an analytic-complex structure such as (51b).

Immediacy of Causation

The question is whether (50c) is equivalent to (56).

(56)	Sue tickled her daughter, and this caused her daughter to go into a frenzy.

Picture scenario (57).

(57)	Sue tickles her daughter, who laughs. Sue stops tickling her daughter and walks away. A few minutes later, her daughter remembers that it is undignified to be tickled, and becomes angry, eventually going into a frenzy.
	a.		Sue tickled her daughter, and this caused her daughter to go into a frenzy.
	b.	#	Sue tickled her daughter into a frenzy.

(57b) is false, but (57a) is true.  This strongly argues against the analytic-complex analysis, and strengthens the argument for a complex structure.

Temporal Modification

As noted in (54c), it does not appear to be possible to attach a temporal modifier to the M event alone.  This argues against the analytic-complex analysis.

Event Structure Summary

Altogether, the evidence is quite strong that the event in (50c) is not analytic-complex.  The overall evidence weakly supports a compound-complex analysis over a simple analysis.

Intransitive, object-control

(58)	a.	Bill sneezed out the window.
	b.	The toilet leaked through the floor to the kitchen below.
	c.	Bill ate off the floor.

The main quality distinguishing (58) from the previous cases is that the result is predicated upon a theme which is not lexically realized.  I will refer to this semantic theme as pro.

Let's look at (58a).  Possible rough interpretations are in (59).

(59)	a.	∃e [ action(e, sneeze) & goal(e, out the window) & agent(e, bill) & theme(e, pro) ]	simple
	b.	∃M ∃R [ cause(M,R) & action(M, sneeze) & goal(R, out the window) & agent(M, bill) & theme(R, pro) ]	analytic- complex
	c.	∃e ∃M ∃R [ cause(e,M,R) & action(M, sneeze) & goal(R, out the window) & agent(e, bill) & theme(e, pro) ]	compound- complex

Event Complexity

Impoverished Lexicon

Because pro is unrealized, it is not immediately obvious whether it is typically present, as in (60).

(60)	a.	(?)	Bill sneezed.
	b.	(?)	Bill sneezed pro.

This can be teased apart by noting that the implicit theme can serve as an antecedent to a deictic pronoun, unlike with a typical intransitive verb, as shown in (61).

(61)	a.	*	Bill ran.  It was fast.
	b.		Bill sneezed.  It was green.

So, apparently, sneeze typically involves an unrealized theme, but no goal.  As with tickle, we can avoid positing an additional lexeme by positing either an additional subevent or an adjunct.  Therefore, this condition is indeterminate.

Thematic Uniqueness

Not relevant: There is only one agent, one (implicit) theme, one goal.

?? Argument per Subevent

The application of this principle depends on whether pro counts as a lexical argument.  L&RH's (1999) analysis specifically refers to "overt direct objects", which would seem to rule out pro.  So, on the strictest reading, this condition would indicate a simple event structure.

Pervasive Theme

The theme of R-out the window is pro, which is also (presumably) the theme of M-sneeze.  A pervasive theme allows, but does not require, a single-event reading.  Thus, this condition is indeterminate.

?? Paraphrase

There does not appear to be an acceptable paraphrase of (58a), as in (62).

(62)	a.	*	A sneeze went out the window by Bill sneezing.
	b.	??	Something went out the window by Bill sneezing.

This would suggest a simple event structure.

?? Lexical Entailment

To follow L&RH (1999), we need to know whether sneeze lexically entails a goal.  They would claim that the Paraphrase condition shows that it does, despite the apparent acceptability of (63).

(63)	Bill sneezed.

They would therefore claim this suggests a simple event structure.

? Temporal Dependence

The two putative subevents are clearly temporally dependent, as shown in (64).

(64)	a.		Bill sneezed out the window.
	b.	#	Bill quickly sneezed out the window by sneezing slowly.
	c.	#	Bill quickly sneezed out the window, although its going out the window took a while.
	d.	#	Bill sneezed out the window, although he'd stopped sneezing when it went out the window.
	e.	#?	Bill sneezed out the window and five more times.

As we saw, this is hardly conclusive evidence of a simple event structure, but it does hint in that direction.

Note that the aspectual difference between (58a) and (58b) dramatically improves the possibility of temporal independence, as shown in (65d,e).

(65)	a.		The toilet leaked through the floor to the kitchen below.
	b.	#	The toilet quickly leaked through the floor to the kitchen below by leaking slowly.
	c.	#	The toilet quickly leaked through the floor to the kitchen below, although it took a while to get to the kitchen below.
	d.		The toilet leaked through the floor to the kitchen below, although it had stopped leaking by the time it got to the kitchen below.
	e.	?	The toilet leaked through the floor to the kitchen below and for another 20 minutes.

This would tend to outweigh the evidence from (64), suggesting a complex event structure.

Complexity Summary

The overall evidence is indeterminate.  Our conclusion would depend on which conditions we are more inclined to trust. 

Event Structure

These arguments show that the structure, if it is complex, is compound-complex.  They also shed some more light on the simple vs. complex question.

Coparticipants

There is a strong intuitive sense that bill and pro are involved in the same event.  This argues against an analytic-complex structure such as (59b).

Immediacy of Causation

The question is whether (58a) is equivalent to something like (66).

(66)	Bill sneezed pro, and this caused pro to go out the window.

As seen in (62) and (66), it is virtually impossible to come up with a meaningful paraphrase, let al.one pose a scenario with indirect causation.  This offers no significant evidence to differentiate between analytic-complex vs. compound-complex, but strengthens the argument for a simple event analysis.

Temporal Modification

As noted in (64c), it does not appear to be possible to attach a temporal modifier to the M event alone.  This argues against the analytic-complex analysis.

Event Structure Summary

Altogether, the evidence is quite strong that the event in (58a) is not analytic-complex.  There is weak evidence that it is, in fact, a simple event.

Transitive, selected, subject-control

(67)	a.	Bill followed the road into the forest.
	b.	We drove Highway 5 from SD to SF.
	c.	John danced mazurkas across the room.

The main quality distinguishing (67) from previous cases is that the result is predicated upon the agent of the M event rather than its (nonexistent) theme.

Let's examine (67a).  Possible rough interpretations are in (68).

(68)	a.	∃e [ action(e, follow) & goal(e, into the forest) & agent(e, bill) & path(e, the road) ]	simple
	b.	∃M ∃R [ cause(M,R) & action(M, follow) & goal(R, into the forest) & agent(M, bill) & theme(R, bill) & path(R, the road) ]	analytic- complex
	c.	∃e ∃M ∃R [ cause(e,M,R) & action(M, follow) & goal(R, into the forest) & agent(e, bill) & path(e, the road) ]	compound- complex

Note that I use the semantic role label path rather than theme for the road, because the road does not undergo a change of state.  However, the choice of label does not directly affect the analysis.

We could also add theme(e, bill) to (68c), but this would not seem to contribute meaningfully to the interpretation.

Event Complexity

Impoverished Lexicon

follow typically takes a path for the relevant reading, but no goal, as in (69).

(69)	a.	#	Bill followed.
	b.		Bill followed the road.
	c.	#	Bill followed into the forest.

We would like to avoid positing a distinct follow lexeme to introduce the goal.  As with previous cases, we can do this by positing either an additional subevent or an adjunct.  Therefore, this condition is indeterminate.

Thematic Uniqueness

Not relevant: There is only one agent, one path, one goal.

?? Argument per Subevent

Not relevant: Multiple arguments allow, but do not entail, multiple subevents.

Pervasive Theme

The theme of R-into the forest is bill, which is also a participant in M-follow.  A pervasive theme allows, but does not require, a single-event reading.  Thus, this condition is indeterminate.

?? Paraphrase

(67a) is readily paraphrased as (70).

(70)	Bill went into the forest by following the road.

This provides weak evidence for a complex event structure.

?? Lexical Entailment

According to L&RH (1999), the possibility of a paraphrase shows that follow does not lexically entail an endpoint, a goal.  They would therefore argue that this reveals a complex event structure (which they would then interpret as a single event, via coidentification).

? Temporal Dependence

The two putative subevents appear to be temporally dependent, as shown in (71).

(71)	a.		Bill followed the road into the forest.
	b.	#	Bill quickly followed the road into the forest by (following, walking) slowly.
	c.	#	Bill quickly followed the road into the forest, although going into the forest took a while.
	d.	#	Bill followed the road into the forest, although he'd already stopped (following, walking) when he went into the forest.
	e.		Bill followed the road into the forest and out the other side.

Only (71e) suggests otherwise.  This provides weak evidence for a simple event structure.

Complexity Summary

Altogether, the evidence is mixed, but weakly favors a compound-complex analysis.

Event Structure

These arguments show that the structure, if it is complex, is compound-complex.  They also shed further light on the simple vs. complex question.

Coparticipants

There is a strong intuitive sense that bill and the road are involved in the same event.  This argues against an analytic-complex structure such as (68b).

Immediacy of Causation

The question is whether (67a) is equivalent to (71).

(71)	Bill followed the road, and this caused Bill to go into the forest.

Picture scenario (72).

(72)	Bill follows the road, walking down the middle line. A car comes along from behind and hits Bill, scooping him up onto the hood. The car drives into the forest, bringing Bill along with it.
	a.		Bill followed the road, and this caused Bill to go into the forest.
	b.	#	Bill followed the road into the forest.

(72b) is false, but (72a) is true.  This strongly argues against the analytic-complex analysis, and argues for a complex structure.

Temporal Modification

As noted in (71c), it does not appear to be possible to attach a temporal modifier to the M event alone.  This argues against the analytic-complex analysis.

Event Structure Summary

Altogether, the evidence is quite strong that the event in (67a) is not analytic-complex, and weakly favors a compound-complex analysis.

Intransitive, subject-control

(73)	a.	Bill rolled out of the room.
	b.	Gretchen swam across the river.
	c.	Bob and Sally drove to the store.
	d.	Trace waltzed (beautifully) to the other side of the ballroom (and onto the balcony).

(73) is similar to (67), except there is no second argument to the M predicate.

Let's examine (73b).  Possible rough interpretations are in (74).

(74)	a.	∃e [ action(e, swim) & goal(e, across the river) & agent(e, gretchen) ]	simple
	b.	∃M ∃R [ cause(M,R) & action(M, swim) & goal(R, across the river) & agent(M, gretchen) & theme(R, gretchen) ]	analytic- complex
	c.	∃e ∃M ∃R [ cause(e,M,R) & action(M, swim) & goal(R, across the river) & agent(e, gretchen) ]	compound- complex

One might use path instead of goal for across the river.  However, this would imply an atelic interpretation, instead of the correct telic interpretation, shown in (75).

(75)	Gretchen will swim across the river (#for 20 minutes, in 20 minutes).

Event Complexity

Impoverished Lexicon

swim occurs most naturally with some sort of complement or adjunct, although this can take virtually any form or specify virtually any semantic role, as shown in (76).

(76)	a.	?	Gretchen swam.
	b.		Gretchen swam frantically.
	c.		Gretchen swam the butterfly stroke.
	d.		Gretchen swam the English Channel.
	e.		Gretchen swam 15 laps.
	f.		Gretchen swam for 10 minutes.
	g.		Gretchen swam along the coastline.
	h.		Gretchen swam across the river.
	i.		Gretchen swam with her eyes closed.

We fervently hope to avoid positing a distinct swim lexeme for each case in (76).

Unlike talk, even when we introduce an NP complement as in (76b-d), it does not intuitively change the meaning of the event or add a participant – for instance, no theme undergoing a change of state.  We can therefore reasonably posit either an additional subevent or a non-thematic role introduced by a construction, even for the NP cases.  Therefore, this condition is indeterminate.

Note the contrast with (77).

(77)	a.	*	Gretchen went.
	b.	?	Gretchen went frantically.
	c.	*	Gretchen went foot.
	d.	*	Gretchen went Jamaica.
	e.	?	Gretchen went 15 laps.
	f.	?	Gretchen went for 10 minutes.
	g.		Gretchen went along the coastline.
	h.		Gretchen went across the river.
	i.	?	Gretchen went with her eyes closed.

go is far more restrictive in its distributional options.  This is compatible with the hypothesis that go is a verb of motion requiring a path/goal, while swim is an activity verb which can optionally take a motion predicate as a resultative.  This would support a complex event analysis for (73b).

Thematic Uniqueness

Not relevant: There is only one agent, one goal.

?? Argument per Subevent

There is only one lexicalized argument.  This principle would then suggest a simple event structure.

Pervasive Theme

The theme of R-across the river is gretchen, which is also a participant in M-swim.  A pervasive theme allows, but does not require, a single-event reading.  Thus, this condition is indeterminate.

?? Paraphrase

(73b) is easily paraphrased as (78).

(78)	Gretchen crossed the river by swimming.

This suggests a complex event structure.

?? Lexical Entailment

According to L&RH (1999), the possibility of a paraphrase shows that swim does not lexically entail an endpoint, a goal.  L&RH would therefore suggest that this reveals a complex event structure (but a single event).

? Temporal Dependence

The evidence for temporal dependence is questionable (79).

(79)	a.		Gretchen swam across the river.
	b.	?#	Gretchen quickly swam across the river by swimming slowly (and letting the current take her).
	c.	#?	Gretchen quickly swam across the river, although going across the river took a while (because of the strong current).
	d.	#	Gretchen swam across the river, although she'd already stopped swimming when she crossed the river.
	e.		Gretchen swam across the river and back again.

This condition is indeterminate.

Complexity Summary

Altogether, the evidence weakly favors a complex analysis.

Event Structure

These arguments show that the structure, if it is complex, is compound-complex.  They also provide more evidence for the simple vs. complex question.

Coparticipants

There is an undeniable intuitive sense that the agent of swim and the theme of across the river (both gretchen) are involved in the same event.  This argues against an analytic-complex structure such as (74b).

Immediacy of Causation

The question is whether (73b) is equivalent to (80).

(80)	Gretchen swam, and this caused Gretchen to go across the river.

Picture scenario (81).

(81)	Gretchen goes for a swim in the river. A log floats along and bumps her on the head, knocking her unconscious. Her body washes up on the far shore.
	a.		Gretchen swam, and this caused Gretchen to go across the river.
	b.	#	Gretchen swam across the river.

(81b) is false, but (81a) is true.  This strongly argues against the analytic-complex analysis, and argues for a complex structure.

Temporal Modification

As noted in (79c), it does not appear to be possible to attach a temporal modifier to the M event alone.  This argues against the analytic-complex analysis.

Event Structure Summary

Altogether, the evidence is quite strong that the event in (73b) is not analytic-complex.  The overall evidence weakly favors a compound-complex analysis over a simple analysis.

Transitive, unselected, subject-control

(82)	a.	The car honked its way down the road.
	b.	Bill whistled his way past the house.

As far as I can tell, the 'way' construction is the only licenser of this configuration.  A detailed analysis is beyond the scope of this paper.

The most significant difference between (82) and previous cases is that there is no causative relationship between M and R.  I therefore use the predicate means rather than cause in the semantic representation.

Also, the M predicate takes a dummy object way, which can be taken either to be semantically vacuous, or to be an anaphor of the car.  Note that way does not appear to refer to any actual entity or concept, nor is there any apparent semantic difference between (82) and (83).  I will therefore ignore way in the semantic representation.

(83)	a.	The car honked down the road.
	b.	Bill whistled past the house.

Let's examine (82a).  Possible rough interpretations are in (84).

(84)	a.	∃e [ action(e, honk) & path(e, down the road) & agent(e, the car) ]	simple
	b.	∃M ∃R [ means(M,R) & action(M, honk) & path(R, down the road) & agent(M, the car) & theme(R, the car) ]	analytic- complex
	c.	∃e ∃M ∃R [ means(e,M,R) & action(M, honk) & path(R, down the road) & agent(e, the car) ]	compound- complex

Using goal instead of path would imply a telic interpretation, instead of the correct atelic interpretation, shown in (85).

(85)	The car will honk down the road (for 20 minutes, #in 20 minutes).

Event Complexity

Impoverished Lexicon

honk does not intrinsically take a path or goal, but can take either argument optionally, as in (86).

(86)	a.		The car honked.
	b.		The car honked along the coastline.
	c.		The car honked down the road.

Unlike swim, adding a path or goal to (86a) does change the intuitive understanding of the event.  We therefore need an explanation to avoid positing a new honk lexeme.  The simplest explanation is that the result predicate introduces a new subevent.  This therefore suggests a complex event structure.

Thematic Uniqueness

Not relevant: There is only one agent, one path.

?? Argument per Subevent

There is only one lexicalized argument.  This principle would then suggest a simple event structure.

Pervasive Theme

The theme of R-down the road is the car, which is also a participant in M-honk.  A pervasive theme allows, but does not require, a single-event reading.  Thus, this condition is indeterminate.

?? Paraphrase

(82a) is easily paraphrased as (87b), although not as (87a).

(87)	a.	#	The car went down the road by honking.
	b.		The car went down the road, honking.

This suggests a complex event structure.

?? Lexical Entailment

honk does not lexically entail motion, and therefore does not entail a path.  This suggests a complex event structure.

? Temporal Dependence

The putative subevents do not appear to be temporally dependent, as shown in (88b,e).

(88)	a.		The car honked down the road.
	b.	?	The car quickly honked down the road, honking slowly.
	c.	#?	The car quickly honked down the road, going down the road slowly.
	d.	#	The car honked down the road, although it had already stopped honking when it went down the road.
	e.		The car honked down the road and into the horizon.

This would suggest a complex analysis.

Complexity Summary

Altogether, the evidence suggests a complex event structure.

Event Structure

These arguments show that the structure is compound-complex.

Coparticipants

There is an intuitive sense that the agent of honk and the theme of down the road (both the car) are involved in the same event.  This argues against an analytic-complex structure such as (84b).

Immediacy of Causation

Not relevant: There is no causation.

Temporal Modification

As noted in (88c), it does not appear to be possible to attach a temporal modifier to the M event alone.  This argues against the analytic-complex analysis.

Event Structure Summary

Altogether, the evidence favors a compound-complex analysis.

Conclusions

In summary, none of these examples allow for an analytic-complex analysis.  Some PP resultatives (89d,f) clearly indicate a compound-complex event structure.  Only one case (89b) (weakly) favors a simple event structure.

(89)		Resultative Type	Event Structure
	a.	intransitive, subject-control	(weak) compound-complex
	b.	intransitive, object-control	(weak) simple
	c.	transitive, selected, object-control	(weak) compound-complex
	d.	transitive, unselected, object-control	compound-complex
	e.	transitive, selected, subject-control	(weak) compound-complex
	f.	transitive, unselected, subject-control	compound-complex

Ockham's Razor suggests a common analysis for all cases, a compound-complex event structure.  However, before this conclusion can be stated emphatically, stronger evidence will need to be found for the questionable cases.