Noname manuscript No.

(will be inserted by the editor)

On Affectedness

John Beavers

the date of receipt and acceptance should be inserted later

Abstract Affectedness (a change in an event participant) has been implicated in ar-

gument realization, lexical aspect, transitivity, and various syntactic operations. How-
ever, it is rarely given a precise, independently-motivated definition. Rather, it is usu-
ally defined either intuitively or diacritically, or reduced to the properties it is meant
to explain, especially lexical aspect. In this paper I provide a semantic definition of
affectedness as a relationship between an argument and a property scale (following
recent work by Beavers 2008 and Kennedy and Levin 2008). I justify this analysis
by re-examining some of the empirical evidence for affectedness, and argue that it is
not reducible to lexical aspect, but is tightly correlated to it in a way that motivates a
definition involving two mutually-interdependent participants. This model also gives
us a precise way to define the pervasive notion of degrees of affectedness, in terms
of monotonically strengthened truth conditions about the result state along what I
call the Affectedness Hierarchy. This in turn captures important subset relations that
dynamic predicates stand in regarding affectedness and aspectual diagnostics. I also
show how this definition of affectedness brings together many of the above phenom-
ena under a single, unified approach.

1 Introduction

The notion of affectedness, i.e. an observable change in an event participant, is fre-

quently evoked in work on lexical semantics and syntax, so much so that it is nearly
impossible to reference all of the relevant work. However, there are several phe-
nomena for which affectedness has been especially key. Affectedness has long been
central to work on argument realization, especially in determining direct objecthood
(Fillmore 1968, 1970, 1977 S. Anderson 1971b, 1977b; Jackendoff 1990; Dowty
1991; Beavers 2006, inter alia). Furthermore, affectedness has been linked to lexical

John Beavers
The University of Texas at Austin
E-mail: jbeavers@mail.utexas.edu
2

aspect, especially as a determinant of telicity (Tenny 1987, 1992, 1994; Jackend-

off 1996; Krifka 1998; Beavers 2006, 2008, in press, inter alia). Affectedness has
also been evoked in constraints on syntactic operations, including DP-passives and
middles (Anderson 1979; Fiengo 1980; Jaeggli 1986; Condoravdi 1989, inter alia).
Finally, affectedness is widely assumed to figure into whether or not a verb is transi-
tive, both within and across languages (Hopper and Thompson 1980; Tsunoda 1981,
1985; Blume 1998; Testelec 1998; Næss 2003, inter alia).
Yet despite widespread appeal to the notion, affectedness rarely receives a pre-
cise, linguistically motivated definition. Rather, it is usually defined intuitively (e.g.
as the property of having undergone some change; see e.g. Fillmore 1970, 125), or
through representational schema that lack predictive force. For example, change-of-
state and change-of-location are sometimes assumed to be distinct, both conceptually
and linguistically (e.g. they license different result state modifiers; Jackendoff 1990,
94). Therefore they are often modeled via separate thematic roles (patient vs. theme)
or separate subevent types (GO vs. BECOME). Yet they figure into object realization
and aspect in nearly identical ways (Tenny 1992). For this reason they are sometimes
modeled in terms of the same semantic primitives (a general theme role or BECOME
subevent; Rappaport Hovav and Levin 1998, 108-109). Still further theories adopt
a multistratal view of lexical meaning that allows for criss-crossing generalizations
to capture the similarities and differences (e.g. the “action” vs. “thematic” tiers of
Jackendoff 1990). All are equally valid intuitions, but a question less often addressed
is whether we can find independent criteria by which we can predict what changes
languages will treat the same and how they will be similar.
Affectedness is also well known to be a matter of degree. Intuitively, in (1) the
patient the apple is increasingly less affected from (1a) to (1d).
(1) a. John ate the apple up. (Apple is completely gone)
b. John cut the apple. (Apple cut, not necessarily to a particular degree)
c. John kicked the apple. (Apple impinged, not necessarily affected)
d. John touched the apple. (Apple manipulated, not necessarily impinged)
Degrees of affectedness have been especially important in work on transitivity (Hop-
per and Thompson 1980, inter alia), where higher degrees of affectedness correspond
to higher transitivity. However, high and low affectedness are hard to define precisely,
and are usually left to intuition. For example, Hopper and Thompson (1980, 252-253)
claim that “[t]he degree to which an action is transferred to a patient is a function of
how completely that patient is AFFECTED” (emphasis theirs), although they never
define degree of transfer, and subsequent work has not improved on this situation.
Finally, rarely is a notion of affectedness defined that brings all of the above phe-
nomena together under a single rubric. This is largely because affectedness is often
appealed to in discussion of other phenomena, but is rarely an object of study itself.
The primary strand of literature that has tied multiple affectedness-based phenom-
ena together is work on lexical aspect, following Carol Tenny (1987, 1992, 1994) in
particular. In this literature, affectedness is usually done away with as a grammatical
primitive, and is instead reduced to a set of aspectual properties, especially telicity
(determining the endpoint of the event) and measuring out (measuring the time course
 3

of the event). But there is a conceptual difference between affectedness and aspect,
and it is a complex empirical question whether they can be distinguished linguisti-
cally, though I believe that they can.
My goal in this paper is thus to address affectedness head on, and to propose a
new way of putting its use in linguistic theorizing on more rigorous empirical and
theoretical grounds. In particular, I have the following four objectives in mind:
1. To reassess the empirical diagnostics that isolate an independent notion of affect-
edness, and clarify how this notion does or does not correlate with lexical aspect;
2. To synthesize recent proposals about aspect and change-of-state into a single,
unified model that captures the empirical diagnostics and aspectual correlations;
3. To use this model to give a precise definition of degrees of affectedness;
4. To sketch how this model applies to syntactic phenomena in a consistent way.
In §2 I review several semantic diagnostics for affectedness, and show that the
diagnostics group predicates into a subset hierarchy, where predicates that pass n
tests are a strict subset of those that pass n − 1 tests, an important property I expand
on later. In §3 I compare these diagnostics with the standard aspectual diagnostics
thought to be associated with affectedness, and show that they only partly coincide.
While affected arguments figure into telicity, they are not the only ones that do so,
and conversely affected arguments do not figure into measuring out, though others
do. I thus argue that affectedness is not a property of just one entity, but requires a
mutually constraining relation between multiple participants in the event.
In §4 I define affectedness as in (2) (following Hay et al. 1999; Wechsler 2001,
2005; Beavers 2002, 2006, 2008; Kennedy and Levin 2008; Rappaport Hovav 2008).
(2) An argument x is affected iff there is an event e and a property scale s such
that x reaches a new state on s through incremental, abstract motion along s.
Thus affectedness is a relational notion, involving both a patient and a property scale
defining the progress of its change, itself a full-fledged argument of the verb. Further-
more, the constraints imposed on each argument are relativized to the other, requiring
a properly ternary thematic relation between the patient, the event, and the scale. I
show that explicitly positing another argument (the property scale) into the equation
has two advantages for understanding affectedness. First, it gives us a principled way
to capture the indirect relation of affectedness and aspect: some aspectual proper-
ties correlate with the scale, some with the affected argument, and some with both.
Second, it allows us to abstract away from real world changes to a general notion
of affectedness in a principled way. Different real world changes are distinguished
in terms of the type of the scale argument (e.g. scales of cleanliness, color, dirtiness,
volume, position, etc.). Similarities across real world changes are due to the structure
of the scale argument, regardless of type (e.g. subpart structure, endpoints, etc.).
In §5 I define degrees of affectedness based on another property of the scale:
(3) How affected x is corresponds to how specific the predicate is about where x
ends up on s.
I identify four degrees of affectedness that form an implicational Affectedness Hierar-
chy based on monotonically weakening truth conditions, where an argument affected
4

to degree m is affected to all degrees less than m. I show how this hierarchy captures
the subset property of affectedness tests discussed in §2, and in §6 I show how it fig-
ures into DP-preposing and middle formation. I conclude in §7, and discuss also the
role of a purely semantic theory of affectedness in a larger theory of lexical semantics.

2 Empirical Diagnostics for Affectedness

As mentioned above, affectedness has to do with change, which is standardly under-

stood as some condition φ obtaining that did not before. Often this is assumed to cor-
respond to a linguistically primitive BECOME subevent type, where [ BECOME φ ]
entails that for some time interval I φ does not hold at the initial point in I but start-
ing at the final point in I φ does hold (Dowty 1979, 140-144, updating von Wright
1963). BECOME is usually assumed to be either part of the lexical decomposition
of a verb’s meaning (Dowty 1979, Foley and Van Valin 1984, Rappaport (Hovav)
and Levin 1988, 1998, Levin and Rappaport Hovav 1995, Pinker 1989, Jackendoff
1990, 1996, Van Valin and LaPolla 1997, inter alia) or a separate syntactic head
(Baker 1997; Travis 2000; Folli and Ramchand 2002; Folli and Harley 2004; Ram-
chand 2008, inter alia), although the differences between these two approaches are
irrelevant here. However, “change” is quite a general notion, and as Dowty has it,
φ could be anything, even an activity or process (e.g. [ BECOME It is raining ] is a
well-formed formula). Affectedness proper is usually conceived of as the more spe-
cific notion that φ is a stative result predicate that obtains and maintains for some
entity x as a result of the event (see Koontz-Garboden 2006, in press). I focus on this
notion here.
Standard evidence for BECOME subevents includes ambiguities with temporal
and restitutive adverbs (see e.g. Dowty 1979, 238ff., inter alia). For example, predi-
cates that entail a change such as open the door in (4a) are ambiguous with again
between a reading where the same agent performed the action again vs. that the
result state obtained again, but not necessarily due to the same agent. Non-change
predicates like laugh have only the former reading as in (4b).
(4) a. John opened the door again.
b. John laughed again.
This is explained if the event structure of (4a) is [ ψ CAUSE [ BECOME φ ] ], where
again can modify either the CAUSE or BECOME subevent, while in (4b) the struc-
ture is just [ DO φ ]. However, these ambiguities are possible with other types of
verbs, including verbs of possession such as want (Dowty 1979, 245-250), which do
not entail a change. For example, John wanted a car again is ambiguous between
a reading where John’s desire for a car has occurred again, or that he wishes he
once again owned a car. Thus adverbs may not test for change per se, but rather for
subevents, of which BECOME subevents are just one type. So in isolating affected-
ness, we must first ask the question of what constitutes a BECOME subevent and
how we know we have one of these as opposed to any other subevent.
I assume that change can only be encoded in dynamic (non-stative) predicates.
But which dynamic predicates indicate changes, and which changes do languages
 5

treat as the “same”? It is not my goal to be exhaustive, but intuitively, the properties
in (5) can or have all been considered types of affectedness for some entity x.1

(5) a. x changes in some observable property. (clean/paint/delouse/fix/break x)

b. x transforms into something else. (turn/carve/change/transform x into y)
c. x moves to and stays at some location. (move/push/angle/roll x into y)
d. x is impinged upon. (hit/kick/punch/rub/slap/wipe/scrub/sweep x)
e. x goes out of existence. (destroy/eliminate/delete/eat/consume/reduce x)
f. x comes into existence. (build/design/construct/create/fashion x)

We can lump predicates entailing these properties into four classes (building on Tenny
1992, 15-18; see also Tenny 1987, 1994; Jackendoff 1996; Krifka 1998, Rappaport
Hovav and Levin 2005; Rappaport Hovav 2008). Predicates that entail changes in
some property of a participant as in (5a-b) are “change-of-state” predicates, while
those that entail change in location as in (5c) are “motion” predicates. Predicates in
(5d) that entail contact but no change (a point I return to below) are “surface con-
tact” (Fillmore 1970; Rappaport Hovav and Levin 2005) and “impact” predicates.
Predicates entailing (5e-f) are “incremental theme” predicates. Predicates of coming-
into-existence are also said to have EFFECTED rather than AFFECTED objects (Quirk
and Greenbaum 1973, 171-175, Lakoff 1976, 52, Hopper 1985, 68; see also Jespersen
1933, 109 on “Objects of Result”). These classes are just rough guidelines for ease
of discussion. There are many subtypes among each, all with their own unique prop-
erties (see e.g. Levin 1993 and Baker and Ruppenhofer in press).
The properties in (5) are not unrelated, since they all are typical properties of
objects (Fillmore 1970, 1977; Anderson 1971b, 1977b; Dowty 1991; Beavers 2006,
inter alia), and almost all have been proposed as conditioning factors on middle and
DP-passive formation (Condoravdi 1989). Thus we have a syntactic reason to lump
them together. But are they also semantically linked? One semantic domain where
nearly all of (5) figure is lexical aspect, especially in determining telicity (Tenny
1992, inter alia). However, the connection of affectedness to aspect is complex, and
is something to be explained, so I set it aside until §3, though I note that telicity
does not always line up with affectedness, but rather picks out a subset of change
predicates. Predicates that entail very specific results (e.g. John shattered the vase)
are necessarily telic, but not those that entail non-specific results (e.g. John cooled
the soup). So is there a simpler link between (5), e.g. a predicate that applies to
all such arguments? I review and re-evaluate some of the evidence that has been
proposed for this below, and argue that it does. However, I also show that the evidence
sorts predicates into a subset hierarchy, which I later analyze in terms of degrees of
affectedness.
Happen/Did To. To my knowledge the world’s only proposed direct test for af-
fectedness is Cruse’s (1973, 13) What happened to X is Y (see also Lakoff 1976,
1 Affectedness may also include changes-in-possession, such as coming-to-possess (give x y), ceasing-

to-possess (take x from y), or being the theme in a possession change (y in the previous two examples).
Since possession presents its own issues (e.g. systematic polysemy; Green 1974; Tham 2005), I set it aside.
6

47-48, Jackendoff 1990, 125-130, Rappaport Hovav and Levin 2001, 786-787).2 Con-
sider (6), where intuitively the object of (6a) is affected and passes What happened to
X is Y as in (6b). This is compared to (7a), where the object is intuitively unaffected,
and likewise does not pass the test in (7b).
(6) a. The Romans destroyed the barbarian city.
b. What happened to the barbarian city was that the Romans destroyed it.
(7) a. They followed the star (out of Bethlehem).
b. *What happened to the star is they followed it (out of Bethlehem).
However, (7b) is acceptable if a special context is established, e.g. in a supernatural
context where the star is there to lead people out of Bethlehem and vanishes once its
mission is accomplished. However, this requires a very rich context and is not entailed
by follow per se, conforming to what Jackendoff (1990, 294, fn. 7) calls a “discourse
patient”. To keep things manageable, I focus only on change that is entailed by the
verb regardless of context (Jackendoff’s “grammatical patient”).3
Applying the test to the objects in (5), we see they nearly all pattern alike. In par-
ticular, (5a-e) are amenable to What happened to X is Y, as shown in (8a-e) respec-
tively, suggesting that (5a-e) are not just syntactically but also semantically related.
(8) a. What happened to the bedroom is John cleaned/painted it.
b. What happened to the wood is John turned/carved it into a toy.
c. What happened to the vase is John moved/pushed it into the house.
d. What happened to the car is John hit/wiped it.
e. What happened to the cake is that John destroyed/ate it.
However, What happened to X is Y does not appear to pick out all of (5), though I
argue that it does when two orthogonal factors are controlled for. First, the test is odd
for effected objects in (5f) (see also Quirk and Greenbaum 1973, 174 and Hopper
1985, 72 for similar data on What Z did to X is Y):
(9) #What happened to the shed is John built/created/fashioned/constructed it.
However, if the object has some degree of prior existence the test sounds significantly
better. This includes model airplanes or things were destroyed and are recreated:4
(10) a. What happened to the model airplane is John built it.
b. What happened to the shed is John rebuilt/refashioned/reconstructed it.
The object need not have prior physical existence. If I design a house, and someone
asks me what my plans are with my design, it is not infelicitous to respond with (11).
2 Cruse also gives What Z did to X is Y, which presupposes an agent Z. Since I am not interested in

agentivity I focus mostly on What happened to X is Y, though the two are usually interchangeable.
3 There is another, colloquial use of What happened to X as in What happened to Mary? which means

Where did Mary go? (as pointed out to me by Steve Wechsler). This may be related, but I set it aside here.
4 Indeed, in Google searches for build in the middle voice (a construction which supposedly only applies

to verbs with affected objects; see §3 and §6.2) most of the examples I found involved model airplane kits.
 7

(11) What happens to the house next is I build it!

This suggests that the problem with effected objects in this test is a prior existence
presupposition, a complicating factor that should be controlled for.
Second, while theme objects of transitive motion verbs pass the test (see (8c)),
theme subjects of intransitive motion verbs differ depending on the verb type. Sub-
jects of unaccusatives pass the test, as do inanimate subjects of unergatives:
(12) a. The ball fell/glided down the hill.
b. What happened to the ball is it fell/glided down the hill.
However, animate subjects of unergative motion verbs do not, even if the same dislo-
cation entailed to occur with transitives also occurs. This is shown in (13).
(13) a. John walked/sauntered up the trail.
b. #What happened to John is that he walked/sauntered up the trail.
This suggests that agentivity is a further complicating factor. Indeed, Cruse presents
What happened to X is Y in opposition to What X did is Y, which picks out agentivity.
Subjects of unergatives are frequently agentive, unlike other theme subjects:
(14) a. What John did was walk/saunter up the trail.
b. #What the boulder did was fall/glide slowly down the hill.
Thus agentivity must also be controlled for. Therefore What happened to X is Y does
pick out all of (5), once we control for prior existence and agentivity.
However, there is one further problem with this test that necessitates looking for
other tests for affectedness. Surface contact/impact verbs in (5d) pass What happened
to X is Y, but are unique among (5) in not entailing any change. For this reason,
Rappaport Hovav and Levin (2001, 786-787) suggest that What happened to X is Y
really picks out “force recipients” (following Croft 1990, 1991, 1998), i.e. things that
take the brunt of an action and may change as a result, but not necessarily. This can be
more abstract than physical contact, as in What the cops did to him was threaten him,
where threaten entails neither change nor physical contact. So perhaps affectedness
is a broader notion than literally undergoing a change (as suggested by Jackendoff
1990, 125). Alternatively, we may want a more restrictive diagnostic that picks out
only participants that undergo a real change (those in (5a-c,e-f)). I turn to this next.
Entailment Tests. Since change involves new conditions obtaining, it is natural
to focus on entailment. An obvious test is past participle predication (Tenny 1992,
22-24), e.g. is X is Ved entailed? Applying this to (5) gives (15), where only (15d)
sounds even remotely natural, although the interpretation must be that is hit means
‘is damaged’ and is wiped means ‘is clean’.5
(15) a. John just cleaned/painted the bedroom, #but it is not clean/painted.
5 I test for entailment by contradiction, which I indicate by a #-mark on the continuation that forms

the test, though properly speaking the entire string is infelicitous by virtue of being a contradiction. I use
just to mitigate against the possibility of a change occurring and being subsequently undone. Alternatively,
Kratzer (2000) proposes the test X is still Ved, which I think gets at the same idea.
8

b. John just turned/carved the wood into a toy, #but it is not turned/carved
into a toy.
c. John just moved the pot into the room, #but it is not moved into the room.
d. John just hit/wiped the car, but it is not ?hit/?wiped.
e. John just destroyed/ate the cake, #but it is not destroyed/eaten.
f. John just built/constructed the house, #but it is not built/constructed.
However, one could argue that this does not show that all of these predicates involve
the same BECOME subevent. Ideally, we need a single property of some high degree
of generality entailed to be true by all of these predicates (akin to happened/did to).
I propose something is different about X, which picks out almost all of (5), but
excludes surface contact/impact predicates:
(16) a. John just cleaned/painted the bedroom, #but nothing is different about it.
b. John just turned/carved the wood into a toy, #but nothing is different
about it.
d. John just hit/wiped the car, but nothing is different about it.
e. John just destroyed/ate the cake, #but nothing is different about it.
f. John just built/constructed the house, #but nothing is different about it.6
Unfortunately, this test does not pick out moving figures, agentive or not (to my ears):
(17) a. John just walked out of the room, but nothing is different about him.
b. The ball just moved out of the room, but nothing is different about it.
Perhaps something is different about X only picks out properties that can be observed
by looking at X itself, divorced from context. Thus we need a special test just for mo-
tion, which involves looking at X and its surroundings, such as X is somewhere else,
which picks out motion predicates in which change-of-location necessarily occurred:
(18) a. John just walked out of the room, #but he is not somewhere else.
b. John just walked, but he is not somewhere else.
Since What happened to X is Y already groups motion with other changes, positing
two different tests is just a refinement within a larger category. Thus entailment tests
pick out a subset of the predicates picked out by What happened to X is Y.
Resultatives. Another test is resultative predication. Although resultatives are of-
ten assumed to only predicate of deep structure objects (see Levin and Rappaport
Hovav 1995, 34), Rappaport Hovav and Levin (2001, 784-790) show that the actual
constraint (for English) is semantic: the subject of the result XP must be a force recip-
ient (see also Goldberg 1995, 188). This explains why subjects of transitive motion
verbs can take result XPs as in (19), where the men and not the star change location.
6 Something is different about X sometimes generates a presupposition failure with certain cre-

ation/consumption predicates eat, since the test presupposes X exists. However, if the entity has some
existence before/after the event, the test yields the appropriate contradiction.
 9

(19) The wise men followed the star to Bethlehem.

This gives us another semantic test for affectedness, and indeed all of the changes
in (5) in principle allow result XPs as in (20), except creation predicates, perhaps
due to a prior existence constraint (although there are further constraints governing
resultatives; see Wechsler 2001, 2005; Beavers 2002, 2008 and Boas 2003):
(20) a. John painted the bedroom a fierce red.
b. John carved the wood into a toy.
c. John pushed the vase into the room.
d. John pounded the metal flat.
e. John reduced the water level to nothing.
However, resultatives vary across languages. Washio (1997) argues that Japanese
resultatives are more restricted, and primarily occur with verbs that independently
entail a result. This is shown in (21), where someru ‘dye’ entails a result and allows
a result XP, but keru ‘kick’ does neither:
(21) a. Mary-ga doresu-o pinku-ni some-ta
Mary-NOM dress-ACC pink-DAT dye-PAST
‘Mary dyed the dress pink.’
b. *Kanozyo-wa musuko-o azadarake-ni ket-ta.
she-TOP son-ACC black and blue-DAT kick-PAST
‘She kicked her son black and blue.’ (Washio 1997, 5-6, (13b), (18b))
Washio (1997, 12-16) also notes that some verbs that do not entail a result but lexi-
cally restrict the possible result states also license resultatives, including nobasu ‘roll
out’, migaku ‘polish’, niru ‘boil, simmer’, and haku ‘wipe (in a removal context)’.
An example for haku is given in (22).
(22) Kare-wa teeburu-o kirei-ni hui-ta
he-TOP table-ACC clean-DAT wipe-PAST
‘He wiped the table clean.’ (Washio 1997, 16, (49))
However, native speakers tell me that result readings do arise by implicature for these
four verbs (which Washio hints at). If so, the generalization is that resultatives pick
out entities that are entailed or implicated to be affected, a stronger condition than
in English, but picking up on the same force recipient vs. actual change contrast as
What happened to X is Y vs. result entailments.
Types of Result XPs. Result verbs also differ from surface contact verbs in pos-
sible result XPs (although this is a hard notion to quantify; see Boas 2003, 150-158,
216-233 for a related discussion). For example, shatter allows only result XPs de-
scribing numbers of pieces, while wipe allows result XPs describing various physical
changes (see also Washio 1997, 7-8 and Rappaport Hovav 2008, 22-23):
(23) a. John shattered the vase into a million/thousand/thirty-six different pieces.
b. #John shattered the vase slightly/in half/into two pieces/silly/flat/red/up.
(24) a. John wiped his face dry/clean/off/raw/to a healthy glow.
b. #John wiped his face flat/red/up.
10

However, verbs that entail results may also vary in how many result XPs they al-
low, which partly correlates with how specific they are inherently about the result
state. While shatter is quite specific (lots of pieces), cut is not, and allows result XPs
indicating many types of damage or shapes:
(25) a. John cut the bread into 15/a million pieces/an amusing shape/open/up.
b. #John cut the bread clean/silly/flat/red.
However, cool is also non-specific, but only permits temperature result XPs:
(26) a. John cooled the soup to 50/100/250 degrees/down/to a palatable level.
b. #John cooled the soup red/congealed/into mush.
Thus we cannot quite correlate specificity directly to variation in result XPs, but we
can at least define a subset of all result verbs based on lower variation in result XPs.7
Summary. In this section I have reviewed several tests for affectedness. Some of
them (What happened to X is Y, English resultatives) pick out a less restricted notion
of being a force recipient, while others (Japanese resultatives, entailments) pick out a
more restricted notion. Some tests that seem to exclude certain changes (What hap-
pened to X is Y for effected or agentive X) may not exclude them when orthogonal
factors are controlled for. I summarize the tests as follows (including telicity):
(27) Dynamic predicate φ entails
Diagnostics specific result non-specific result force recipient non-force recipient
Telic X × × ×
Change entailed X X × ×
Permit result XP X X X/× ×
Happened/did to X X X ×
Dynamic X X X X/×
Result XP variation Low Low/High High N/A

Crucially, as (27) shows, the tests group dynamic predicates into subset relations: a
subset of dynamic predicates take force recipients, a subset of these entail change,
and a subset of these are telic. This is an important property that I do not believe
has been discussed before, and I expand on this and provide an analysis of it in §5.
At this point, whether we apply the term “affected” to force recipients or the more
narrow notion of having a change is purely terminological, though only the latter
conforms to Dowty’s definition of BECOME. In §3 and §4 I equate “affectedness”
with the stronger notion and focus solely on this, but I incorporate force recipients
back into the picture in §5. Regardless, we now have empirical reasons for lumping
these notions together, and for assuming that there is an isolatable class of BECOME
subevents. In the next section I turn to lexical aspect, and in so doing I address the
question of what the crucial ingredients are of a BECOME subevent.

3 The Partial Correlation of Affectedness to Lexical Aspect

Carol Tenny (1987, 1992, 1994) has argued that affectedness is subsumable under a
more general aspectual category (see also Cornips and Hulk 1999; Egerland 2000).
7 Rappaport Hovav and Levin (1998, 102-103, 2005, 281-282) also note that surface contact and result

verbs differ in allowing resultatives with non-subcategorized objects (e.g. Sandy wiped/*broke the dishes
off the table). This may follow from a more general property that non-result verbs allow object deletion
more readily than result verbs (Rappaport Hovav 2008, 22-26). I return to this briefly below.
 11

Part of the argument is that some operations that have been tied to affectedness in
previous literature also pick out some unaffected arguments, suggesting a broader
notion is at play. In particular, DP-passives and middles were thought to be restricted
solely to affect arguments (M. Anderson 1979, 14-15, Fiengo 1980, 37-38, Jaeggli
1986, 607-608). Indeed, entailment tests seem to correlate with preposable DPs, as
shown for DP-passives in (28)-(31) and middles in (32)-(35).
(28) a. The Romans just destroyed the city, #but nothing is different about it.
b. The city’s destruction
(29) a. Dan just converted the missionaries, #but nothing is different about them.
b. The missionaries’ conversion
(30) a. John just avoided the traffic jam, but nothing is different about it.
b. #The traffic jam’s avoidance by John
(31) a. The cops just pursued the criminals, but nothing is different about them.
b. #The criminals’ pursuit by the cops
(32) a. John just opened the door, #but nothing is different about it.
b. This door opens easily.
(33) a. John just tightened the cinch, #but nothing is different about it.
b. This cinch tightens easily.
(34) a. Bob just avoided the traffic jam, but nothing is different about it.
b. #Traffic jams avoid easily.
(35) a. Fred just pursued the burglars, but nothing is different about them.
b. #Burglars pursue easily. (cf. Tenny 1994, 8-9, (11), (12), (15), (16))
However, affectedness does not quite cut it (Tenny 1994, 9-10, Jackendoff 1996,
312, fn.7). Path objects are unaffected, as shown in (36a), but also prepose as in (36b).
(36) a. The settlers crossed the desert, but nothing is different about it.
b. The desert crosses easily for settlers with large wagons.
Location objects that are not paths (i.e. that do not define a trajectory) do not prepose:
(37) #The desert wanders easily for settlers with large wagons.
Thus there is something about affected arguments and paths that lumps them together.
Tenny argues that the relevant supercategory is purely aspectual in nature:8
“An affected argument has been generally described as an argument which
undergoes some change. Undergoing change is a temporal process. An af-
fected argument can be more adequately described in aspectual terms, as an
argument which measures out and delimits the event described by the verb.”
“...[t]he term ‘measure out’ is used here in an informal sense, as a convenient
metaphor for uniform and consistent change, such as change along a scale...
A delimited event is one that the language encodes as having an endpoint in
time.” (Tenny 1992, 9, 4-5, emphasis mine)
8 This is part of the Aspectual Interface Hypothesis (Tenny 1992, 2-3, inter alia) that the only lexical

semantic information visible to syntax is aspect; see Rappaport Hovav and Levin (2005) for a rebuttal.
12

Thus all affected arguments on the intuitive sense plus paths are “affected” on Tenny’s
sense. But how can we tell that all preposable DPs measure out and delimit the event?
For delimitation, we must ask whether the predicate is telic and whether this hinges
on the object. Telicity can be determined by in/for tests (Dowty 1979, 56-58). To tell
if an argument delimits the event, we can check whether a definite, specific vs. bare
plural/mass object has an effect on telicity (Tenny 1992, 24-29; see also Garey 1957;
Verkuyl 1972, 1993; Krifka 1989, 1992, 1998; Dowty 1991). For example, in (38)
the expression of the object influences telicity, suggesting that it delimits the event.
(38) a. Sandy ate the peach in/?for an hour.
b. Sandy ate peaches for/??in an hour.
Measuring out is less well-defined, but Tenny (1992, 15-29) offers several tests
for it. One of the clearest is the halfway/half of test: are V X halfway and V half of
X synonymous? If so, then the object measures out the event, since progress through
half of the object corresponds to half of the event. For example, (39a,b) are synony-
mous, suggesting that the patient measures out the event.
(39) a. Sandy ate half of the peach.
b. Sandy ate a peach halfway.
Patients and paths do seem to pattern alike by these diagnostics. The object of
drink seems to measure out the event as in (40a), and delimit it as in (40b,c), and
similarly for paths, as in (41).
(40) a. Jim drank half of the wine ↔ Jim drank the wine halfway.
b. Jim drank the wine in/?for an hour.
c. Jim drank wine for/??in an hour.
(41) a. John crossed half of the desert ↔ John crossed the desert halfway.
b. John crossed the desert in/?for ten days.
c. John crossed deserts for/??in ten days.
These tests do not pick out non-patients as in (42), nor non-paths as in (43).
(42) a. John avoided half of the reunion. = #John avoided the reunion halfway.
b. John avoided the reunion for/*in an hour.
c. John avoided reunions for/*in an hour.
(43) a. John wandered half of the desert. = #John wandered the desert halfway.
b. John wandered the desert for/#in three days.
c. John wandered deserts for/#in three days.
This suggests that patients and paths are indeed two of a kind.
However, I argue that this is an illusion, and suggest that we do need to separate
affectedness from aspect, although they are intertwined. The crucial comparison is
between the affectedness tests outlined in §2 and the aspectual tests outlined here. I
start with motion, where the diagnostics do not pick out the same arguments, as noted
by Jackendoff (1996, 310-311). This is shown in (44) for the ball rolled down the hill,
where the ball is affected but not the path.
 13

(44) a. The ball just rolled down the hill, #but it is not somewhere else.
b. The ball just rolled down the hill, but nothing is different about it.
Conversely, the path but not the ball measures out the event, as shown in (45).
(45) a. Half of the ball rolled down the hill. = The ball rolled down the hill
halfway.
b. The ball rolled down half of the hill. ↔ The ball rolled down the hill
halfway.
Thus the figure is affected and the path measures out, contradicting Tenny’s assertion
that affectedness can be subsumed under measuring out. However, Tenny was right
about one thing. Affected arguments do delimit the event. However, so do paths, even
with the same predicate (see Dowty 1979, 63, Jackendoff 1996, 340-341, Filip 1999,
100-101, Rothstein 2004, 99, Beavers in press). Consider (46a), where a definite,
specific figure moving on a delimited path yields telicity, but a bare plural figure in
(46b) yields atelicity, as does a non-delimited path as in (46c).
(46) a. The ball(s) rolled down the hill in/?for an hour.
b. Balls rolled down the hill for/??in an hour.
c. The ball rolled (further) for/??in an hour.
Thus to fully understand the aspectual properties of the predicate, we need both ar-
guments: the figure is affected and the path measures out, yet both delimit the event.
All of this applies equally well to other predicate types. Patients of change-of-
state verbs also pass affectedness tests but do not measure out, as in (47) for dim.
(47) a. Bill dimmed the lights, #but nothing is different about them.
b. Bill dimmed half of the lights. = Bill dimmed the lights halfway.
However, something does measure out the event in this case, although it is not the pa-
tient argument per se. Rather, as Tenny (1992, 15-16) herself suggests, it is a property
of the patient argument, namely its relative dimness (from fully bright to completely
out). This is shown in (48), where dim half dim and dim halfway are synonymous.9
(48) Bill dimmed the lights half dim ↔ Bill dimmed the lights halfway.
Thus the patient is affected and the property measures out, like figures and paths
(cf. Dowty 1991, 569 on holistic vs. incremental theme). Likewise, telicity is again
doubly determined. A definite, specific object and a specific result determine telicity,
but a bare plural object and/or a vague result (e.g. a comparative resultative; Goldberg
and Jackendoff 2004, 542-543) determines atelicity:
(49) a. Bill dimmed the lights half dim in/?for an hour.
b. Bill dimmed lights half dim for/??in an hour.
c. Bill dimmed the lights dimmer and dimmer for/??in an hour.

9 Context also plays a role, since half dim means different things in different contexts (see Kearns 2007;

Kennedy and Levin 2008 on the role of context in the interpretation of deadjectival “degree achievement”
verbs). To me, both clauses in (48) have the same reading regardless of what constitutes half dim in context.
14

Thus change-of-state predicates and motion predicates pattern alike in separating af-
fectedness and measuring out, but distributing delimitation across two participants.10
So why do affected arguments sometimes measure out the event as in (40)? The
only predicates where this happens are incremental theme (creation/consumption)
predicates such as drink and eat. However, I argue that it is nonetheless a property
of the patient that measures out the event even here. In particular, incremental theme
predicates also show double determination of telicity. This is seen by looking at the
conative alternation (Levin 1993, 41-42, van der Leek 1996; Broccias 2003, Beavers
2006, 58-63, 115-120, 157-163, in press). A transitive version of drink with a definite,
specific object is telic as in (50a). As expected, replacing the object with a mass noun
yields telicity as in (50b). However, keeping a definite, specific DP patient but using
the conative drink at produces an atelic predicate, as in (50c).

(50) a. John drank a pint of ale in/?for an hour.

b. John drank ale for/??in an hour.
c. John drank at a pint of ale (slowly) for/??in an hour.

So something else is relevant for telicity. I suggest it is the property of physical extent
(Hay et al. 1999, 141, Rappaport Hovav 2008, 17). Following Beavers (2006, 104),
(50c) is atelic because it is vague as to how much of the patient is drunk, just as in
(46c) the exact goal is left open or in (49c) the exact result is left unspecified. The
reason the patient seems to measure out the event is because it is hard conceptually to
distinguish the patient from its physical extent, although (50) shows we need to. Thus
in summary, for all result predicates affectedness is a property of the theme/patient,
measuring out is a property of a path/property, and telicity is contingent on both.
However, as Beavers (in press) shows, the double telicity facts are surprising in
light of standard analyses of the correlation of telicity to event participants. For sim-
plicity I assume a predicate φ is telic iff for any event it describes it does not describe
any subpart of that event.11 For example, John built the house is telic, since for any
event this sentence describes, no subevent is also a building of the house (it would be
part of building a house, or building part of a house). However, John slept is atelic,
since any event of John sleeping consists of subevents of John sleeping.
Telicity is often partly contingent on the expression of a privileged argument, usu-
ally called the “incremental theme”. Incremental themehood is standardly analyzed
in terms of a homomorphism that preserves certain mereological properties of incre-
mental theme in the event (and vice versa) (Tenny 1987, 1992, 1994; Krifka 1989,
1992, 1998; Dowty 1991; Jackendoff 1996, inter alia). For creation/consumption,
Krifka (1998, 213, (51)) defines a strictly incremental homomorphism between theme
x and event e that I recast informally as follows:
10 Rappaport Hovav and Levin (2005) argue that change-of-state is a better determinant of argument

realization than aspect. Although they do not discuss this, it implicitly follows that the two should not be
reducible to each other.
11 Krifka (1998, 207, (37)) gives a more sophisticated definition which relies more properly on a notion

of shared beginning and ending points (see also Rothstein 2004, 8). For purposes of this paper I focus on
the simpler notion of telicity as nothing hinges on this per se, though everything I say here can be modified
to satisfy a more sophisticated notion of telicity with no dramatic consequences; see also fn. 12.
 15

(51) Strictly Incremental Relation (SINC): every unique part of e corresponds

to a unique part of x and vice versa.
For predicates that assign this thematic role to their theme arguments, the event pro-
gresses in an isomorphic fashion through the theme. For example, in John drank the
wine, each part of the drinking event corresponds to a unique part of the wine. For any
event of drinking a specific amount of wine that can be described as John drank the
wine, any subevent of that event is a subevent of drinking a smaller unique quantity
of wine, and cannot be described by John drank the wine, which is thus telic. Con-
versely, for John drank wine a subevent of drinking wine is also an event of drinking
wine, and thus the predicate is atelic. The crucial difference is the overt quantization
of the theme: a quantized (e.g. definite, specific) theme induces a telic predicate, and
a non-quantized (e.g. bare plural or mass noun) theme induces an atelic predicate.
For motion and change-of-state predicates, Krifka (1998, 224, (69)) defines a
slightly different relation, given in (52) for path or property scale p in event e.
(52) Strict Movement Relation (SMR): every unique part of e corresponds to
a unique part of p and vice versa; temporal adjacency in e corresponds to
spatial adjacency in p.
For an SMR the event progresses as the figure moves along the path/scale, and each
part of the event corresponds to a unique part of the path/scale. If John walked from
his house to the store describes John crossing a certain path, any subevent of this is
an event of John crossing a strict subpart of that path, and cannot be described by
the same sentence, which is thus telic. For John walked the path is heavily undeter-
mined, so any subevent of John walking is an event of John walking, and the pred-
icate is atelic. The difference hinges on the explicit boundedness of the path. Like
a SINC, the SMR is isomorphic in number of subparts. But an SMR also preserves
adjacency, where temporally adjacent subparts of the event correspond to spatially
adjacent traversal of the path (i.e. telekinesis is ruled out).12
However, the fact that both themes and paths are incremental themes simulta-
neously is a problem for this approach. Naı̈vely, we could assume that each stands
in an independent homomorphic relation to the event. However, this turns out to be
untenable. Consider again (46), repeated here:
(53) The balls rolled down the hill in/?for an hour.
Suppose the event stands in a SINC to the figure and an SMR to the path. Now,
consider a context for (53) in which I stand on top of the hill and throw two balls
down the hill at once. At some point both balls will move simultaneously. Yet this
contradicts the SINC: one part of the event corresponds to two parts of the theme.
However, there is a crucial caveat. While two parts of the theme move at the same
12 Krifka (1998, 219, (59)) defines a more general Incremental Relation (INC) that embeds a SINC but
allows two different parts of the theme to correspond to the same part of the patient in certain contexts
(necessary for John read the book). Likewise, Krifka (1998, 225, (71)) also defines a general Movement
Relation (MR) that embeds an SMR but allows backtracking, loops, and stop-n-go motion. I adopt the
stricter SINC and SMR relations here for expository purposes, though my analysis could easily be recast
in terms of INCs and MRs using the same embeddings. It is primarily for these more complex relations
that the more sophisticated notion of telicity discussed in fn. 11 is required.
16

time, they must do so on different parts of the path. If two balls are entirely co-located
they are the same ball. Thus the violation of the SINC can be mitigated if we make
the homomorphism relative to different parts of the path, a point I return to below.
Furthermore, in the same scenario, one ball could move faster and end up at the
bottom before the other. In this case at some time t the faster ball is rolling along
one point near the end of the path, while at t + 1 the slower ball is rolling along
a non-adjacent point nearer the beginning. This violates the SMR, where temporally
adjacent motion subevents must happen at spatially adjacent points on the path. How-
ever, again, there is an issue of relativity. This problem only arises if we look at dif-
ferent parts of the theme. If we focus on just one (indivisible) part of the theme, it
of course never moves along spatially non-adjacent parts of the path at temporally
adjacent times. This suggests again that we can mitigate the problem if we take both
arguments into account in a mutually constraining way: the SINC only holds for par-
ticular parts of the path, while the SMR only holds for particular parts of the theme.
To sum up so far, I have isolated an empirical notion of affectedness in terms
of affectedness diagnostics, which define subset hierarchies among dynamic predi-
cates. I then compared affectedness diagnostics to aspectual diagnostics, and showed
that they only partly coincide, but are not reducible to one another. Rather, predi-
cates entailing affectedness involve two necessary arguments: an affected theme and
a path/scale that measures out the event, which interact significantly to delimit the
event. I next outline a model that relates all of these facts together.

4 A Ternary Analysis of Affectedness, Measuring Out, and Telicity

I propose a model of affectedness that builds on proposals by Tenny (1987, 1992,

1994), Dowty (1991), Jackendoff (1996), Krifka (1998), Hay et al. (1999), Wech-
sler (2001, 2005), Beavers (2002, 2006, 2008a, in press), Kennedy and Levin (2008),
Ramchand (2008), and Rappaport Hovav (2008). The crucial metaphor is motion.
Intuitively, there is no path without a figure and vice versa, and defining separate the-
matic roles for them is ultimately ill-conceived. Non-motion change is the same (on a
Localist Hypothesis; Gruber 1965, J. Anderson 1971, 1977, Jackendoff 1976, 1983;
DeLancey 2000; Talmy 2000). All entities have properties (e.g. ‘height’), which are
arranged along scales that consist of (a) a set of possible values (e.g. number of
inches) for that property and (b) an ordering relation between them (e.g. taller/shorter)
(Kennedy and McNally 2005, 351). When an entity changes in that property, it “moves”
along the scale. There is no patient without such a scale and vice versa.
Hay et al. (1999) and Kennedy and Levin (2008) apply this analysis primarily
to deadjectival degree achievements (e.g. lengthen, widen), where the scale comes
from the adjective. They focus on how telicity corresponds to boundedness of the
scale. For example, straight has an inherent maximal value (e.g. completely straight),
while wide does not (e.g. #completely wide), and this is preserved in the telicity of
the deadjectival verbs, where straighten is inherently telic but not widen:
(54) a. The workers straightened the pole in/?for an hour.
b. The workers widened the road for/??in an hour.
 17

Following suggestions by Hay et al., Wechsler (2001, 2005) and Beavers (2002,
2008a, in press) extend this analysis to other result verbs, and show that gradability of
the underlying scale correlates with durativity, so that transitions along non-gradable
scales (such as the binary contrast between dead and not dead; cf. #deader or #more
dead) determine punctual events, while transitions along gradable scales (such as the
many degrees of flatness; cf. flatter) can also determine durative events:
(55) a. Wyatt shot the outlaw dead with a single shot/#with multiple shots.
b. Wyatt flattened the tulips with several stomps/with one stomp.
Thus properties of scales (boundedness, complexity) correspond to aspectual proper-
ties of predicates that describe changes along them (see Kennedy and McNally 2005,
361-367 for more on the relationship of event structure and scalar structure). Cru-
cially, this model admits two entities (the theme and scale) into the change relation,
exactly the number of entities needed to capture the indirect correlation of affect-
edness and aspect above. I outline a version of this model here which makes these
predictions, and thereby also provide further support for a scalar analysis of change.
I adopt the mereological event semantic model of motion and property change
of Krifka (1998) (see also Beavers 2006, 2008, in press), where a change involves
a (Strict) Movement Relation of theme x between SOU RCE and GOAL states on
scale s during event e.13 Since I focus only on results here, I assume the SOU RCE is
contextually inferred and pick out results via the following substitution (c=context):
(56) For all dynamic predicates φ, themes x, events e, states g, and scales s:
[[φ(x, s, e) ∧ result′ (x, s, g, e)] ↔ [φ(x, s, e) ∧ SOU RCE(s, bc , e) ∧ GOAL(s, g, e)]]

We can now adopt the following definition for (the strong notion of) affectedness:
(57) An argument x is affected iff there is an event e, a property scale s, and a
state g such that result′ (x, s, g, e) necessarily holds.
Thus result′ defines a BECOME subevent as a necessary transition of an affected
theme to a new state along a scale. What differentiates motion, change of state, and
creation/consumption is only the type of scale selected by the predicate: a location
scale (i.e. a path), a property scale, or an extent scale (Beavers 2008; Rappaport Ho-
vav 2008). Otherwise, these predicates are formally parallel, i.e. they have parallel
semantic decompositions as in (58), where the first conjunct represents the process
(qua a CAUSE subevent), and the second the result (qua a BECOME subevent).
(58) a. John walked to the cafe. (scale s of position of John)
∃e∃s[walk ′ (john, s, e) ∧ result′ (john, s, cafe, e)]
b. John wiped the table clean (scale s of cleanliness of the table)
∃e∃s[wipe′ (john, s, table, e) ∧ result′ (table, s, clean, e)]
c. John ate the apple. (scale s of volume/existence of the apple)
∃e∃s[eat′ (john, s, apple, e) ∧ result′ (apple, s, gone, e)]
13 In this model entities fall into domains of objects U , events U , and connected, directed
P E
paths/scales PH , and form mereological part/whole structures. For any x, x′ ∈ UX , x′ may be a subpart
of x (x′ <X x) or a subpart or equal to x (x′ ≤X x). The sum (or join) of x and x′ is x ⊕X x′ . I assume
definite DPs have direct reference, and I represent constants in boldface. See Krifka (1998, 227-228, (73))
for formal definitions of SOU RCE and GOAL.
18

a. The event: e

b. By xi <P x: e1 ... ei ... en

c. By xi <P x, sj <H s: e11 ... ej1 ... em 1 j

1 ... ei ... ei ... em 1 j m
i ... en ... en ... en

d. By sj <H s, xi <P x: e11 ... e1i ... e1n ... ej1 ... eji ... ejn ... em m m
1 ... ei ... en

e. By sj <H s: e1 ... ej ... em

f. The event: e

Fig. 1: Breakdown of an event in a Figure/Path Relation

On this approach the indirect correlation of affectedness and aspect follows almost
for free, once we define the θ-relation that holds between e, x, and s. The crucial
innovation is that the θ-relation is ternary, and the constraints are all mutually inter-
dependent. For this I adopt the Figure/Path relation proposed by Beavers (in press).
In an FPR, e is decomposed into a series of motion subevents, one for each part
of x crossing all of s. Thus in The balls rolled down the hill the event decomposes
into a series of motion subevents, one for each ball moving along the entire path
down the hill. This replicates the isomorphism of the SINC up to individual motion
events. However, each subevent is itself related to the path via an SMR, meaning each
subevent is further decomposed in terms of the path. Thus each ball crosses each part
of the path at a unique point in the event. This is defined recursively in (59), where an
FPR “slices off” a unique motion subevent ei <E e across s for all xi <P x, where
an SMR holds between ei and s, meaning ei is further decomposable for all sj <H s:
(59) A relation θ is an FPR iff for every e, x, s such that θ(e, x, s):
a. e and s stand in an SMR or
b. there exists xi <P x for which there exists unique ei <E e such that ei
stands in an SMR to s and the rest of x and e stand in an FPR with s.
Beavers offers more explicit details of this breakdown, but I represent it informally
by the diagram in Fig. 1. Line (a) represents the entire event, which is broken down in
(b) into subevents by parts of the theme. By the SMR, each of these subevents can be
broken down again by parts of the scale in (c), holding the part of the theme constant.
These subevents can be rearranged by holding each part of the scale constant and
grouping together each subevent of each part of the theme crossing that part of the
scale as in (d). Joining these produces (e), a series of subevents of the theme crossing
each part of the scale. Joining these produces the entire event again as in (f).
Thus the FPR provides several different perspectives on the event. Regarding
telicity, the two most important are in lines (a) and (f), where the event can be broken
 19

down by either the path or the theme, predicting double telicity effects. For example,
telicity is guaranteed for (46a) as in (60) (the λ-term represents the meaning prior to
existential closure over the event; I ignore tense).
(60) The ball rolled down the hill.
λe∃s[roll′ (ball, s, e) ∧ result′ (ball, s, bottom, e)]
For all e described by (60), for no e′ <E e is there an x′ <P ball nor an s′ <H s
satisfying this description, since x′ 6= ball and s′ is not the full path described by
(60). Thus e′ is not in (60), and the predicate is telic. If the quantity of the theme or
boundedness of the scale is left open, as in (46b,c), the predicate is atelic:
(61) a. Balls rolled down the hill.
λe∃s∃x[roll′ (x, s, e) ∧ ball′ (x) ∧ result′ (x, s, bottom, e)]
b. The ball rolled (further).
λe∃s∃g[roll′ (ball, s, e) ∧ result′ (ball, s, g, e)]
For any e described by (61a), for some e′ <E e there is x′ <P x such that ball′(x),
meaning e′ satisfies (61a), making it atelic. Likewise, for all e described by (61b), for
some e′ <E e there is a s′ <H s satisfying this description (since no constraints hold
for s other than having some goal), making (61b) atelic (see Krifka 1998, 224-225).
This extends directly to change-of-state and creation/consumption. Thus double
telicity for dim in (49) and drink in (50) follow in exactly the same way, since the
representations are formally parallel to (60)-(61):
(62) a. Bill dimmed the lights half dim.
λe∃s[wipe′ (bill, s, lights, e) ∧ result′ (lights, s, 21 dim, e)]
b. Bill dimmed lights half dim.
λe∃s∃x[wipe′ (bill, s, x, e) ∧ light′(x) ∧ result′ (x, s, 12 dim, e)]
c. Bill dimmed the lights (dimmer and dimmer).
λe∃s∃g[wipe′ (bill, s, lights, e) ∧ result′ (lights, s, g, e)]
(63) a. John drank a pint of ale.
λe∃s[drink ′ (john, s, ale, e) ∧ result′ (ale, s, gone, e)]
b. John drank ale.
λe∃s∃x[drink ′ (john, s, x, e) ∧ ale′ (x) ∧ result′ (x, s, gone, e)]
c. John drank at a pint of ale.
λe∃s∃g[drink ′ (john, s, ale, e) ∧ result′ (ale, s, g, e)]
Thus telicity is not about the entire theme being affected, nor is it about traversal of
an entire path or scale. Telicity is instead a property a predicate has when all of the
theme crosses all of the path/scale, an inherently relational definition.
Second, something not discussed by Beavers (in press) is that the theme/path
distinction gives us the space to distinguish measuring out from affectedness, so that
in V half of X half of the theme is predicated of, but in V X halfway half of the scale is,
i.e. halfway is a scalar modifier like completely. This is illustrated in (64) for scanning
a brain, in a context where a complete scan takes ten passes over the brain. Scanning
half of the brain entails scanning half of the brain ten times, while scanning the brain
halfway entails scanning the entire brain five times:
20

(64) a. McCoy scanned half of Spock’s brain.

∃e∃s[scan′ (mccoy, s, 12 brain, e) ∧ result′ ( 21 brain, s, scanned, e)]
b. McCoy scanned Spock’s brain halfway.
∃e∃s[scan′ (mccoy, s, brain, e) ∧ result′ (brain, s, 21 scanned, e)]
Thus we should not expect halfway and half of to correlate, contra Tenny. The degree
to which they do is simply the degree to which the scale represents the physical extent
of the theme, where it is hard to tell the theme from the scale.
In summary, the scalar model of change I adopted recognizes two entities, the
affected argument and the scale of change, which stand in a mutually constraining
relation to each other and the event (as opposed to a binary θ-relation á la Parsons
1990). This provides a natural way to capture the indirect connection of affectedness
and aspect, and thus provides additional motivation for a scalar analysis of change
beyond the scale/event structure correlation noted in previous work. Furthermore, we
have posited one core element (result′ ) in common for all result predicates. In the
next section I return to the affectedness diagnostics of §2 and the subset hierarchy they
form, and argue that they can be modeled naturally in terms of a theory of degrees of
affectedness which hinges on this analysis.

5 Modeling Degrees of Affectedness

As mentioned in §1, affectedness is often assumed to be a matter of degree, with a

contrast between high vs. low affectedness. However, high and low affectedness are
hard to define precisely, since this putatively involves defining what sorts of changes
constitute more of an effect than others. Of course, on a reduction of affectedness
to aspect à la Tenny, one can define high affectedness as telicity and low affected-
ness as atelicity. However, as argued above, we do not want to reduce affectedness
to aspect, although I ultimately suggest that telicity does figure into identifying high
affectedness. In the non-aspect literature, high and low affectedness are rarely given
precise or consistent definitions, likely due to the fact that they are usually evoked to
explain other phenomena but are rarely studied themselves. As noted above, Hopper
and Thompson (1980, 252-253) claim that “[t]he degree to which an action is trans-
ferred to a patient is a function of how completely that patient is AFFECTED” (em-
phasis theirs), although they never define degree of transfer. Conversely, Næss (2003,
1202) defines higher effects as those that either affect more of the object or are more
salient. Næss motivates the latter property by comparing kill to break, where kill en-
tails higher affectedness because it has more dramatic consequences for humans. I
have no idea how to codify this, and I doubt it has any real linguistic consequences.
The part/whole relation is also complicated, since it is as much a question of
amount as of effect. Do drink beer vs. drink the beer differ in degree of affectedness
or in how much beer was affected to the same degree? Likewise, changes that affect
small parts of entities might be greater than those that effect more. If I chip the rim of
a wine glass is it less affected than if I smudge it all up? In principle we should keep
quantity distinct from degree of affectedness, and the FPR gives us the tools to do
this, by separating themes from scales. I propose here a principled, linguistically mo-
tivated definition of degree of affectedness. The crucial factor is specificity: the more
 21

specific the predicate is about the end of the theme’s progress on the scale, the higher
the degree of affectedness. Furthermore, we do not need to decide whether killing
or breaking yields more of an effect. Instead, we can isolate equivalence classes of
changes through affectedness tests and aspect, side-stepping the fuzziness issue.
I start with the quantized vs. non-quantized change contrast of Hay et al. (1999,
132-138), illustrated in (65a,b) respectively (see also the indefinite/definite change
contrast of Dowty 1979, 168-170). In (65a) the jeans become exactly five inches
longer, so that a specific result is entailed, and the predicate is telic. In (65b) they are
known to have become longer, but not how much, and the predicate is atelic.
(65) a. The tailor lengthened the jeans 5ins in/?for an hour.
b. The tailor lengthened the jeans for/??in an hour.
However, both predicates pass affectedness tests, so that both entail an effect.
(66) The tailor lengthened the jeans (5ins), #but nothing is different about them.
Following Hay et al. (1999, 132), quantized and non-quantized changes both involve
transition of a theme x along a scale s in event e, but differ in how specific the
predicate φ is about the result.14 For quantized change φ supplies a specific endpoint
gφ on s, while in non-quantized change φ only entails that an endpoint exists.
But what about the unaffected force recipients in §2, e.g. objects of surface contact
and impact verbs that permit result XPs and pass What happened to X is Y but do not
necessarily change? Here I follow Tenny (1992, 20) and assume that such predicates
have a scale argument (and thus are compatible with change), but do not entail a
result′ and thus an FPR may not hold.15 Finally, a non-force recipient is anything
that has a thematic role, with or without a scale argument Building on the notion of
result′ defined in the last section, I codify these four thematic relations as in (67) for
predicate φ and theme x, with examples of verbs that assign them:
(67) a. x undergoes a quantized change iff φ → ∃e∃s[result′ (x, s, gφ , e)]
(e.g. accomplishments/achievements: break, shatter, destroy, devour x)
b. x undergoes a non-quantized change iff φ → ∃e∃s∃g[result′ (x, s, g, e)]
(e.g. degree achievements/cutting: widen, lengthen, cut, slice x)
c. x has potential for change iff φ → ∃e∃s∃θ[θ(x, s, e)]
(e.g. surface contact/impact: wipe, scrub, rub, punch, hit, kick, slap x)
d. x is unspecified for change iff φ → ∃e∃θ[θ(x, e)]
(e.g. all others: see, smell, follow, play (as children), ponder x)
Each type of affectedness is built around the notion of result′ , but differs in how
certain arguments are bound. This defines a classification of predicates that differs
from the one in (5) in §2. The first was based on types of changes (motion, creation,
14 Hay et al. (1999) and Kennedy and Levin (2008) analyze this via the “difference degree” between the
initial and final states on the scale. As result′ here stands for a transition from a contextually supplied
initial state to a final state, the approach I outline is easily translatable into those terms.
15 Alternatively, we could assume potential is about a change, but the relevant truth conditions have a

probability less than 1 but greater than 0 of obtaining, or obtain at some possible world but not necessarily
the real one. These proposals would work just as well in the foregoing discussion.
22

consumption, property), and thus on the type of the scale. This classification is based
on scalar structure (where the theme ends up), regardless of type.
However, Rappaport Hovav (2008) argues for a simpler classification of dynamic
predicates, between those that have a scale and those that do not. The former includes
predicates that entail scalar change (cool, widen) and the latter includes mainly activ-
ities (laugh, see).16 Her scalar predicates are my (non-)quantized change predicates,
where I simply distinguish between specific and non-specific results, an uncontro-
versial subdistinction. Her non-scalar predicates are my potential and unspecified
predicates, and here our classifications differ more dramatically. While we agree that
most activities such as laugh and see have no scalar structure (my unspecified change
class), my class of potential change predicates consists of surface contact and impact
predicates, which have unaffected force recipient arguments. I analyze unaffected
force recipients as being associated with an underlying, latent scale. However, Rap-
paport Hovav proposes that surface contact verbs like scrub do not lexicalize scales
(she does not discuss impact predicates). She argues for this based on the fact that
surface contact verbs allow a relatively wide range of result XPs (including results
for non-subcategorized objects), permit object deletion, and allow out- prefixation,
none of which apply to result verbs like dim, suggesting they do not form a class (cf.
ibid.: 22-24, (12a,c), (16a,c), (18b,c)):
(68) a. Cinderella scrubbed her knees sore/the table clean.
b. We dimmed the room half dim/*empty.
(69) a. All last night, Cinderella scrubbed.
b. All last night, we dimmed *(the lights in the house).
(70) a. Cinderella outscrubbed her stepsisters.
b. *Our stage-hand outdimmed your stage-hand.
This is explained, Rappaport Hovav suggests, if verbs that lexicalize scales only take
result XPs compatible with the lexicalized scale, and furthermore must overtly realize
the affected argument (ruling out object deletion and non-subcategorized objects).
Verbs that do not lexicalize scales ipso facto allow more variability in objects and
result XPs. This explains the difference between scrub and dim.
However, the fact that these verbs allow result XPs and pass What happened to
X is Y (i.e. have force recipient arguments) suggest that they nonetheless share at
least something in common with results verbs. I propose that what they share is in-
deed a latent scale argument, though they differ in that result verbs entail a necessary
transition along the scale and potential change verbs do not, like non-scalar verbs.
Thus I predict that impact/surface contact verbs should share properties in common
with both result verbs and non-scalar verbs. While I do not have any new insights on
the behavior of non-scalar verbs or Rappaport Hovav’s data, I show below that my
analysis is useful for understanding how these verbs pattern like result verbs.17
16 Rappaport Hovav also argues that incremental theme verbs (eat, drink, read) do not have scale argu-
ments by default, but rather the scale is provided by their object (which in the case of eat is also affected
non-incrementally) (see also Ramchand 2008, 25-33). I offer a similar analysis for read in §6.2, but the
conative data I discuss in §3 argues against this analysis of consumption predicates.
17 Beth Levin (p.c.) suggests that the ease with which surface contact and impact predicates allow result

XPs may be because the manners they encode lead naturally to certain types of results, but that this is a
 23

One of the biggest advantages of my definitions in (67) is that they are related to
one another in terms of monotonically weakening truth conditions, defined by exis-
tential generalization. Non-quantized change is an existential generalization over the
goal state of a quantized change, potential for a change is an existential generaliza-
tion over the relation between the theme, scale, and event, and being unspecified for a
change is an existential generalization over the thematic role of the theme, producing
the following implicational Affectedness Hierarchy:
(71) The Affectedness Hierarchy: for all x, s, φ, e,
∃e∃s[result′ (x, s, gφ , e)] → ∃e∃s∃g[result′ (x, s, g, e)] → ∃e∃s∃θ[θ(x, s, e)] → ∃e∃θ[θ(x, e)]

Because this hierarchy is implicational, if x bears degree of affectedness n it bears all

degrees less than n. On these definitions, undergoing a quantized change (reaching a
definite result state) entails undergoing a non-quantized change (reaching some result
state), which entails potential for change (there is a scale argument), which entails
being in the event. Since this is based on weakening truth conditions, the hierarchical
arrangement of (67) into (71) follows for free, independent of any data that lead us to
posit these degrees of affectedness. This is an important property of (67). Recall that
the affectedness tests in §2 classify predicates into subset relations, where predicates
that pass m tests are a subset of those that pass m − 1 tests:
(72) Dynamic predicate φ entails
Diagnostics specifc result non-specific result force recipient non-force recipient
Telic X × × ×
Change entailed X X × ×
Permit result XP X X X/× ×
Happened/did to X X X ×
Dynamic X X X X/×
Result XP variation Low Low/High High N/A

I now show how the subset property follows directly from (71).
As discussed in §4, explicit boundaries on s generate telicity (so long as x is ex-
pressed by a definite, specific DP), so that telicity follows for free from (67a). None
of (67b-d) meet this condition, thus they determine atelicity. Next, (67a,b) differ from
(67c,d) in that only the former entail a result. This means (non-)quantized changes
should pass something is different about X or X is somewhere else. Stating the rel-
evant truth conditions is simple due to the Affectedness Hierarchy. Since quantized
changes entail non-quantized changes, all quantized change predicates satisfy all of
the truth conditions for non-quantized change predicates. Thus we need only state
the conditions for affectedness tests in terms of non-quantized changes, and this will
naturally subsume all quantized changes (where the difference between motion and
other types of changes has only to do with scale type):
(73) a. x is somewhere else is true iff there exists e such that x undergoes a non-
quantized change along a physical path in e.
b. something is different about x is true iff there exists e such that x under-
goes a non-quantized change along an abstract scale in e.
matter of real world knowledge rather than lexicalization. However, it is difficult to tell real world knowl-
edge from lexicalization when it affects the grammatical behavior of the verb in question, and as I show
below my analysis captures the mixed behavior of these verbs in a more perspicuous way.
24

Due to the entailment property, (73) pick out (67a,b), but exclude (67c,d). Participle
tests can be defined similarly for the specific scale lexicalized by a specific verb.
Next, What happened to X is Y picks out (67a-c) but excludes (67d). We can again
use the Affectedness Hierarchy to pick out the the strongest set of common truth con-
ditions, namely potential for change (ignoring non-agentivity and prior existence):
(74) What happened to x is φ is true iff x has potential for change in φ.
Due to the Affectedness Hierarchy, this picks out all (non-)quantized and potential
changes. Without latent scales, this would be a difficult class to define.
Finally, recall from §2 that English allows result XPs for force recipients, while
Japanese only allows result XPs for (implicated) affected arguments. The English
case follows from independent properties of resultatives. As Wechsler (2001, 2005)
and Beavers (2002, 2008) have argued, result XPs impose constraints on the end-
point and mereological complexity of the scale argument of the verb. This requires
a scale argument to be present, ruling in all of (67a-c) but ruling out (67d) (unless
recontextualized to involve a scale). Japanese simply has a stronger condition:
(75) A Japanese predicate φ permits a result XP iff it has a theme argument x that
is implicated to undergo a non-quantized change.
Furthermore, recall that some verbs entailing results take fewer result XPs than those
with unaffected force recipients. For (non-)quantized change, shatter and cool allow
only result XPs describing large numbers of pieces or degrees respectively, reflecting
the degree of specificity these verbs already put on their result/scale arguments:
(76) a. John shattered the vase into a million/thousand/thirty-six pieces.
b. #John shattered the vase slightly/into two pieces/in half/silly/flat/red/up.
(77) a. John cooled the soup to 50/100/250 degrees/down/to a palatable level.
b. #John cooled the soup red/congealed/into mush.
However, recall also that some non-quantized change predicates allow more types of
results than cool, such as cut, which allows various result XPs of physical rending:
(78) a. John cut the bread into ten/a million pieces/an amusing shape/slightly/up.
b. #John cut the bread clean/silly/flat/red.
Both cut and cool are non-quantized change predicates, so how do we capture the dif-
ference? This can be accommodated in terms of weaker or stronger sortal constraints
on scale/result arguments. Any predicate that selects a result or scale from the do-
main PH of scales may restrict that result/scale to some X ⊂ PH . For two predicates
selecting result/scale from X ⊂ PH and Y ⊂ PH , if |X| < |Y | then the former
predicate imposes stronger conditions on possible results/scales. Indeed, this may be
found even among quantized change verbs. As Boas (2003, 216-233) notes kill al-
lows at best only dead as a result XP, while shatter may allow either to (X) pieces
(for large X) or a very narrow set of result locations (all over the table), so that two
quantized change predicates differ in allowing one or two types of results/scales.18
18 Alternatively, the difference may have to do with total vs. partial ordering. Degrees for cool are totally

ordered (for any two different degrees one is always cooler than the other), but for cut some results of
 25

Furthermore, since potential change verbs take latent scale arguments, we might
expect sortal constraints on possible results. A glance through the BNC resultative
data collected by Boas (2003, 321-240, Appendix A) suggests that surface contact
and impact verbs largely take two or three types of result XPs.19 First is motion,
which for surface contact verbs involves removal/dispersal of some element on the
surface of the force recipient, and for impact verbs involves this or motion of the
force recipient (apart, into the ground, off, open, over the edge, over the top, shut).
Both classes also allow result XPs denoting states that follow from removal (dry,
empty, clean) and thus might be reduced to an extension of motion. Finally, both
classes also involve various types of physical change (flat, to pieces, soft, smooth) or
mental change (awake, unconscious, silly, stupid, sore, dead, to death, over the edge
(metaphorically)). The physical/mental change classes seem to be a hodge-podge,
though most fit the mold of adverse change or “damage” (one possible exception is
kick awake, attested once in the sample). More detailed corpus work is needed to
determine what, if any, sortal constraints these verbs place on their scale argument.
Finally, an unspecified result predicate allows a result XP only if it can be con-
strued of as having a scale (e.g. He sneezed the napkin off the table). However, as Boas
(2003, 260-277) discusses, this involves a very rich context and a process of analogy
(cp. He blew the napkin off the table (with a sneeze)), and is thus best thought of as
wholesale coercion of the verb’s meaning to another type.
I thus revise (72) as follows, where the tests follow the Affectedness Hierarchy:
(79) Degrees of Affectedness
Diagnostics Quantized Non-quantized Potential Unspecified
Telic (for quantized theme) X × × ×
Change entailed X X × ×
Permit result XP X X X/× ×
Happened/did to X X X ×
Dynamic X X X X/×
Result XP variation Low Low/High High N/A

The Affectedness Hierarchy therefore gives us the subset property of affectedness

tests for free, since it defines subset relations through weakening truth conditions
which different tests are sensitive to. To my knowledge, this notion has never figured
into work on affectedness, yet seems crucial for understanding the subset relations.
Furthermore, assuming latent scalar arguments for force recipients allows us to slot
these verbs into a position along the hierarchy along with other types of verbs, giv-
ing us a way to both group them together with result verbs (the existence of a scale
argument) and also separate them from result verbs (no necessary change). Indeed,
if the conditions on obligatory realization of affected arguments are stated on non-
quantized predicates, this predicts that potential and unspecified change predicates
will pattern alike according to object deletion and out- prefixation, while still admit-

physical rending may not be orderable, e.g. cut a piece of paper into a dog shape vs. cut a piece of paper
into a cat shape may reflect change along two different branches of a partially ordered space of cutting.
19 Attested impact verbs in Boas’s sample are ax, bang, bash, batter, bayonet, beat, bludgeon, bump,

bust, chip, club, dab, gun, hack, hammer, kick, knock, pat, paw, peck, pound, pummel, punch, ram, shoot,
slam, smack, spear, stab, stamp, stomp, stone, strike, trample, whip, while attested surface contact verbs
are brush, caress, comb, dust, lick, mold, polish, press, rinse, rub, sand, scour, scrape, scrub, skim, smooth,
swab, towel, wash, wipe, taking a liberal view of what counts in each class.
26

ting that potential change predicates have something in common with result verbs
that leads to result XP predication in English and What happened to X is Y.
Now that we have defined an independent ranking of types of affectedness, we can
analyze degrees of affectedness. In §4 I defined affectedness as entailing a result′ .
This subsumes quantized or non-quantized changes, which we might relabel “high”
and “low” affectedness. What distinguishes them is telicity, but this is just a side
effect of how specific the predicate is about what change occurs. In other words,
linguistically speaking, high affectedness is whatever specificity leads to telicity, and
low affectedness is everything else. This side-steps the thorny issue of identifying
which real world changes constitute high or low affectedness, a welcome result. Of
course, one may wish to call potential change a very low degree of affectedness.
This is a terminological issue, but the Affectedness Hierarchy gives us the relevant
categories and relates them together so that this choice is nonetheless principled.
Conversely, unspecified change predicates do not have a scale. Thus while they are
not incompatible with an affected reading, it involves a more dramatic coercion of
their argument structure, making them less compatible with affected readings.
The test case is to compare the Affectedness Hierarchy back to work in transitiv-
ity, where Hopper and Thompson (1980) note that clauses that entail higher degrees
of affectedness are more likely to be transitive than those showing lower degrees
of affectedness.20 Tsunoda (1981, 1985) and Malchukov (2005) take this further by
positing hierarchies of basic verb meanings corresponding to their relative degree of
transitivity across languages. Malchukov (p.81) gives the following hierarchy from
most to least likely to be transitive, based on properties of the patient argument:

(80) effective action > contact > pursuit > (motion)

(‘break’) (‘hit’) (‘wait/search for’) (‘go’)

His motion class consists entirely of non-causative motion verbs, which have only
one core, obligatory argument (the theme) and thus are inherently intransitive (ex-
cept when taking a path object, in which case they are categorically transitive by def-
inition and correspond to non-quantized change). But the other three classes clearly
follow the Affectedness Hierarchy (though no distinction is noted between quantized
and non-quantized change). Thus the Affectedness Hierarchy conforms to hierarchies
independently motivated by transitivity, and in fact provides an independent explana-
tion for such hierarchies based on strength of truth conditions about the effect.
Finally, an interesting related proposal is given in Washio (1997, 38-42), who pro-
poses five types of thematic roles given in Figure 2, which he argues form a contain-
ment relation: Patient1 s are a subset of non-Patients, and for all 1 ≥ i ≥ 3 Patienti+1 s
are a subset of Patienti s. Washio argues that English allows result XPs for all Patients,
but Japanese only allows them for Patient3 s and Patient4 s, i.e. a subset of their dis-
tribution in English. Likewise, the increasing specificity of these Patients figures into
possible result XPs, just as discussed above (see Washio 1997, 7-14).
However, there are a few problems with Washio’s classification. First, there is
clearly not a containment relation in Fig. 2, since non-Patients are stipulated to not
20 Hopper and Thompson (1980) do not define transitivity grammatically, but it is usually taken to mean

nominative/accusative or ergative/absolutive case-frames (Tsunoda 1981, 1985; Malchukov 2005).

 27

non-Patient the verb lexically specifies that it is not affected; hence it may not
undergo any change of state; e.g., see her (*stiff).
Patient1 the verb, being intransitive, lexically specifies nothing about [result
state]; it may be interpreted as “affected” by virtue of discourse or
pragmatics; Jackendoff’s [(1990, 294)] discourse patient; e.g. run
(the pavement) thin.
Patient2 the verb lexically specifies that it is affected; hence it may undergo
some change of state; but the verb does not specify whether or how
it changes; e.g., drag the logs (smooth).
Patient3 the verb lexically specifies that it is affected; hence it may undergo
some change of state; the verb does not specify whether it actually
changes its state or not; but the verb specifies that, if it does change,
then it changes in certain fixed directions (the verb has a disposition
toward certain states); e.g., wipe the table (clean).
Patient4 the verb lexically specifies that it undergoes some specific change
of state; hence it is also affected; e.g., sharpen the pencil (pointy).

Fig. 2: Washio’s Patients (Washio 1997, 40, (116), (115))

be affected, Patient1 is stipulated to be unspecified for an affect, and Patient2 through

Patient4 are stipulated to be affected, all of which are inconsistent. Second, the Patient1
vs. Patient2 distinction hinges on a grammatical property (transitivity), but this has
nothing necessarily to do with affectedness (and indeed the examples illustrating the
contrast are simply unaccusative vs. causative motion verbs).21 Third, the Patient2
vs. Patient3 contrast is not motivated by anything other than the resultative facts it is
meant to explain. My approach avoids these pitfalls since (a) the Affectedness Hier-
archy is based on weakening truth conditions and thus properly defines a containment
relation, (b) the hierarchy is thus independently motivated (determined by existential
generalization), and (c) it is based solely on semantics, where we should expect a
theory of affectedness to be based.
Nonetheless, the intuition behind Washio’s classification is clear, and it can eas-
ily be related to mine. (Non-)quantized change constitute his Patient4 , making an
additional distinction between specific results vs. non-specific results. His Patient1
through Patient4 constitute my potential change, divided by how specific specific a
predicate is about possible changes, which I captured above in terms of sortal con-
straints. Finally, being unspecified for a change corresponds to Washio’s non-Patient.
Thus the classifications are quite compatible, though they differ in details.
In summary, the degree of affectedness reflects how specific a predicate is about
the result, and the Affectedness Hierarchy captures four linguistically relevant levels
of specificity, picked out by clusters of properties. As discussed in §2, the relevant
properties classify predicates into subset relations, and the Affectedness Hierarchy
gives us an independently-motivated way to capture this. I turn next to two applica-
tions of this theory to other syntactic phenomena implicated in affectedness.
21 Washio (1997, 8) assumes without motivation that all intransitive verbs by definition do not encode

result states, but never discusses unaccusatives, the classic case of intransitives with patient arguments.
28

6 Applications and Extensions

6.1 Object/Oblique Alternations

Beavers (2006) shows that this model of affectedness provides a uniform, consistent
way to analyze certain object/oblique alternations. In general, more affected argu-
ments are more likely to be objects. Consider the conative alternation for eat in (81).
When the theme is an object, it must be entirely consumed (to a contextually or
conventionally determined degree). When it is an oblique, it need only be partly con-
sumed (the “holistic/partitive” contrast of S. Anderson 1971b, 389).22 In both variants
the theme is affected, as in (82).
(81) a. Kim just ate a slice of pie, #but most of it is still there.
b. Kim just ate at a slice of pie, but most of it is still there.
(82) a. Kim just ate a slice of pie, #but nothing is different about it.
b. Kim just ate at a slice of pie, #but nothing is different about it.
There is also telicity contrast. The object variant is telic but the oblique variant is not:
(83) a. Kim ate a slice of pie in/?for 30 seconds flat.
b. Kim ate at a slice of pie for/??in five minutes.
Given these facts, complete vs. non-complete change can be analyzed as quantized
vs. non-quantized change, where the relevant specific result is “completely changed”.
However, there are other contrasts. Consider the conative with cut/slice verbs:
(84) a. John cut/sliced the tire.
b. John cut/sliced at the tire.
When the theme is an object it is affected, but not when it is an oblique:
(85) a. John just cut/sliced the tire, #but nothing is different about it.
b. John just cut/sliced at the tire, but nothing is different about it.
Furthermore, both variants are atelic:
(86) a. John cut/sliced the tire for/??in an hour.
b. John cut/sliced at the tire for/??in an hour.
Thus (84a) represents non-quantized change; the tire could be more or less cut in
context. The oblique undergoes no necessary change. However, to my ears, the theme
in both cases passes What happened to X is Y (though (87a) sounds slightly more
natural), and thus is always a force recipient.
(87) a. What happened to the tire was that John cut/sliced it.
b. ?What happened to the tire was that John cut/sliced at it.
22 The exact notion of completeness is also contingent on context (e.g. whatever constitutes complete

for food of that type in a given context), which I set aside here (see Beavers 2006, 50-57).
 29

Next, consider the conative with impact verbs. In neither variant is the theme
affected, and both are atelic:
(88) a. John just hit/kicked the wall, but nothing is different about it.
b. John just hit/kicked at the wall, but nothing is different about it.
(89) a. John hit/kicked the wall for/??in an hour.
b. John hit/kicked at the wall for/??in an hour.
Crucially, only the object variant passes What happened to X is Y, suggesting that the
theme of hit/kick is a force recipient while the theme of hit/kick at is not:
(90) a. What happened to the wall is John hit/kicked it.
b. #What happened to the wall is John hit/kicked at it.
Finally, there are some verbs that simply show no semantic contrast in conative
alternations (e.g. chew (at), gnaw (at/on)). But among the ones that do, different
object/oblique alternations pick out different contrasts, summarized as follows:
(91) a. Incremental theme verbs: eat (at) a slice of pie.(quantized/non-quantized)
b. Cutting verbs: cut/slice (at) the rope. (non-quantized/force recipient)
c. Impact verbs: hit/kick (at) Defarge. (force recipient/not a force recipient)
As the affectedness and aspect tests show, semantically contentful object/oblique al-
ternations reflect minimal contrasts along the Affectedness Hierarchy, as in (92).
(92) Contrast : quantized → non-quantized → potential → unspecified
eat conative : DO ⇔ OBL
slice conative : DO ⇔ OBL
kick conative : DO ⇔ OBL

Beavers (2006) gives a detailed analysis of these facts in terms of the lexical entailment-
based approach of Dowty (1991), which I recast informally as follows:
(93) Morphosyntactic Alignment Principle (for objects): The theme is affected
to degree n in the object variant and degree n(−1) in the oblique variant.
This is not a complete analysis of any particular object/oblique alternations (e.g.
Guerssel et al. (1985) suggest that having an underlying instrument is also a con-
dition on conatives). Rather, (93) is a constraint on possible alternations, i.e. what a
possible semantic contrast is. Furthermore, this analysis suggests that the condition-
ing factor for these alternations is strength of truth conditions, not different subevent
structures as is commonly assumed, something only predicted on an analysis like this.

6.2 The Affectedness Hierarchy and DP-Preposing

As discussed in §3, it was typically thought that only affected arguments undergo
DP-preposing (middle and DP-passives), but as Tenny noted, paths also prepose, sug-
gesting that we need a category encompassing both. However, if we look not at the
DPs that prepose, but at the overall predicates, a different picture emerges. First, I
30

take it as given that only monotransitive verbs allow preposing (ditransitives do not;
cf. *This book gives/sells him easily/*He sells/gives the book easily).23 The mono-
transitives that allow preposing (focusing here on middles) include the following (the
classes can be verified by the affectedness and aspectual tests above):
(94) a. Predicates with objects that undergo quantized changes:
Bob carved the wood into a toy/This wood carves easily (with a penknife).
b. Predicates with objects that undergo non-quantized changes:
Bob pushed the mower/This mower pushes easily (on short grass).24
c. Predicates with subjects that undergo quantized changes:
Bob crossed the desert/This desert crosses easily (with large wagons).
Verbs with force recipient objects do not undergo preposing (cf. *These footballs
kick easily), nor do verbs with objects unspecified for a change (cf. *Soft clothes
touch easily). Although the preposed themes in (94a,b) are different from the pre-
posed path in (94c), the predicates all share one semantic property: they all entail
(non-)quantized change. The difference is that in (94a,b) the theme is the object, but
in (94c) the theme is the subject and the scale (a physical path) is the object. If we
ignore argument realization for now, we can give the concise definition of predicates
that allow preposing in (95), though as noted in fn. 23 further constraints may apply.25
(95) Affectedness Constraint (Preliminary): If a transitive predicate entails a non-
quantized change, its external argument may be eliminated.
Just as in §5, all quantized changes entail non-quantized changes, so (95) picks out
all predicates that entail either kind of change. Note that there is one missing class in
(94): no predicates with non-quantized theme subjects and path objects allow prepos-
ing. However, such predicates are never transitive since path objects always corre-
sponds to quantized change (Beavers 2006, 152-156):
(96) a. The hiker walked the mountain trail in/?for three hours.
b. The hiker walked up/along/down the mountain trail for/*in three hours.
So why do both themes and paths prepose? This is due to independent argument
realization constraints. By the FPR, verbs with theme arguments also take a path/scale
argument and vice versa. Since themes and paths/scales co-occur, the primary point
of variation in the argument structure of such verbs is the presence of an agent:
23 Fagan (1992, 76-81) also notes that preposed objects must be responsible for the event in question

(following van Oosten 1977, 1986). This predicts the contrast between buy and sell shown in (i):
(i) This car sells/*buys easily.
With sell a property of the car makes it sell well, but with buy this is usually not the case. For example,
it is appropriate to answer How did Alex manage to sell the Jaguar? with It’s a great car, a real bargain,
but less appropriate to give the same answer to How did Alex manage to buy the car? Tenny (1992, 10, fn.
6) likewise notes that middles are subject to considerable speaker variability and are more restricted than
DP-passives. I other constraints on DP-preposing aside for now, focusing just on affectedness.
24 This use of push necessitates motion. The static push on reading does not allow a middle.
25 I assume DP-preposing is an operation of external argument elimination followed by raising of the

object, to distinguish it from passives, which are external argument absorption (see Jaeggli 1986).
 31

(97) Possible dynamic predicate θ-grids: < (Agent,) Theme, Path/Scale >
Crucially, both themes and paths can be objects. However, it is uncontroversial that
agents are selected as subject over themes by default, and themes are objects over
paths by default. Thus if there is an agent, the theme can be object. If not, the theme
can be subject and the path can be object. Whichever is object is the thing that pre-
poses, so that nothing semantic unifies preposable DPs. Rather, the relevant con-
straints are not about the semantics of particular arguments, but about predicates as
a whole, statable in terms of the Affectedness Hierarchy. Pure syntax (e.g. how Case
checking and A-movement work) determines the rest.
However, there are three apparent exceptions to (95) brought up by Fagan (1992,
68-72), who argues against affectedness as the main factor for middle formation.
These include certain activity predicates in (98), performance verbs such as read,
play, recite, utter, and perform (cf. Tenny 1992, 17-18), and certain manner of motion
predicates that are transitive but do not entail a result as in (100) (see also Fiengo
1980, 49-50, (89) for more such examples):
(98) a. This pipe smokes nicely.
b. A person who isn’t self-conscious photographs well.
(99) a. This piano plays easily.
b. This book reads quickly.
(100) a. The car drives easily
b. This racetrack drives smoothly.
I suggest that these data do not contradict (95), and in fact even support it. In partic-
ular, (98) actually do entail a (non-)quantized change. In (98a) there must exist some
substance that is smoked, and in (98b) a photograph must have been created. Verbs
like smoke include verbs where the object is the instrument used in the change-of-
state, such as wield X against Y, and verbs like photograph include representation-
source verbs (e.g. paint, copy, translate). Thus (98) support (95), since the relevant
degree of change is entailed, even if it is not associated with an overt argument.
The performance verbs in (99) form an interesting class regarding my analysis of
affectedness. Tenny (1992, 73) analyzes these as abstract motion, where the theme
(e.g. a performer) traverses a piece (a series of notes or words) from one end to the
other, much like a path (see also Jackendoff 1996, 232-233). In (99a) this “path” is
left unexpressed, while in (99b) the “path” is the subject. On this analysis it is natural
that an FPR is the underlying θ-relation.26 Indeed, such predicates do display the
same aspectual properties as motion verbs, including the halfway/half of correlation
for the performance object (cp. path objects) and double telicity effects:
(101) a. We performed half of the play. ↔ We performed the play halfway.
b. The troupe performed the play in a record two hours.
c. Troupes performed the play ??in a record two hours.
d. The troupe performed plays ??in a record two hours.
26 Krifka (1998, 218) notes that with read backtracking is allowed, suggesting that strict incrementality

must be relaxed slightly as in fn. 12.

32

(102) a. The student read half of the book. ↔ The student read the book halfway.
b. The student read the book in five hours.
c. Students read the book ??in five hours.
d. The student read books ??in five hours.
These facts are explained if we assume an FPR. My analysis differs slightly from
Rappaport Hovav (2008, 24-26). She argues that these verbs do not lexicalize a scale,
which is instead supplied by the object. I suggest that the verb selects for a “piece”
scale (qua a path), but like other path objects it is deletable (e.g. walk (the plank)).
However, there is a hitch: the relevant arguments do not pass affectedness tests:
(103) a. We just performed the play, but we are not anywhere else/nothing is dif-
ferent about us.
b. Kim just read the book, but she is not anywhere else/nothing is different
about her.
The problem lies in the nature of the scales: “traversing” a piece does not result in a
lasting physical change. Thus the relevant scale types needs to be distinguished again.
An appropriate test for performance may be X is at the end of the piece, and this does
seem to generate the appropriate sort of contradiction:
(104) a. We just performed the play, #but we are not at the end of it.
b. Kim just read the book, #but she is not at the end of it.
Finally, consider the verbs in (100), which are transitive because they take instru-
ment objects or circular path objects, but do not entail an effect. In (100a) the car
and driver could end up at the same spot, and in (100b) the figure must necessarily
end up at its original position. However, in the middle of the event some temporary
change must have occurred. To drive a car or drive a racetrack, the driver must leave
her initial position, even if she return there later. If we put these predicates in the
progressive to focus on the middle of the event, entailment tests work much better:
(105) a. John is driving his car, #but he is not somewhere else.
b. John is driving the racetrack, #but is not somewhere else.
Other such predicates that involve temporary but not lasting change are jiggle (the
handle), shake (the rug), rock (the boat), and flap (the wings) (brought to my attention
by Malka Rappaport Hovav, p.c., though she suggests that they do not lexicalize
scales). These predicates also show middles (unlike surface contact verbs):
(106) a. These handles jiggle easily, but won’t open without a good shove.
b. Due to its unstable stand, this monitor shakes easily.
c. This boat rocks violently even in very mild storms.
d. These wings flap quickly for better hovering and sustained flight.
To capture “temporary change”, we could revise the constraint in (95) as follows:
(107) Affectedness Constraint (Revised): If a transitive predicate describing event e
entails a non-quantized change in some e′ ≤ e, its external argument may be
eliminated.
 33

This picks out (non-)quantized changes (where e′ = e), and potential change pred-
icates in which some temporary change occurs. Potential change verbs that do not
have this property do not undergo middles (cf. *Smooth tables wipe easily). Thus a
slight change to the Affectedness Constraint captures (98)-(100) as well, but again
capitalizes on the notion of monotonically weakening truth conditions.

7 Conclusion

Although affectedness has been implicated in numerous linguistic phenomena, it is

rarely addressed as its own linguistic primitive. I have reiterated empirically that af-
fectedness is a true linguistic primitive, encompassing changes in states, locations,
and existence, based on a revision of the more general evidence for the existence
of BECOME subevents and their grammatical relevance. I have further argued that
while affectedness is connected to aspect, they do not wholly overlap, and there are
reasons to keep at least some aspectual notions (e.g. measuring out, types of telicity)
separate. This suggests that BECOME subevents necessarily involve multiple par-
ticipants. I propose that the underlying θ-relation borne by affected argument is a
ternary one that relates it in a mutually constraining way to an event and a property
scale that the theme “moves” along in the event. Positing another entity into the BE-
COME relation gives us (a) the wiggle room needed to separate measuring out and
affectedness, as well as different parts of telicity, (b) a principled way to distinguish
real world changes but also relate them (types of scales vs. the structure of scales),
and (c) a precise way to characterize degrees of affectedness in terms of how specific
a predicate is about the endpoint of the theme’s movement on the scale.
The relevant degrees of affectedness fall into an implicational hierarchy based on
monotonically weakening truth conditions, which gives us a way to analyze affected-
ness tests and other phenomena. Semantically contentful argument/oblique alterna-
tions reflect various minimal contrasts along this hierarchy, while other phenomena
pick out particular ranges on this hierarchy as in (108), which combines both affect-
edness tests and the syntactic phenomena contingent on affectedness.
(108) Degrees of Affectedness
Diagnostics Quantized Non-quantized Potential Unspecified
Telic (for quantized theme) X × × ×
Change entailed X X × ×
Permit DP-preposing X X X/× ×
Permit result XP X X X/× ×
Happened/did to X X X ×
Dynamic X X X X/×
Result XP variation Low Low/High High N/A
Relative transitivity High High Lower Lowest

Some of these phenomena are implicational in nature, so that applying to predicates

that entail a certain degree of affectedness means applying to predicates entailing
stronger degrees of affectedness, supporting this analysis. Other phenomena are in-
stead gradient in nature, where the degree of specificity correlates with a higher de-
gree of some other property (restrictions on result XPs, transitivity), again implicating
a hierarchy of this sort. The fact that no phenomenon picks out a discontinuous range
34

is also telling. If degrees of affectedness were not aligned in this way, this would
not be expected. Furthermore, no phenomenon I am aware of picks out a continuous
range that does not include quantized change. This suggests that more generally lan-
guages do not contain constraints that rule out certain phenomena from applying to
particular higher degrees of affectedness (i.e. there are no constraints of the form “X
applies to entities affected to degree n but not degrees greater than n”). It is a matter
of future work to see what other phenomena show the same type of behavior (e.g. see
Gaylord 2007 on an application of this to auxiliary selection in split-intransitive lan-
guages). But now we have a more fine-grained theory than one based on the simple
lexicalization or not of a BECOME subevent, as well as empirical ways of probing
for how specific the predicate is about the BECOME subevent.
Of course, there may be other degrees of affectedness. The approach of Washio
(1997) suggests that quantitative restrictions on possible result states may define sub-
classes of these types, for example distinguishing deadjectival degree achievements
from other non-quantized changes in terms of possible result XPs. Likewise, the dis-
cussion of middles suggests that some notion of temporary change may be grammat-
ically relevant, which would fall somewhere between non-quantized and potential for
a change, though this has so far not proved relevant for any other phenomenon. Like-
wise, there may be change that is non-scalar in nature. Rappaport Hovav (2008) as-
sumes that pure activities such as play (baseball) do involve some (temporary) change
(since something must occur in the middle of the event), though it does not seem to
follow any trajectory that can be measured, something for future investigation.
Returning to the larger issue of what affectedness is, we can also ask how af-
fectedness fits into a larger theory of lexical semantics. As noted in the introduction,
affectedness is usually defined either intuitively or through representational diacritics.
Representationally, the most common approach is to posit some primitive component
of decompositional verbal structure such as a BECOME subevent type, or a syntactic
head that predicates a result state, and tie analyses of specific phenomena to these
primitives. Syntactic approaches have been especially prevalent of late (see in par-
ticular Ramchand 2008), where the head itself has syntactic properties (such as de-
termining c-command relationships between arguments) from which other syntactic
properties follow directly. On these approaches an argument is simply identified by its
position in the decomposition (i.e. being a particular argument of the relevant prim-
itive). However, I have suggested that affected arguments have positive, semantic,
identificational properties. This raises the question of whether these properties are al-
ways associated with an independently defined set of syntactic properties, pointing to
a one-to-one correspondence with some decompositional primitive. If so, this would
further justify decompositional approaches. If not, it suggests that there is an inde-
pendent, linguistically relevant semantic tier of information that makes a different
classification of arguments, something that must be reconciled with decompositions.
Furthermore, the Affectedness Hierarchy is implicated in a range of phenomena as
in (108). But do different degrees of affectedness correspond to different subeven-
tal primitives, variations on the same primitive, or a wholly separate component of
lexical meaning? This again depends on identifying syntactic correlates, but if they
do correspond to subevental primitives, we must ask what it means for primitives to
be semantically related to one another in terms of strength of truth conditions. These
 35

are questions for future investigation. My goal here is instead simply to (re)define
the semantic terms upon which further investigation of affectedness can proceed, by
putting the intuitive basis of a measurable change at center stage.

References

Anderson, John M. 1971a. The Grammar of Case. Cambridge: Cambridge University Press.
Anderson, John M. 1977a. On Case Grammar. Croom Helm.
Anderson, Mona. 1979. Noun Phrase Structure. Ph.D. thesis, The University of Connecticut.
Anderson, Stephen R. 1971b. On the role of deep structure in semantic interpretation. Foundations of
Language 7:387–396.
Anderson, Stephen R. 1977b. Comments on the paper by Wasow. In P. W. Culicover, T. Wasow, and
A. Akmajian, eds., Formal Syntax, pages 361–378. New York: Academic Press.
Baker, Colin and Josef Ruppenhofer. In press. Framenet’s frames vs. levin’s verb classes. In BLS 28.
Baker, Mark C. 1997. Thematic roles and syntactic structure. In L. Haegeman, ed., Elements of Grammar,
pages 73–137. Dordrecht: Kluwer.
Beavers, John. 2002. Aspect and the distribution of prepositional resultative phrases in English. LinGO
Working Paper #2002-7, CSLI, Stanford University, Stanford.
Beavers, John. 2006. Argument/Oblique Alternations and the Structure of Lexical Meaning. Ph.D. thesis,
Stanford University.
Beavers, John. 2008. Scalar complexity and the structure of events. In J. Dölling, T. Heyde-Zybatow,
and M. Schäfer, eds., Event Structures in Linguistic Form and Interpretation, pages 245–265. Berlin:
Mouton de Gruyter.
Beavers, John. In press. Multiple incremental themes and figure/path relations. In Proceedings of SALT
XVIII.
Blume, Kerstin. 1998. A contrastive analysis of interaction verbs with dative complements. Linguistics
36:253–280.
Boas, Hans C. 2003. A Constructional Approach to Resultatives. Stanford: CSLI Publications.
Broccias, Cristiano. 2003. The English Change Network: Focusing Changes into Schemas. Berlin: Mouton
de Gruyter.
Condoravdi, Cleo. 1989. The middle: Where semantics and morphology meet. In MIT Working Papers in
Linguistics 11, pages 16–31. MIT Press.
Cornips, Leonie and Aafke Hulk. 1999. Affected objects. Languges in Contrast 1:191–210.
Croft, William. 1990. Possible verbs and the structure of meaning. In S. L. Tsohatzidis, ed., Meanings
and Prototypes, pages 48–73. London: Routledge.
Croft, William. 1991. Syntactic Categories and Grammatical Relations: The Cognitive Organization of
Information. Chicago, IL: University of Chicago Press.
Croft, William. 1998. Event structure in argument linking. In M. Butt and W. Geuder, eds., The Projection
of Arguments: Lexical and Compositional Factors, pages 21–63. Stanford: CSLI Publications.
Cruse, D.A. 1973. Some thoughts on agentivity. Journal of Linguistics 9:11–23.
DeLancey, Scott. 2000. The universal basis of case. Logos and Language 1:1–15.
Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67:547–619.
Egerland, Verner. 2000. The Affectedness Constraint and AspP. Studia Linguistica 52:19–47.
Fagan, Sarah M. B. 1992. The Syntax and Semantics of Middle Constructions. Cambridge: Cambridge
University Press.
Fiengo, Robert. 1980. Surface Structure: The Interface of Autonomous Components. Cambridge, MA:
Harvard University Press.
Filip, Hana. 1999. Aspect, Eventuality Types, and Nominal Reference. New York: Garland.
Fillmore, Charles J. 1968. The case for case. In E. Bach and R. T. Harms, eds., Universals in Linguistic
Theory, pages 1–90. New York: Holt.
Fillmore, Charles J. 1970. The grammar of Hitting and Breaking. In R. Jacobs and P. S. Rosenbaum, eds.,
Readings in English Transformational Grammar, pages 120–133. Waltham: Ginn.
Fillmore, Charles J. 1977. The case for case reopened. In P. Cole and J. M. Sadock, eds., Grammatical
Relations, pages 59–82. New York: Academic Press.
36

Foley, William and Robert D. Van Valin. 1984. Functional syntax and universal grammar. Cambridge:
Cambridge University Press.
Folli, Raffaella and Heidi Harley. 2004. Consuming results: Flavors of little-v. In P. Kempchimsky and
R. Slabakova, eds., Aspectual Enquiries, pages 1–25. Dordrecht: Kluwer.
Folli, Raffaella and Gillian Ramchand. 2002. Event structure composition: The case of goal of motion
and resultative constructions in Italian and Scottish Gaelic. In H. J. Verkuyl, ed., Proceedings of
Perspectives on Aspect Conference. Utrecht: OTS.
Garey, Howard B. 1957. Verbal aspects in French. Language 33:91–110.
Gaylord, Nicholas. 2007. Auxiliary Selection and the Typical Properties of Subjects. Master’s thesis, The
University of Texas at Austin.
Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure.
Chicago, IL: University of Chicago Press.
Goldberg, Adele E. and Ray Jackendoff. 2004. The English resultative as a family of constructions. Lan-
guage 80:532–569.
Green, Georgia. 1974. Semantic and Syntactic Regularity. Bloomington: Indiana University Press.
Gruber, Jeffery. 1965. Studies in Lexical Relations. Ph.D. thesis, MIT. Reprinted as part of Lexical
Structures in Syntax and Semantics, North-Holland, Amsterdam, 1976.
Guerssel, Mohamed, Kenneth Hale, Mary Laughren, Beth Levin, and Josie White Eagle. 1985. A cross-
linguistic study of transitivity alternations. In W. H. Eilfort, P. D. Kroeber, and K. L. Peterson, eds.,
CLS 21, Part 2: Papers from the Parasession on Causatives and Agentivity, pages 48–63. Chicago
Linguistic Society.
Hay, Jennifer, Christopher Kennedy, and Beth Levin. 1999. Scalar structure underlies telicity in degree
achievements. In The Proceedings of SALT IX, pages 127–144.
Hopper, Paul J. 1985. Causes and affects. In CLS 21, pages 67–88.
Hopper, Paul J. and Sandra A. Thompson. 1980. Transitivity in grammar and discourse. Language 56:251–
299.
Jackendoff, Ray. 1976. Toward an explanatory semantic representation. Linguistic Inquiry 7:89–150.
Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.
Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press.
Jackendoff, Ray. 1996. The proper treatment of measuring out, telicity, and perhaps event quantification
in English. Natural Language and Linguistic Theory 14:305–354.
Jaeggli, Osvaldo A. 1986. Passive. Linguistic Inquiry 17:587–622.
Jespersen, Otto. 1933. Essentials of English Grammar. London: George Allen & Unwin Ltd.
Kearns, Kate. 2007. Telic senses of deadjectival verbs. Lingua 117:26–66.
Kennedy, Chris and Beth Levin. 2008. Measure of change: The adjectival core of degree achievements. In
L. McNalley and C. Kennedy, eds., Adjectives and Adverbs: Syntax, Semantics, and Discourse, pages
156–182. Oxford, UK: Oxford University Press.
Kennedy, Christopher and Louise McNally. 2005. Scale structure, degree modification, and the semantics
of gradable predicates. Language 81.
Koontz-Garboden, Andrew. 2006. On the typology of state/change of state alternations. In G. Booij and
J. Marle, eds., Yearbook of Morphology, pages 83–117. Dordrecht: Springer.
Koontz-Garboden, Andrew. To appear. The states in changes of state. In BLS 32.
Kratzer, Angelika. 2000. Building statives. In BLS 26, pages 385–399.
Krifka, Manfred. 1989. Nominal reference, temporal constitution and quantification in event semantics.
In R. Bartsch, J. van Benthem, and P. van Emde Boas, eds., Semantics and Contextual Expressions,
pages 75–115. Dordrecht: Foris Publications.
Krifka, Manfred. 1992. Thematic relations as links between nominal reference and temporal constitution.
In I. A. Sag and A. Szabolcsi, eds., Lexical Matters, pages 29–53. Stanford: CSLI Publications.
Krifka, Manfred. 1998. The origins of telicity. In S. Rothstein, ed., Events and Grammar, pages 197–235.
Dordrecht: Kluwer.
Lakoff, George. 1976. Towards generative semantics. In J. D. McCawley, ed., Notes from the Linguistic
Underground, pages 43–62. New York: Academic Press.
Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago, IL:
University of Chicago Press.
Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity: At the Syntax-Lexical Semantics Inter-
face. Cambridge, MA: MIT Press.
Malchukov, Andrej. 2005. Case pattern splits, verb types and construction competition. In M. Amberber
and H. de Hoop, eds., Competition and Variation in Natural Languages: The Case for Case, pages
 37

73–118. Amsterdam: Elsevier.

Næss, Åshild. 2003. What markedness marks: The markedness problme with direct objects. Lingua
114:1186–1212.
Parsons, Terence. 1990. Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge,
MA: MIT Press.
Pinker, Stephen. 1989. Learnability and Cognition. Cambridge, MA: MIT Press.
Quirk, Randolph and Sidney Greenbaum. 1973. A Concise Grammar of Contemporary English. New
York: Harcourt Brace Jovanovich.
Ramchand, Gillian. 2008. Verb Meaning and the Lexicon: A First Phase Syntax. Cambridge: Cambridge
University Press.
Rappaport, Malka and Beth Levin. 1988. What to do with θ-roles. In W. Wilkins, ed., Thematic Relations,
pages 7–36. San Diego, CA: Academic Press.
Rappaport Hovav, Malka. 2008. Lexicalized meaning and the internal structure of events. In S. Roth-
stein, ed., Theoretical and Crosslinguistic Approaches to the Semantics of Aspect. Amsterdam: John
Benjamins.
Rappaport Hovav, Malka and Beth Levin. 1998. Building verb meanings. In M. Butt and W. Geuder, eds.,
The Projection of Arguments: Lexical and Compositional Factors, pages 97–133. Stanford: CSLI
Publications.
Rappaport Hovav, Malka and Beth Levin. 2001. An event structure account of English resultatives. Lan-
guage 77:766–797.
Rappaport Hovav, Malka and Beth Levin. 2005. Change-of-state verbs: Implications for theories of argu-
ment projection. In N. Erteschik-Shir and T. Rapoport, eds., The Syntax of Aspect, pages 274–286.
Oxford: Oxford University Press.
Rothstein, Susan. 2004. Structuring Events. Oxford: Blackwell.
Talmy, Leonard. 2000. Toward a Cognitive Semantics: Typology and Process in Concept Structuring.
Cambridge, MA: MIT Press.
Tenny, Carol. 1987. Grammaticalizing Aspect and Affectedness. Ph.D. thesis, MIT.
Tenny, Carol. 1992. The aspectual interface hypothesis. In I. A. Sag and A. Szabolcsi, eds., Lexical
Matters, pages 490–508. Stanford: CSLI Publications.
Tenny, Carol. 1994. Aspectual Roles and the Syntax-Semantic Interface. Dordrecht: Kluwer.
Testelec, Yakov G. 1998. On two parameters of transitivity. In L. Kulikov and H. Vater, eds., Typolgy of
Verbal Categories, pages 29–45. Tübingen: Niemeyer.
Tham, Shiao Wei. 2005. Representing Possessive Predication: Semantic Dimensions and Pragmatic Bases.
Ph.D. thesis, Stanford University.
Travis, Lisa. 2000. Event structure in syntax. In C. Tenny and J. Pustejovsky, eds., Events as Grammatical
Objects: The Converging Perspectives of Lexical Semantics and Syntax, pages 145–185. Stanford:
CSLI Publications.
Tsunoda, Tasaku. 1981. Split case-marking in verb-types and tense/aspect/mood. Linguistics 19:389–438.
Tsunoda, Tasaku. 1985. Remarks on transitivity. Journal of Linguistics 21:385–396.
van der Leek, Frederike. 1996. The English conative construction: A compositional account. In Proceed-
ings of CLS 32, pages 363–378.
van Oosten, Jeanne. 1977. Subjects and agenthood in English. In W. A. Beach, S. E. Fox, and S. Philosoph,
eds., CLS 13, pages 451–471.
van Oosten, Jeanne. 1986. The Nature of Subjects, Topics, and Agents: A Cognitive Explanation. Bloom-
ington: Indiana University Linguistics Club.
Van Valin, Robert D. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning, and Function. Cambridge,
UK: Cambridge University Press.
Verkuyl, Henk J. 1972. On the Compositional Nature of the Aspects. Dordrecht: Reidel.
Verkuyl, Henk J. 1993. A Theory of Aspectuality: The Interaction between Temporal and Atemporal
Structure. Cambridge, UK: Cambridge Unversity Press.
von Wright, Georg Henrik. 1963. Norm and Action. London: Routledge & Kegan Paul.
Washio, Ryuichi. 1997. Resultatives, compositionality, and language variation. Journal of East Asian
Linguistics 6:1–49.
Wechsler, Stephen. 2001. A fresh aspect on resultatives. Talk given at University of California at Berkeley.
Wechsler, Stephen. 2005. Resultatives under the ‘event-argument homomorphism’ model of telicity. In
N. Erteschik-Shir and T. Rapoport, eds., The Syntax of Aspect, pages 255–273. Oxford: Oxford Uni-
versity Press.