If You Want A Quick Kiss, Make It Count: How Choice Of Syntactic Construction Affects Event Construal

Abstract

When we hear an event description, our mental construal is not only based on lexical items, but also on the message's syntactic structure. This has been well-studied in the domains of causation, event participants, and object conceptualization. Less studied are the construals of temporality and nu-merosity as a function of syntax. We present a theory of how syntax affects the construal of event similarity and duration in a way that is systematically predictable from the interaction of mass/count syntax and verb semantics, and test these predictions in six studies. Punctive events in count syntax (give a kiss) and durative events in mass syntax (give advice) are construed as taking less time than in transitive frame (kiss and advise). Durative verbs in count syntax (give a talk), however, result in a semantic shift, orthogonal to duration estimates. These results demonstrate how syntactic and semantic structure together systematically affect event construal.

Full text

If You Want A Quick Kiss, Make It Count:
How Choice Of Syntactic Construction
Affects Event Construal
Eva Wittenberga, Roger Levya,b
aDepartment of Linguistics
University of California, San Diego
9500 Gilman Drive
La Jolla, CA 92093-0108
bDepartment of Brain and Cognitive Sciences
Massachusetts Institute of Technology
77 Massachusetts Avenue
Cambridge, MA 02139-4307
Abstract
When we hear an event description, our mental construal is not only based
on lexical items, but also on the message’s syntactic structure. This has
been well-studied in the domains of causation, event participants, and object
conceptualization. Less studied are the construals of temporality and nu-
merosity as a function of syntax. We present a theory of how syntax affects
the construal of event similarity and duration in a way that is systematically
predictable from the interaction of mass/count syntax and verb semantics,
and test these predictions in six studies. Punctive events in count syntax
(give a kiss) and durative events in mass syntax (give advice) are construed
as taking less time than in transitive frame (kiss and advise). Durative verbs
in count syntax (give a talk), however, result in a semantic shift, orthogonal
Email address: ewittenberg@ucsd.edu (Eva Wittenberg)
Preprint submitted to Journal of Memory and Language November 29, 2016

to duration estimates. These results demonstrate how syntactic and semantic
structure together systematically affect event construal.
Keywords: Events, Lexical Aspect, Light Verb Constructions, Mass–Count
distinction, Individuation, Linguistic Framing
Introduction
When people talk to each other about what happened, they usually don’t
need to specify how long it took. Everybody knows from experience that a
kiss lasts a few moments, a conference talk may carry on for about twenty
minutes, and giving professional advice takes maybe half an hour, so there is
typically no need to explicitly mention the duration. Duration is also usually
not encoded grammatically. However, grammatical cues in event descrip-
tions often significantly influence other aspects of event representations in
listeners, such as causation, event structure, and the semantic roles of event
participants (Fausey & Boroditsky, 2010; Johnson & Goldberg, 2013; Wit-
tenberg & Snedeker, 2014). It would be all the more interesting, thus, if very
subtle grammatical choices were to reliably affect how long listeners think an
event takes.
In this article, we explore how encoding event descriptions in simple verbs
(to kiss, to advise) versus count or mass noun light verb constructions (to
give a kiss, to give advice) has repercussions on the temporal encoding of
these events. Based on the fundamental observation that the reference prop-
erties of syntactic objects can change the reference properties of the whole
predicate (Krifka, 1992), we predict that nominalizing an event can help di-
viding experience into countable units, influencing duration estimates in a
2

way that is systematically predictable from the interaction of verb semantics
and nominal syntax.
This hypothesis was inspired by a previous study on how events are indi-
viduated, depending on mass and count syntax. Barner et al. (2008) found
that using count syntax (to do climbs), but not mass syntax (to do climbing),
affects how events are quantified; and that atomic, punctive events (kissing,
kicking) are more readily quantified by counting over individual subevents
(kisses, kicks) than non-atomic, durative events. This is in line with the
Number Asymmetry hypothesis (Barner & Snedeker, 2006): whereas count
syntax specifies individuation, mass syntax is underspecified.
If it is true that mass and count syntax contribute to event individuation,
then we should expect predictable influences of mass or count syntax also on
estimates of event duration. We distinguish between two types of events:
Atomic, telic, mostly punctive events, like kissing or kicking; and non-atomic,
atelic, mostly durative events, like talking or advising (Dowty, 1991; Vendler,
1957, see Footnote 1).
Punctive events are distinct from durative events not only in that they
are conceptually short and bounded by a natural end point (telic), but also
in that sentences in which they appear are often conventionally understood
to describe several instances of the same punctive event, that is, they are
understood iteratively (Barner et al., 2008; Kim & Kaiser, 2015; Paczynski
et al., 2014): For instance, you may find that John kissed Mary evokes the
image of not one, but multiple kisses, each of which can be categorized as a
subevent of kissing. Thus, punctive events can have a distinct substructure.
Durative events, in contrast, are atelic and , they do not possess a distinct
3

××××
punctive
events
kiss
× × × ×
to give
a kiss
atomic, telic
count
syntax
durative
events
advise/talk
to give
advice
to give
a talk
non-atomic,
atelic
count
syntax
mass
syntax
punctive count durative mass durative count
Predictions: kiss →give a kiss advise →give advice talk →give a talk
temporal
construal shorter shorter no prediction
number
of events fewer same fewer
event
similarity similar similar different
Figure 1: Predictions of how mass versus count syntax interacts with verb semantics, with
regards to event duration construals, number of events, and event similarity; see pp. 13ff..
4

substructure, and they do not receive an iterative reading, even if the du-
ration of the event is explicitly extended beyond a conventionally accepted
time frame (cf. Senator Cruz talked all night).
Many of the aforementioned events, like kiss, advise or talk, can either
be expressed as transitive verbs, or as so-called light verb constructions. In
light verb constructions, the verb contributes little semantics beyond tense,
number agreement, and aspect, while the meaning of the expression comes
from the deverbal noun (Brugman, 2001; Butt, 2003, 2010; Jackendoff, 1974;
Jespersen, 1954; Wiese, 2006). These light verb constructions appear either
with count syntax, such as to give a kiss and to give a talk, or mass syntax,
such as to give advice. Thus, light verb constructions offer us an opportunity
to study the interaction of verb type and mass versus count syntax with an
existing alternation, as opposed to unusual constructions such as to do climbs
(Barner et al., 2008), or using novel lexical items (Wellwood et al., 2016):
Light verb constructions, like to give a kiss, and their full verb counterparts,
like to kiss, are in a relationship of syntactic alternation with minimal dif-
ference in meaning (Allerton, 2002; Glatz, 2006). In our study of punctive
and durative events, we use light verb constructions with give, which is itself
telic (Newman, 1996).
The mass–count distinction and verbal aspect
Ever since Bach (1986), linguistic theory has been fascinated by the par-
allels between kinds of objects vs. substances on the one hand, and kinds
of atomic vs. non-atomic events on the other hand (Casati & Varzi, 2008;
Hale & Keyser, 1993; Harley, 2005; Jackendoff, 1991; Krifka, 1992; Quine,
1969; Rothstein, 2008; Verkuyl, 1972). One of the defining differences be-
5

tween objects and materials is that labels for objects denote atomic units,
which cannot be partitioned arbitrarily: Only a whole apple, not a piece of
an apple, can be described with the count noun an apple. A piece of an ap-
ple, on the other hand, will need to be further described with a quantifier or
specific expression, such as slice of an apple, or apple core. Objects can also
be individuated and counted (three apples). Materials, however, are non-
atomic, and can be partitioned in an arbitrary fashion: a quart of applesauce
can be divided into many portions, yet each individual portion still denotes
applesauce (Bale & Barner, 2009; Rips & Hespos, 2015, and many others).
Introducing individuability to mass nouns, however, is easily accomplished
when they are quantized (a bottle of wine, a quart of applesauce; see Krifka,
1992; Wiese & Maling, 2005).
Events have the property of atomicity or non-atomicity, too: Some events
are atomic, and some events are non-atomic. For example, if Mary kissed
John, then she stopped just for a moment, and then started kissing him
again, the post-interruption kiss would be a new event, even if the people
and location are the same: an event of kissing is atomic in that it cannot
be broken apart. (Note also that the character of the start and end points
is constitutive of the event: if there is not contact between a set of lips and
a surface, with a a clearly defined onset and a clearly defined, voluntary or
involuntary, offset, the term kiss does not apply.) This not true for all events
(or processes, see Wellwood et al., 2016): If the president talked to a crowd,
stopped for a moment, and started talking again, it could still be the same
event of talking. Similarly, advising can be partitioned and spread over many
advising sessions, but the overarching event of advising is the same, as long
6

as there is some degree of spatial or character continuity (Anderson et al.,
1983; Magliano & Zacks, 2011; Zacks & Tversky, 2001). Talking and advising
are thus non-atomic: they can thus be broken up and still count as the same
talking and advising events.
The atomicity and non-atomicity of events is highly correlated with no-
tions of telicity, boundedness, and aspect in verbs or predicates, as well as the
distinction between events and processes in some approaches.1For the pur-
pose of this article, atomicity as described above will be the defining criterion
1Telic events are said to involve some kind of natural endpoint (Andersson, 1972; Garey,
1957; Bauer, 1970; Klein, 1994; Vendler, 1957, among others). This definition covers
accomplishments, like to draw a circle, but not all punctive events. The classical definition
of punctive events is that they only take a moment in time. How long this moment takes,
however, is underspecified: The duration of to sneeze depends entirely on the sneezer; to
explode can conceptually take more than a few seconds; and the event described by to
break might last for a few minutes. Thus, temporal properties make up one part of the
diagnostics; the other part is contributed by the intuition that there is a natural endpoint
to a given event.
Linguistic diagnostics, such as test for aspectual types, are of limited help. Even the
classic test for a durative, namely using a temporal for -PP to detect atelic events, is
not entirely reliable: John talked for an hour is understood as one continuous event and
thus classified as durative; John sneezed for an hour is understood as iterated, and thus
classified as punctive. But the time frame defined by the prepositional phrase matters
immensely: John kissed Mary for a minute is understood as continuous, and thus classi-
fied as durative; John kissed Mary for an hour is likely understood as iterated, and would
thus be classified as punctive (note, in contrast, the unavailability of an iterative reading
for achievements: *John discovered the error for an hour, e.g. Bott, 2010). Further com-
plicating the grammatical picture is that iterated punctive events pattern with (durative)
activities in some tests (such as allowing for nonsubcategorized objects in reflexive resul-
tatives or in out-prefixation, or in some tense entailment relationships). When punctive
events are understood non-iteratively, they pattern with (durative) achievements in some
other tests (such as in onset repair readings; Kearns, 2000).
Thus, there are at least two world-knowledge factors at play in defining the aspectual
class of a verb describing an event: the existence of a natural endpoint of an event and
connected to that, the inherent duration of an event; and an event’s tendency to occur
several times in a row and so its availability for an iterative interpretation. In addition, lin-
guistic diagnostics are not always straightforward and might involve pragmatic inferences
that are beyond the lexical semantics of the verb.
7

for an event to be classified as either punctive (atomic, e.g., kiss) or durative
(non-atomic, e.g., talk, advise), even though the overlap of terminologies is
not perfect.
In this study, we focus on the interaction of count and mass syntax with
punctive and durative events in a syntactic ditransitive frame provided by
the verb give, such as Mary gave Douglas a kiss,the professor gave her stu-
dent advice, or the president gave a talk to the audience. Given the parallels
between between mass/count syntax and verbal aspect, and given the telic
verb give, we expect count and mass syntax to interact differently with punc-
tive and durative event nouns: Count syntax with a punctive deverbal noun
like giving a kiss should pick out one single instance of kissing; mass syntax
with a durative deverbal noun like giving advice should carve out a portion
of advising. If this is true, then the grammar conveys a subtle difference in
event duration between the simple transitive verb and the light verb con-
struction, for any given event: the light verb construction communicates an
event that lasts a shorter time.
For durative events in count syntax in combination with the telic light
verb give, however, the predictions are not as straightforward. Just consider
what happens when one packages substances like beer, glass, string, stone or
iron into count syntax: In some cases, the count noun denotes portions (of
arbitrary size) of the substance (a string, a stone, a beer ), but in other cases,
the count noun phrase happens to encode objects or units that are related
to the mass noun, yet in an arbitrary way, such as in a glass or an iron.
Crucially, the resulting denotation for mass nouns in count syntax is variable,
and each case conveys aspects of meaning that cannot be predicted from the
8

intrinsic structure of the underlying substance (Gordon, 1985; Srinivasan &
Rabagliati, 2015).
So what should we expect in the case of durative verbs entering count
syntax, like talk in give a talk? If the analogy between durative verbs and
substance nouns really holds, one would predict a certain degree of conven-
tionalization, similar to the some glass →a glass case: Core parts of a given
durative event would remain the same, but the event type would shift in
meaning from verb to count noun construction, possibly closing a lexical gap
in doing so (Allerton, 2002; Glatz, 2006; Grimshaw & Mester, 1988; Miya-
gawa, 1989). For example, there is a strong intuition that giving a talk, albeit
still retaining the core meaning of utterance production, is conceptually fur-
ther from talking than giving a kiss is from kissing.
But then, if it is true that durative events in count syntax undergo a
conceptual shift in event kind, predictions about event duration are up in
the air, since the change induced by count syntax would be orthogonal to
changes in temporal conceptualization. In this article we explore both sides:
Whether event duration estimates are modulated by the introduction of mass
and count syntax, and whether there is an effect on how similar events are
judged as depending on the syntactic construction. In the next section, we
will discuss the link between event representation and linguistic encoding in
more detail.
Event construal via linguistic encoding
How linguistic framing influences people’s event conceptualization, mem-
ory, and recall has long been a topic of interest in science, such as in be-
havioral economics (Halkjelsvik et al., 2011; Kahneman & Tversky, 1977;
9

Kruger & Evans, 2004; Roy et al., 2005), criminal justice (Boltz, 1995; Loftus
et al., 1987; Macar et al., 1994; Tse et al., 2004), and cognitive science (Mad-
den & Zwaan, 2003; Magliano & Schleich, 2000). These studies, however,
were mainly concerned with behavioral or memory consequences of linguis-
tic encoding accompanying visual scenes, and less with representational or
grammatical issues.
In psycholinguistics, using grammatical alternations to study their influ-
ence on event construal started with Gropen et al. (1991), who found that
subtle changes to event structure affected which form of the locative alter-
nation people used in production (cover a surface with marbles or dropping
marbles onto a surface). A later study confirmed the intuition that syntac-
tically omitting agents from an event description reduces how much blame is
assigned to them (Fausey & Boroditsky, 2010). Directly related to the con-
structions used in this article, we know that using a light verb construction
with give influences the construal of thematic roles (Wittenberg et al., under
review; Wittenberg & Snedeker, 2014). And finally, there is evidence from a
production study that the naturalness of event divisions predicts the choice
of mass or count syntax (Wellwood et al., 2016).
The question of whether differences in event descriptions cause differences
in duration estimates is less well studied. So far, there is a small experimen-
tal literature showing that verbal aspect or the verb itself influence duration
estimates. For example, human locomotion events seen on video are remem-
bered as taking longer when they are described as walking events than when
they are described as running events (Burt & Popple, 1996); and a Dutch
study has shown that describing a short event in progressive verb aspect (is
10

kissing) makes people think that the kiss took longer than if simple present
is used (Flecken & Gerwien, 2013). Pedersen & Wright (2002), in contrast,
found only small effects of event descriptions on the duration estimates; how-
ever, their manipulation was on writing style and purposefully not as tightly
controlled for semantic and syntactic factors as other studies. Importantly,
Coll-Florit & Gennari (2011, Study 4) found that event duration estimates
are tightly linked to aspectual nature of verbs.
Thus, there is some evidence that linguistic choices influence the way
people think about the temporal dimensions of events. Yet, it is not sur-
prising that one should find this influence by grammatical means that by
default operate in the temporal dimension, like aspect, or by choosing lexical
items according to the speed of an action that they express. The alternation
between transitive verbs and their light verb construction counterparts, how-
ever, affords a way to look beyond the more obvious aspects of how syntactic
and semantic regularities work together to create a rich, full, and detailed
representation of an event.
How could this interaction work? Most current theoretical approaches
have accounted for the fact that some lexical items can appear both as
nouns and verbs with the stipulation that prior to lexical insertion, their
grammatical status is neutral. That means, a word like kiss only acquires
its grammatical category upon insertion into a syntactic tree; either way,
regardless of whether it is inserted as a noun (such as He gave her a kiss)
or a verb (He kissed her ), it conceptually refers to the same kind of event
(Barner & Bale, 2002; Halle & Marantz, 1994).
Taking this as a starting point, we should be able to observe a systematic
11

interaction between syntax, the lexicon, and event construal. We assume that
syntactic, conceptual, lexical, and phonological structure interact and predict
upcoming features and structures every step of the way (Garrod & Pickering,
2003; Jackendoff, 2002, 2007; Levy, 2008; Martin, 2016).2This architecture of
the linguistic system allows for an interaction between representations on the
levels of syntax (mass and count syntax), semantics (the telicity of give), and
event knowledge (how long kissing usually takes). Other approaches (Gold-
berg, 1995, 2003) explicitly allow for different meaning shades of grammatical
constructions; presumably, the distinct meaning of a ditransitive construc-
tion as a telic event would be even more straightforwardly predicted from
this perspective. Crucially, both of these models would predict differences in
duration estimates due to the interaction of verbal and nominal semantics
and syntax.
Key theoretical predictions and the current studies
The current studies investigate whether describing an event with mass or
count syntax, as opposed to a simple transitive verb, affects the construal
of event duration, event similarity, and event repetitions in a comprehender.
Figure 1 shows a graphical representation of our theory.
Starting from the top of Figure 1, we look at punctive events like kissing.
2Another option would be a strictly linear model: A lexical item is inserted into a
syntactic tree, its phonological form created, and its meaning read off by the comprehender,
without any feedback loop to the semantic or conceptual system. In this case, both forms
of the lexical items (e.g., kiss,talk or advise as either noun or verb) are ontologically linked
to the same event structure regardless of grammatical status, and it should not make a
difference for duration estimates whether they appear as a verb or in either mass or count
noun syntax, since the comprehender’s lexicon is set up the same way – kiss conveys the
same event regardless of whether it appears as verb or noun. Thus, this linear model
would not predict any differences in duration estimates.
12

Previous studies have shown that punctive events are often construed as
occurring more than once (Barner et al., 2008; Kim & Kaiser, 2015; Paczynski
et al., 2014), e.g., a comprehender might hear John kissed Mary and imagine
more than one kiss. In Figure 1, this is represented by four crosses, which
are distinctive, atomic subevents within one iterated, bounded kissing event.
Count syntax, according to our theory, should encourage event individuation
in iterative events: In the case of our example, giving a kiss describes only
one atom of a kiss. This in turn should have repercussions in the construal
of event temporality, leading to conceptualizations of shorter event duration.
Another prediction is that punctive events in count syntax will be construed
as consisting of fewer event iterations than in transitive syntax.
The bottom of Figure 1 visualizes our predictions for durative events. Du-
rative events, represented by the continuous snake line, have no set endpoint
(and in many cases, no set starting point either). They are also non-atomic;
for example, advising is a process that consists of many points in time, but
it is hard to conceptually delimit when advising starts and ends based on
these time points. However, it is possible to carve out a particular portion
of this process, and we claim that this is linguistically done when the event
appears in mass syntax: to give advice refers to a chunk of advising whose
boundaries are limited (although not quite as strictly as in the case of to give
a kiss). In terms of event counts, the predictions are weaker than for the
punctive event counts, since mass syntax does not aid in event individuation
(although the telicity of the verb give might).
When durative events occur in count syntax, such as in to give a talk
(bottom right on Figure 1), we hypothesize that, analogous to the cases of
13

using mass nouns in count syntax (glass, iron versus a glass, an iron), there
are arbitrary and unpredictable changes in the kinds of events these nouns
then describe. For example, giving a talk still retains a sense of utterance
production, but in a very different context and with different event parame-
ters (this difference is represented by a change in color, and the zigzag line,
in Figure 1). This creates another prediction: Talking and giving a talk
should be conceptually less closely related than punctive count and durative
mass pairs. In terms of event counts, we predict that count syntax form will
help event individuation and possibly lead to a reduction in event counts.
Crucially, both event similarity and event count differences would operate
entirely orthogonally from the construal of event duration.
We present six experiments that test these predictions. To test the claim
that the construal of event duration is predictable from the interaction of
mass versus count syntax, and verb semantics, we present two studies that
elicited open estimates of event duration (Experiment 1a and 1b); in order
to ensure that the estimates obtained in Experiments 1a and 1b are not due
to task difficulty, we then present a temporal categorization experiment, in
which participants categorized event descriptions into predefined time bins
(Experiment 2). Then we present two studies that establish whether using
count or mass syntax affects how many events people imagine: Experiments
3a and 3b investigate whether the determiner ”a” in ditransitive count syntax
picks out one particular instance of an event, which would expound shorter
temporal estimates in count syntax. Finally we investigate whether describ-
ing durative events in count syntax indeed leads to conceptual shifts in event
type, which would explain the inconsistent pattern observed for durative
14

events (Experiment 4).
Stimuli used for all studies
count mass
punctive to kiss – to give a kiss
to embrace – to give an embrace
to hug – to give a hug
to kick – to give a kick
to poke – to give a poke
to shake – to give a shake
to cuddle – to give a cuddle
durative to talk – to give a talk to advise – to give advice
to address – to give an address to thank – to give thanks
to lecture – to give a lecture to assure – to give assurance
to present – to give a presentation to encourage – to give encouragement
to speak – to give a speech to recognize – to give recognition
to scold – to give a scolding to support – to give support
to assist – to give assistance
Table 1: Experimental item pairs used for all experiments.
The verbs we used cluster together in a number of semantic and syntac-
tic factors: For example, all events in the ”punctive” category are atomic,
short and in themselves telic. They are, incidentally, also all contact verbs.
In a light verb construction, they appear in count syntax. The events in
the ”durative count” category are all non-atomic, of medium duration, and
atelic. Incidentally, they are also all actions of utterance towards others.
When appearing in a light verb construction, they do so with count syn-
tax. Their semantic common denominator may be best captured as social
actions with the direct object as beneficiary. When appearing in a light verb
construction, they take on mass syntax. Above and beyond other semantic
aspects, however, atomicity of the verbs is the most important factor for the
15

purpose of this article, and we used the notions ”punctive” and ”durative”
to distinguish between those categories of events.
We used punctive verbs (kiss) and durative verbs (advise, talk) either in
a transitive frame (After their first date, John kissed Mary ) or in a ditran-
sitive light verb construction with the telic verb give (After their first date,
John gave a kiss to Mary) such as in Table 1 (see Appendix A for a full list
of stimuli).3The ditransitive frame introduces a distinction between count
syntax (give a kiss/talk) and mass syntax (give advice). We expected count
syntax to force event individuation in punctive verbs, such that, when asked
about event duration, people should judge the same event to be shorter in the
ditransitive than in the transitive frame. For durative verbs, we predicted
the same pattern for mass syntax, albeit to a lesser degree, because only
the light verb give would encourage telicity. For durative events that enter
count syntax, we predicted a different pattern: Since there are no distinc-
tive subevents that can be counted, applying count syntax to durative verbs
should not lead to differences in duration. Instead, it should open the door to
different event construals, orthogonal to changes in temporal structure (see
Figure 1 for an overview of predictions).
Of the twenty experimental items, seven encoded punctive events, e.g.
kissing/giving a kiss, six were durative events and could be used with count
syntax, e.g. talking/giving a talk, and seven encoded durative events that
3In fact, talk and speak do not occur in transitive frames, but rather with prepositional
objects (talk *(to) the students). In the context of this article, however, we use ”transitive”
to mean two-place argument structures with either a Direct Object or a Prepositional
Object.
16

could be used with mass syntax, e.g. advising/giving advice.4Note that
in this last category, the noun describing the action (advice) was not pre-
ceded by a determiner. None of the light verb constructions that we used
can alternate between mass and count syntax (e.g. to give *(a) kiss; to give
(*an) advice). In short, there were three different verb alternations: punctive
transitive vs. ditransitive with count syntax (punctive count); durative tran-
sitive vs. ditransitive count syntax (durative count); and durative transitive
vs. ditransitive mass syntax (durative mass).
In all studies except for Experiment 4, the items were embedded in a
sentence context. All sentences used the simple past. Each sentence included
a temporal or local adjunct phrase, to encourage non-repetitive readings (e.g.
The professor advised her student on his paper yesterday). In addition to the
experimental items, we also created 27 filler sentences.
Experiment 1a: Open estimates of event duration
In Experiment 1a, we asked participants to rate how long events took.
The events were described by simple transitive or light ditransitive construc-
tions, and we were interested in whether the predictions displayed in Figure
1 would be confirmed. If so, we expect punctive events like kissing to be
estimated as taking less time in the ditransitive construction (giving a kiss).
We should also observe the same trend for durative events when they occur
in mass syntax (advising – giving advice), but not durative events in count
4The missing item in the durative count category was condole – giving condolences,
which we ultimately decided to exclude because condole is used only very rarely as a verb
in American English. In English, there are no punctive events in mass syntax with give.
17

syntax (talking – giving a talk). Experiment 1b served as a replication, with
event category as a between-subjects factor.
Methods
Participants
We recruited 100 unique individuals on Amazon Mechanical Turk, an on-
line crowd-sourcing tool. Mechanical Turk allows access to a large number of
study participants, who participate anonymously for reasonable compensa-
tion, in our case for about $6 an hour (Buhrmester et al., 2011; Crump et al.,
2013). Our participants had IP addresses within the United States and were
self-reported native speakers of English.
Stimuli
We created two lists out of the sentences described above and distributed
the experimental items across them with a fully within-subjects Latin-square
design, such that each participant saw each sentence in only one of the two
constructional forms (transitive, or ditransitive light verb). The fillers were
the same across lists.
Procedure
Participants read each sentence and then estimated how long the event
described in the sentence probably took. The exact instructions can be found
under https://github.com/ewittenberg/QuickKissing.
For each item, participants were able to enter their estimated event du-
ration in a set of three text boxes, one for hours, one for minutes, and one
for seconds, e.g., a participant could respond “1 hour(s), 13 minute(s), 7
18

second(s)”. Empty boxes were treated as a response of zero for that unit of
time. Completing the study took about 17 minutes on average.
Results
Responses in which the estimated duration was zero were discarded. This
affected less than .1% of the data. Since effects of grammatical structure on
event duration would be likely to operate proportionally to intrinsic event
duration, we transformed all responses to log-seconds for purposes of data
summarization and analysis. Figure 2 shows the pattern of results, with
responses back-transformed to hours, minutes, and seconds, for convenience
of interpretation. In all figures in the article, bar plots show means and
Standard Errors of by-subject means unless otherwise stated.
For punctive count events, using a ditransitive light verb construction
instead of the transitive verb cut the time estimates in half, from about
40 to about 20 seconds. For durative count items, the effect was smaller
(transitive µ=31 minutes, ditransitive µ=27 minutes), but in durative mass
items, it was even stronger than for punctive count events (transitive µ=50
minutes, ditransitive µ=29 minutes).
For data from this and all following experiments (with the exception
of Experiment 4, which has a different design), we conducted two types of
statistical analysis. The first is a 2 ×3 ANOVA-style analysis of the main
effect of construction (transitive or ditransitive light verb), the omnibus main
effect of event category (punctive count, durative count, and durative mass),
and the omnibus interaction between the two. The second is a set of planned
pairwise tests of the effect of construction within each event category. The
reason for this latter set of planned tests is that the strength of evidence
19

1 second
10 seconds
1 minute
10 minutes
1 hour
punctive count
(kiss − to give a kiss) durative count
(talk − to give a talk) durative mass
(advise − to give advice)
Estimated Event Duration in Seconds, log scale
Construction
transitive verb
ditransitive
light verb
Figure 2: Experiment 1a: Duration estimates per item pair.The y-axis is represented in
log scale, but labeled with reader-friendly time estimates for convenience.
for an effect of construction for each event category is relevant to assessing
the overall support of the data for our main hypothesis regarding effects of
syntactic construction on event construal.
In all analyses we used mixed-effects regression models (Baayen et al.,
2008; in this experiment, linear mixed-effects models) with R’s lme4 pack-
age (Bates et al., 2014), using maximal random effects structure justified by
the design (Barr et al., 2013; where noted, random correlation parameters
were dropped to ensure model convergence) and computing p-values through
likelihood-ratio tests between models differing only in the presence or ab-
sence of the fixed-effect parameter(s) being tested. We used Helmert coding
for both fixed-effects predictors, grouping punctive count and durative mass
20

items together as one Helmert contrast pair, and their average contrasted
with durative count items as the second Helmert contrast pair.5
The top half of Table 2 shows the results of the 2 ×3 ANOVA-style
analyses (these analyses involved random by-participant intercepts and slopes
for all fixed effects and random by-item intercepts and slopes for construction,
with all random correlation parameters removed). We see significant main
effects of construction and event category; the interaction does not reach
statistical significance.
The bottom half of Table 2 shows results of the planned pairwise tests
within each event category (with random intercepts and slopes for both par-
ticipants and items; no random correlation parameters needed to be omitted).
Whenever punctive events appear in a count light verb construction with give
(ditransitive frame, such as to give a kiss), they are estimated to take less
time than when they appear in a simple transitive (to kiss). This effect was
marginally significant. For durative events in mass syntax (to give advice –
to advise), the same pattern was statistically significant; the numeric pattern
for durative events in count syntax is far from statistically significant (to give
a talk – to talk).
Replication: Experiment 1b
We replicated this study with 300 participants on Amazon Mechanical
Turk by making event type (punctive count, durative count or durative mass)
a between-subjects factor, reasoning that thinking about events on vastly
5We implemented this coding in Ras numeric predictors rather than as factors, which
allows us to test main effects in the presence of the interaction through likelihood-ratio
tests; see Levy (2014), among others, for details.
21

Df χ2p-value
construction 1 7.9 0.005 ***
event category 2 23.02 0.000 ***
construction ×event category 2 3.28 0.196 n.s.
punctive count – construction 1 3.54 0.059 .
durative count – construction 1 0.15 0.690 n.s.
durative mass – construction 1 6.17 0.013 *
Table 2: Table of likelihood estimation results for duration estimates in Experiment 1a,
testing the main effects and their interaction (upper part), and the results of testing the
main effect of construction in planned pairwise comparisons (lower part).
different time scales might wash out sharper judgments. For example, the
literature on prospective time estimation shows that in events lasting more
than a few moments, people count internally to ”track time”, while for very
short events, this strategy is not used (Grondin et al., 1999; Zakay & Block,
1995). Similarly, short events like kissing or hugging might be imagined by
conjuring up an image of this event, while for longer events like talking or
advising, abstract world knowledge might be employed to estimate duration.
Experiment 1b served to exclude this potential mix of strategies in temporal
estimation. The exact instructions can be found under https://github.
com/ewittenberg/QuickKissing.
Figure 3 shows the pattern of results; as in Experiment 1a, punctive
count item pairs were estimated to take roughly half as long when they
were presented in ditransitive light verb frames (transitive µ=27 seconds,
ditransitive µ=14 seconds). Durative count items showed no trace of an effect
of syntactic construction, with mean estimates of 27 minutes in the transitive
frame and 31 minutes in the ditransitive frame. Durative mass items showed
the same numeric pattern as punctive count items, with mean estimates of
22

1 second
10 seconds
1 minute
10 minutes
1 hour
punctive count
(kiss − to give a kiss) durative count
(talk − to give a talk) durative mass
(advise − to give advice)
Estimated Event Duration in Seconds, log scale
Construction
transitive verb
ditransitive
light verb
Figure 3: Experiment 1b (replication with event category as between-subjects factor):
Duration estimates per item pair. The y-axis is represented in log scale, but labeled with
reader-friendly time estimates for convenience.
76 minutes in the transitive frame and 54 minutes in the ditransitive frame,
although this difference was not statistically significant.
Table 3 shows the results of the regression analyses, conducted identically
as those in Experiment 1a except that random by-subjects slopes for event
category and its interaction with construction were excluded, since event
category was a between-subjects manipulation. The top half of Table 3
shows the results of the ANOVA-style analyses likelihood-ratio tests (since
event category was both between subjects and between items, random by-
subject slope for event category, or the random interaction, were not needed).
Both main effects of construction and event category, and their interaction,
23

were significant.
The planned pairwise comparison results are shown in the bottom half of
Table 3 (here, random correlation parameters did not need to be removed).
As in Experiment 1a, whenever punctive events appear in a count light verb
construction with give (ditransitive frame, such as to give a kiss), they are
estimated to take less time than when they appear in a simple transitive (to
kiss). The effect was not significant for durative events with mass syntax (to
give advice – to advise), or for durative events with count syntax (to give a
talk – to talk).
Df χ2p-value
construction 1 4.98 0.026 *
event category 2 27.52 <0.0001 ***
construction ×event category 2 6.10 0.047 *
punctive count – construction 1 9.64 0.026 **
durative count – construction 1 0.40 0.531 n.s.
durative mass – construction 1 0.95 0.330 n.s.
Table 3: Table of likelihood estimation results for duration estimates in Experiment 1b,
testing the main effects and their interaction (upper part), and the results of testing the
main effect of construction in pairwise comparisons (lower part).
Discussion of Experiments 1a and 1b
As described above, the theory we proposed of how linguistic encoding
and event type interact in event construal strongly predicted an effect of con-
struction (transitive versus light-verb) on inferred event duration for punctive
events whose light-verb encoding involves count syntax (kissing vs. giving a
kiss): light-verb syntax should shorten inferred event durations. This pre-
diction was borne out in both Experiments 1a and 1b. The theory predicted
no systematic effect on inferred event duration for durative events encoded
24

with count syntax, as this encoding requires more substantial, conventional-
ized shifts (talk vs. give a talk ). This prediction was also borne out: there
was no trace of a systematic effect of construction on inferred event duration
across Experiments 1a and 1b. We suggested that the theory’s predictions
were less clear for durative events encoded with mass syntax (advise vs. give
advice), and indeed our results across the two experiments were less clear.
We found a significant shortening effect of light-verb syntax in Experiment
1a; in Experiment 1b, the same numeric pattern was evident but did not
reach significance.
As a methodological note, one might notice that the duration estimates
seemed fairly high. We know from previous studies that estimating the du-
ration of an event is often influenced by its pleasantness or desirability (Kah-
neman & Tversky, 1977; Roy et al., 2005). Although the numerical values of
the estimated event durations are not crucial here, since we are only inter-
ested in the difference in estimates that is attributable to syntactic structure,
it is conceivable that our participants’ estimates were influenced by factors
not due to the grammatical construction alone. For this reason, we decided
to replicate this study as a categorization task with predefined time windows
as answer options. The predefined time windows were based on previous
answers, to validate the results obtained in Experiment 1a and 1b.
Experiment 2: Categorizations of event duration
This experiment gave participants the opportunity to estimate event du-
rations without needing to come up with their own time estimates, and in-
stead being able to select among predefined time bins for each individual
25

event.
Method
Participants
We recruited 80 self-reported native speakers of English on Amazon Me-
chanical Turk with IP addresses within the United States.
Stimuli
We used the same sentences as in the previous studies. Participants read
each sentence and then had to categorize the event for duration by clicking
one of four options. These options were created from the quartiles of the
empirical distribution of estimated durations of each item pair in Experiment
1b; thus, every item had different answer options, as in Examples (1) and
(2):
(1) Laughing nastily, the thug kicked the victim.
How long did this take?
a) up to 5 seconds
b) between 5 seconds and 13 seconds
c) between 13 seconds and 10 minutes
d) more than 10 minutes
(2) The professor advised her student on his paper yesterday.
How long did this take?
a) up to 25 minutes
b) between 25 minutes and 1 hour
26

c) between 1 hour and 2 weeks
d) more than 2 weeks
The same lists as in Experiment 1a were used.
Results
Figure 4 shows the proportional distribution of categorizations into the
four time bins, and the proportion of counts in each category, depending on
event category and grammatical construction. In general, we observed a ten-
dency to categorize all events into the shorter time bins. More theoretically
crucial, light verb constructions were categorized as being shortest.
To assess the strength of evidence for the light verb construction shifting
responses systematically toward the shorter quartiles (and the strength of
evidence for any interaction of such an effect with event category), we used
a mixed-effects cumulative logit model using R’s ordinal package (Chris-
tensen, 2015). For ordered response categories (in our case, four bins rang-
ing from shortest to longest), a mixed-effects cumulative logit model speci-
fies response probabilities for a given data point as a function of predictor
variables. Instead of the intercept, ordered logit models provide a set of
threshold parameters, which describe the boundaries from one bin to the
next, and the probability of being drawn from one particular bin is esti-
mated by the linear predictors with the inverse logit function (see Appendix
B for an in-depth explanation of the mathematical underpinnings). The pre-
dictor variables in our model were construction (transitive or ditransitive),
event category (punctive count, durative count, or durative mass), and their
interaction. We Helmert-coded predictors and used maximal random effects
27

structure as in Experiment 1a. Under this coding, construction is coded with
transitive=−1, ditransitive (light verb)=1. Response categories were coded
1 to 4 in order of increasing duration. For this model and predictor coding,
if the light verb construction yields systematically shorter inferred event du-
rations than the transitive construction, it should manifest as a significantly
negative parameter estimate for the fixed effect of construction.
Table 4 shows the statistical results of the categorization task. The first
three rows indicate the threshold coefficients from one bin into the next. The
middle part of the table shows the regression coefficient for construction –
the –0.195 value indicating that the ditransitive construction is associated
with shorter event durations – and the results of the likelihood-ratio tests
for the main effects and their interaction, using likelihood-ratio tests as in
Experiments 1a and 1b. Construction had a significant effect on duration
estimates in the model comparison, and the interaction between construction
and event category was marginally significant, but the main effect of event
category was not. The lack of a main effect of event category is reassuring
given that the time bins were constructed based on duration estimates from
Experiment 1a.
Finally, the bottom part of Table 4 displays the results of analyses of the
main effect of construction within each event category. For punctive count
pairs, there was a significant difference in categorizations depending on con-
struction. For durative count and durative mass items, this difference was
not significant; however, a look at the β-estimates tells us that the effect of
construction went in opposite directions for durative count items, compared
to punctive count and durative mass items: whereas the ditransitive con-
28

struction resulted in more “shortest” categorizations for punctive count and
durative mass items, this effect was absent (numerically: reversed) for the
durative count items.
βSE
shortest|short -0.114 0.256
short|long 2.201 0.265
long|longest 4.669 0.320
Df βLR.stat p-value
construction 1 -0.195 4.079 0.043 *
event category 2 4.46 0.107 n.s.
construction ×event category 2 5.21 0.070 .
punctive count – construction 1 −0.811 4.995 0.025 *
durative count – construction 1 0.290 1.263 0.262 n.s.
durative mass – construction 1 −0.607 2.156 0.142 n.s.
Table 4: Experiment 2: Regression table for categorizations.
Discussion of Experiment 2
This study gave participants the opportunity to estimate event durations
using predefined time bins as choices, which might have been easier for them
than coming up with time estimates on their own. The answer options in
Experiment 2 were based on each item pair’s averaged estimates from Ex-
periment 1a, which, as we discussed above, seemed fairly high. Nevertheless,
we found significant differences for punctive count events: When presented
in ditransitive light verb constructions (give a kiss), they were estimated
to last less time than in transitive syntax (kiss). For durative count and
durative mass events (talking/giving a talk, advising/giving advice), we did
not observe such a difference, although the β-estimates indicate that durative
mass events follow the same trend as punctive count events, with ditransitive
29

(0.67)
(0.53)
(0.28)
(0.36)
(0.04)
(0.08)
(0.01)
(0.02)
(0.35)
(0.36)
(0.42)
(0.49)
(0.22)
(0.14)
(0.01)
(0.01)
(0.56)
(0.46)
(0.26)
(0.27)
(0.12)
(0.2)
(0.06)
(0.06)
punctive count durative count durative mass
shortest
short
long
longest
0 0.5 1 0 0.5 1 0 0.5 1
Proportion of duration category chosen
Constructions
transitive verb
ditransitive light verb
Figure 4: Experiment 2: Proportion of duration categories chosen, per event type and
construction.
30

structures pushing categorizations towards shorter bins.
Interestingly, the largest proportion of choices in both grammatical con-
structions and all three event types fell to the short or shortest options.
Given that the choice options were based on previously obtained quartiles,
the answers should have been roughly equally distributed. Thus, our data
contribute to the literature on over- and underestimation of event duration
in an interesting way: Open guesses as in Experiment 1a and 1b tend to
overestimate (at least for the event types we used here, which, unlike in clas-
sical studies on the planning fallacy, did not include unpleasant chore-like
events), whereas predefined categorizations are closer to more realistic event
durations.6
As stated above, though, we were not per se interested in how long events
are estimated to take, but in the influence of mass and count syntax on es-
timated event durations. Crucially, as in Experiment 1a and 1b, we found
that count syntax shortens the time estimate for punctive, but not durative
events. We had predicted this pattern because punctive verbs are often in-
terpreted iteratively, and by using count syntax, one picks out one particular
subevent. For durative verbs with count syntax, however, there is no distinct
subevent to be picked out: Thus, we hypothesized that durative verbs might
undergo a larger conceptual shift than when put in count syntax (to give a
talk) than do punctive count or durative mass verbs.
The remaining studies test these possible mechanisms of how temporal
6We tested this explanation by giving participants shorter answer options that were
not based on previous estimates. We found a much more equal distribution of answers.
Since this study is tangential to the goal of this article, it is not reported in detail, but
the data are accessible under https://github.com/ewittenberg/QuickKissing.
31

estimates are affected by syntax: Experiments 3a and 3b investigate whether
count syntax serves to single out one particular instance of an event. This
could explain why events are imagined to be shorter in punctive count syn-
tax. It would not, however, explain why durative count time-estimates were
unaffected by changes in syntactic structure; in this case, we hypothesized
that the lack of an effect in this condition is due to a conceptual shift be-
tween transitive verbs and count syntax, similar to the change from ”iron”
to ”an iron”. This conceptual shift would be orthogonal to event duration.
We investigated the existence of such a conceptual shift in Experiment 4.
Experiment 3a: Event repetition
This section presents two studies that establish whether using count or
mass syntax affects how many times an event is understood as occurring.
The theoretically most crucial prediction applies to punctive events. Punc-
tive events, like kissing, are often understood iteratively, even if their lexical
semantics merely conveys one single, telic, event (Kim & Kaiser, 2015). For
example, people might interpret John kissed Mary as him kissing her more
than once. However, when presented in count syntax (to give a kiss), only
one kiss should be singled out. This would explain the consistently lower
time estimates obtained in Experiments 1a, 1b, and 2.
Obviously, in order to be counted, a given event needs to be individuated,
and we expect individuation to be easier in count than in mass syntax (Barner
et al., 2008). So, for durative events in mass syntax, predictions are not
quite as straightforward: The professor advised her student consists of many
individual advising situations. The telic verb give in The professor gave
32

advice to her student introduces a boundary to the advising event (Krifka,
1992); however, no definite article aids in the event individuation, so we
might expect a weaker effect than for the punctive count events.
However, for durative events that can enter count syntax, we might pre-
dict the same trend as for punctive count events: talking could convey more
events than giving a talk. Note that this prediction is independent of our
second prediction, namely that there is a conceptual shift involved from a
transitive durative verb to the same lexical item in ditransitive count syntax;
and that any changes in event repetition counts would still be orthogonal to
changes in event duration.
Methods
Participants
For this study, we recruited 80 self-described English native speakers from
Amazon Mechanical Turk.
Procedure
Participants read each sentence and then noted how many events they
imagined reading the sentence. The exact instructions can be found under
https://github.com/ewittenberg/QuickKissing.
Stimuli
We used the stimuli described in Table 1, counterbalanced across two
lists.
33

Results
Our data spanned several orders of magnitude, (e.g., responses for punc-
tive events in transitive frame ranged from 1 to 60), effects of construc-
tion are likely to operate proportional to intrinsic construal of numbers of
events, as was the case with construed event durations in Experiments 1a
and 1b. Therefore we log-transformed event-count responses for purposes of
data summarization and analysis.
Mean count of events was lower for ditransitive light verb constructions
in each event category. Figure 5 shows the pattern of results, with log event
counts back-transformed to raw event counts for convenience of interpreta-
tion: For punctive count events, using a ditransitive light verb construction
instead of the transitive verb reduced the mean count from 1.7 to 1.5 (log-
averages). These effects were also numerically present for durative count and
durative mass items, but variances in these constructions were higher.
Df χ2p-value
construction 1 7.011 0.008 **
event category 2 5.326 0.069 .
construction ×event category 2 0.016 0.992 n.s.
punctive count – construction 1 7.187 0.007 **
durative count – construction 1 1.086 0.297 n.s.
durative mass – construction 1 1.030 0.310 n.s.
Table 5: Table of likelihood estimation results for event counts in Experiment 3a, testing
the main effects and their interaction (upper part), and pairwise comparisons (lower part).
Statistical analysis followed the same procedures as in Experiment 1a.
The upper part of Table 5 shows results of the 3 ×2 ANOVA-style analyses
(here, all random correlation parameters were removed to ensure model con-
vergence). We find that construction had a significant main effect on event
34

1
1.5
2
2.5
punctive count
(kiss − to give a kiss) durative count
(talk − to give a talk) durative mass
(advise − to give advice)
Event Counts, on log scale
Construction
transitive verb
ditransitive
light verb
Figure 5: Count of imagined events in Experiment 3a. The y-axis is represented in log
scale, but labeled with back-transformed counts (log-averages) for convenience.
35

counts, event category was marginally significant, and their interaction was
not. The lower half of Table 5 shows results of the planned tests of the simple
effect of construction within each event category (random correlation param-
eters did not need to be removed). We see a significant effect of construction
in punctive count pairs, but not in durative count or durative mass pairs.
Replication: Experiment 3b
We replicated this study by asking 80 native speakers on Amazon Me-
chanical Turk for the number of specific event types imagined after each
sentence, for example “How many kisses did you just imagine?”, “How many
talks did you just imagine?”, or “How many advices did you just imagine?”
(instead of, as in Experiment 3a, asking the generic “How many events did
you just imagine?”). Note that predictions for durative count items might
be stronger in this study than in Experiment 3a, since we asked for the
specific event (e.g., talks), thus excluding any super- or subevents people
might have imagined as well, such as ”climbing the podium”, ”adjusting the
microphone”, etc. We obtained a similar pattern of results (see Figure 6).
Data analysis procedures were identical to those in Experiment 3a. Table
6 displays the results of the event count estimation test for the main effects,
which were both significant, and their interaction, which was not. Pairwise
comparisons (bottom part of Table 6) show main effects of construction for
both punctive count and durative count, but not for durative mass events.
Discussion of Experiment 3a and 3b
This study investigated whether phrasing an event with mass or count
syntax, instead of with a transitive verb, affects any iterative readings that
36

1
1.5
2
2.5
punctive count
(kiss − to give a kiss) durative count
(talk − to give a talk) durative mass
(advise − to give advice)
Event Counts, on log scale
Construction
transitive verb
ditransitive
light verb
Figure 6: Count of imagined events in Experiment 3b. The y-axis is represented in log
scale, but labeled with back-transformed counts (log-averages) for convenience
37

Df χ2p-value
construction 1 11.034 0.000 ***
event category 2 8.297 0.016 *
construction ×event category 2 1.001 0.606 n.s.
punctive count – construction 1 5.391 0.020 *
durative count – construction 1 4.063 0.043 *
durative mass – construction 1 2.234 0.135 n.s.
Table 6: Table of likelihood estimation results for event counts in Experiment 3b, testing
the main effects and their interaction (upper part), and pairwise comparisons (lower part).
are present.
Crucially, we see a significant reduction in imagined number of events as
a measure of implicit iterativity from transitive verb encoding to ditransitive
light verb encoding – but, as predicted, only consistently in punctive count
events, and to a lesser degree, in durative count events. Thus, the more
pronounced effect for durative count items in Experiment 3b, and the overall
reduction in event counts in Experiment 3b, confirm our hypothesis that
count syntax encourages individuating over subevents, leading to shorter
event conceptualization.
Contrary to what the previous literature has claimed, durative events
resulted in iterative readings, as well; we attribute that to the fact that events
like advising or talking, while not easily segmentable by pieces of advice or
specific identifiable talking events, do carry an element of interactivity: There
is a back-and-forth between the advisor or talker, and the addressee; and
there might also be habitual interpretations available. Thus, it could be that
our participants conceptualized the “number of events” question as “number
of subevents” – which would explain the occasional “more than one” answer
for count syntax like give a talk.
38

However, we had hypothesized that count syntax might encourage a con-
ceptual shift in durative verbs that acts orthogonally to any changes in tem-
poral conceptualization. The last experiment investigates this possibility.
Experiment 4: Event Similarity
This study tested another prediction of our theory: that durative events in
count syntax (to give a talk) should be conceptually further apart from their
transitive verb counterparts (to talk ) than punctive events in count syntax,
or durative events in mass syntax. This prediction is drawn from the analogy
to mass nouns, such as glass or iron into count syntax: In many cases, the
count noun denotes objects that are conceptually related, yet different, from
the noun in mass syntax, such as in a glass or an iron. If the durative events
in count syntax behave in parallel, we should expect them to be conceptually
further apart from their transitive counterparts than punctive events in count
or durative events in mass syntax.
Methods
Participants
We recruited 40 self-described English native speakers from Amazon Me-
chanical Turk for this study.
Procedure
We asked participants to rate event similarity between transitive and
ditransitive frames on a 7-point Likert scale where 1 indicated “same event”,
and 7, “completely different event”. The exact instructions can be found
under https://github.com/ewittenberg/QuickKissing.
39

Stimuli
We used the 20 item pairs shown in Table 1, without the sentence context.
In addition, we created 26 filler pairs that ranged from very close synonyms
(repairing – fixing) to very different events (working – being lazy).
Results
For filler items, the average rating was 3.62 (SD=2.1), with the full range
of the scale being used.
Figure 7 shows the rating results for critical pairs. Punctive events (to kiss
vs to give a kiss) received a mean rating of 1.6 (SD=1.0). Pairs of durative
verbs and durative verbs in mass syntax (to advise vs to give advice) received
a similarly low rating (mean: 1.5, SD=.9). This means that for both punctive
count and durative mass items, both constructions denote very similar events.
As predicted, difference ratings for pairs of durative verbs and durative verbs
in count syntax (to talk vs to give a talk ) were higher: 2.1 (SD=1.5).
Df χ2p-value
event category 2 6.819 0.033 *
punctive count vs durative count 1 3.799 0.046 *
durative count vs durative mass 1 4.869 0.027 *
punctive count vs durative mass 1 0.000 0.984 n.s.
Table 7: Experiment 4: Table of likelihood estimation results for event similarity in Exper-
iment 4, testing the main effect of event category (upper part), and pairwise comparisons
between event categories (lower part).
Like in Experiments 1a, 1b, and 3, we used linear mixed models with
maximal random effects structure (by-subjects intercepts and event-category
slopes, by-items intercepts; no random correlation parameters needed to be
removed). Table 7 displays the results from 3-level omnibus ANOVA-style
40

0.0
0.5
1.0
1.5
2.0
punctive count
(kiss − to give a kiss) durative count
(talk − to give a talk) durative mass
(advise − to give advice)
Similarity Rating
Figure 7: Experiment 4: Punctive count and durative mass item pairs were rated more
similar to each other than durative count item pairs.
analysis, as well as simple comparisons between all pairs of event categories.
There was a significant main effect of event category in the omnibus analysis.
The pairwise comparisons show no significant difference between similarity
ratings for punctive count and durative mass event pairs; durative count
pairs, on the other hand, were rated significantly less similar to each other
than pairs in the other two event categories.
Discussion of Experiment 4
Experiment 4 confirms the intuition that while giving a kiss and kissing,
as well as giving advice and advising, belong ontologically to the same event,
41

giving a talk and talking are conceptually further apart – albeit with con-
siderable overlap. This result contributes to claims made in the literature
that light verb constructions often help to close lexical gaps (Glatz, 2006;
Grimshaw & Mester, 1988; Miyagawa, 1989), and it draws our attention to
interesting parallels to using count syntax on mass nouns (Gordon, 1985;
Srinivasan & Rabagliati, 2015; Wiese & Maling, 2005).
These results are crucial to our question of how using mass or count
syntax affects the construal of event duration: If a durative lexeme in count
syntax covers a conceptually different event than the same lexical item in a
transitive verb frame, the conceptualization of event duration applies to the
new concept, and thus any changes in duration estimates would be largely
coincidental.
General Discussion
We have presented a family of experiments that tested the hypothesis
that describing an event with mass versus count syntax affects the construal
of event similarity and duration in a way that is systematically predictable
from the interaction of mass versus count syntax and verb semantics (see
Figure 1). These predictions were built on insights from formal semantics
that has pointed out similarities between mass nouns and durative, non-
atomic events on the one side, and count nouns and punctive, atomic events
on the other side (Bach, 1986; Casati & Varzi, 2008; Harley, 2005; Jackendoff,
1991; Krifka, 1992; Wellwood et al., 2016).
Specifically, we had predicted that punctive events in count syntax (give
a kiss) are construed as taking less time than in transitive verb frame (kiss);
42

this pattern was predicted because in punctive events, which are often inter-
preted iteratively (Kim & Kaiser, 2015), one atomic subevent is singled out
by count syntax.
In durative events, the same pattern was predicted for events in mass
syntax, such as giving advice. In combination with the telic verb give, the
mass noun carves out a limited portion of the event structure, leading to
shorter event conceptualization.
A different pattern was predicted for durative verbs in count syntax (give
a talk versus talk ): We expected durative events in count syntax to be seman-
tically further from their verbal counterparts (to talk – a talk) than punctive
events (to kiss – a kiss). Since these shifts were presumably orthogonal to
the temporal structure of the event, we did not make any predictions about
duration estimates. These predictions stemmed from insights about the par-
allels between durative events and mass nouns: When some mass nouns are
forced into count syntax (a glass; an iron), there is a semantic shift from the
mass noun meaning (glass; iron).
Our experimental results broadly confirm these predictions. In Experi-
ments 1a and 1b, we elicited open duration estimates, which were consistently
lower for punctive events in count syntax and durative events in mass syn-
tax than when they occurred in transitive frames; durative events in count
syntax did not show this effect. Experiment 2 replicated these results by
using the quartiles of each event’s individual duration estimates obtained in
Experiment 1a as answer choices; again, punctive count and durative mass
ditransitive structures were judged to take less time than transitive struc-
tures, while there was no difference for durative events in count syntax.
43

On our theory, a key factor for the reduction of duration estimates in
punctive verbs was that people should imagine fewer events taking place than
in the ditransitive count frame, due to the combination of a telic verb (give)
and nominal count syntax (a kiss). However, this effect should be weaker for
durative events in either count or mass syntax when participants are asked for
number of events. Experiment 3a confirmed these predictions. Experiment
3b showed that when people are asked for how many specific events they
imagined – how many talks, for example – the reduction in event counts is
significant even for durative events in count syntax. This confirms both the
observation that event individuation is easier in count syntax (Barner et al.,
2008), and that durative verbs behave similarly to mass nouns (Krifka, 1992).
Experiment 4 showed, consistent with our predictions, that durative events
undergo a semantic shift in count syntax: The semantic differences between
transitive and ditransitive frames were larger in durative count pairs (talk –
give a talk) than in punctive-count pairs (kiss – give a kiss) or durative-mass
pairs (advise – give advice).
Thus, the shift from a transitive to a ditransitive frame has systematically
predictable repercussions, depending on whether the event type was durative
or punctive, and depending on whether the event was described with a mass
or with a count noun.
These results provide psycholinguistic evidence for the observation in for-
mal semantics that reference properties of syntactic objects change the refer-
ence properties of the whole predicate (Krifka (1992); also see Quine (1969);
Verkuyl (1972)). In our case, the nominalization of eventive verbs using light
verb constructions with a telic verb helped to divide experience into count-
44

able units (for punctive verbs in count syntax and durative verbs in mass
syntax), or introduced a semantic shift (for durative verbs in count syntax),
similar to what happens with mass nouns in count syntax.
Our results also complement studies that suggest that people conceptu-
alize events differently depending on subtle choices among syntactic alterna-
tions (Fausey & Boroditsky, 2010; Johnson & Goldberg, 2013; Wittenberg &
Snedeker, 2014) and that conversely, subtle changes in event structure result
in changing preferences between syntactic alternations (Gropen et al., 1991).
This growing family of experiments, together with a number of corpus studies
(Benor & Levy, 2006; Bresnan et al., 2007), contributes to our understanding
how syntactic and semantic structures, as well as processing pressures, such
as a preference for using frequent, accessible lexical items, lead to a speaker’s
decision between two seemingly equivalent constructions.
These findings may also help interpret results from less explicit tasks. We
know from behavioral, ERP, and MEG studies that light verb constructions
are processed differently from non-light constructions (Briem et al., 2009;
Pi˜nango et al., in press; Wittenberg et al., 2014; Wittenberg & Pi˜nango,
2011). So far, however, only semantic role mismatches had been identified as
a factor contributing to the processing difference (Wittenberg et al., under
review; Wittenberg & Snedeker, 2014). Based on the present work, we might
hypothesize that the calculation of temporal structure also plays a role in
the real-time processing of light verb constructions.
A question this article raises is how well these results might generalize to
other constructions, for example, other light verb constructions with atelic
verbs. We would predict that other light verb constructions with boun-
45

ded verbs would lead to the same effect; for example, John took a shower
should be estimated as taking less time than John showered, whereas John
had a shower should affect estimated event durations to a lesser degree,
since both verbs (to shower and to have ) are atelic. Another area where the
interaction of count/mass syntax and verbal aspect might have repercussions
for the temporal conceptualization of events are idioms. In the literature
on idiom comprehension there is a consensus that the lexical items in an
idiomatic phrase such as to kick the bucket are accessed individually as well
as holistically (Cutting & Bock, 1997; Holsinger, 2013; Sprenger et al., 2006;
Tabossi et al., 2009; Titone & Libben, 2014). So if an idiom consists of
a semelfactive verb (kick) and a count noun (bucket), and its semantically
transparent counterpart is a semantic achievement (to die) we would predict
that John kicked the bucket will be estimated to take a shorter time than
John died. These predictions remain to be addressed in future studies.
In sum, this article showed that using mass versus count syntax affects
the construal of event similarity and duration in a way that is systemati-
cally predictable from the interaction of mass versus count syntax and verb
semantics. Our results confirm observations from formal semantics that the
properties of objects and events are mirrored in count/mass syntax and ver-
bal aspect, and are advancing our understanding of the effects of syntactic
choices on subtle aspects of event construal.
46

Acknowledgments
We would like to thank David Barner, Victor Ferreira, Richard Gerrig,
Ray Jackendoff, Titus von der Malsburg, Emily Morgan, Jesse Snedeker,
David Townsend, Alexis Wellwood, Heike Wiese, as well as three anonymous
reviewers for helpful discussion, and Suhas Arehalli for assistance in running
the studies. This work was supported by NSF Grant IIS-0953870, NIH Grant
HD065829, and an Alfred P. Sloan Fellowship to Roger Levy, as well as by
an German Academic Exchange Service (DAAD) postdoctoral scholarship to
Eva Wittenberg.
47

References
Allerton, D. (2002). Stretched Verb Constructions in English. London.
Anderson, A., Garrod, S. C., & Sanford, A. J. (1983). The accessibility of
pronominal antecedents as a function of episode shifts in narrative text.
Quarterly Journal of Experimental Psychology,35 , 427–440.
Andersson, S.-G. (1972). Aktionalit¨at im Deutschen: Die Kategorien Aspekt
und Aktionsart im Russischen und im Deutschen. Almqvist & Wiksell.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects model-
ing with crossed random effects for subjects and items. Journal of Memory
and Language,59 , 390–412.
Bach, E. (1986). The algebra of events. Linguistics and philosophy ,9, 5–16.
Bale, A. C., & Barner, D. (2009). The interpretation of functional heads:
Using comparatives to explore the mass/count distinction. Journal of Se-
mantics,26 , 217–252.
Barner, D., & Bale, A. (2002). No nouns, no verbs: psycholinguistic argu-
ments in favor of lexical underspecification. Lingua,112 , 771–791.
Barner, D., & Snedeker, J. (2006). Children’s early understanding of mass-
count syntax: individuation, lexical content, and the number asymmetry
hypothesis. Language Learning and Development,2, 163–194.
Barner, D., Wagner, L., & Snedeker, J. (2008). Events and the ontology
of individuals: Verbs as a source of individuating mass and count nouns.
Cognition,106 , 805–832.
48

Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects
structure for confirmatory hypothesis testing: Keep it maximal. Journal
of Memory and Language,68 , 255–278.
Bates, D., Maechler, M., Bolker, B., Walker, S., Christensen, R. H. B.,
Singmann, H., Dai, B., Eigen, C., & Rcpp, L. (2014). Package ‘lme4’.
R Foundation for Statistical Computing, Vienna, .
Bauer, G. (1970). The English ?perfect?reconsidered. Journal of Linguistics ,
6, 189–198.
Benor, S. B., & Levy, R. (2006). The chicken or the egg? A probabilistic
analysis of English binomials. Language, (pp. 233–278).
Boltz, M. G. (1995). Effects of event structure on retrospective duration
judgments. Perception & psychophysics,57 , 1080–1096.
Bott, O. (2010). The processing of events volume 162. John Benjamins
Publishing.
Bresnan, J., Cueni, A., Nikitina, T., Baayen, R. H. et al. (2007). Predicting
the dative alternation. Cognitive foundations of interpretation, (pp. 69–
94).
Briem, D., Balliel, B., Rockstroh, B., Butt, M., im Walde, S. S., & As-
sadollahi, R. (2009). Distinct processing of function verb categories in the
human brain. Brain Research,1249 , 173–180.
Brugman, C. (2001). Light verbs and polysemy. Language Sciences,23 ,
551–578.
49

Buhrmester, M., Kwang, T., & Gosling, S. D. (2011). Amazon’s Mechanical
Turk a new source of inexpensive, yet high-quality, data? Perspectives on
psychological science,6, 3–5.
Burt, C. D., & Popple, J. S. (1996). Effects of implied action speed on
estimation of event duration. Applied cognitive psychology,10 , 53–63.
Butt, M. (2003). The Light Verb Jungle. Harvard Working Papers in Lin-
guistics,9, 1–49.
Butt, M. (2010). The light verb jungle: still hacking away. In B. B. M. Amber-
ber, M. Harvey (Ed.), Complex Predicates in Cross-Linguistic Perspective
(pp. 48–78). Cambridge University Press.
Casati, R., & Varzi, A. C. (2008). Event concepts. Understanding events:
From perception to action, (pp. 31–53).
Christensen, R. H. B. (2015). ordinal—Regression Models for Or-
dinal Data. R package version 2015.1-21. http://www.cran.r-
project.org/package=ordinal/.
Coll-Florit, M., & Gennari, S. P. (2011). Time in language: Event duration
in language comprehension. Cognitive Psychology,62 , 41–79.
Crump, M. J., McDonnell, J. V., & Gureckis, T. M. (2013). Evaluating
Amazon’s Mechanical Turk as a tool for experimental behavioral research.
PloS one,8, e57410.
Cutting, J. C., & Bock, K. (1997). That’s the way the cookie bounces: Syn-
50

tactic and semantic components of experimentally elicited idiom blendsß.
Memory & Cognition,25 , 57–71.
Dowty, D. (1991). Thematic Proto-Roles and Argument Selection (Pre-
publication version). Language,67:3 , 547–619. URL: http://www.ling.
ohio-state.edu/~dowty/trld-pdf/.
Fausey, C. M., & Boroditsky, L. (2010). Subtle linguistic cues influence
perceived blame and financial liability. Psychonomic Bulletin & Review,
17 , 644–650.
Flecken, M., & Gerwien, J. (2013). Grammatical aspect modulates event
duration estimations: findings from Dutch. In the 35th Annual Meeting of
the Cognitive Science Society (CogSci 2013) (pp. 2309–2314). Cognitive
Science Society.
Garey, H. B. (1957). Verbal aspect in French. Language,33 , 91–110.
Garrod, S., & Pickering, M. J. (2003). Linguistics fit for dialogue. Behavioral
and Brain Sciences,26 , 678–678.
Glatz, D. (2006). Funktionsverbgef¨uge–semantische Doubletten von ein-
fachen Verben oder mehr. Dom¨anen der Lexikalisierung kommunikativer
Konzepte,33 , 129.
Goldberg, A. (1995). Constructions: A Construction Grammar Approach to
Argument Structure. University of Chicago Press.
Goldberg, A. (2003). Constructions: a new theoretical approach to language.
Trends in Cognitive Sciences,7, 219–224.
51

Gordon, P. (1985). Evaluating the semantic categories hypothesis: The case
of the count/mass distinction. Cognition,20 , 209–242.
Grimshaw, J., & Mester, A. (1988). Light Verbs and Theta-Marking. Lin-
guistic Inquiry,19 , 205–232.
Grondin, S., Meilleur-Wells, G., & Lachance, R. (1999). When to start
explicit counting in a time-intervals discrimination task: A critical point
in the timing process of humans. Journal of Experimental Psychology:
Human Perception and Performance,25 , 993.
Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Affectedness
and direct objects: The role of lexical semantics in the acquisition of verb
argument structure. Cognition,41 , 153–195.
Hale, K., & Keyser, J. (1993). On argument structure and the lexical ex-
pression of syntactic relations. In K. Hale, & J. Keyser (Eds.), The View
from Building 20 (pp. 53–109). Cambridge: MIT Press.
Halkjelsvik, T., Jørgensen, M., & Teigen, K. H. (2011). To read two pages,
I need 5 minutes, but give me 5 minutes and I will read four: How to
change productivity estimates by inverting the question. Applied Cognitive
Psychology,25 , 314–323.
Halle, M., & Marantz, A. (1994). Some key features of Distributed Morphol-
ogy. MIT working papers in linguistics ,21 , 88.
Harley, H. (2005). How do verbs get their names? Denominal verbs, manner
incorporation, and the ontology of verb roots in english. In The Syntax of
Aspect (pp. 42–64). University Press.
52

Holsinger, E. (2013). Representing idioms: Syntactic and contextual effects
on idiom processing. Language and speech,56 , 373–394.
Jackendoff, R. (1974). A Deep Structure Projection Rule. Linguistic Inquiry ,
5.4 , 481–505.
Jackendoff, R. (1991). Parts and boundaries. Cognition,41 , 9–45.
Jackendoff, R. (2002). Foundations of Language. Oxford University Press.
Jackendoff, R. (2007). A Parallel Architecture perspective on language pro-
cessing. Brain Research,1146 , 2–22.
Jespersen, O. (1954). A Modern English Grammar on Historical Principles.
London: Allen & Unwin and Copenhagen: Ejnar Munksgaard.
Johnson, M. A., & Goldberg, A. E. (2013). Evidence for automatic access-
ing of constructional meaning: Jabberwocky sentences prime associated
verbs. Language and Cognitive Processes,28 , 1439–1452. URL: http://
dx.doi.org/10.1080/01690965.2012.717632. doi:10.1080/01690965.
2012.717632.
Kahneman, D., & Tversky, A. (1977). Intuitive prediction: Biases and cor-
rective procedures. Technical Report DTIC Document.
Kearns, K. (2000). Semantics. Modern Linguistics Series. England: Macmil-
lan, .
Kim, L., & Kaiser, E. (2015). The number of times a motion repeats influ-
ences sentence processing. In Proceedings of the Cognitive Science Society
Conference.
53

Klein, W. (1994). Time in language. Psychology Press.
Krifka, M. (1992). Thematic relations as links between nominal reference and
temporal constitution. In I. Sag, & A. Szabolcsi (Eds.), Lexical Matters
(pp. 29–53). CSLI Publications, Chicago University Press.
Kruger, J., & Evans, M. (2004). If you don’t want to be late, enumerate:
Unpacking reduces the planning fallacy. Journal of Experimental Social
Psychology,40 , 586–598.
Levy, R. (2008). Expectation-Based Syntactic Comprehension. Cognition,
106 , 1126–1177.
Levy, R. (2014). Using R formulae to test for main effects in the presence of
higher-order interactions. arXiv:stat.ME/1405.2094v1 , .
Loftus, E. F., Schooler, J. W., Boone, S. M., & Kline, D. (1987). Time went
by so slowly: Overestimation of event duration by males and females.
Applied Cognitive Psychology,1, 3–13.
Macar, F., Grondin, S., & Casini, L. (1994). Controlled attention sharing
influences time estimation. Memory & Cognition,22 , 673–686.
Madden, C. J., & Zwaan, R. A. (2003). How does verb aspect constrain event
representations? Memory & Cognition,31 , 663–672.
Magliano, J. P., & Schleich, M. C. (2000). Verb aspect and situation models.
Discourse processes,29 , 83–112.
Magliano, J. P., & Zacks, J. M. (2011). The impact of continuity editing in
narrative film on event segmentation. Cognitive Science,35 , 1489–1517.
54

Martin, A. E. (2016). Language processing as cue integration: Grounding
the psychology of language in perception and neurophysiology. Frontiers
in psychology,7.
Miyagawa, S. (1989). Light verbs and the ergative hypothesis. Linguistic
inquiry, (pp. 659–668).
Newman, J. (1996). Give: A cognitive linguistic study volume 7. Walter de
Gruyter.
Paczynski, M., Jackendoff, R., & Kuperberg, G. (2014). When Events
Change Their Nature: The Neurocognitive Mechanisms Underlying As-
pectual Coercion. Journal of Cognitive Neuroscience, .
Pedersen, A. C., & Wright, D. B. (2002). Do differences in event descriptions
cause differences in duration estimates? Applied Cognitive Psychology,16 ,
769–783.
Pi˜nango, M. M., Mack, J., & Jackendoff, R. (in press). Semantic Com-
binatorial Processes in Argument Structure: Evidence from Light-Verbs.
Proceedings of Berkeley Linguistics Society, .
Quine, W. V. (1969). Word and object. Cambridge, Mass, .
R Core Team (2014). R: A Language and Environment for Statistical Com-
puting. R Foundation for Statistical Computing Vienna, Austria. URL:
http://www.R-project.org/.
Rips, L. J., & Hespos, S. J. (2015). Divisions of the physical world: Concepts
of objects and substances. Psychological bulletin,141 , 786.
55

Rothstein, S. (2008). Structuring events: A study in the semantics of aspect
volume 5. John Wiley & Sons.
Roy, M. M., Christenfeld, N. J., & McKenzie, C. R. (2005). Underestimating
the duration of future events: memory incorrectly used or memory bias?
Psychological bulletin ,131 , 738.
Sprenger, S. A., Levelt, W. J., & Kempen, G. (2006). Lexical access during
the production of idiomatic phrases. Journal of Memory and Language,
54 , 161–184. URL: http://www.sciencedirect.com/science/article/
B6WK4-4J1410G-2/2/257706a2bcebae5e29e13cc838ad7320.
Srinivasan, M., & Rabagliati, H. (2015). How concepts and conventions
structure the lexicon: Cross-linguistic evidence from polysemy. Lingua, .
Tabossi, P., Wolf, K., & Koterle, S. (2009). Idiom syntax: Idiosyncratic or
principled? Journal of Memory and Language,61 , 77–96.
Titone, D., & Libben, M. (2014). Time-dependent effects of decomposability,
familiarity and literal plausibility on idiom priming: A cross-modal priming
investigation. The Mental Lexicon,9, 473–496.
Tse, P. U., Intriligator, J., Rivest, J., & Cavanagh, P. (2004). Attention
and the subjective expansion of time. Perception & psychophysics,66 ,
1171–1189.
Vendler, Z. (1957). Verbs and Times. The Philosophical Review,66 , 143–160.
Verkuyl, H. J. (1972). On the compositional nature of the aspects volume 15.
Springer.
56

Wellwood, A., Hespos, S. J., & Rips, L. (2016). The object : substance ::
event : process analogy. M.s. Northwestern University.
Wiese, H. (2006). “Ich mach dich Messer” – Grammatische Produktivit¨at in
Kiez-Sprache (“Kanak Sprak”). Linguistische Berichte,207 , 245–273.
Wiese, H., & Maling, J. (2005). Beers, kaffi, and Schnaps – Different gram-
matical options for ‘restaurant talk’ coercions in three Germanic languages.
Journal of Germanic Linguistics,17:1 , 1–38.
Wittenberg, E., Khan, M., & Snedeker, J. (under review). How mismatches in
the syntax-semantics interface influence conceptualizations of event struc-
ture. Manuscript.
Wittenberg, E., Paczynski, M., Wiese, H., Jackendoff, R., & Kuperberg,
G. (2014). The difference between “giving a rose” and “giving a kiss”:
Sustained neural activity to the light verb construction. Journal of Memory
and Language,73 , 31–42.
Wittenberg, E., & Pi˜nango, M. M. (2011). Processing light verb construc-
tions. The Mental Lexicon,6, 393–413.
Wittenberg, E., & Snedeker, J. (2014). It takes two to kiss, but does it take
three to give a kiss? Categorization based on thematic roles. Language,
Cognition and Neuroscience,29 , 635–641.
Zacks, J. M., & Tversky, B. (2001). Event structure in perception and con-
ception. Psychological bulletin ,127 , 3.
57

Zakay, D., & Block, R. A. (1995). An attentional gate model of prospective
time estimation. Time and the dynamic control of behavior, (pp. 167–178).
58

Appendix A: Stimuli used for Experiments 1-3
Stimuli are shown in transitive frame. In the ditransitive light verb con-
struction, the underlined verb would serve as the light noun, i.e., After their
first date, Douglas kissed Mary →After their first date, Douglas gave a kiss
to Mary.
1. Punctive Count
(a) After their first date, Douglas kissed Mary.
(b) Laughing nastily, the thug kicked the victim.
(c) When they met up, Nathasha hugged Cynthia.
(d) Noam embraced Jennifer before they split up.
(e) Julius cuddled his brother at bedtime.
(f) Laura poked Owen because he was so annoying.
(g) Martin shook the cocktail before he served it.
2. Durative Count
(a) The professor lectured on injustice yesterday.
(b) The CEO talked about his latest sales strategy last night.
(c) The President spoke about affordable education on Wednesday.
(d) During the community art show, the composer presented his latest
work.
(e) After the graduation ceremony, Tom Hanks addressed the stu-
dents.
(f) The mother scolded the child today.
3. Durative Mass
(a) The professor advised her student on his paper yesterday.
(b) Yesterday, the nurse assisted Dr. Kohler in the emergency room.
59

(c) After the job interview, Sheila assured Keith.
(d) Before Kelly left for a year abroad, her friends encouraged her.
(e) After the devastating flood, the mayor recognized the rescue work-
ers.
(f) His new girlfriend supported Sam after his divorce.
(g) At the summer party, the university official thanked the academic
community for their efforts.
60

Appendix B: The mixed-effects cumulative logit model used for
Experiment 2
Data were analyzed using R (R Core Team, 2014), specifically with mixed-
effects cumulative logit model using the ordinal package (Christensen, 2015).
For ordered response categories 1,2, . . . , N , a mixed-effects cumulative logit
model specifies response probabilities for a given datum ias a function of
predictor variables as follows. There are N−1 linear predictors, each of the
following form:
ηij =αj+X
k
βkxik +X
k
bkzik j∈ {1,2, . . . , N −1}
where the {xik}and {zik }are fixed- and random-effects predictors respec-
tively, the {βk}and {bk}are fixed- and random-effects regression parameters
respectively, and αiis a threshold parameter for the boundary between
the j-th and (j+1)-th response categories. (The threshold parameters play a
role analogous to the intercept in an ordinary mixed logit model.) The linear
predictors are related to cumulative probabilities {γij}through the inverse
logit function:
γij =P(Yi≤j) = eηij
1 + eηij
61

The probability of datum ihaving response category jis thus
P(Yi=j) =













γi1j= 1
γij −γij−1j∈ {2, . . . , N −1}
1−γij−1j=N
That is, the {γij }carve up the interval [0,1] into a set of category response
probabilities, as illustrated in Figure 8. In mixed-effects cumulative logit
models, the random-effects regression parameters are assumed to be drawn
from some multivariate normal distribution, the covariance matrix of which
is estimated jointly along with the threshold parameters and fixed-effects
regression parameters (here, via Laplace-approximated maximum likelihood).
62

−4 −2 0 2 4
η1η2η3
γ1
γ2
γ3
0
1
Figure 8: Cumulative logit models. The N−1 linear predictors {ηj}induce a set of N
multinomial response category probabilities through the inverse logit transform.
63