Doing it again and again may be difficult – but it depends on what you are doing…
Keywords: aspectual coercion, semantic processing, experimental semantics, Event Calculus
Psycholinguistic experimentation has recently started to investigate processing difficulty in aspectual
coercion. A number of studies focus on the computation of iterative readings in sentences like John
coughed all night. The findings are rather mixed. While some studies report difficulty (Piñango et al.,
1999, Todorova et al. 2000, Husband et al., 2006, Piñango et al., 2006, Brennan & Pylkkänen, submitted)
others don’t find any effect (Pickering et al., 2006). Brennan & Pylkkänen (submitted) argue that this may
be due to disregarding aspectual properties of the materials. They carefully selected clear instances of
point-action verbs (points) and for these they obtained enhanced processing load in reading times and
MEG data. The present study scrutinizes the processing of iterative readings contingent on aspectual
class. We are especially interested in iterated accomplishments where we expect initial construction of an
implausible model and subsequent recomputation of a more plausible situation. We describe two reading
time experiments in German which compared the processing of aspectual coercion.
Experiment 1 tested two different kinds of coercion. The short for-adverbial in (1a) is an instance of
subtractive coercion - the accomplishment is coerced into an activity1 by “subtracting” the culmination
from the eventuality. The long for-adverbial in (1c) coerces the accomplishment into an iterative or
habitual reading because a loading activity cannot last for such a long time. The long in-adverbial in (1d)
is a pragmatic mismatch that cannot be resolved. Short in-adverbials in (1b) are used as control.
Experiment 2 investigated points in single event (2a) vs. iterative readings (2b) in constructions similar to
the one used in Brennan & Pylkkänen (submitted).
To make sure that the materials of Experiment 1 and 2 in fact exhibit the intended readings, we
pretested conditions (1a-c) and (2) in a questionnaire study (40 acc. and 20 points). 24 subjects judged
whether the sentence expresses that the (with acc.: culminating) event happened and if so, whether it
happened only once or repeatedly. Short in-adverbials (71%) and long for-adverbials (86%) over-
whelmingly received “yes” responses compared to 15% for short for-adverbials. Thus, short for triggered
subtractive coercion. All of the “yes” answers in the short in condition were “only once” while in the long
for condition 97% were “repeatedly”. Thus, long for coerced accomplishments into an iterative reading.
The points received 100% “yes” in both conditions. Single events had “only once” judgments in 94%,
whereas iterated events were 94% judged “repeatedly” indicating the computation of coerced readings.
In Experiment 1 participants read the sentences phrase-by-phrase and judged if they made sense. All
conditions except mismatch (23%) were overwhelmingly judged sensible: subtractive coercion (87%),
iteration (70%) and control (84%). Shifting an accomplishment into an iterative reading slowed down
reading times of the long for-adverbial compared to those of short in. The former were numerically
roughly equal to mismatch. Interestingly, subtractive coercion (short for) didn’t differ significantly from
control (short in). The reading times of Experiment 1 are depicted in Figure 1.
Experiment 2 revealed no difference in reading time between single event and iterated event
sentences (see Figure 2). The latter were numerically even faster than the former. Taken together, the
results show that computing an iterated reading doesn’t introduce difficulty per se but rather, difficulty
depends on aspectual class.
The findings can be explained using the Event Calculus (EC) by Hamm & van Lambalgen (2005), an
algorithmic semantic framework that constructs situation models in a computationally tractable way. EC
allows us to model difficulty in iterative accomplishments due to reanalysis of the initial, implausible
model. Both subtractive coercion and the iterative coercion of a point involve completely different
computations and are predicted to be less demanding. We argue that it is this computational difference
that corresponds to the differential effect in processing.
1 This is shown by the different entailments: The worker loaded the cart entails the cart was completely loaded but
when modified by for twenty minutes the entailment doesn’t hold anymore.
