=Paper=
{{Paper
|id=Vol-1845/paper4
|storemode=property
|title=The Cost of Content Repair: Presupposition Accomodation
|pdfUrl=https://ceur-ws.org/Vol-1845/paper4.pdf
|volume=Vol-1845
|authors=Filippo Domaneschi,Simona Di Paola
}}
==The Cost of Content Repair: Presupposition Accomodation==
https://ceur-ws.org/Vol-1845/paper4.pdf
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
The Cost of Context Repair: Presupposition
Accomodation
Simona Di Paola and Filippo Domaneschi
Department of Educational Sciences - University of Genoa
filippo.domaneschi@unige.it
simona.dipaola@edu.unige.it
Abstract. In the present experiment, the processing costs and time-
course of presupposition accommodation were studied, as compared to
presupposition satisfaction and independently of the presupposition
trigger in use. Two main results emerged from the data collected. First,
presupposition accommodation requires greater processing costs than
satisfaction, reflecting a process of context repair where both a linking
and an updating process are needed. Second, presupposition
accommodation takes places immediately just as the trigger becomes
available and proceeds incrementally during the sentence processing.
This result suggests that presuppositions are processed on-line and that,
independently of the type of trigger in use, they are accommodated
before the asserted content is computed.
Keywords. Context, experimental pragmatics, presupposition,
accommodation, satisfaction, triggers.
1. Repairing context by presupposition accomodation
Presuppositions are background information communicated as taken for granted. They
are carried by presupposition triggers, that is lexical items and syntactic constructions
that activate a presupposition when used in an utterance - e.g. definite descriptions,
change of state verbs, etc. (Karttunen, 1974; Levinson, 1983). For example, the
utterance
36
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
(1) Mark has given up smoking
introduces the presupposition
(1a) Mark has been smoking
which is activated by the change of state verb to give up.
According to the traditional Stalnaker-Karttunen-Heim semantic account
(Heim,1990; Stalnaker, 1974; Heim & Kratzer 1998), presuppositions typically re-
strict the context update. According to Stalnaker (2002), a sentence p presupposes q if
the use of p would be inappropriate and if q did not belong to the background of com-
mon presuppositions in a conversation. If the presupposition q is entailed by the con-
text, then q is said to be satisfied. Conversely, if q does not belong to the common
ground this leads to presupposition failure. In such a case, speakers are supposed to
repair the failure to make sense of the presupposing utterance. The mechanism
underlying failure repairing with a presupposed utterance is represented by
accommodation (Lewis, 1979; Heim 1982), that is the process whereby the
presupposition that is not satisfied is introduced in the context set to make the context
update possible.
Within this semantic framework, therefore, presuppositions constitute a precondition
for the comprehension of an utterance and for the update of the context with the
assertive content of the presupposing utterance.
2. Processing presupposition accommodation vs. satisfaction
Compared to other topics in experimental pragmatics, the processing dynamics of pre-
suppositions are still rather underexplored (see Schwarz, 2015 for a comprehensive
review) and this is even more so for the study of presuppositions when they are sup-
ported by the context (i.e. satisfaction) versus when they need to be processed within
a defective context which has to be repaired (i.e. accomodation).
When compared to satisfaction, presupposition accommodation has been shown to
elicit longer reading times (Schwarz, 2007; Tiemann et al., 2015) and this has been
taken as evidence that presupposition accomodation requires extra cognitive costs
than satisfaction. However, up to now, presupposition accomodation has been studied
mainly in relation to a specific trigger type - e.g. auch (too) in Schwarz (2007) and
wieder (again) in Tiemann et al. (2015). Thus, overall, what we know is that certain
37
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
trigger types are harder to process in a condition of accomodation than satisfaction.
We do not know yet what the genuine costs of presupposition accomodation per se
are, independently of trigger type and as the overall process of context repair. In other
words, we do not know yet whether the processing of presupposition accomodation is
costlier than satisfaction.
Furthermore, how such a process unfolds over time (i.e. its time-course) is still un -
clear. In Schwarz (2007) only the reading times of the whole sentence were collected
and, even though they were longer for accomodation than satisfaction, this does not
provide direct evidence on the exact time-course of processing accommodation.
Similarly, in Tiemann et al. (2015), though within a word-by-word paradigm,
participants were found not to have accommodated the presupposition of wieder since
the frequency of correct answers to the verification questions about the presupposition
in the neutral condition - requiring accomodation - was very low. As a result, the
greater reading times for wieder in the neutral condition (vs. satisfaction) indicate that
tracking down an antecedent in the preceding context is costlier indeed, but - again -
this does not provide evidence on the cost of repairing a defective context by
accommodation.
Overall, then, what the processing costs and time-course of genuine presupposition
accommodation are is still an underexplored issue.
3. The present study
The present study addressed two questions: (i) Is presupposition accommodation cog-
nitively costlier than presupposition satisfaction independently of the trigger in use or
is this difference related to specific trigger types? And (ii) When are presuppositions
accomodated, on-line during the sentence processing or off-line after the processing
of the assertive content of the utterance? That is, what is the time-course of presuppo-
sition accomodation? A secondary issue concerned potential variations in processing
due to different categories of triggers.
As for research question (i), we predicted that - within a word-by-word reading
times paradimg, the processing of accommodation (vs. satisfaction) should elicit
longer reading times independently of the trigger in use. This would be so because,
according to the traditional semantic framework, in presupposition accomodation both
a linking and an updating process are involved: the presupposed antecedent activated
by the trigger needs to be tracked down in the preceding context (i.e. linking) and,
then, the discourse mental model has to be updated with the presupposed information
(i.e. updating).
As for research question (ii), following Tiemann et al. (2011), two crucial regions
of interest can be identifyied in a presupposing sentence: the triggering point (e.g.
38
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
give up in (1) above), where the hearer is alerted to track down in the preceding con-
text an antecedent for the presupposition, and the computation point (e.g. smoking in
(1) above), where the content of the presupposition is actually processed. Therefore,
we predicted that eventual extra reading times for accomodation (vs. satisfaction)
might be observed (i) on the whole sentence only, (ii) at a single sentence region only
(e.g. the triggering point) or (iii) at both the triggering and the computation points,
with no increased reading times for the whole sentence. Each of these possibilities
would reveal a different scenario about the time-course of presupposition accomoda-
tion: (i) would show that a presupposition is accommodated off-line, after the compu-
tation of the assertive content; (ii) would indicate that presuppositions are processed
on-line; and (iii) would show that presupposition accomodation is both an on-line and
incremental process, which takes place in different phases while the sentence unfolds.
4. Experiment
Methods and material
Thirty-seven native speakers of Italian [Mean age= 24.08; SD = 4.94; 16 M; 21 F]
volunteered in the experiment after providing their informed consent. They were all
University students.
Fourty short stories in Italian were created, each composed of 2 context sentences
and a target sentence. The target sentence contained one of four presupposition trig-
gers: definite descriptions (DD, n: 10); change of state verbs (CSV, n: 10); iterative
expressions (IT, n: 10); and focal particles (FC, n: 10).
The stories were presented in two conditions created by manipulating the content of
context sentence 1 while keeping context sentence 2 and the target sentence un-
changed between conditions: in the satisfaction condition (SAT), the context sentence
1 made explicit the information of the presupposition activated by the trigger in the
target sentence. In the neutral condition (NEU), the information provided by context
sentence 1 did not satisfy the presupposition of the target sentence and needed accom-
modation.
Each story was followed by one target question, aimed at verifying if participants
had accommodated the presuppositions and two distractor questions – see Fig. 1 for
an item example1 and Fig. 2 for an example of target sentence with each trigger type.
1
English (literal) translation:
Context sentences 1: Before her pregnancy Gaia smoked ten cigarettes per day (SAT)/ Gaia is
at the third month of her first pregnancy (NEU).
Context sentence 2: The possible fetal diseases scare her a lot.
Target sentence: From the very beginning she has given up smoking but her
worries remained the same.
39
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
The experimental material was controlled for: (i) words number of the target
sentences (M: 15.15; SD: 0.66); (ii) type (nouns and verbs only) and position of the
words for the triggering and computation points (5 th and 7th position in the target
sentence, respectively); (iii) avoidance of conversational and conventional implica-
tures on or before the trigger; (iv) plausibility and predictability, normed with two on-
line rating studies (5-points-Likert scale) on the basis of which only items with high
plausibility (i.e. 3-to-5) and low predictability (i.e. 1-to-3) were selected.
Fig. 1. Example of an item with CSV in condition SAT and NEU.
Fig. 1. Example of target sentence for each trigger type – Italian and English (literal)
translation.
Procedure
40
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
Materials were administered according to a latin-square design such that partici -
pants who read a story in the accommodation condition never read the same story in
the satisfaction condition. The procedure consisted of a self-paced reading times para-
digm were participants were asked to read the stories and answer three verification
questions at the end of each story. Context sentences 1 and 2 were presented as a
whole on a computer screen, one after the other. The target sentences were presented
word-by-word.
The order of stimuli presentation was randomized across participants, as was the order
of the verification questions within each trial.
The independent variables in this experiment were Presupposition Condition (Satis-
faction vs. Neutral) and Trigger Type (DDs, CSVs, ITs and FCs). The word-by-word
reading times on the target sentences and accuracy on the verification task (i.e. correct
responses to target questions) were collected.
Three main regions of interest were identifyied for the word-by-word reading times:
(i) the triggering point T1; (ii) the word following T1 (T1+1); and (iii) the computa-
tional point T2 for CSVs, ITs and FCs. For example, with ‘ha smesso di fumare’ (Eng.
has given up smoking) the three critical regions were: T1 = smesso; T1+1 = di; T2 =
fumare.
Reading times at the region before the trigger (T1-1), the final word (FW) and the to-
tal reading times of the whole sentence were collected as well.
Statistical analyses were carried out by Linear-Mixed models statistics (LMM) and
post-hoc comparisons were adjusted with Tukey method to correct for multiple com-
parisons.
Results
The overall frequency of correct responses to target questions was 83.92% across
conditions (i.e. SAT and NEU). In condition NEU, our participants provided correct
responses 74.89% of the time, thus showing that the presuppositions triggered in the
target sentences have been mostly accommodated.
LMM statistics on the word-by-word reading times has revealed that an effect of
condition emerged on two critical regions of the sentence: the triggering point T1
(F(1, 74.13)= 4.72; p< 0.05) and its subsequent region T1+1 (F(1, 72.05)= 4.39; p<
0.05), with longer reading times for presupposition accomodation than satisfaction –
see Figure 3. In addition, a significant interaction of ConditionXTrigger Type
emerged at T1+1 (F(6, 71.98)= 2.97; p< 0.05) and the computation point T2 (F(4,
979.36)= 2.40; p< 0.05).
Finally, post-hoc comparisons revealed a significant difference for CSVs vs. DDs
(t= - 2.68; DF =73.43; p< 0.05), DDs vs. FCs (t= 3.93; DF= 73.82; p< 0.005) and
41
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
DDs vs. ITs (t= 3.06; DF=73.54; p< 0.05) at T1+1 and for CSVs vs. ITs (t= -2.42;
DF= 54.06; p< 0.05) and FCs vs. ITs (t= -2.57; DF = 49.06; p< 0.05) at T2.
Fig. 1. Mean reading times in conditions NEU vs SAT.
5. Discussion
5.1. Accomodation takes longer than Satisfaction: processing costs
An effect of the condition was observed at two critical regions for all the presupposi-
tion triggers at stake, with longer reading times in accommodation than satisfaction:
the triggering point T1 and its subsequent region T1+1. Taken together, this result
suggests that presupposition accommodation takes longer than satisfaction and, im-
portantly, this is so independently of the type of trigger in use, thus reflecting the
costs associated with the process of context repair.
42
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
Following the traditional semantic framework, according to which the processing of
presupposition accommodation involves a linking and an updating process, and since
our participants mostly accommodated the presuppositions, the longer reading times
in accommodation seem to reflect indeed the costs associated with one of the involved
processing steps. In other words, accommodation might be costlier than satisfaction
either because of the linking process or because of the updating process.
The methods used in the present study are not suitable enough to tease apart the ex-
act contribution of each of these processes. Nonetheless, the regions of interest where
the condition effect emerged provides some interesting cues to this purpose.
Accommodation was costlier at the triggering point T1, and at the subsequent region
T1+1. This suggests that the linking process rather than the updating process is likely
to increase the cognitive costs of processing presupposition accommodation. That is,
what makes accommodation costlier than satisfaction seems to be the process of
tracking down in the preceding defective context a proper antecedent for the presup-
position. A process, this latter, which is triggered by the trigger itself, where in fact
longer reading times were found. Such an interpretation seems compatible with Tie-
mann et al. (2015) findings, where the longer reading times on wieder in the neutral
condition, together with the low accuracy rates on the verification questions (i.e. no
updating process), provide evidence for extra processing costs related to the linking
process when the context is defective (i.e. accommodation vs. satisfaction).
5.2. Accomodation takes longer than Satisfaction: Time-course
Our reading times data revealed significant effects at different regions of the presup-
posing target sentence. First, accommodation elicited longer reading times than satis-
faction at T1 and T1+1 (i.e. effect of condition). Second, the condition effect was dif-
ferently distributed at T1+1 and T2 depending on trigger type (i.e. significant interac-
tion ConditionXTrigger Type): at T1+1, the longest reading times were elicited by
DDs; at T2, they were elicited by ITs (vs. CSVs and FCs).
Overall, these results support two main points about the time-course of presupposition
accommodation. First, since the reading times for all trigger types in the neutral con-
dition were longer at the triggering point and its subsequent region (T1 and T1+1),
presupposition accommodation seems to take place on-line during sentence process-
ing, immediately just as the trigger becomes available to the hearer.
Second, with certain types of triggers, accommodation is incrementally processed
while the sentence unfolds. In fact, the increased reading times on the computation
point T2 for ITs (vs. CSVs and FCs) suggest that, at least with this trigger type, pre-
supposition accommodation proceeds incrementally in two phases during the sentence
processing. First, the triggering point (e.g. di nuovo) alerts the reader to track down in
the discourse mental model an antecedent event (i.e. linking process). Second, the
43
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
content of the antecedent event becomes available on the computation point (e.g. chi-
avi, see Fig. 2), where the presupposition is finally accommodated and the context up-
dated (i.e. updating process).
Overall, these data provide compelling evidence that presupposition accommoda-
tion is not only processed on-line, but also its processing occurs incrementally while
the sentence unfolds. This, in turn, fits well with those traditional semantic accounts
according to which presuppositions are lexically encoded meanings that constitute a
condition for context update and for the comprehension of a presupposing utterance.
5.3. Presupposition Triggers
The main aim of the present study was to investigate the processing costs and time-
course of presupposition accommodation per se, hence independently of trigger type.
We looked at different categories of triggers to generalize as much as possible the pro-
cessing dynamics of the overall process of presupposition accommodation. Therefore,
the differences among trigger types were a secondary issue.
Nonetheless, our data revealed that presupposition accommodation elicits longer
reading times with DDs at the region T1+1 and with ITs at the computation point T2.
Overall, this suggests that different triggers differently affect the cognitive load of
presupposition processing. In particular, accommodating presuppositions triggered by
DDs and ITs is more cognitively demanding than other triggers at different phases of
the sentence processing. The why and how of such differences need to be properly ad-
dressed in future works with more targeted methods, but results emerged from our
data allow for some tentative speculations.
At T1+1, that is the region of the sentence immediately following the triggering
point, DDs elicited the longest reading times with respect to ITs, FCs and CSVs. In
our experiment, we used DDs consisting of a noun only (e.g. the designer) and mean-
ing that, in the target sentences where the presuppositions were triggered by DDs, the
triggering and the computation points coincided. The longest reading times emerged
with DDs might then be explained by the fact that, in such cases, the linking and the
updating processes associated with the presupposition accomodation of DDs take
place simultaneously during the sentence processing when the definite description be-
comes available to the hearer. As a consequence, it is possible that the simultaneous
occurrence of both the linking and the updating processes be responsible for the in -
creased reading times at the region following the triggering point (i.e. T1+1), due to
the demanding processing of the preceding information.
Longer reading times were found for ITs than CSVs and FCs at the computation
point T2 of the presupposing target sentence. In other words, this suggest that, con-
trary to CSVs and FCs, the processing of ITs required higher cognitive costs in this
44
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
region of the sentence. We hypothesize that such increased processing times might be
due to two main differences between ITs on the one hand and CSVs and FCs on the
other hand.
First, following Zeevat (1992), ITs (e.g. again) are cases of resolution triggers to-
gether with FCs (e.g. too) and unlike CSVs (e.g. to give up), which are instead consid-
ered as lexical triggers. Resolution triggers are said to require the anaphoric retrivial
of an entity or event from the common ground and this might make the linking
process for ITs more demanding than for lexical triggers, such as precisely CSVs,
whose conventional meaning directly encode a precondition for their asserted content.
As a result, then, the more demanding linking process associated to the retrivial of an
anaphoric entity presupposed by ITs might explain - at least partially - its longer read-
ing times than CSVs.
Finally, it is possible that ITs required longer reading times than FCs because even
though they both are resolution triggers, differently from FCs, the processing of ITs
requires the construction of a more complex mental representation where temporally
displaced events are included. In previous experimental research, the representation of
temporally displaced event has been associated to higher cognitive costs for ITs (Do-
maneschi et al., 2014; Tiemann et al., 2015).
To summarize, then, it is possible that DDs were harder to process because the trig -
gering and the computation points concided and this caused the linking and the updat-
ing processes to co-occur.
Finally, processing the presuppositions triggered by ITs might have resulted in longer
reading times for two reasons. On the one hand, linking the information carried by ITs
to the previous context is more demanding since it involves the retrivial of an
anaphoric entity/event. On the other hand, the processing of ITs requires the mental
representation of temporally displaced events, which is therefore more complex.
6. Conclusions
In the present experiment, the processing costs and time-course of presupposition ac-
comodation were studied, as compared to presupposition satisfaction and indepen-
dently of the presupposition trigger in use. Data collected revealed two main results.
First, presupposition accommodation requires greater processing costs than satisfac-
tion, reflecting a process of context repair where both a linking and an updating
process are needed. Second, presupposition accommodation is an on-line process, it
takes place immediately just as the trigger becomes available and proceeds incremen-
tally during the sentence processing.
To the best of our knowledge, these results represent a first evidence that presuppo-
sition accomodation, independently of the trigger in use, requires longer processing
45
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
times mirroring the underlying process of context repair. Future work is needed to de-
compose the associated cognitive costs and tease apart the exact contribution of the
linking and the updating processes on the one hand and of the type of trigger on the
other hand.
Funding: This study was funded by the Italian Ministry of Education, University and
Research within the three-year project SIR_2014 - EXPRESS – Experimenting on
Presuppositions directed by Filippo Domaneschi, project code RBSI147WM0.
References
1. Domaneschi, F., Carrea, E., Penco, C., & Greco, A. (2014). The cognitive
load of presupposition triggers: mandatory and optional repairs in
presupposition failure. Lang Cogn Neurosci, 29(1), 136-146.
2. Heim, I. R. (1982). The semantics of definite and indefinite noun phrases.
Amherst: University of Massachussetts (Doctral dissertation).
3. Heim, I. (1990). Presupposition projection. In R. van der Sandt (ed.),
Reader for the Nijmegen Workshop on Presupposition, Lexical Meaning,
and Discourse Processes. University of Nijmegen.
4. Heim, Irene & Angelika Kratzer. 1998. Semantics in generative grammar.
Oxford University Press.
5. Karttunen, L. (1974). Presupposition and linguistic context. Theor Lin-
guist, 1, 181-194.
6. Levinson, S. C. (1983). Pragmatics. Cambridge: Cambridge University
Press.
7. Lewis, D. (1979). Scorekeeping in a language game. J Philos Logic, 8,
339-359.
8. Schwarz, F. (2007). Processing presupposed content. J Semant, 24(4),
373-416.
9. Schwarz, F. (2015) (Ed.), Experimental Perspectives on Presupposition,
Studies in Theoretical Psycholinguistics, Dordrecht: Springer.
10. Stalnaker, R. (1974). Pragmatic Presuppositions. In M. Munitz, & P. Un-
der (Eds.), Semantics and Philosophy (pp. 197–213). New York: New
York University Press.
11. Stalnaker, R. (2002). Common Ground. Linguistics and Philosophy,
25:701-721.
12. Tiemann, S., Schmid, M., Bade, N., Rolke, B., Hertrich, I., Ackermann,
H., Knapp, J., & Beck, S. (2011). Psycholinguistic evidence for
46
CEUR Proceedings of the Workshop on Contexts in Philosophy - Paris, June, 20, 2017
presuppositions: on-line and off-line data. In I. Reich et al. (Eds.),
Proceedings of Sinn & Bedeutung, 15 (pp. 581-595). Saarbrücken:
Saarland University Press.
13. Tiemann, S., Kirsten, M., Beck, S., Hertrich, I., Rolke, B. (2015). Presup-
position Processing and Accommodation: An Experiment on wieder
(‘again’) and Consequences for Other Triggers, in Schwarz (2015), 39-65.
14. Zeevat, H. (1992). Presupposition and accommodation in update seman-
tics. Journal of Semantics, 9, 379-412.
47