ArticlePDF Available

Abstract and Figures

Previous research (e.g., Cutler, Perception & Psychophysics, 20, 55-60, 1976) has shown that detection of the initial phoneme of a word is speeded when the word is pronounced with a focal accent. This "accent advantage" is also observed when the accented word is replaced by a neutrally accented one. The present two experiments were designed to identify what aspect of the context preceding the target word is the source of this advantage. Both indicated that the advantage can be ascribed to the syllable immediately preceding the target word, rather than some possibly global but more distal attribute of the context. The first experiment used the recordings that had been used by Cutler Perception & Psychophysics, 20, 55-60, (1976) with the addition of a between-subjects manipulation of the local context. In one condition, the syllable immediately before the target word was the one that had been recorded in the sentence context (preceding an accented or an unaccented target word). In the other, cross-spliced, condition, the preceding syllable was exchanged between accented and unaccented contexts. The second (pre-registered) experiment used new recordings and a within-subject manipulation of the pre-target syllable. The studies confirmed and extended the observation that the pre-target syllable rather than some other prosodic aspect of the preceding context is the source of the faster phoneme detections.
Content may be subject to copyright.
The prosodic accent advantage in phoneme detection: Importance
of local context
Charles Clifton Jr
1
&Amanda Rysling
2
&Jason Bishop
3
Accepted: 26 August 2021
#The Psychonomic Society, Inc. 2021
Abstract
Previous research (e.g., Cutler, Perception & Psychophysics, 20,55-60,1976) has shown that detection of the initial phoneme of
a word is speeded when the word is pronounced with a focal accent. This accent advantageis also observed when the accented
word is replaced by a neutrally accented one. The present two experiments were designed to identify what aspect of the context
preceding the target word is the source of this advantage. Both indicated that the advantage can be ascribed to the syllable
immediately preceding the target word, rather than some possibly global but more distal attribute of the context. The first
experiment used the recordings that had been used by Cutler Perception & Psychophysics, 20,55-60,(1976) with the addition
of a between-subjects manipulation of the local context. In one condition, the syllable immediately before the target word was the
one that had been recorded in the sentence context (preceding an accented or an unaccented target word). In the other, cross-
spliced, condition, the preceding syllable was exchanged between accented and unaccented contexts. The second (pre-registered)
experiment used new recordings and a within-subject manipulation of the pre-target syllable. The studies confirmed and extended
the observation that the pre-target syllable rather than some other prosodic aspect of the preceding context is the source of the
faster phoneme detections.
Keywords Attention .Psycholinguistics .Speech perception
Introduction
Listenersprocessing of words is affected in various ways
when they are linguistically focused. Placing prosodic prom-
inence on a word increases the magnitude of the N400 ERP
response, especially to informationally old as opposed to new
words (Li et al., 2008). It also increases the N400 made to an
unexpected or anomalous word (Brunellière et al., 2018;Li&
Ren, 2012). Contrastively focusing a word in English, using a
nuclear L+H* (Beckman & Hirschberg, 1994) accent, speeds
lexical decisions to a semantically related word in a cross-
modal priming task (Husband & Ferreira, 2015;seealso
Bishop, 2017). Accenting a word increases accuracy of
identifying a word given a choice of two phonetically similar
words (e.g., mouse/mouth) for normal-hearing subjects listening
to sentences in moderate noise (van Donselaar & Lentz, 1994).
However, when the masking noise increased and the target
words accent was inappropriate (the target word was given
by being introduced in a preceding yes/no question), the effect
was reduced. Various ways of indicating focus (using it-clefts or
pseudoclefts, italicization of a written word, contrastive pitch
accenting, etc.) increase accuracy of memory for focused words
(Birch & Garnsey, 1995; Fraundorf et al., 2010;Kemberetal.,
2019; Sanford et al., 2006; Sturt et al., 2004).
Arguably, the clearest and most straightforward evidence
for the attentional effect of linguistic focus comes from one of
the earliest demonstrations. Cutler (1976) showed that the
time to detect a targeted word-initial phoneme is speeded
when the word was pronounced with a contrastive accent, as
for DIRT in (1a) below, as compared to when it had been
unaccented, as in (1b) (also Cutler & Foss, 1977).
(1) a. (accented) She managed to remove the DIRT
from the rug, but not the berry stains.
b. (unaccented) She managed to remove the dirt
from the RUG, but not the clothes.
*Charles Clifton, Jr
cec@umass.edu
1
Department of Psychological and Brain Sciences, University of
Massachusetts, 135 Hicks Way, Amherst, MA 01003, USA
2
Department of Linguistics, University of California Santa Cruz,
Santa Cruz, CA, USA
3
Program in Linguistics, The Graduate Center, City University of
New York, New York, NY, USA
Attention, Perception, & Psychophysics
https://doi.org/10.3758/s13414-021-02371-5
c. (neutral) She managed to remove the dirt from
the rug.
We refer to this pattern as the accent advantage. This result,
by itself, may not be surprising. The accented word is acous-
tically prominent, and its acoustic properties may be enough
to speed phoneme detection. There is evidence, though, that
the effect cannot be based solely on the acoustic properties of
the focused word itself. It appears to be sufficient to focus the
word via the syntactic, prosodic, or informational context, as
several of the demonstrations cited above suggest. A clear
example of the sufficiency of informational context appears
in Cutler and Fodor (1979) (and in Akker & Cutler, 2003).
These researchers presented subjects with a wh-question (e.g.,
Which man was wearing the hat?) followed by a sentence
with a word that began with a target phoneme. When this
target word addressed the question (e.g., corner in The man
on the corner...), it was responded to faster than when it did
not address the question, even though the target sentences
including the target word were identical.
Another instance of the apparent sufficiency of the context,
holding the target word accent constant, appears in Cutler
(1976). In her cross-spliced experimental condition, the
accented word (e.g., DIRT from (1)) was replaced by splicing
in a non-accented version that had been recorded in a prosod-
ically neutral context, as in (1c), that was lexically identical to
the accented and unaccented contexts up through the end of
the phrase that included the target item. Although the size of
the focusing advantage was reduced in the spliced condition
compared to the unspliced condition, it remained substantial,
at 39 ms. This advantage had to be attributed to some acoustic
property of the context preceding the target word, since the
lexical context was identical in all cases, as was the target
word itself. Further, the acoustic property appears to be pro-
sodic, since the advantage even appears when the preceding
material is presented as a delexicalized melody (Cutler &
McQueen, 2014). Cutler attributes the effect to some aspect
of the full prosodic contour of the sentence, writing "listeners
monitored the prosodic contour of sentences and directed at-
tention to where accent would fall...the first six words [pre-
ceding the target word] have a prosodic contour suggesting
that accent will fall on [the target word]" (Cutler,2012, p 243).
We acknowledge that whether the effective prosodic property
is global or limited to some more localized aspect of the pro-
sodic contour is an open question.
Listeners have often been shown to use prosody to guide
their processing of upcoming material. For example, the
rhythmic pattern of an utterance can strongly influence how
listeners parse upcoming material into lexical items (Dilley &
McAuley, 2008). Listeners can also quickly use contrastive
accents to anticipate upcoming referents in a visual world task
(Ito & Speer, 2008). Some American English listeners appar-
ently can use early, non-local intonational patterns to guide
selection of a referent in a mouse-tracking task (Roettger et al.,
2020). Mauchand et al. (2021) provide event-related potential
(ERP) evidence that listeners can use prosody to predict
whether upcoming material is going to be ironic or literal.
Fiveash et al. (2021) provide evidence that attention can be
rhythmically entrained: detection of an accented target sylla-
ble in sentences where accented syllables appear at regular
600-ms intervals is speeded for listeners who are exposed to
trains of tones at 600-ms intervals early in an experimental
session. Even readers seem to use (implicit) prosody to antic-
ipate the stress pattern of upcoming words (Breen & Clifton.,
2011).
The present research asks what property or properties of the
preceding context speeds phoneme detection in experiments
like Cutlers(1976) phoneme-monitoring experiment. It is
tempting to think that some property of the contextsrhythm
or tune alerts the listener to upcoming material that deserves
attention (Cutler, 2012). However, previous attempts to iden-
tify such a property have proved inconclusive. Cutler and
Darwin (1981), Experiment 2), for example, found that the
accent advantage remained even when the fundamental fre-
quency (F
0
) of the context that preceded the target word was
set to a constant value, indicating that intonational variation is
not a necessary basis for the effect. Ip and Cutler (2018)found
faster phoneme detection times in accented contexts recorded
by four different Australian speakers (three of whom showed a
statistically significant effect), even though the speakers dif-
fered substantially in the acoustics ofthe context preceding the
target word. One speaker varied in duration, mean, maximum,
and variability of F
0
, one varied in mean F
0
only
,
and two
varied in durational measures only. In a study demonstrating
the phenomenon in Chinese, Ip and Cutler (2017)reported
greater F
0
range in contexts preceding an accented versus an
unaccented target. Cutler (1987) manipulated F
0
, duration,
and intensity contours by imposing the contour of the
accented context on the unaccented context and vice versa.
When speech hybridization techniques were used to swap
the rhythm (durational pattern) between contexts, the accent
advantage was reduced by approximately 50% (not reaching
significance). The accent advantage fully disappeared when
F
0
and intensity were swapped but rhythm was maintained,
and reversed when all three were swapped. It therefore
appears that no one prosodic property of the preceding
context is necessary or sufficient to support the accent
advantage.
In addition to which particular cues in the preceding con-
text matter, it is also unclear where in the preceding context
they matter. Listening to digitized versions of the materials
used in the original Cutler (1976)study,
1
the present authors
were struck by apparent differences in the sound of the sylla-
ble immediately preceding the target word. This syllable was
1
We are deeply grateful to Anne Cutler for providing us with these files.
Atten Percept Psychophys
retained from the original context after the neutral target word
was spliced in, and seemed to be subjectively crisperwhen
it came from the accented than from the unaccented context.
Cutler and Darwin (1981), Experiment 1), possibly having
made a similar observation, noted that the time between the
offset of the syllable preceding the target item and the start of
the release burst that began the target item was longer for the
accented context than for the unaccented context (mean = 31
ms). They speculated that this extra closure duration might
have provided a cue to the identity of the consonant or simply
provided the listener with time to prepare to recognize the next
word. However, they disconfirmed this possibility in a new
experiment in which they spliced the onset of the target word
immediately at the onset ofthe stop closure, setting the closure
duration to zero for both accented and unaccented conditions.
The accent advantage remained.
It is possible that the difference in closure duration was not
the only difference between the syllables immediately preced-
ing the target item (we refer to such syllables as pre-target
syllables). Rysling et al. (2020) reported an experiment very
similar to Cutler (1976) but adding a between-subject splicing
manipulation. One group of subjects heard sentences that were
spliced as Cutler had done, with a neutral target word spliced
in at the end of the closure after the last word of the preceding
context (which had been recorded preceding an accented vs.
an unaccented word). The other group of subjects heard
sentences where the neutral material that was spliced in in-
cluded the syllable that had preceded the target word in the
neutral recording. The only difference between these condi-
tions was that the pre-target syllable in the first group had been
recorded preceding an accented versus unaccented target
word, while the pre-target syllable in the second group had
been recorded preceding a neutral instance of the target word.
The remainder of the preceding material (which we refer to as
the preamble) was the same accented versus unaccented
context for both groups. The first group showed reliably faster
phoneme detection times following the accented than the un-
accented context (although the effect was smaller than in most
previous reports, for a difference of 21 ms in the means and
25 ms in medians). The second group (with the neutral pre-
target syllable as well as target) showed a reversed effect:
phoneme detection times were 17 ms and 14 ms slower
(means and medians) following the accented than the unac-
cented context (but a Bayesian mixed effects analysis indicat-
ed that the 95% credible interval (CI) did include a zero-ms
effect for this condition, meaning that this difference cannot
be securely trusted). While the disappearance of the accent
advantage in the latter condition does indicate a critical con-
tribution of the pre-target syllable, the possible reversal is
puzzling. The authors speculated that listeners might have
begun forming an expectation about the presence of an up-
coming contrastive accent(Rysling et al., 2020,p.EL288)
that was disrupted by the occurrence of an unaccented pre-
target syllable. However, the solid conclusion from Rysling
et al. is that the accent advantage disappears when the pre-
target syllable is replaced by a neutral recording of the same
syllable. The accent advantage observed with the original pre-
target syllable may involve any of several mechanisms. One
possibility is that acoustic properties of this syllable could
trigger increased attention. Another is that there is more antic-
ipatory co-articulation of the initial segment of the accented
target word (a possibility suggested by Salverda et al., 2014).
While the present research was not designed to explore alter-
native mechanisms for the effect observed by Rysling et al.,
we return to a brief discussion of the possibilities in the
General discussion.
The two experiments reported here were designed to fur-
ther explore the possible reversal of the accent advantage in
the neutral pre-target syllable condition in Rysling et al.
(2020) as well as to address some concerns about that exper-
iment. One concern is simply that the recorded sentences used
by Rysling et al. may not have fully realized whatever pro-
sodic properties the original Cutler (1976) materials had, a
possibility made salient by the reduced size of the accent ad-
vantage in Rysling et al. (2020)comparedtoCutler(1976). To
address this concern, the present Experiment 1used the same
recordings that were used by Cutler (1976), with two splicing
manipulations similar to those used by Rysling et al. (2020).
The present Experiment 2used materials based on those used
by Rysling et al. (2020), in a pre-registered within-subjects
design. In this design, each participant was tested both on
utterances in which the pre-target syllable was consistent with
the remainder of the pre-target context and utterances in which
it was inconsistent. Our intention was to avoid possible effects
that may result from listeners strategically ignoring prosody
when all the sentences they heard had this inconsistency (as in
Rysling et al., 2020, and the present Experiment 1).
Both experiments were designed as straightforward tests of
whether the accent advantage reflects (a) some aspect of the
prosodic contour of the material preceding the target word, up
to but not including the pre-target syllable, or (b) some aspect
of the pre-target syllable itself. The experiments crossed
accented versus unaccented prior contexts with pre-target syl-
lables that had been recorded in accented versus unaccented
utterances rather than replacing the original pre-target syllable
with a neutral one (whose prosody could have varied among
items since in some but not all neutral sentences, nuclear ac-
cent likely fell on the target item).
Experiment 1
The goal of the first experiment was to replicate Cutler (1976)
using her recordings, and to add new conditions that were
identical except that the pre-target syllables were exchanged
between the accented and unaccented preceding context
Atten Percept Psychophys
conditions. We expected to find results similar to Cutler
(1976) in the original conditions. If the accent advantage
was due to the overall prosody of the preceding context, it
should remain when the originally accented pre-target syllable
is replaced by the originally unaccented one. However, if the
accent advantage is instead simply due to some property of the
pre-target syllable, it should be replaced by an accent disad-
vantage when that syllable is replaced by one taken from the
unaccented condition.
Method
Participants One hundred and six University of Massachusetts
undergraduates, all at least 18 years old, were tested in one
individual 20-min session each. Each participant completed
an Institutional Review Board-approved consent form and
received course credit. Three were eliminated for reporting
that their first language was not English and three others were
eliminated because of technical problems. Another three were
eliminated for having excessively long mean detection times
(two or more SDs above the grand mean phoneme detection
time). An additional participant (the last one tested in their
experimental condition) was also eliminated to obtain the
targeted 96 final participants with complete counterbalancing.
Materials Dr. Anne Cutler kindly provided digitized (22 kHz,
16 bit, monaural) copies of the sentences used in the original
Cutler (1976) study. These were spoken by a male with a
moderate Texas accent. Each experimental sentence contained
a target word that began with the segment that participants
were cued to listen for on a given trial (always a stop conso-
nant, either /b/ /d/ or /k/ for the experimental items). Each
experimental sentence had been recorded in two versions,
one with a pitch accent on the target word, and one with the
target word unaccented but with a pitch accent on a different
word (sometimes preceding the target word). Additional pitch
accents sometimes appeared in the sentence after the target
word. An example is shown in (2) below, where the target
word appears in italics and the accented word appears in
UPPERCASE. The target segment for both versions of this
example was a /b/. Both members of a sentence pair were
lexically (but not phonetically) identical up to the target word,
but differed after it. The /marks indicate segmentation points
used in cross-splicing, described later. Transcriptions of the 20
pairs of experimental sentences appear in Appendix 1.
(2) a. (accented) The head cashier said /the /BAG/
contained money, and so did the strongbox.
b. (unaccented) The head CASHIER said /the /bag/
contained money, although his colleagues disagreed
with him.
In the materials supplied by Cutler, the accented or unac-
cented target word had always been replaced by a word that
had been recorded in a neutral context (e.g., The head cashier
said the bag contained money for the example in (2)), begin-
ning at the start of the burst that indicated the release of the
initial segment of the target word.
These 40 sentences were divided, using the program
Audacity (http://audacityteam.org/), into four regions,
indicated by the /marks in (2). The first region was the
preceding context, i.e., all the words in the sentence up to
but not including the syllable before the target word. We
henceforth refer to this region as the preamble.The
second region was the localregion, i.e., the syllable that
preceded the target word.
2
The third region was the target
word itself, beginning at the zero crossing immediately
before the initial burst of the target words initial stop
consonant (in Cutler, 1976, the beginning of the target word
was taken to be as close as possible to the release of the stop
consonant burst) (quoted from Cutler & Darwin, 1981,p.
218). The final region was the conclusion.This was defined
as all the recording from the end of the target word to the end
of the sentence. Audacity was then used to create separate files
for each region and name them by their region and their source
(e.g., P01A.wav for the preamble of accented file 1, D02U.
wav for the local region –“determiner”–of the unaccented
version of file 2, etc.); the target segment was always labeled
with the same Aor Ulabel as the conclusion of its record-
ing, but is referred to as N,since it was always originally
taken from a neutral recording). Audacity was then used to
recombine them. For the originalcondition, the four A seg-
ments of each file were spliced together, resulting in the
AANA condition; similarly, the four U segments were joined,
resulting in the UUNU condition. For the cross-splicedcon-
dition, the A and U local regions were swapped, resulting in
AUNA and UANU conditions.
In addition to these 40 experimental sentences, there were
40 filler sentences, 20 of which contained a word that began
with the segment a participant was cued to listen for on a trial.
Further, there were three practice sentences taken from the
original Cutler (1976) recordings, together with 12 additional
feedbackpractice sentences, spoken by a female, to be used
to provide participants with speed and accuracy feedback, as
described below (this practice was not used by Rysling et al.,
2020, or by Cutler, 1976). Two of the three original practice
sentences and eight of the feedback practice sentences
contained a word that began with a designated target
segment. Finally, the segment cue recordings used in Cutler
(1976) were used. These were recorded by the same male
speaker as the experimental and filler sentences, and took
the form buh as in boy, kuh as in cat, duh as in did, and
for filler items only guh as in get.
2
In one case, namely item 9, it was impossible to identify the beginning of this
syllable so the localregion was taken to be the disyllable nother.
Atten Percept Psychophys
Acoustic analysis Praat (Boersma & Weenink, 2019)scripts
were created to conduct acoustic analyses of the AANA and
the UUNU files. The scripts were provided with measures of
the start times of each of the local and target regions. They
returned measures of the duration of the preamble and the
local region, together with F
0
and intensity information for
each region. The scripts also calculated the minimum, maxi-
mum, and standard deviation of F
0
and intensity of the pre-
amble and of the local region (eliminating the F
0
measures for
two items, items 4 and 6, where F
0
could not be measured
because of lack of voicing). All measurements are given in
Table 1, with the outcomes of paired-sample t-tests.
It is clear thatthe A local regions are longer than the U local
regions, by an average of 27 ms, a difference very similar to
that reported by Cutler and Darwin (1981). Further, the local
regions seem to be more variable in intensity, and possibly in
F
0
/pitch. In addition, the U preambles are longer, more vari-
able in F
0
(and so likely pitch), and perhaps in intensity, than
the A preambles, a difference that can be attributed to the
presence of an accented word in some of the U but none of
the A preambles.
Procedure The program PsychoPy (Pierce et al., 2019)was
used to present the acoustic materials to participants and to
record responses. Participants were seated in a cubicle and
listened to materials over headphones, adjusted to a comfort-
able level. After the experimenter gave a very short introduc-
tion to the experiment, participants read instructions that on
each trial they would see and hear a target segment (B / buh as
in boy, etc.), and then hear a sentence that might or might not
contain a word that began with that segment. If they detected
such a word, they were to press the space bar on a computer
keyboard as fast as possible. They were told that they should
try to respond in less than half a second (500 ms), and that they
were to ignore any occurrences of the target segment that did
not appear at the start of the word. They were told to ignore
spelling.
The experiment began with 12 practice trials, recorded in a
female voice. On these trials, the participant received feedback
on the computer screen. They were shown their detection time
in milliseconds if they correctly detected a target segment, and
messages saying oops you missed the target or oops there was
no target if they made an error. The experimenter remained in
the room during these practice trials and encouraged the
participant to respond quickly. After these practice trials, the
experimenter left the room. The participant heard three more
practice trials, taken from Cutler (1976) and recorded by the
same male speaker who recorded the experimental sentences.
They then heard 20 experimental and 40 filler sentences in an
individually randomized order, for a total of 60 trials.
Participants were assigned to two different conditions,
Originaland Cross Spliced.Forty-eight were assigned
to each condition, after elimination of rejected participants
data. The Original participants heard 20 AANA and UUNU
experimental sentences, while the Cross Spliced participants
heard 20 AUNA and UANU sentences. Within each group,
there were two 24-participant counterbalancing conditions, in
each of which a counterbalanced half of the sentence appeared
in the accented (AANA, AUNA) condition and the other half
in the unaccented (UUNU, UANU) condition. The experi-
mental script measured detection time from the start of the
sentence. The duration of the preamble plus the local segment
was subtracted offline, to yield detection times from the start
of the target phoneme.
Results
The data were analyzed using R 4.0 (R development Core
Team, 2019). Trials with misses were eliminated (44 of
1,920; 2.3%), as were trials with detection times < 100 ms
Table 1 Acoustic measures of experimental materials, Experiment 1
Region Condition, measure Dur Mean Hz Min Hz Max Hz SD Hz Mean dB Min dB Max dB SD dB
Preamble A1.538 135 102 179 18.8 76 57 83 5.5
U1.582 135 98 189 22.6 76 56 83 5.8
df 19 19 19 19 19 19 19 19 19
T2.84 0.26 2.57 1.73 2.27 0.34 0.58 1.49 1.66
P0.01** 0.80 0.02* 0.10 0.03* 0.74 0.57 0.15 0.11
Local A0.171 110 102 118 5.5 70 60 77 6.3
U0.144 108 103 114 4.2 70 62 75 5.1
df 19 17 17 17 17 19 19 19 19
T4.47 1.05 0.51 2.01 1.76 0.18 2.49 1.95 4.27
P0.001*** 0.31 0.61 0.06 0.10 0.86 0.02* 0.07 0.001***
SD standard deviation, df degrees of freedom
*=p<.05,**=p< .01, *** = p< .001
Atten Percept Psychophys
or > 1,500 ms (76 of the remaining 1,876 trials; 4.1%). Mean
response times (RTs) and their standard errors (SEs) are given
in Table 2. Box-and-whisker plots of the by-subject mean RTs
as a function of local context, separated by original versus
cross-spliced recording conditions, are shown in Fig. 1.
The detection times in ms were analyzed using brms
(Bürkner, 2017), which supports a Bayesian analysis modeled
on the familiar frequentist LME4 (Bates et al., 2015).
3
Alog-
normal analysis of the detection times in ms was conducted.
The fixed (population) effects were accented versus unaccent-
ed preamble and accented versus unaccented local context.
Each effect was sum-coded (ANOVA-style), using the
contr.sum() function of R (resulting in contrast weights of -1
and +1). The interaction of these effects can be viewed as the
overall effect of original versus cross splicing. Participant and
item intercepts and non-interacting slopes of preamble and
determiner were used as the random (group-level) terms.
4
Reasonable but minimally informative priors were used for
the intercept (lognormal(6,5)) and the fixed effects (nor-
mal(0,20)); otherwise, default priors were used.
The estimated effects and their 95% CIs are given in
Table 3. The CI for local context does not overlap zero (the
Bayesian equivalent to saying that the local context had a p <
0.05 significant effect), but the CI for preamble accent and for
the interaction included zero (i.e., cannot be trusted). The sep-
arate effects of local context at each level of preamble were
tested using the R function emmeans(). When the preamble
was accented, the accented local context speeded detection
times; the 95% highest probability density interval (HPD)
was from -0.108 to -0.002. With the unaccented preamble,
the accented local context resulted in a nearly trustworthy
speed-up; the HPD was from -0.108 to 0.003.
An exploratory analysis (conducted because a similar
analysis of Experiment 2yielded potentially interesting
results) examined the effect of trial order. Each experimental
trial was assigned a sequence number of 120, indicating the
order in which it occurred for a given participant. These values
were centered and added to the Bayesian analysis reported
above. The previously reported effects did not change sub-
stantially. There was evidence of slowing across trials (the
95% CI of the effect of centered trial on log RT ranged from
0.0004 to 0.0040, barely not overlapping zero; a similar effect
was also seen in the 20 filler items with target phonemes).
However, all three interaction effects involving centered trial
clearly overlapped zero, meaning that they cannot be trusted.
Discussion
The results are remarkably clear: Phoneme detection times were
faster when the syllable immediately preceding the neutral target
item (i.e., the local context) had been recorded together with an
accented version of the target item. Whether or not the remainder
of the material that preceded the target item had been recorded
before an accented or an unaccented target item did not affect
times. The apparent conclusion is that, while the accent advan-
tage may reflect the prosody of the preceding material, it does not
reflect an extended preparation for a focused item based on non-
local aspects of this prosody. Rather, any prosodic effect is local,
perhaps reflecting the greater duration and variability in intensity
(and thus perhaps perceived pitch) of the accented pre-target
syllable. However, we note that this may contrast with the results
reported in Rysling et al. (2020), who found a marginal tendency
towards slower times following an accented than following an
unaccented preamble when the local context was replaced by a
neutral syllable. The 95% CI of their effect included zero ms, so it
cannot be fully trusted; in contrast, there was a trustworthy ad-
vantage for the accented condition when the local context had not
been replaced. We return to this in the General discussion.
Before exploring the effect more thoroughly, we report a
second experiment, using different recordings of the Cutler
(1976) items and a within-subject design. It is possible to
suspect that subjects in the cross-spliced condition of
Experiment 1somehow became sensitized to any clash be-
tween the prosody of the preamble and thatof the local context
and came to disregard the former. However, if subjects hear
both original and cross-spliced materials, and still show the
accent advantage in the former (and a reversed effect in the
latter), this possibility would be ruled out.
Experiment 2
Experiment 2was conceptually the same as Experiment 1,
except for the within-subject manipulation of splicing and
the use of newly recorded materials. The experiment was
Table 2 Mean (and SE) of detection times, in milliseconds, for
Experiment 1
Preamble Accented local context Unaccented local context
Accented 566 (7.5) 598 (8.0)
Unaccented 568 (7.8) 596 (7.7)
Note: Cells onthemajor diagonal came from the original recordings; cells
on the minor diagonal came from the cross-spliced recordings
3
Parallelanalyses were conducted usingLME4. Their resultsclosely mirrored
the Bayesian analyses reported here, but maximal analyses did not converge,
and it is possible that the less-than-maximal analyses that did converge were
anticonservative (Barr et al., 2013).
4
The brms formula that was used was phmonOK.brm1x <- brm(RTms ~
preamble * determiner+ (1+determiner+preamble|participant) + (1+determin-
er+preamble|item), prior <- myprior, iter = 4000,chains = 4, family =
lognormal,data = phmonOK)
0
The pre-registration was done before the advantage of Bayesian analyses
was realized. It instead specified frequentist analyses. These analyses are given
in Appendix 3. However, the text reports their Bayesian equivalents.
Atten Percept Psychophys
pre-registered on the Open Science Framework (https://osf.io/
u7qpk).
5
Methods
Participants A power analysis indicated that 160 participants
would be needed to reach 75% power to detect a difference of
30 ms between two (accented and unaccented) conditions.
6
164 participants who reported English to be their native lan-
guage were tested. All were University of Massachusetts un-
dergraduates, aged 18 years or more, who participated for
course credit. As indicated below, three were rejected for
missing too many trials and one more was rejected to preserve
counterbalancing (40 subjects in each of four conditions).
Materials A female speaker recorded modified versions of the
materials used in Cutler (1976).
7
Each of 20 sentences had a
version in which the target word (the word that began with the
segment that was cued on its trial) was accented, and a version
in which some other word was accented. All experimental
sentences (with accented and unaccented targets) are given
in Appendix 2. One major change from the Cutler (1976)
materials was that none of the sentences in which the target
word was unaccented contained an accented word before the
target word (five of the 20 original sentences contained such a
word). The example in (3) below is representative of the items.
The recordings were modified by having a neutral version of
the target word (beginning at the zero-crossing immediately
preceding the initial release burst) spliced in to replace the
same word in both the accented and the unaccented sentences.
(3) a. (accented) I understand that the cash register at
/the /BACK /contains money, and the one at the front
doesn't.
b. (unaccented) I understand that the cash register at
/the /back/ contains money, and so does the
STRONGBOX.
The recordings were segmented, using Audacity, into a
preamble,”“local context,
8
the targetword, and the con-
clusion(see (3) for an example). These separate segments
were then labeled as Aor U,depending on whether they
came from an accented or an unaccented recording (the target
was always a neutral version, but was labeled according to the
sentence from which it was segmented; it is referred to here as
N). These segments were then re-combined, using
Audacity, resulting (as in Experiment 1) in AANA and
UUNU originalsentences, and AUNA and UANU cross-
splicedsentences.
6
The R package simr(Green & McLeod, 2016) was used to estimate power,
based on the output of the lmer analyses conducted on Experiment 1, with data
reduced to five observations/condition/subject. It estimated power for a
frequentist linear mixed model, rather than the Bayesian model reported in
the text.
7
These recordings were the same as those used in Rysling et al. (2020).
However, they were resegmented, following slightly different rules for a few
sentences. For example, in sentence 4 (see Appendix 2), the local syllable
contained the second syllable of overas well as the following adue to the
difficulty in identifying the start of the latter.
8
As in Experiment 1, there were a very few deviations from thissegmentation.
In item 4, the syllablecontained the second syllable of over as well as the
following a, and in item 12, it contained the vocalic part of hire as well as the
following a.
Fig. 1 Box-and-whisker plots of by-subject mean response times (RTs; ms) in Experiment 1
Atten Percept Psychophys
As in Experiment 1, there were 40 filler sentences (20 with
a target word that began with a cued segment), all recorded by
the same female speaker who recorded the experimental
sentences. In addition, both sets of practice sentences used in
Experiment 1were used in Experiment 2, and the spoken
versions of the target segment cue were also the same as used
in Experiment 1.
Acoustic analysis As in Experiment 1, Praat (Boersma &
Weenink, 2019) scripts were created to conduct acoustic anal-
yses of the AANA and the UUNU files. The resulting mea-
surements (after eliminating items for which F
0
could not be
determined) are shown in Table 4, with values of paired t-
tests.
As in Experiment 1, the accented local regions were longer
than the unaccented local regions, by an average of 21 ms. In
contrast to Experiment 1, differences between accented and
unaccented local regions were apparent in F
0
variability rather
than intensity variability. There was some indication that the
accented local regions may have been lower in mean or min-
imum intensity. The differences between the preambles that
appeared in Experiment 1were not present in Experiment 2,
presumably because the former but not the latter sometimes
contained pitch accents in the unaccented condition.
Procedures The procedures were the same as in Experiment 1,
with one major change: the original versus cross-splicing ma-
nipulation was a within-participants manipulation in
Experiment 2. Four counterbalancing conditions were used.
Each condition had five of the 20 experimental items in each
of the four conditions of the experiment: original versus cross-
spliced, crossed with the accented versus unaccented local
context. Thus, each participant received five experimental
sentences in each of the conditions AANA and UUNU (the
original conditions) and AUNA and UANU (the cross-spliced
conditions), and across participants each sentence was tested
equally often in each of these conditions.
Data analysis ThedatawereanalyzedinR4.0(R
Development Core Group, 2019). (Appendix 3contains a
frequentist analysis done in R 3.9, using LME4 (Bates et al.,
2015), as specified in the OSF pre-registration.) Trials in
which the target was missed were discarded (90 of 3,280;
2.7%), as were trials with detection times < 100 ms or >
1,500 ms (146 of 3,170; 4.6%). Three subjects who had more
than 25% (5) of their 20 trials discarded were eliminated from
the experiment, as was one additional subject, so as to main-
tain counterbalancing (40 subjects in each of 40 conditions).
Results
The mean detection times (and their standard errors) are
shown in Table 5, and a box-and-whisker plot of by-subjects
means (separated by original vs. cross-spliced recordings) in
shown in Fig. 2. As in Experiment 1, the data were analyzed
using brms (Bürkner, 2017). A lognormal analysis of the de-
tection times in milliseconds was conducted. The fixed
(population) effects were accented versus unaccented pream-
ble and accented versus unaccented local context. Contrasts
were sum-coded (ANOVA-style) using the R function
contr.sum(),whichresultedin-1versus1contrasts.
Participant and item intercepts and slopes were used as the
random (group-level) terms.
9
Reasonable but minimally infor-
mative priors were used for the intercept (lognormal(6,5)) and
the fixed effects (normal(0,20)); otherwise, default priors were
used. The estimated effects and their 95% CIs are shown in
Table 6. The CI for local context does not overlap zero, indi-
cating that its effect can be trusted, but the effect of preamble
accent does overlap zero (i.e., cannot be trusted). The 95% CI
for the interaction of these effects did not span zero, indicating
that it is trustworthy. Follow-up tests were performed by using
the R function emmeans. While the effect of local context can
be trusted when the preamble is unaccented (its 95% HPD
extended from -0.10 to -0.03), it is questionable when the
preamble is accented. Here, the 95% HPD extends from -
0.056 to 0.017. Examination of Table 5indicates that the
interaction can be described as a 29-ms effect of accented local
context in the original materials (where the preamble matched
the local context) but an 18-ms effect in the cross-spliced
materials. Alternatively, the interaction can be described as
longer overall times when the preamble and local context
matched (576 ms) versus when they mismatched (562 ms).
Finally,itcouldalsobeviewedassimplyreflectinga
9
The brms formula used was phmonOK.brm1x <- brm(RTms ~ preamble *
determiner + 1+determiner+preamble|participant) + (1+determiner+
preamble|item), prior <- myprior, iter = 4000, chains = 4, family =
lognormal,data = phmonOK)
Table 3 Estimated effect and lower and upper bounds of 95% credible interval (CI) for Experiment 1
Effect Estimate Error Lower bound Upper bound
Preamble (accented vs. unaccented 0.0003 0.008 -0.0161 0.0173
Local Context (accented vs. unaccented -0.0275 0.0066 -0.0408 -0.0147
Interaction Preamble × Local Context -0.0007 0.0121 -0.0246 0.0231
Atten Percept Psychophys
deviantly long delay when both preamble and local context
were unaccented.
An exploratory analysis of changes over the course of the
experiment was conducted by adding experimental trials 120
as a centered fixed effect. A Bayesian analysis like the one
reported above, but with centered trial number added, resulted
in a similarly trustworthy effect of local context. While the
overall effect of centered trial could not be trusted (β=-
0.001, error = .001, 95% CI -0.003 to 0.0004), there was a
trustworthy interaction of local context and trial number (β=-
0.002, error = .001, 95% CI -0.004 to -0.0003). The interac-
tion of local context and trial number can be seen in Fig. 3.
Detection time to accented local context items declined over
trials, but detection time to unaccented ones did not. In fact,
the advantage of accented local items largely appeared only
after a participant had experienced several accented local
items. This suggests that participants may have been learning
something about the value of a local item that had been re-
corded preceding an accented target. Examination of the 20
filler items indicated that they did not change significantly
over trials: a separate frequentist mixed model analysis (using
lmer) of the centered trial effect on filler trials alone yielded a t
<1.0).
General discussion
The present data confirm the conclusion made in Rysling et al.
(2020) about the importance of the preceding syllable in the
phoneme-monitoring accent effect. In that research, detection
of the initial phoneme of a neutrally accented target item was
speeded when the preceding syllable (the local context) had
been recorded together with the accented target item. The
novel results in the present data are (a) the advantage of having
an accented local context generally remains when it is com-
pared with a local context recorded before an unaccented tar-
get item (rather than one recorded in a neutral context, as in
Rysling et al.); (b) the effect is found using the original mate-
rials from Cutler (1976) as well as in newly recorded mate-
rials; and (c) the effect is found when the splicing manipula-
tion involving the local context and the rest of the preceding
context is tested both between (Experiment 1) and within
(Experiment 2) participants.
The theoretical conclusion of primary interest is a negative
one: one must question the tempting conclusion from earlier
studies of prosodic context on phoneme detection (e.g., Akker
& Cutler, 2003; Cutler, 1976; Cutler & Darwin, 1981)that
some non-local aspect of the prosodic context preceding a
focused item prepares a listener for that item, increasing atten-
tion and enhancing phoneme detection. However, it is also
important not to over-extend this conclusion. There is no rea-
son to doubt that semantically focusing a target item by ma-
nipulating the content of the preceding context speeds detect-
ing the initial phoneme of the target item (Akker & Cutler,
2003; Cutler & Fodor, 1979). The fact that Akker and Cutler
(2003) found that the content and the prosody of the preceding
context interacted in an under-additive fashion indicates that a
single mechanism, presumably involving anticipatory alloca-
tion of attention, could be operating in both. Further, it is clear
that a strong prosodic manipulation like the alternating
stressed syllables used by Dilley and McAuley (2008) sets
up an expectation for a words stress pattern (see also the
additional demonstrations of the effects of prosodic context
Table 4 Acoustic measures of experimental materials in Experiment 2
Region Condition, measure Dur Mean Hz Min Hz Max Hz SD Hz Mean dB Min dB Max dB SD dB
Preamble A1.528 218 160 293 34.2 75.6 49.8 87.1 8.71
U1.547 215 162 288 33.7 74.8 50.3 86.7 8.56
df 19 19 19 19 19 19 19 19 19
t3.28 0.69 0.38 1.22 0.49 2.15 0.61 1.47 0.56
p0.01 0.50 0.71 0.24 0.63 0.05 0.57 0.15 0.60
Local A0.151 165 157 179 6.2 71.3 64 79 4.98
U0.130 175 158 189 10.7 72.1 65 79 4.68
df 19 16 16 16 15 19 19 19 19
t3.81 1.97 0.22 1.71 2.77 1.71 2.24 0.19 1.40
p0.002** 0.07 0.83 0.11 0.02* 0.11 0.04* 0.85 0.18
SD standard deviation, df degrees of freedom
*=p<.05,**=p< .01, *** = p< .001
Table 5 Mean (and SE) of detection times, in milliseconds, for
Experiment 2
Condition Accented Local
Context
Unaccented
Local Context
Accented Preamble 560 (6.3) (AANA) 571 (7.5) (AUNA)
Unaccented Preamble 553 (6.4) (UANU) 589 (6.8) (UUNU)
Note: Cells onthemajor diagonal came from the original recordings; cells
on the minor diagonal came from the cross-spliced recordings
Atten Percept Psychophys
described in the Introduction section). Our findings should not
preclude attempting to find other ways in which the prosody
of a context can influence language comprehension. But we
note that those studies in which longer-distance prosodic pre-
diction effects have most clearly been found were studies in
which expectations were generated by presenting listeners
with repeating structure. This is importantly different from
cases like the present studies, in which intonation is likely
used as a cue to information structure. It might not be worth
the cost of generating longer-distance predictions about up-
coming information structure on the basis of intonational cues
in real-time listening, because producers can choose to change
the syntactic and prosodic (and, so, information) structures of
their utterances while in the middle of producing them. Thus,
generating a prediction about the upcoming information struc-
ture of a sentence well before hearing it may only be useful in
cases where the cue to the location of the focus of an upcom-
ing sentenceis non-intonational, i.e., a preceding wh-question.
Otherwise, allocating attention forward to focused material
from relatively farther-away intonational cues may result in
more disconfirmed predictions than it would be worth
generating.
The data reported here differ somewhat from those reported
by Rysling et al. (2020). Their Experiment 1showed a clear
phoneme detection advantage when the full preceding context
(preamble plus local context) came from a recording in which
the target item was accented. In contrast, in their Experiment
2, when the local context had been replaced by one taken from
a neutral context as opposed to one taken from an unaccented
context, the effect of preamble accent became untrustworthy:
its 95% CI overlapped zero. In fact, detection times with a
neutral local context were numerically (17 ms) slower when
the preamble came from an accented utterance than when it
came from an unaccented one. This difference in direction of
the preamble accent effect resulted in a trustworthy interaction
between accent and experiment. Rysling et al. suggested that
this pattern of results indicated that an accented local context
was required to find the phoneme-monitoring effect, but that it
was still possible that the preceding context set up an expec-
tation for an accented phrase. This might result in disruption
Table 6 Estimated effects and lower and upper bounds of 95% credible interval for Experiment 2
Effect Estimate Error Lower bound Upper bound
Preamble (accented vs. unaccented -0.001 0.011 -0.023 0.021
Local Context (accented vs. unaccented -0.021 0.008 -0.037 -0.005
Interaction Preamble × Local Context 0.011 0.005 0.002 0.021
Fig. 2 Box-and-whisker plots of by-subject mean phoneme detection times for Experiment 2
Atten Percept Psychophys
and slowed phoneme detection times when a neutral local
context is encountered. It is possible that the apparent reduc-
tion (but not reversal) of the size of the local context effect in
the accented preamble condition of Experiment 2of the cur-
rent experiment reflects a similar disruption. However, given
the questionable trustworthiness of some of the effects, the
likely lower power of the Rysling et al. (2020) experiment,
and the differences in experimental procedure between the
experiments, any possible effects of prosodic conflict between
the preamble and the local context must be left for further
experimental study.
Leaving such effects aside, the present experiments do pro-
vide reasonably consistent evidence that local context can have
an effect on phoneme detection time, while the more global
context of the preamble at best modulates this effect. The re-
maining questions are: what are the relevant acoustic properties
of the local context, and what is the mechanism by which it has
its effect? We cannot provide secure answers to either of these
questions, but offer some suggestions.
The one consistent acoustic difference between accented
and unaccented local contexts is that the former are longer,
on the order of 2025 ms (with a difference of 11 ms between
accented and neutral local contexts in Rysling et al., 2020).
The similarity of this time difference and the size of the
phoneme detection time effect tempts one to suggest that it
is the sole source of the latter effect. However, this suggestion
is discouraged by the Cutler and Darwin (1981) finding that
the detection time advantage for the accented prior context
condition (with a neutral target item) remained even when
the time difference was eliminated.
There may be other acoustic differences between the
accented and unaccented local contexts. In the materials used
in Experiment 1(and by Cutler, 1976), the accented local con-
texts were more variable in amplitude (and possibly in the pitch
that listeners perceived) than the unaccented ones. In
Experiment 2, the difference in F
0
variability was significant,
but the difference in amplitude variability was not. Comparing
the accented with the neutral local contexts used by Rysling
et al. (2020), the accented contexts were again more variable
in amplitude and had a lower mean F
0
(without a significant
difference in F
0
variability). All these differences could have
resultedinthesubjective difference in crispnessof the local
context, mentioned in the Introduction, but there may have
been other relevant acoustic differences that are difficult to
detect, especially given the shortness and often reduced nature
of the local context. Nonetheless, we suggest that it is not an
Fig. 3 Estimated log response time (RT)(with 95% credible interval) as afunction of experimental trial number, separately for accented and unaccented
local context items
Atten Percept Psychophys
accident that these specific acoustic properties duration, F
0
,
and amplitude are the ones that have been found to differ
across conditions in the studies on the accent advantage effect.
A growing body of work supports the conclusion that pitch and
timing cues are in a trading relation in prosodic parsing, so that
either can suffice to signal accent (Beach, 1991; Brugos &
Barnes, 2012a,2012b; Cumming, 2011;Jeon&Nolan,
2013). Further, trading relations between duration and ampli-
tude have long been recognized for segmental and non-speech
perception (e.g., Repp, 1979; Turk & Sawusch, 1996). If this is
the case, then it is unsurprising that identifying one specific
necessary acoustic property for the accent advantage to obtain
has been so elusive, because trading relations are themselves
sensitive to context effects (Repp, 1982,1983),andsowhich
specific cues traded with which others in a particular experi-
ment would only be understandable from careful analysis of the
individual production of the speaker of a stimulus item.
Conclusions about the mechanism by which local context
has its effect are similarly uncertain. It is tempting to say that
some property (perhaps its crispness)oftheverybrieflocal
context increases the listeners attention to the utterance (much
as the content of the full context appeared to do in Cutler &
Fodor, 1979). It is also possible that there was more anticipatory
co-articulation of the target segment when the target word had
been accented than when it was unaccented. There is evidence
(e.g., Salverda et al., 2014) that such anticipatory co-articulation
can very quickly affect word recognition. Moreover, a large
body of phonetic work has shown that vowels that are not
prosodically prominent, i.e., those that are unstressed or unac-
cented as in the pre-target syllable before an L+H* accented
target in our studies, are coarticulated more strongly with
vowels that are prosodically prominent than prominent vowels
are co-articulated with non-prominent ones (Beddor et al.,
2002;Cho,2004;Fivelaetal.,2011; Fletcher, 2004;
Harrington et al., 2013; Mok, 2012a,2012b; Turco et al.,
2016). If the accented local contexts contained more co-
articulatory information about the upcoming word than the un-
accented local contexts did, this could account for their effects
on phoneme detection speed without appeal to effects of atten-
tion. However (and as noted by a reviewer), the observation that
the accent advantage remains when the recorded context is de-
lexicalized (Cutler & McQueen, 2014) suggests that co-
articulation may not be the only basis for the accent advantage.
The secure conclusion to be drawn from the present exper-
iments (and from Rysling et al., 2020) may be that, while
several previous lines of research have shown that prosody
can guide some aspects of the comprehension of utterances,
it is unsafe to take at face value apparent demonstrations that
the full prosodic contour of an utterance can increase a lis-
teners attention to a particular portion of an utterance. It re-
mains possible that the prosody of the material immediately
preceding the target portion of an utterance can have such an
effect on attention, but further research is needed to determine
whether the effect actually involves attention or some other
mechanism, such as increased anticipatory coarticulation.
Appendix 1: Transcriptions
of the experimental sentences used
in Experiment 1.Accentedversion first.
Accented word IN UPPERCASE; target word
italicized; target segment indicated at the end
of the first sentence
1. The head cashier said the BAG contained money, and so
did the strongbox (B)
The head CASHIER said the bag contained money,
although his colleagues disagreed with him.
2. John decided that he'd take the BOAT down the river,
although flying would have saved him a day. (B)
John decided he'd take the boat DOWN the river, since he
had enjoyed the trip up so much.
3. The top experts are all unable to break the CODE the spy
had used. (K)
The top EXPERTS had been unable to break the code the
spy had used.
4. The agent was not surprised to find a BUG hidden in his
telephone. (B)
The agent was not SURPRISED to find a bug hidden in his
telephone.
5. My brother in law still drives a CAB Sundays, so he'll be
able to take us to the airport. (K)
My bother in law still drives a cab SUNDAYS, but
Saturdays he takes off.
6. The couple had quarreled over a BOOK they had recently
read. (B)
The couple had quarreled over a book they hadn't even
READ.
7. College students and young people in general are
drinking more BEER this year, and also more pop
wine. (B)
College students and young people in general are drinking
more beer THIS year than in any of the previous five years
Atten Percept Psychophys
8. That summer four years ago I ate roast DUCK for the first
time. (D)
That summer four years ago I ate roast duck for EVERY
meal
9. It occurred to me that I'd better buy another COAT before
the winter. (K)
It occurred to me that I'd better buy another coat before you
MENTIONED it.
10. Unfortunately not enough attention was paid to the
COST before the party. (K)
Unfortunately not enough attention was paid to the cost
BEFORE the party, and too much afterwards
11. I wanted to hire the BAND to give the place some atmo-
sphere, and now my partner wants to hire some topless
dancers as well.
Iwantedtohiretheband to give the place some
ATMOSPHERE, not to drive everyone away.
12. Down on the farm we used a CART for bringing feed to
the animals, since we didn't have anything else to use.
Down on the farm we used a cart for bringing feed to the
ANIMALS, but we never rode around in it.
13. She managed to remove the DIRT from the rug, but not
the berry stains (D)
She managed to remove the dirt from the RUG, but not
from their clothes.
14. The newly-elected senators wife decided to hire a new
COOK before they moved to Washington, and perhaps a
gardener as well. (K)
The newly-elected senators wife decided to hire a new
cook BEFORE they moved to Washington, rather than try to
find one there.
15. The politician reminded his friend of a BET they had
made. (B)
The politician reminded his friend of a bet they had made
about the ELECTION.
16. The phys ed teacher made a point of riding her BIKE to
work to set a good example. (B)
The phys ed teacher made a point of riding her bike to
WORK, although she never took it to the shopping center.
17. The policeman investigating the crime suggested that the
DOOR could have been forced open, as well as the win-
dow. (D)
The policeman investigating the crime suggested that the
door could have been FORCED open, not left unlocked by the
owner.
18. In many New England houses they still sound a BELL
for dinner, and expect their dinner guests to wear eve-
ning clothes. (B)
In many New England houses they still sound a bell for
DINNER, although other meals are less formal.
19. She felt very guilty about the DEBT she owed him, and
felt she should try harder to pay it. (D)
She felt very GUILTY about the debt she owed him, but
she still made no attempt to pay it.
20. My father was very fond of the DOG that he had, but he
couldnt stand the cats.
My FATHER was very fond of the dog that he had, but my
mother hated it.
Appendix 2: Materials used in Experiment 2.
Accented word IN UPPERCASE; target word
italicized; target segment was
as in Experiment 1.
1. I understand that the cash register at the BACK contains
money, and the one at the front doesn't.
I understand that the cash register at the back contains
money, and so does the STRONGBOX.
2. John decided that he would take the BOAT down the river,
although flying would have saved him a day.
John decided that he would take the boat down the river,
since he enjoyed the trip UP so much.
3. My brother-in-law still drives the CAB Sundays, so hell
be able to take us to the airport.
My brother-in-law still drives the cab Sundays, but
SATURDAYS he takes off.
Atten Percept Psychophys
4. The couple had quarreled over a BOOK they had recently
read.
The couple quarreled over a book they hadn't READ.
5. College students and young people in general are drinking
more BEER this year, and also more pop wines.
College students and young people in general are drinking
more beer THIS year than in any of the previous five years.
6. That summer four years ago I ate roast DUCK for the first
time.
That summer four years ago I ate roast duck for EVERY
MEAL.
7. It occurred to me that Id better buy another COAT before
you mentioned it.
It occurred to me that I'd better buy another coat before the
winter.
8. Unfortunately not enough attention was paid to the COST
before the party, which created tension.
Unfortunately not enough attention was paid to the cost
BEFORE the party, and too much afterwards.
9. I wanted to hire the BAND to give the place some atmo-
sphere, and now my partner wants to hire some topless
dancers as well.
Iwantedtohiretheband to give the place some
ATMOSPHERE, not to drive everyone away.
10. Down on the farm we used a CART for bringing
feed to the animals, since we didnt have anything
else to use.
Down on the farm we used a cart for bringing feed
to the ANIMALS, but we never rode around in it.
11. She managed to remove the DIRT from the rug but not
the berry stains.
She managed to remove the dirt from the RUG but not
from her clothes.
12. The newly elected senatorswifedecidedtohirea
COOK before they moved to Washington, and per-
haps a gardener as well.
The newly elected senators wife decided to hire a cook
BEFORE they moved to Washington, rather than try to find
one there.
13. The politician reminded his friend of a BET they had
made about the election.
The politician reminded his friend of a bet he had LOST.
14. The physical education teacher made a point of riding
her BIKE to work to set a good example.
The physical education teacher made a point of riding her bike
to WORK, although she never took it to the shopping center.
15. The policeman investigating the crime suggested that the
DOOR could have been forced open, as well as the
window.
The policeman investigating the crime suggested that the door
could have been FORCED open, not left unlocked by the owner
16. In many New England houses they still sound a BELL
for dinner, and expect their dinner guests to wear eve-
ning clothes.
In many New England houses they still sound a bell for
dinner, although OTHER meals are less formal.
17. She felt very guilty about the DEBT she owed him, but
she still made no attempt to pay it.
She felt very guilty about the debt she owed him, and felt
she should try HARDER to pay it.
18. My father was very fond of the DOG that he had, but he
couldnt stand the cats.
My father was very fond of the dog that he had, but my
MOTHER hated it.
19. The layoffs didn't affect the CORPORATE report, they
wreaked havoc on the market index.
The layoffs didn't affect the corporate report, the allega-
tions of FRAUD did.
20. The over-worked envoy from DENMARK was caught
off-guard by the Venezuelan representative.
The over-worked envoy from Denmark was caught off-
guard, which was UNUSUAL for him.
Atten Percept Psychophys
Appendix 3: Frequentist analysis
of Experiment 2.
Log RTs were analyzed in a linear mixed model using R3.9 and
LME4 (Bates et al., 2019). Separate analyses of the original and
the cross-spliced conditions indicated a significant effect of
determiner in each case (β= 0.045, SE = 0.013, t = 3.347 for
the original condition, β= 0.039, SE = 0.014, t = 2.85). When
both conditions were combined, the maximal analysis that con-
verged without a singular fit had random participant intercepts
and noninteracting item slopes by preamble and local context. It
resulted in a significant effect of local context (A faster than U,
β= -0.021, SE = 0.0073, t = -2.848), a nonsignificant effect of
preamble (β= -0.001, SE = 0.010, t = -0.125), and a significant
interaction between preamble and local context (reflecting
faster responses in the cross-spliced than the original condition),
β= 0.011, SE = 0.0047, t =-2.35). Because of concern that the
less-than-maximal analysis might be anti-conservative (Barr
et al., 2013), separate ANOVAS were conducted by-subject
and by-item, using the R package ezAnova. The results were
generally congruent. The main effect of local context was sig-
nificant by subjects (F(1,159) = 18.59, p < .001) and by items
(F(1,19) = 8.05, p = 0.01) as were the interactions of local
context and preamble (F(1,159) = 4.93, p < .03; F(1,19) =
6.28, p < .03) while the effect of preamble had F < 1.
Acknowledgements We would like to thank Anne Cutler for providing her
original stimulus materials, John Kingston for discussion of these results, and
Elizabeth McGarvey for extensive assistance in conducting this research.
Funding No funding sources were used in conducting or reporting this
research.
Declarations
Conflict of interest The authors have no relevant financial or non-
financial interests to disclose.
Ethics approval All experimental procedures were approved by the
University of Massachusetts Amherst Human Subjects IRB. The proce-
dures used in this study adhere to the tenets of the Declaration of Helsinki.
Consent to participate Written informed consent was obtained from all
individual participants in this study.
Consent for publication Not applicable.
References
Akker, E., & Cutler, A. (2003). Prosodic cues to semantic structure in native
and nonnative listening. Bilingualism: Language and Cognition, 6(2),
81-96. https://doi.org/10.1017/s1366728903001056
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(3), 255-278. https://doi.org/
10.1016/j.jml.2012.11.001
Bates, D., Maechler, M, Bolker, B, & Walker, S (2015). Fitting linear
mixed-effects models using lme4. Journal of Statistical Software,
67(1), 1-48. https://doi.org/10.18637/jss.v067.i01
Beach, C. M. (1991). The interpretation of prosodic patterns at points of
syntactic structure ambiguity: Evidence for cue trading relations.
Journal of Memory and Language,30(6), 644663. https://doi.
org/10.1016/0749-596X(91)90030-N
Beckman, M. E., & Hirschberg, J. (1994). The ToBI annotation conven-
tions. : Ohio State University.
Beddor, P. S., Harnsberger, J. D., & Lindemann, S. (2002). Language-
specific patterns of vowel-to-vowel coarticulation: acoustic struc-
tures and their perceptual correlates. Journal of Phonetics,30,
591-627. https://doi.org/10.1006/jpho.2002.0177
Birch, S., & Garnsey, S. M. (1995). The effect of focus on memory for
words in sentences. Journal of Memory and Language, 34(2), 232-
267. https://doi.org/10.1006/jmla.1995.1011
Bishop, J. (2017). Focus projection and prenuclear accents: evidence
from lexical processing. Language, Cognition and Neuroscience,
32(2), 236-253. https://doi.org/10.1080/23273798.2016.1246745
Boersma, P., & Weenink, D. (2019). Praat: doing phonetics by computer
[Computer program]. Version 6.1.07, retrieved 26 November 2019
from http://www.praat.org/
Breen, M., & Clifton, C. Jr. (2011). Stress Matters: Effects of Anticipated
Lexical Stress on Silent Reading. Journal of Memory and Language
64 (2), 153-170. https://doi.org/10.1016/j.jml.2010.11.001
Brugos, A. & Barnes, J. (2012a). Pitch trumps duration in a grouping
perception task. 25th Annual CUNY Conference on Human
Sentence Processing, New York, NY.
Brugos, A. & Barnes, J. (2012b). The auditory kappa effect in a speech
context. In: Qiuwu Ma, Hongwei Ding, and Daniel Hirst (Eds):
Proeedings of Speech Prosody 2012. (pp 151-154). International
Speech Communication Association. http://www.isca-speech.org/
archive/sp2012
Brunellière, A., Auran, C., & Delrue, L. (2018). Does the prosodic em-
phasis of sentential context cause deeper lexical-semantic process-
ing? Language, Cognition and Neuroscience,1-14.https://doi.org/
10.1080/23273798.2018.1499945
Bürkner, P. C. (2017). brms: An R package for Bayesian multilevel
models using Stan. Journal of Statistical Software, 80(1), 1-28.
https://doi.org/10.18637/jss.v080.i01
Cho, T. (2004). Prosodically conditioned strengthening and vowel-to-
vowel coarticulation in English. Journal of Phonetics,32, 141-
176. https://doi.org/10.1016/S0095-4470(03)00043-3
Cumming, R. E. (2011). The interdependence of tonal and durational
cues in the perception of rhythmic groups. Phonetica, 67, 219
242. https://doi.org/10.1159/000324132/html
Cutler, A. (1976). Phoneme monitoring reaction time as a function of
preceding intonation contour. Perception & Psychophysics, 20,
55-60. https://doi.org/10.3758/BF03198706
Cutler,A.(1987).Components of prosodic effects in speech recognition.
Paper presented at the Eleventh International Congress of Phonetic
Sciences (ICPhS), Tallinn, Estonia, Aug 1-7, 1987).
Cutler, A. (2012). Native Listening. : MIT Press.
Cutler, A., & Darwin, C. J. (1981). Phoneme monitoring reaction time
and preceding prosody: effects of stop closure duration and of fun-
damental frequency. Perception and Psychophysics, 29, 217-224.
https://doi.org/10.3758/BF03207288
Cutler, A., & Fodor, J. A. (1979). Semantic focus and sentence compre-
hension. Cognition, 7, 49-59. https://doi.org/10.1016/0010-
0277(79)90010-6
Cutler, A., & Foss, D. J. (1977). On the role of sentence stress on sentence
processing. Language and Speech, 20,1-10.https://doi.org/10.
1177/002383097702000101
Cutler, A., & McQueen, J. M. (2014). How prosody is both mandatory
and optional. In J. Caspers, Y. Chen, W. Heeren, J. Pacilly, N. O.
Schiller & E. V. Zanten (Eds.), Above and Beyond the Segments:
Atten Percept Psychophys
Experimental linguistics and phonetics (pp. 71-82). : Benjamins.
https://doi.org/10.1075/z.189.06cut
Dilley, L., & McAuley, D. (2008). Distal prosodic context affects word
segmentation and lexical processing. Journal of Memory and
Language 59(3), 294-311. https://doi.org/10.1016/j.jml.2008.06.006
Fiveash, A., Falk, S., & Tillmann, B. (2021). What you hear first, is what
you get: Initial metrical cue presentation modulates syllable detec-
tion in sentence processing. Attention, Perception, & Psychophysics,
83(4), 1861-1877. https://doi.org/10.3758/s13414-021-02251-y
Fivela, B. G., Stella, A., D'Apolito, S., & Sigona, F. (2011).
Coarticulation across prosodic domains in Italian: An ultrasound
investigation. In P. Cosi, R. D. Mori, G. D. Fabbrizio, & R.
Pieraccini (Eds.), Interspeech 2011 (p. 393-396). : ISCA Archive.
Fletcher, J. (2004). An EMA/EPG study of vowel-to-vowel coarticulation
across velars in Southern British English. Clinical Linguistics and
Phonetics,18(6-8), 577-592. https://doi.org/10.1080/
02699200410001703619
Fraundorf, S. H., Watson, D. G., & Benjamin, A. S. (2010). Recognition
memory reveals just how CONTRASTIVE contrastive accenting
really is. Journal of Memory and Language, 63(3), 367-386.
https://doi.org/10.1016/j.jml.2010.06.004
Green, P., & MacLeod, C. J. (2016). SIMR: an R package for power
analysis of generalized linear mixed models by simulation.
Methods in Ecology and Evolution, 7(4), 493-498. https://doi.org/
10.1111/2041-210X.12504
Harrington, J., Kleber, F., & Reubold, U. (2013). The effect of prosodic
weakening on the production and perception of trans-consonantal
vowel coarticulation in German. Journal of the Acoustical Society of
America,134(1), 551-561. https://doi.org/10.1121/1.4808328
Husband, M., & Ferreira, F. (2015). The role of selection in the comprehen-
sion of focus alternatives. Language, Cognition and Neuroscience, 31,
217-235. https://doi.org/10.1080/23273798.2015.1083113
Ip, M. H. K., & Cutler, A. (2017). Intonation facilitates prediction of
focus even in the presence of lexical tones. Paper presented at the
Interspeech 2017, Stockholm, Sweden, August 20-24, 2017.
Ip, M. H. K., & Cutler, A. (2018). Cue equivalence in prosodic entrain-
ment for focus detection. Paper presented at the Seventeenth
Australasian International Conference on Speech Science and
Technology, Sydney, Australia.
Ito, K. & Speer, S. (2008). Anticipatory effects of intonation: Eye movements
during instructed visual search. Journal of Memory and Language
58(2), 541-573. https://doi.org/10.1016/j.jml.2007.06.013
Jeon, H.-S., & Nolan, F. (2013). The role of pitch and timing cues in the
perception of phrasal grouping in Seoul Korean. The Journal of the
Acoustical Society of America,133(5), 3039-3049. https://doi.org/10.
1121/1.4798663
Kember, H., Choi, J., Yu, J., & Cutler, A. (2019). The processing of
linguistic prominence. Language and Speech, 0023830919880217.
https://doi.org/10.1177/0023830919880217
Li, X., Hagoort, P., & Yang, Y. (2008). Event-related potential evidence
on the influence of accentuation in spoken discourse comprehension
in Chinese. Journal of Cognitive Neuroscience, 20(5), 906-915.
https://doi.org/10.1162/jocn.2008.20512
Li, X.-q, & Ren, G.-q (2012). How and when accentuation influences tem-
porally selective attention and subsequent semantic processing during
on-line spoken language comprehension: An ERP study.
Neuropsychologia, 50(8), 1882-1894. https://doi.org/10.1016/j.
neuropsychologia.2012.04.013
Mauchand, M., Caballero, J. A., Jiang, X., & Pell, M. D. (2021).
Immediate online use of prosody reveals the ironic intentions of a
speaker: neurophysiological evidence. Cognitive, Affective, &
Behavioral Neuroscience, 21(1), 74-92. https://doi.org/10.3758/
s13415-020-00849-7
Mok, P. P. K. (2012a). Does vowel inventory density affect vowel-to-
vowel coarticulation? Language and Speech,56(2), 191-209.
https://doi.org/10.1177/0023830912443948
Mok, P. P. K. (2012b). Effects of consonant cluster syllabification on
vowel-to-vowel coarticulation in English. Speech Communication,
54, 946-956. https://doi.org/10.1016/j.specom.2012.04.001
Pierce, J. W., Gray, J. R., Simpson, S., MacAskill, M. R., Höchenberger,
R., Sogo, H.,. Lindeløv, J. (2019). PsychoPy2: experiments in be-
havior made easy. Behavior Research Methods.https://doi.org/10.
3758/s13428-018-01193-y
R_Development_Core_Team (2019). A language and environment for
statistical computing. : R Foundation for Statistical Computing.
Repp, B. H. (1979). Relative amplitude of aspiration noise as a voicing
cue for syllable-initial stop consonants. Language and Speech,
22(2), 173-189. https://doi.org/10.1177/002383097902200207
Repp, B. H. (1982). Phonetic trading relations and context effects: new
experimental evidence for a speech mode of perception.
Psychological Bulletin,92(1), 81. https://doi.org/10.1037/0033-
2909.92.1.81
Repp, B. H. (1983). Trading relations among acoustic cues in speech
perception are largely a result of phonetic categorization. Speech
Communication, 2(4), 341-361. https://doi.org/10.1016/0167-
6393(83)90050-X
Roettger, T. B., Franke, M., & Cole, J. (2020). Positional biases in predictive
processing of intonation. Language, Cognition and Neuroscience,
36(3), 342-370. https://doi.org/10.1080/23273798.2020.1853185
Rysling, A., Bishop, J., Clifton,C. Jr., & Yacovone,A. (2020). Preceding
syllables are necessary for the accent advantage effect. Journal of
the Acoustical Society of America, 148(3), EL285-288. https://doi.
org/10.1121/10.0001780
Salverda, A. P., Kleinschmidt, D., & Tanenhaus, M. K. (2014).
Immediate effects of anticipatory coarticulation on spoken-word
recognition. Journal of Memory and Language, 71(1), 145-163.
https://doi.org/10.1016/j.jml.2013.11.002
Sanford, A. J. S., Sanford, A. J., Molle, J., & Emmott, C. (2006). Shallow
processing and attention capture in written and spoken discourse.
Discourse Processes, 42(2), 109-130. https://doi.org/10.1207/
s15326950dp4202_2
Sturt, P., Sanford, A., Stewart, A., & Dawydiak, E. J. (2004). Linguistic
focus and good-enough representations: An application of the
change-detection paradigm. Psychonomic Bulletin & Review, 11,
882-888. https://doi.org/10.3758/bf03196716
Turco, G., Fougeron, C., & Audibert, N. (2016). The effects of prosody
on French V-to-V coarticulation: A corpus-based study. In
Interspeech 2016 (p. 998-1001). San Francisco: ISCA Archive.
https://doi.org/10.21437/Interspeech.2016-1323
Turk, A.E., & Sawusch, J.R. (1996). The processing of duration and
intensity cues to prominence. Journal of the Acoustical Society of
America,99(6), 3782-3790. https://doi.org/10.1121/1.414995
van Donselaar, W., & Lentz, J. (1994). The function of sentence accents and
given/new information in speech processing: Different strategies for
normal-hearing and hearing-impaired listeners? Language and Speech,
37(4), 375-391. https://doi.org/10.1177/002383099403700403
Publishersnote Springer Nature remains neutral with regard to jurisdic-
tional claims in published maps and institutional affiliations.
Atten Percept Psychophys
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
Using a phoneme detection task, the present series of experiments examines whether listeners can entrain to different combinations of prosodic cues to predict where focus will fall in an utterance. The stimuli were recorded by four female native speakers of Australian English who happened to have used different prosodic cues to produce sentences with prosodic focus: a combination of duration cues, mean and maximum F 0 , F 0 range, and longer pre-target interval before the focused word onset, only mean F 0 cues, only pre-target interval, and only duration cues. Results revealed that listeners can entrain in almost every condition except for where duration was the only reliable cue. Our findings suggest that listeners are flexible in the cues they use for focus processing.
Article
Full-text available
Real-time speech comprehension is challenging because communicatively relevant information is distributed throughout the entire utterance. In five mouse tracking experiments on German and American English, we probe if listeners, in principle, use non-local, early intonational information to anticipate upcoming referents. Listeners had to select a speaker-intended referent with their mouse guided by intonational cues, allowing them to anticipate their decision by moving their hand toward the referent prior to lexical disambiguation. While German listeners (Exps. 1-3) seemed to ignore early pitch cues, American English listeners (Exps. 4-5) were in principle able to use these early pitch cues to anticipate upcoming referents. However, many listeners showed no indication of doing so. These results suggest that there are important positional asymmetries in the way intonational information is integrated, with early information being paid less attention to than later cues in the utterance. Open data, scripts, and materials can be retrieved here: https://osf.io/xf8be/. ARTICLE HISTORY
Article
Full-text available
The accent advantage effect in phoneme monitoring-faster responses to a target phoneme at the beginning of an L þ H*-accented word than to a target phoneme at the beginning of an unaccented word-is viewed as a product of listeners' predictive capabilities [Cutler (1976). Percept. Psychophys. 20(1), 55-60]. However, previous studies have not established what information listeners use to form these predictions [Cutler (1987).]. This article presents evidence that at least the information in the syllable immediately preceding a target phoneme is necessary to cue the predictive attention allocation that underlies the accent advantage effect.
Article
Full-text available
Prominence, the expression of informational weight within utterances, can be signaled by prosodic highlighting ( head-prominence, as in English) or by position (as in Korean edge-prominence). Prominence confers processing advantages, even if conveyed only by discourse manipulations. Here we compared processing of prominence in English and Korean, using a task that indexes processing success, namely recognition memory. In each language, participants’ memory was tested for target words heard in sentences in which they were prominent due to prosody, position, both or neither. Prominence produced recall advantage, but the relative effects differed across language. For Korean listeners the positional advantage was greater, but for English listeners prosodic and syntactic prominence had equivalent and additive effects. In a further experiment semantic and phonological foils tested depth of processing of the recall targets. Both foil types were correctly rejected, suggesting that semantic processing had not reached the level at which word form was no longer available. Together the results suggest that prominence processing is primarily driven by universal effects of information structure; but language-specific differences in frequency of experience prompt different relative advantages of prominence signal types. Processing efficiency increases in each case, however, creating more accurate and more rapidly contactable memory representations.
Article
Full-text available
PsychoPy is an application for the creation of experiments in behavioral science (psychology, neuroscience, linguistics, etc.) with precise spatial control and timing of stimuli. It now provides a choice of interface; users can write scripts in Python if they choose, while those who prefer to construct experiments graphically can use the new Builder interface. Here we describe the features that have been added over the last 10 years of its development. The most notable addition has been that Builder interface, allowing users to create studies with minimal or no programming, while also allowing the insertion of Python code for maximal flexibility. We also present some of the other new features, including further stimulus options, asynchronous time-stamped hardware polling, and better support for open science and reproducibility. Tens of thousands of users now launch PsychoPy every month, and more than 90 people have contributed to the code. We discuss the current state of the project, as well as plans for the future.
Article
Full-text available
The brms package implements Bayesian multilevel models in R using the probabilistic programming language Stan. A wide range of distributions and link functions are supported, allowing users to fit – among others – linear, robust linear, binomial, Poisson, survival, ordinal, zero-inflated, hurdle, and even non-linear models all in a multilevel context. Further modeling options include autocorrelation of the response variable, user defined covariance structures, censored data, as well as meta-analytic standard errors. Prior specifications are flexible and explicitly encourage users to apply prior distributions that actually reflect their beliefs. In addition, model fit can easily be assessed and compared with the Watanabe-Akaike information criterion and leave-one-out cross-validation.
Article
Full-text available
Theories of Focus Projection claim that a single pitch accent on a verb’s argument is sufficient to prosodically mark that verb as part of the focus, negating the need for a prenuclear accent on the verb itself. The present study employed on-line lexical processing to test this claim empirically. In three cross-modal associative priming experiments, listeners heard English SVO sentences with/without prenuclear accenting on the verb in both broad (VP) and narrow (object) focus contexts. Results showed that the absence of a prenuclear accent in broad focus contexts did not disrupt priming, but the presence of one in narrow focus contexts did. This disruption was found to be somewhat modulated by individual differences in “autistic traits”. Overall, the findings are interpreted as supporting a model that includes both (a) a Focus Projection mechanism and (b) an information structural function for prenuclear accents, with the latter possibly subject to cross-listener variation.
Article
In social interactions, speakers often use their tone of voice (“prosody”) to communicate their interpersonal stance to pragmatically mark an ironic intention (e.g., sarcasm). The neurocognitive effects of prosody as listeners process ironic statements in real time are still poorly understood. In this study, 30 participants judged the friendliness of literal and ironic criticisms and compliments in the absence of context while their electrical brain activity was recorded. Event-related potentials reflecting the uptake of prosodic information were tracked at two time points in the utterance. Prosody robustly modulated P200 and late positivity amplitudes from utterance onset. These early neural responses registered both the speaker's stance (positive/negative) and their intention (literal/ironic). At a later timepoint (You are such a great/horrible cook), P200, N400, and P600 amplitudes were all greater when the critical word valence was congruent with the speaker’s vocal stance, suggesting that irony was contextually facilitated by early effects from prosody. Our results exemplify that rapid uptake of salient prosodic features allows listeners to make online predictions about the speaker’s ironic intent. This process can constrain their representation of an utterance to uncover nonliteral meanings without violating contextual expectations held about the speaker, as described by parallel-constraint satisfaction models.
Article
Speakers use prosodic emphasis to express the content of their message in order to help listeners to infer meaning. By measuring event-related potentials (ERPs) to semantically congruent and incongruent final words embedded in a sentential context that was emphasised or de-emphasised, we investigated whether prosodic emphasis conveyed by a sentential context leads listeners to a finer semantic analysis. The negative shift (N400) triggered by the difficulty to combine the incongruent word with the sentence representation was increased by prosodic emphasis at an early stage. Over the later stages, the amplitude of the N400 wave was increased by prosodic emphasis of the sentential context, whatever the semantic congruency of final words. As shown by the N400 wave, emphasising a sentential context affected the lexical-semantic processing of the following word. This study provides clear evidence that prosodic emphasis plays a role in the semantic analysis of sentences by inducing a deeper analysis.