tivity of producing timing patterns. Since the earlier reported
study provided produced timing data and there is a model that
can reproduce it, the entrainment hypothesis can begin to be put
in test.
3.1. Hypothesis Statement
If the shape of duration is important to the process of percep-
tion, there should be a definite relation among changes in dura-
tion and changes in the way duration is perceived. In order to
verify if this is the case it was investigated how well each con-
secutive V-to-V unit is attended to and if the changes observed
in responding level can somehow be related to changes of dura-
tion pattern in the stimuli. Attention was indirectly measured at
individual V-to-V unit position in a stress group from the means
of reaction time to clicks associated to them.
3.2. Stimuli Preparation
Sixteen sentences were chosen from the production corpus that
best fitted duration contours in figures 1 and 2. The choice was
made so as to pick two phrasing conditions and two target word
sizes. The phrasing conditions are illustrated below:
(a) “[A target]NP parece menor hoje.” (dσ = 0)
(b) “[A target bicolor ]NP parece menor hoje.” (dσ = 4)
In condition (a) the target word bears main phrasal stress
and in (b) the boldfaced syllable bears main phrasal stress. Eight
target words were selected, four with four syllables and four
with five syllables. 2.5 kHz pulse-like clicks were added in the
original sound files in successive V-to-V positions following the
schema below. In (c), target word is “patarata” and in (d) tar-
get word is “jaratacaca”. Slashes enclose segments in V-to-V
units in orthographic representation. Inside each V-to-V unit,
the click was always inserted at the right end of the consonantal
segment.
(c) four-syllable words:
/ap/1 /at/2 /ar/3 /at/4
/ap/1 /at/2 /ar/3 /at/ a bicol /orp/4
(d) five-syllable words:
/aj/1 /ar/2 /at/3 /ac/4 /ac/5
/aj/1 /ar/2 /at/3 /ac/4 /ac/ a bicol /orp/5
For each word in each phrasing condition four — as in (c)
— or five — like in (d) — clicked sound files were generated.
A total of 72 test sentences were generated applying this proce-
dure. An additional contextualizing sentence was added prior to
each test sentence. Another 68 pairs of sentences were recorded
by the same speaker of the production corpus to be used as
fillers. A click was randomly placed always in the first sentence
on filler items.
3.3. Stimulus Presentation
Items consisting of a pair of sentences were presented to the
subjects who were instructed to hear them and answer yes or
no to a content question following stimuli presentation. Sub-
jects were also warned that at some point during their listening
of the pair of sentences a click would appear and they were in-
structed to press a joystick button as fast as they could after the
click. The answer to the yes or no question was also recorded
by pressing a joystick button.
DMDX software was used to present the audio files and
record RT to click monitoring. Instructions were presented in
written form to each subject and a training section was run in
the presence of the experimenter so that any doubts could be
solved. Items were randomized prior to each round and blocked
in three parts with breaks between blocks. Sound files were
presented over closed headphones in a quiet room. Eighteen
college students voluntarily participated in the experiment and
signed a consent term. Subjects took about 20 minutes to do the
entire test.
3.4. Statistical Procedures
Raw RT data was log-transformed and then z-score normaliza-
tion was applied according to the following procedure: mean
RT (Xj) and standard deviation (sj∙) was computed for each
subject j and then each i RT sample ofj (xij) was transformed
into zij by means of equation 1.
zij =
χij - X
sj
(1)
There are two reasons to apply such operations. First, RT
is taken here as a measure of attention deployment and we are
interested in differences due to treatment and not in absolute
values of RT. Besides, log-transformation and z-score normal-
ization help fitting highly right-skewed distributions such as RT
raw data into the normal distribution.
Separate two-way ANOVAs were run in two groups: (A)
the test sentences with four-syllable target words and (B) the
group with five-syllable target words. Position in the stress
group and phrasing condition were independent variables and
normalized RT was the dependent variable. A α level of 5%
was fixed for statistics carried out in the experiment.
3.5. Results
Figures 3 and 4 show mean normalized RT for each V-to-V po-
sition in groups (A) and (B) respectively.
For group (A), factors Position (F (3, 509) = 10.688, p <
10-5) and Phrasing (F (1509) = 4.0343, p < 0.05) reached
significance level but not the interaction. Post-hoc testing with
Scheffe shows that position 4 (phrasal stress bearer) has sta-
tistically lower mean than positions 1 (p < 10-5 ) and 2
(p < 0.02). Likewise, position 3 it is significantly lower than
position 1 (p =< 0.03) yet marginally higher that position 4
(p < 0.09).
As for group (B), only Position factor yielded significance
(F (4, 632) = 5.0132, p < 10-5). Post-hoc Multiple com-
parisons with Scheffe indicate position 5 (phrasal stress bearer)
and 2 are significantly different (p < 0.05). Besides, position 4
differs significantly from position 2 (p < 0.02) and marginally
from position 1 (p < 0.06).
It is worthy noting that position 2 in figure 4 has a slightly
higher mean RT compared to position 1, yet this difference
is non-significant. The same comparison reached significance
when duration means were compared (cf. section 2.1). It is pos-
sible, though, that in longer stress groups a larger RT difference
between first and second positions shows up.
These results amount to classical findings that during sen-
tence processing significant reduction in RT is shown in rhyth-
mically crucial points, which can be seen as a positive answer
to question (1) stated above. In addition, these results, specially
those concerning group (A), bring new evidence that perception
facilitation is a gradient process not confined to rhythmically
salient spots. The closer a V-to-V unit is to stress group bound-
ary (as represented by duration maxima in figures 1 and 2), the