Visual Prosody Supports Reading Aloud Expressively

Abstract

Type is not expressive enough. Even the youngest speakers are able to express a full range of emotions with their voice, while young readers read aloud monotonically as if to convey robotic boredom. We augmented type to convey expression similarly to our voices. Specifically, we wanted to convey in text words that are spoken louder, words that drawn out and spoken longer, and words that are spoken at a higher pitch. We then asked children to read sentences with these new kinds of type to see if children would read these with greater expression. We found that children would ignore the augmentation if they weren’t explicitly told about it. But when children were told about the augmentation, they were able to read aloud with greater vocal inflection. This innovation holds great promise for helping both children and adults to read aloud with greater expression and fluency.

Full text

Ann Bessemans, Maarten Renckens, Kevin Bormans, Erik Nuyts, Kevin Larson
Abstract
Type is not expressive enough. Even the youngest speakers are able to express a full
range of emotions with their voice, while young readers read aloud monotonically as if to
convey robotic boredom. We augmented type to convey expression similarly to our voices.
Specifically, we wanted to convey in text words that are spoken louder, words that drawn out
and spoken longer, and words that are spoken at a higher pitch. We then asked children to
read sentences with these new kinds of type to see if children would read these with greater
expression. We found that children would ignore the augmentation if they weren’t explicitly
told about it. But when children were told about the augmentation, they were able to read
aloud with greater vocal inflection. This innovation holds great promise for helping both
children and adults to read aloud with greater expression and fluency.
1. Introduction
Reading is magical. It allows us to communicate over unlimited time and distance.
More immediately, when we read, we need to convert the letters into sounds. Successfully
making this mapping is key step in learning to read. Once the alphabetical code can be
cracked, a child is able to independently decode words (Stanovich, 1986; Rayner & Pollatsek,
1989). Once individual words can be read aloud, it takes another step to read in a way that
sounds natural. Beginning readers often struggle to read aloud in a fluent, expressive
manner. Reading fluency is defined not only by speed and accuracy but also by proper
expression and the naturalness of reading (NAEP, 1995; National Institute of Child Health &
Human Development, 2000). Expressive oral reading can be quantified in terms of prosodic
variation in pitch, duration, and volume (Patel & Furr, 2011). These features can be of crucial
importance in understanding exactly what the speaker or narrator is trying to tell us.

Expressive reading is an increasingly valued component of literacy. The first focus of
reading must be on speed and accuracy of decoding. In Belgium and the Netherlands, the
reading levels are expressed by the AVI-levels (abbreviation of ‘Analyse van
Individualiseringsvormen’, translated as ‘Analysis of forms of individualization’). This kind
of standardization makes it possible to judge the reading level of the child, but these tests are
measured on reading speed and orthographical errors only. Techniques that are aimed at
improving expressive oral reading should be an integral part of reading fluency for ultimately
reading success (Hudson et al., 2005; Rasinski, 1990; Samuels, 1988; Schreiber, 1980).
There are several reasons why children’s prosodic oral reading fluency is important (Duong
et al., 2011; Gussenhoven 2004). Prosodic readers are not only easier to understand, but they
also have the ability to improve decoding, word recognition, reading accuracy, reading speed
and comprehension skills as they are able to segment text into meaningful units (Dowhower,
1991; Miller & Schwanenflugel, 2006; Ashby, 2006; Binder et al., 2013; Young-Suk Grace,
2015). Better prosody correlates to greater reading achievement.
In the Netherlands reading aloud competitions are a tradition for almost 22 years
(Stichting Lezen, 2016). Their most important reasons of being are encouraging children to
read and awaken their enthusiasm for literature. Reading aloud competitions’ main focus is
on the pleasantness of listening to reading aloud. Although there are many criteria which
judges look for, an important one is the use of the own voice (without using artificial voices).
A good reader is able to make use of small changes in tempo, can use a change of pitch and
can read louder and softer to convey a mood or emotion (Stichting Lezen, 2016).
The speech of beginning readers appears flat and laborious when reading aloud
(Miller & Schwanenflugel, 2006; NIH, 2000). Earlier approaches to aid children’s prosodic
reading aloud have focused on repetition and imitation of an adult-repeated-reading model
(Read Naturally, 2015) and guided oral reading (Kuhn & Stahl, 2003; Playbooks, 2013;
PROJECT LISTEN, 2009; Beck & Mostow, 2008). And while beginning readers typically
employ prosody in conversational speech, written text does not provide information about
the intended prosody. Nonetheless, text could indicate prosodic variations by means of the
typeface. We call this visual prosody.
The most well-known ways in which prosody is visualized in typography is in the
punctuation of normal typefaces, and in the phonetic transcriptions in comic books (i.e.
BAM!). Comic artists take into account a visual form of prosody to liven up the text.
However, the text accompanies the image that is rudimentary. It is through the image that
the meaning of the visual text can be determined and understood. Some experimental type
projects have explored the use of phonetic qualities. The acclaimed poet Paul van Ostaijen
made use of visual poetry in his ‘sound poems’. His ‘ritmiese typografie’ was designed by
artist Oscar Jespers. The poem Boem Paukeslag (1921) is probably the most well-known.

There also were more artistically inclined type projects which incorporated facets of spoken
language, such as the ‘New Alphabet’ by Tschichold (1930). These typefaces were developed
based on an idealism, dogma or philosophy, in this case during the Bauhaus. Conceptual type
projects in which aspects of phonetics form the foundations of the typefaces are seen in the
work of Kurt Schwitters ‘Systemschrift’ (1927) and more direct relations to the language itself
in the various projects of the French/Italian type designer Pierre di Sciullo.
Researchers have also thought about introducing visual prosody within text. Both van
Uden (1973) and Patel and Furr (2011) treat visual prosody by adding a second layer to the
text. Van Uden does this in the form of melody bows. Patel and Furr (2011) used two
methods to improve visual prosody: manipulated text cues and augmented text cues. In the
manipulated text cues they shift letter placement horizontally to indicate duration and
vertically to indicate pitch and used grey level to indicate loudness. In the augmented text
cues, they add graphs, lines and vertical bars behind the text to indicate the visual prosody.
They found that both methods are effective, but that the manipulated version is harder to
read, especially by shifting the words vertically. Both van Uden’s and Patel & Furr’s forms of
visual prosody show additional information on top of the text, which reduces the legibility of
the text.
It is also worth looking at the intuitive character of visual prosody. It is not
unreasonable to assume that children intuitively, without additional explanation,
spontaneously interpret certain adjustments as intended by the designers. Evidence for a
common sense or intuitive feeling presented within type design can be found in research
(Shaik, 2009; Lewis & Walker, 1989). For example, a bold or black typeface is perceived
louder against a lighter or greyer one (Shaik, 2009). The intuitive character also gives us
information about het learnability of visual prosody.
The goal of this project is to help children read aloud with more expression.
Specifically, we want to show with type the three main components that people already use
in spoken language: volume, duration or word length and pitch (Sitaram & Mostow, 2012;
Peppé, 2009; Schwanenflugel et al., 2006; Cutler et al., 1997; Dowhower, 1991; Bollinger,
1989; Lehiste, 1970). We want to do this without reducing legibility and in a way that will be
easy to learn. To explore whether it is possible to make prosody visible in type to guide
children’s reading aloud, we formulate four research questions:
A. Will children read text aloud with greater expression with text that is designed
to show the components of prosody?
B. Will children read the cues as intended: the volume cue read with greater
volume, the pitch cue read at a higher pitch, and the duration cue for a longer amount
of time?

C. Will the children intuitively understand the visual prosody or is instruction of
the visual cues needed?
D. After using the visual prosody, will the children be able to correctly describe
what each of the components of visual prosody mean? What does visual prosody tell
us about its learnability?
2. Methodology
Participants
118 children participated in the study. No participants were disqualified. The
participants in this study were Flemish children aged eight to ten years old and were enrolled
in regular elementary school. All children were reading normally for their age (reading level
of at least AVI 5). The tests were conducted at the elementary school ‘Sint-Rita’, located in
Sint-Truiden, Belgium. The children’s parents were informed about the research by a formal
letter. After the parents’ insight into the research, a written approval was asked if their child
was allowed to participate in the test. The children were randomly divided into two groups,
an information group (61 children) and a no-information group (57 children).
Fonts
The typeface Matilda was selected because its legibility has been extensively studied
for use with normal and low vision children (Bessemans, 2012). 8 new versions of Matilda
were designed for this study to show the volume, duration, and pitch features of prosody. All
conditions were shown at a sufficiently large 18 point size.
Volume
The boldness of letters was modified to indicate that a word should be read with
increased volume. Figure 1 shows a word with the normal Matilda font, a half bold font, and
a full bold font.
Figure 1: from top to bottom: ‘beer’ in the normal variation, ‘half bold’ and ‘full bold’. The Dutch sentence translates
to “The bear is in the garden.”
Duration

The width of letters was modified to indicate that a word should be read slower, or for
a longer amount of time. Figure 2 shows the normal Matilda font, a half wide font, and a full
wide font.
Figure 2: from top to bottom: ‘alleen’ in the normal variation, ‘half wide’ and ‘full wide’. The Dutch sentence translates to “The
poor man was left alone.”
Pitch
Visually describing pitch was the most challenging aspects of prosody. Two attempts
were made in order to test if one would work better than the other. In a first version of pitch,
letters were raised above the baseline to show that pitch should be raised. Figure 3 shows the
normal Matilda font, a half raised font, and a full raised font. In second version of pitch,
letters were stretched vertically to show that pitch should be higher. Figure 4 shows the
normal Matilda font, a half stretched font, and a full stretched font.
Figure 3: from top to bottom: ‘op’ in the
normal variation, ‘half raised’ and ‘full raised’.
Figure 4: from top to bottom: ‘ezel’ in the
normal variation, ‘half stretched’ and ‘full
stretched’.
In total 9 fonts (variations on one typeface) were used in the study, namely the
normal Matilda (n) and its 8 prosodic type design parameters aimed at influencing volume
(‘half bold’, ‘full bold’), duration (‘half wide’, ‘full wide’) and pitch (‘half raised’, ‘full raised’,
‘half stretched’, ‘full stretched’).
Sentences
5 unique sentences were examined in this project, each with a key word that would
appear in the studied conditions. The reading level for the sentences were slightly below the
reading level for the children. This was done in order to assure that the measurements could
be focused solely on the children’s reading aloud and not on reading difficulties of words that
otherwise might have occurred. The creation of these sentences were done in collaboration
with the teachers of the respective classes. Each of the 5 sentences was repeated 9 times,

once in each of the conditions. This made the pronunciation of each of the conditions directly
comparable.
45 sentences in total were presented in A5 size booklets with 5 sentences per page on
slightly off-white to yellow paper. Figure 5 shows a sample page in one of the booklets.
Figure 5: An example of how the sentences are presented to the participants in the booklet.
Procedures
The participants were told that we were investigating ways of making reading easier and
more fun. The study was conducted one participant at a time in a quiet, familiar room. The
participants were assigned to either the information group or to the no-information group.
The no-information group received no introduction to the volume, duration, and pitch
conditions they were asked to read, while the information group was shown the different
conditions and was given examples of reading the sentences with increased prosody. These
instructions were taught in a playful manner, where the child had to effectively look at the
testing material and search for prosodic cues. The children discovered the prosodic cues and
were taught the envisioned way to read them out aloud.
Figure 8A: A child pointing to noticed parameters in the sentence during the talk before or after the test.
Figure 8B: The actual reading test in which the child is reading after getting used to the microphone and
the design researcher pointing at the sentence that the child should read.

Each participant was then given a booklet with the 45 sentences presented in a
different random order. Their task was to read the sentences aloud the best way they could.
Only after a participant showed understanding of the task, the test was started. Some shy
children in the information group which were afraid of pronouncing the parameter clearly,
were asked to exaggerate a little the pronunciation. If during the reading session, a
participant in the no-information group asked about the conditions, he/she was told we were
making changes to the text but nothing could be said about it until the end of the
experiment. During the test, in order to ensure all sentences were read, the administrator
indicated the sentence the child had to read by pointing at it. This was also done to ensure
that, during the recordings, there were pauses to indicate clearly the start and ending of
every sentence. The participant read all sentences in the book in his/her own pace and was
allowed to correct when desired. If deemed necessary, a break in the middle of the book was
taken. For the youngest children this break was necessary, as more children than expected
lacked the concentration to read 45 sentences consecutively.
At the end of the first day of the experiment, each participant was debriefed. The no-
information group received information about the prosodic cues in the same way as the
information group got the clues. The purpose of the experiment was explained, and the child
was given the chance to ask questions about the study.
One or two days after each participant read the test sentences and was debriefed, the
participants were given a questionnaire as a whole class assignment. There were four tasks as
part of the questionnaire. The first was to look at prosody marked sentences and identify
which words have special prosody. The second task was to write down how they would
pronounce the prosody marked words. The third task was to state a preference between the
two kinds of pitch conditions. And the fourth task was an open-ended request for feedback.
Measurements
Digital audio recording was done with the Neumann U87ai microphone, designed for
voice recording. The digital processing and saving of the audio file was executed via the
program Praat, developed at the Department of Phonetic Sciences, University of Amsterdam
(Boursma, 2001; Boursma & Weenink, 2016). All data was collected on the most important
vowel of the test word. This is in line with Moneta et al. (2008) who focussed only on the
vowels to measure voice quality, emotions, in terms of amplitudes and frequencies.
Statistics
With X as the volume, duration or pitch, results are calculated as {average X of one
vowel of one specific word} divided by {average X of all the same vowels of the same word of

the same child}. E.g. the average pitch of the ‘ee’ in the word ‘beer’ written in vt_f, compared
with the average of {all the average pitches of all the ‘ee’ of all the words ‘beer’ the same child
has pronounced}.
The impact of a learning effect was avoided by following procedure. (i) There are 5
sentences and 9 fonts, resulting in 45 sentence-font-combinations. Every booklet has exactly
these 45 sentences. (ii) These 45 sentences were randomized. After this randomization, the
order was manually adapted (with as minimal changes as possible) in such a way that the
same sentence was presented maximally twice immediately after one another. Also the same
font was presented maximally twice after one another. (iii) By the combination of this
randomization and the limited manually adaptions, sentences nor fonts were too much
clustered in the beginning, the middle or the end of the reading task. (iv) This procedure was
done 20 times. Hence, there were 20 different booklets with each an unique order of
sentence-font-combinations. Every child read one booklet. (v) Since the statistics are
performed on these 20 different booklets, that are quite equally spread in the data set, in the
overall dataset sentences and fonts are even more equally spread over the beginning, the
middle or the end of the reading task. (vi) It is assumed that the learning effect hardly differs
between the fonts. By the steps described in the former steps, all fonts were equally spread
over the order of the 45 sentences, and thus the learning effect is measured in all fonts in a
same way. (vii) By calculating statistics on a ratio: {average of a vowel in a specific
font}/{average of this vowel of all fonts} the impact of the learning effect disappears as it is
both in nominator and denominator.
The effect of the fonts on the parameters of visual prosody is measured using a
Generalized Linear Model with repeated measures in SAS, procedure mixed. This procedure
includes adapted Tukey post hoc comparisons that takes the Bonferroni correction into
account to test simultaneously the set of all pairwise comparisons {μi−μj}. A Generalized
Linear Model is an extension of the classical ANOVA, but it has the extra options (repeated
measures, Tukey, Bonferroni correction) required for this dataset. For the present paper,
only the comparisons of the different fonts with the normal font are used.
3. Results
81 out of 5310 sentences were not processed due to an unknown error from the
speech recognition application. The error happened mainly in one sentence, resulting in a
highly underrepresentation of this sentence in the sample. The other 44 sentences were
equally spread in the sample (average 359 ± standard deviation 12). Not all words recorded

could be used, e.g. if a child stuttered the automatic data recognition program could not
recognize the word. In total 14457 words were included in the analyses.
The no-information group
The no-information group showed little difference between the nine fonts (see table 1
and graph 1 till 3).
font
Volume
Duration
Pitch
full bold
100%
101%
100%
full raised
101%
102%
100%
full stretched
100%
101%
101%
full wide
100%
102%
100%
half bold
100%
100%
101%
half raised
100%
101%
97% **
half stretched
100%
98%
99%
half wide
100%
98%
101%
normal
100%
100%
100%
Table 1: Average of one condition divided by the average of the normal condition for volume, duration and pitch
when no instructional information was provided to the participants. *’s indicates significant difference from normal font:
*=p<0.05; **=p<0.01; ***=p<0.001
There are no statistically significant differences between the fonts for volume, nor for
duration. For pitch, only ‘half raised’ differs statistically significant from the normal font
(p=0.01).
The information group
In the information group, all fonts differed significantly for all three measures with
the normal font (table 2).
typeface
Volume
Duration
Pitch
Volume
example dB
Pitch
example Hz
Duration
example sec
full bold
103% ***
134% ***
107% ***
72
257
0,19
full raised
103% ***
135% ***
111% ***
72
266
0,19
full
stretched
102% ***
133% ***
106% ***
71
254
0,19
full wide
103% ***
149% ***
106% ***
72
254
0,21
half bold
103% ***
129% ***
108% ***
72
259
0,18
half raised
102% ***
116% ***
103% **
71
247
0,16
half
stretched
101% *
118% ***
103% **
71
247
0,17
half wide
102% ***
117% ***
103% ***
71
247
0,16
normal
100%
100%
100%
70
240
0,14

Table 2: Average of one condition divided by the average of the normal condition for volume, duration and pitch
when instructional information was provided to the participants. *’s in the 3 first columns indicate significant difference from
normal font: *=p<0.05; **=p<0.01; ***=p<0.001.
To have a feeling of the impact in a realistic situation, we added for each component
an example in the last three columns of table 2. E.g. assume for volume a word which is
spoken, when using the normal font, with a volume of 70 dB (which is very near to the
average of the volume measured in this experiment). In the condition full bold, this can be
on average multiplied with 103%, hence the volume of the pronunciation would be 72dB. For
duration a word of 0.14 seconds and for pitch a word of 240 Herz are given as examples. Also
these absolute values are very close to the averages found in the dataset when normal font
was used.
Children in the information group read words louder when presented in 7 of the fonts
compared to the normal font see graph 4). Volumes of all fonts differed statistically
significant from the volume of words in the normal font. The largest effect on volume was for
the full bold, full wide, half bold, full raised conditions, an increase of 3% over the normal
font. The following graph represents the effect of font the volume on a word spoken by a
child.
Graph 4: Visualization of the effect of the font on a word that, with a normal font would be expressed with a volume of
70dB
Children in the information group showed a statistically significant increase in
duration of reading time for all test fonts compared to the normal font (see graph 5). The
60.0
62.0
64.0
66.0
68.0
70.0
72.0
74.0
Decibel
Information group, impact of font on loudness

largest increase in duration was for the full wide font, which was read 49% longer than the
normal font. The full wide font was read statistically significant longer than all other fonts.
Graph 5: Visualization of the effect of the font on a word that, with a normal font would be expressed with a duration
of 0.14 seconds
Children in the information group showed a statistically significant increase in pitch
for all the fonts compared to the normal font (see graph 6). The largest increase in pitch was
for the full raised font, which had vowels spoken at a 11% higher pitch than the normal font.
Full raised was read at a statistically significant higher pitch than all other fonts.
Graph 6: Visualization of the effect of the font on a word that, with a normal font would be expressed with a pitch of
240 Hz.
0.12
0.13
0.14
0.15
0.16
0.17
0.18
0.19
0.2
0.21
0.22
seconds
Information group, impact of font on duration
200
210
220
230
240
250
260
270
280
Herz
Information group, impact of font on pitch

Questionnaire
Two days after the test, participants were asked to identify words that should be spoken
differently and to say how they should be spoken. The half bold and full bold fonts were
correctly recognized 99.2% of the time. 81% of the participants said they should be read loud,
louder, or harder. Only 7% gave no answer or a very unclear answer. The half wide font was
recognized 55% of the time and the full wide was recognized 78% of the time. 93% of the
participants correctly said they should be read long or longer. The half raised font was
recognized 22% of the time, while the full raised font was recognized 99% of the time. 83% of
the participants correctly said that they should be read high or higher. 7% of the participants
incorrectly said they should be read louder. The half stretched font was recognized 49% of
the time and the full stretched font was recognized 99% of the time. 61% of the participants
correctly said they should be read high or higher. 15% of the participants incorrectly said
they should be read longer.
Graph 7: The proportion of words containing a specific parameter that are marked within a sentence, two days after
the test.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
full bold half bold full wide half wide full
raised
half
raised
full
stretched
half
stretched
proportion of marked prosodic parameters

Pie Chart 1: The answers that the children gave on the question ‘how to read the given prosodic parameter?’ two days
after the test.
At the end of the questionnaire, the participants were given an opportunity to provide
their opinions about the fonts. As design researchers, we believe that these rather subjective
81,36%
0,85%
3,39%
0,85%
3,39%
3,39%
0%
0%
6,78%
81%
BOLD
% CHILDREN SAY HOW TO READ A PROSODIC PARAMETER
increase intensity
decrease intensity
increase duration
decrease duration
increase pitch
decrease pitch
increase pitch and duration
do nothing
no anwer or very unclear answer
increase intensity
decrease intensity
increase duration
decrease duration
increase pitch
decrease pitch
increase pitch and duration
do nothing
no anwer or very unclear answer
1,69%
0%
93,22%
0%
2,54%
0,85%
0%
0%
1,69%
81%
WIDE
7,63%
0,85%
14,41%
1,69%
61,02%
1,69%
0,85%
0,85%
11,02%
STRETCHED
6,78%
1,69%
3,39%
0%
83,90%
1,69%
0%
0%
2,54%
81%
RAISED
81%93%
61%
14%
8%
7%
11%
84%

(when compared to the statistical output) opinions from the participants are of great
importance in determining whether or not the test material has an actual chance of being
used in real reading material for children, which they find pleasing to use. The reactions were
generally positive with the participants enjoying the increased variety in the text and the
additional support to help reading aloud. One participant said that it was “easier for reading
because you don’t have to read on the same tone and then it does not become boring.”
Another said that reading was “easier because you know if you have to read longer, higher or
louder.” Yet another one described the experience as if you were communicating in real life
instead of reading.
4. Discussion
The goal of this project was to help children read aloud with more expression. We
focused on the prosodic components volume (louder pronunciation), duration (slower
pronunciation) and pitch (higher pronunciation). While we hoped that the cues would be
understood without any explanation, we found that the children who received no explanation
choose to focus on reading the sentences quickly instead of with greater expression. Reading
for speed is the most common form of reading assessment, so it’s not entirely surprising that
the kids attempted to read quickly (Mostow & Duong, 2009). It was predicted by some of the
teachers that children, due to the intensive testing for measuring their reading level based on
speeds and accuracy, would interpret the test in this way and thus read as fast and correct as
possible, while ignoring prosody. Consequently, no statistically significant differences in the
prosody measures were found.
The children who were given an explanation of the prosody conditions read them
aloud quite clearly. Interestingly, prosody marked words tended to be spoken with increased
amounts of all kinds of prosody. For example, the words marked with increased volume were
read with statistically significant higher volume, but were also read slower and at a higher
pitch. The same was true for the other conditions.
All of the conditions caused the children to read the target word louder. The effect
was strongest for the full conditions and both full and half bold. Patel and Furr (2011) found
no effect of using grey levels to change kids’ volume. But in our study, we found a reliable
increase of volume for the same conditions that increase pitch reliably (changing the x-
height). Pitch and volume are strongly related as usually people speak louder when they raise
their pitch. This relationship was not intentional as we hoped to typographically convey each
prosody factor independently.

Widening the letters in a word was effective for getting the kids to pronounce a word
for a longer amount of time. This is in line with the findings of Patel and Furr (2011). But in our
study we found that all conditions increased duration statistically significant, though not as
dramatically as wide letters. It is surprising that the other conditions also led to increased
durations. This might have happened because the children needed additional cognitive
resources to correctly pronounce all of the prosody conditions, causing an increase in
duration. We observed this most clearly with the stretched and raised conditions (meant to
increase pitch) as the children found raising their pitch effortful.
Both the full raised and full stretched pitch conditions caused a statistically
significant increase in spoken pitch. The half raised and half stretched conditions did
increase pitch less. The fact that the higher voice is more difficult to understand may be
comprehended out of practice. Children do learn the difference between low and high in
nursery school, however this is treated as a spatial phenomenon. It was often seen that when
children needed to go higher in voice, they moved their body upwards. Because of the way
high and low is taught, they often didn’t know what to do with their voice. When a child had
almost no problems in the pronunciation of a higher voice he was asked afterwards if he had
a musical background. Often this was the case.
The half wide, half raised and half stretched font may be too subtle. Two days after
the test with the explanation, they are recognized in less than 75% of the situations.
5. Conclusion and further research
Only the data of the information group show differences in recordings of those words
that were highlighted with the prosodic cues. We believe that the no-information group
experienced this test as a regular reading test for measuring their reading level. This test
evaluates the child only on its reading speed and accuracy.
Within the information group, the analysis of the reading tests proves that the
prosodic design parameters have the intended effect on the oral reading of children.
Reading aloud of a single prosodic component hardly happened without an
interaction with the other prosodic components. However, when isolating each parameter
regarding their hypothesized effects, all parameters differed statistically significant from the
normal condition for pitch, volume and duration of speech. For each comparison with ‘n’, the
full variation of the intended parameter gave the most significant results for the intended
prosodic component. Thickening increased volume the most, widening the duration, raising
x-height increased the pitch the most. The effect of ‘full raised’ on pitch was an average
increase of 9%. The effect of ‘full bold’ on the volume was an average increase of 2%. The

effect of ‘full wide’ on the duration was an average increase of 37%. The effects of parameters
on the prosodic cues that were not intended were lower, and not always significant.
Based on the findings we recommend type designers to implement a thickened font
when they would like children to guide in their speaking aloud with a louder voice by means
of a typeface. Both ‘full bold’ and ‘half bold’ are good references for designers when they
would like to implement a volume parameter within their typeface. Type designers involved
in visual prosody are advised to widen the font, like our parameter ‘full wide’ when children
should be guided to read with a slower voice. From a designer’s point of view, we question if
it would be possible to design an even more extended type that is still aesthetically justified
in terms of letter shapes and text color. When type designers want to implement a design
parameter to guide the children in reading aloud with a higher voice, they can raise the x-
height, as in ‘full raised’.
The hypothesis that visual prosody in type is able to influence children’s reading
aloud with more expression is confirmed by this study. However, based on this research we
can not conclude whether visual prosodic cues are sensed in an intuitive manner. Thus,
instruction is needed. It is important to note that this research was conducted by Belgian
children and that in general, Belgians are known to be rather reluctant when it comes to
trying out things differently and rather do it in ways that are familiar to them (Hofstede,
2001). For example, when compared to the Dutch, Belgians are in general more introvert
(Laurent, 1973; Portzky et al., 2008; Gerritsen, 2014). This characteristic may attribute to
the fact that without instruction, the children may have seen the prosodic cues, but didn’t
execute them when reading aloud. There is a chance that, when other nationalities conduct
the same test, results may differ regarding the intuitive reading aloud of the parameters.
All in all, these type design parameters have the potential to influence the reading
aloud of children, and can therefore assist type and typographic designers to create new
typefaces and educational materials that aim to influence expressive reading. Within the new
technology of OpenType Font Variations (introduced in 2016) these parameters can be more
easily applied by typographers and usable by type designers.
Furthermore, the research to visualizing prosody through text proves to be promising
for further research, not only on printed matter but also in digital reading. There is a great
deal of enthusiasm for this work by publishing houses as it has the potential of making text
more expressive and can teach children more consciously reading aloud skills. Expressive
type may reduce the cause of confusion in written communication and might improve
reading comprehension. Additionally, prosody also has a diverse range of other uses
including making expressive captioning for the deaf community and teaching expression for
the autistic community.

Acknowledgements
We would like to thank Wouter Vanmontfort for the adaptation of Praat; Tom
Wollaert for optimising the audio recordings and Microsoft Advanced Reading Technologies
for the grant.

Bibliography
∗ ASHBY, J. (2006). “Prosody in skilled silent reading: evidence from eye movements.” Journal of Research in Reading. 29, (3), 318-
333.
∗ BECK, J.E., MOSTOW, J. (2008). “How who should practice: Using learning decomposition to evaluate the efficacy of
different types of practice for different types of students.” 9th International Conference on Intelligent Tutoring Systems, Montreal,
June 23-27, 353-362.
∗ BESSEMANS, A. (2012) Letterontwerp voor kinderen met een visuele functiebeperking. PhD dissertation, Leiden University &
Hasselt University. http://hdl.handle.net/1887/20032
∗ BOURSMA, P., & WEENINK, D. (2016). Praat: doing phonetics by computer (Version 6.0.08), Computer program.
http://www.fon.hum.uva.nl/praat
∗ BOURSMA, P. (2001). "Praat, a system for doing phonetics by computer." Glot International. vol. 5, 341-345.
∗ BINDER, S. K., TIGHE E., JIANG Y., KAFTANSKI K., QI C., & ARDOIN, S. P. (2013). “Reading expressively and understanding
thoroughly: An examination of prosody in adults with low literacy skills.” Read Writ. 26, 665-680.
∗ BOLINGER, D. (1989). Intonation and its uses: Melody in grammar and discourse. Stanford, CA: Stanford University Press.
∗ CUTLER, A., DAHAN, D., & VAN DONSELAAR, W. (1997). “Prosody in the comprehension of spoken language: A literature review.”
Language and speech. 40, (2), 141-201.
∗ DOWHOWER, S. (1991). “Speaking of prosody: Fluency’s unattended bedfellow.” Theory into Practice. 30, (3), 165-175.
∗ DUONG, M., MOSTOW, J., & SITARAM, S. (2011). “Two methods for assessing oral reading prosody ACM.” Transactions on Speech
and Language Processing (Special Issue on Speech and Language Processing of Children’s Speech for Child -machine Interaction
Applications) 7, (4), 11-22. doi 10.1145/1998384.1998388.
∗ GERRITSEN, M. (2014.). “Vlaanderen en Nederland: één taal, twee culturen?” Neerlandia/Nederlands van Nu. (1), 26-29.
∗ GUSSENHOVEN, C. (2004). The phonology of tone and intonation. New York, NY: Cambridge University Press.
∗ HOFSTEDE, G. (2001). Culture’s consequences: comparing values, behaviors, institutions and organizations across nations.
Thousand Oaks/London/New Delhi: Sage Publications.
∗ HUDSON, R.F., LANE, H.B., & PULLEN, P.C. (2005). “Reading fluency assessment and instruction: what, why, and how?” The
Reading Teacher. 58, (8), 702–714.
∗ KÖHLER, W. (1929). Gestalt psychology. New York, NY: Liveright.
∗ KUHN, M.R., & STAHL, S.A. (2003). “Fluency: A review of developmental and remedial practices.” Journal of Educational
Psychology. 95, (1), 3–21.
∗ LAURENT, P-H. (1973). “The Benelux States and the New Community.” Current History (pre-1986). April, 64, 166-182.
∗ LEHISTE, I. (1970). Suprasegmentals. Cambridge, MA, USA: MIT Press.
∗ LEWIS, C., & WALKER, P. (1989). “Typographic influences on reading.” British Journal of Psychology. 80, (2), 241-257.
∗ PROJECT LISTEN. (2009). LISTEN’s Reading Tutor. [Website] http://www.cs.cmu.edu/~./listen/ [Assessed: 2017, September, 24th]
∗ MILLER, J., & SCHWANENFLUGEL, P. J. (2006). “Prosody of syntactically complex sentences in the oral reading of young children.”
Journal of Educational Psychology. 100, 310-321.
∗ MONETA, M. E., PENNA, M., LOYOLA, H., BUCHHEIM, A., & KÄCHELE, H. (2008). Measuring emotion in the voice during
psychotherapy interventions: A pilot study. Biological Research. 41, (4), 389-395. DOI: /S0716-97602008000400004
∗ MOSTOW, J., & DUONG, M. (2009). “Automated Assessment of Oral Reading Prosody.” Proceedings of the 2009 conference on
Artificial Intelligence in Education: Building Learning Systems that Care: From Knowledge Representation to Affective Modelling,
July 20, 189-196. doi 10.3233/978-1-60750-028-5-189
∗ NATIONAL INSTITUTE OF CHILD HEALTH AND HUMAN DEVELOPMENT (NIH). (2000). Report of the National Reading Panel.
Teaching Children to Read: an evidence-based assessment of the scientific research literature on reading and its implications for
reading instruction. NIH Publication 00-4769. Washington, D.C., USA: US Government Printing Office.
∗ NATIONAL ASSESSMENT OF EDUCATIONAL PROGRESS (NAEP). (1995). “Listening to children read aloud: Oral fluency.”
NAEPFacts. 1, (1), 2-5.
∗ PATEL, R., & FURR, W. (2011, May, 7-12). “ReadN’Karaoke: Visualizing prosody in children’s books for expressive oral reading.” CHI
Session: Books & Language, 3203-3206.
∗ PEPPÉ, S. J. E. (2009). “Why is prosody in speech-language pathology so difficult?” International Journal of Speech-Language
Pathology. 11 (4), 258-271.
∗ PLAYBOOKS. (2013). Playbooks; roleplay reader. [Website] http://www.readerstheater.com [Accessed: 2015, Januari, 2nd].
∗ PORTZKY, G., WILDE DE, E. J. & HEERINGEN VAN, K. (2008.) “Deliberate self-harm in young people: differences in prevalence and
risk factors between The Netherlands and Belgium.” European Child & Adolescent Psychiatry. 17 (3), 179-186.
∗ RASINSKI, T. V. (1990). “Effects of repeated reading and listening while reading on reading fluency.” Journal of Educational
Research. 83, 147-150.
∗ RAYNER, K., & POLLATSEK, A. (1989). The psychology of reading. New Jersey: Prentice Hall.
∗ READ NATURALLY, INC. (2015). Read naturally. [Website] http://www.readnaturally.com [Accessed: 2015, Januari, 2nd].
∗

∗ SAMUELS, S. J. (1988). “Decoding and automaticity: Helping poor readers become automatic at word recognition.” The Reading
Teacher. 41, 756–760.
∗ SCHREIBER, P. A. (1980). “On the acquisition of reading fluency.” Journal of Reading Behavior. 7, 177–186.
∗ SCHWANENFLUGEL, P.J., MEISINGER, E.B., WISENBAKER, J.M., KUHN, M.R., STRAUSS, G.P., & MORRIS, R.D. (2006).
“Becoming a fluent and automatic reader in the early elementary school years.” Reading Research Quarterly. 41, (4), 496-522.
∗ SHAIKH, D. (2009). “Know your typefaces! Semantic differential presentation of 40 onscreen typefaces.” Usability News. 11, (2).
∗ SITARAM, S., & MOSTOW, J. (2012). “Mining data from project LISTEN’s reading tutor to analyze development of children's oral
reading prosody.” Proceedings of the Twenty-Fifth International Florida Artificial Intelligence Research Society Conference. 478-
483.
∗ STANOVICH, K. E. (1986). “Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy.”
Reading Research Quarterly. 21, (4), 360-407.
∗ STICHTING LEZEN. (2016). The National Reading Aloud Competition. [Website] https://cdn.denationalevoorleeswedstrijd.nl/wp-
content/uploads/2018/05/National-Reading-Aloud-Competition-Rule-Book-for-web.pdf
∗ VAN UDEN, A. (1973). Taalverwerving door taalarme kinderen. Rotterdam: University press Rotterdam.
∗ YOUNG-SUK GRACE, K. (2015). “Developmental, component-based model of reading fluency: An investigation of predictors of word-
reading fluency, text-reading fluency, and reading comprehension.” Reading Research Quarterly. 50, (4), 459–481.