An Alternative Way to Analyze and Predict Consonant Clusters

Productions in Brazilian Portuguese Phonological Assessments

ao V

ıctor B. Marques

1 a

, Jo

ao Carlos D. Lima

1 b

, M

arcia Keske-Soares

2 c

and Fabr

ıcio Andr

e Rubin

3 d

Centro de Tecnologia, Universidade Federal de Santa Maria, Santa Maria, Brazil

Centro de Ci

encias da Sa

ude, Universidade Federal de Santa Maria, Santa Maria, Brazil

Petroleo Brasileiro S.A., Rio de Janeiro, Brazil

Keywords:

Inference Logic, Speech Therapy, Phonological Assessments Tools, Consonant Clusters, Predicting Phonetics.

Abstract:

To conduct phonological assessments in children, it is necessary to have a set of words that contains a repre-

sentative sample of adult vocabulary. One of the obstacles to obtain a minimal set is the need to include words

with various consonant clusters so that such complex phonetic structures can be validated. In the current liter-

ature, there is only one way to determine whether a child is capable of producing a consonant cluster: through

the application of a phonological assessment, which contains several words with diverse phonetic structures

to be evaluated. In this context, logical inferences are one of the fundamental pillars in any learning area,

as they establish logical connections between information to form knowledge about a speciﬁc subject. This

work proposes an alternative way to indirectly assess a child’s ability to produce consonant clusters, based on

their ability to articulate similar clusters. The proposed algorithm is fed with the consonant clusters produced

and not produced by the child during the assessment. The goal is to discern which other clusters the child

is capable or incapable of producing, using the separation of consonant clusters into simpler phonetic struc-

tures. The method was validated with a database containing over 1200 phonological assessments conducted

in school-age children, native speakers of Brazilian Portuguese. The accuracy of our approach was 97% with

12% false positives and 8% false negatives, indicating that the method is interesting and signiﬁcantly faithful

to real-world results but still leaves room for future improvements. Nevertheless, it is believed that it can

be used to reduce the number of words needed in a phonological assessment, through indirect evaluation of

speciﬁc phonetic structures.

1 INTRODUCTION

In the context of speech therapy, phonological as-

sessments are conducted in school-age children to

identify the phonemes they can produce satisfactorily

and to detect possible phonological disorders (Usha

and Alex, 2023). However, this task demands time

from both the therapist and the child, often extend-

ing to approximately 1 to 1.5 hours (Combiths et al.,

2022), and the process is inﬂuenced by the number

of target words included in the assessment. In the

southern region of Brazil, where this work was con-

ducted, Brazilian Portuguese is used as the native lan-

https://orcid.org/0009-0007-3206-725X

https://orcid.org/0000-0001-9719-3205

https://orcid.org/0000-0002-5678-8429

https://orcid.org/0009-0009-5154-7843

guage, and there are different phonological assess-

ment tools in the country containing different sets of

words (Yavas et al., 2001; Savoldi et al., 2013; Ceron

et al., 2020). Each of these tools is adapted to the

regional context and incorporates words representing

familiar ﬁgures in children’s vocabulary (Sotero and

Pagliarin, 2018).

Despite advances in the ﬁeld, traditional meth-

ods of phonological assessment still rely on pen-and-

paper approaches, lacking software-based solutions

odi and T

oth, 2023). Such solutions are compatible

with speech therapy due to their audiovisual nature

(Uberti et al., 2022) and represent cost-effective alter-

natives to traditional methods (Usha and Alex, 2023).

Internationally, there is a growing effort to develop

computational solutions aimed at making this area of

pediatric speech therapy more efﬁcient and accessi-

ble to society (R

anen et al., 2021; Str

ombergsson

216

Marques, J., Lima, J., Keske-Soares, M. and Rubin, F.

An Alternative Way to Analyze and Predict Consonant Clusters Productions in Brazilian Portuguese Phonological Assessments.

DOI: 10.5220/0012555600003690

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 26th International Conference on Enterprise Information Systems (ICEIS 2024) - Volume 1, pages 216-223

ISBN: 978-989-758-692-7; ISSN: 2184-4992

et al., 2022). However, it is noteworthy that these

solutions are often applied mainly for research pur-

poses and lack effective commercial implementation

(Uberti et al., 2020).

In the ﬁeld of speech therapy, determining an in-

dividual’s ability to produce a phoneme requires that

the phoneme to be articulated correctly on at least

two occasions (Stoel-gammon, 1985). This require-

ment implies the need for equivalent words in phono-

logical assessment, i.e., words that possess the same

phonemes in the same positions. However, this ap-

proach contributes to an increased number of words

in the assessment (Marques. et al., 2023), resulting in

a potential redundancy in the evaluation of phonemes.

Thus, consonant clusters also represent a signiﬁ-

cant challenge in the quest for a reduced set of words,

as assessing each consonant cluster demands the iden-

tiﬁcation of two other words with the same occur-

rence. Additionally, it is important to note that conso-

nant clusters are complex linguistic structures, mak-

ing the task of ﬁnding words that not only share these

structures but also represent familiar elements in chil-

dren’s vocabulary non-trivial. This aspect adds an ad-

ditional layer of complexity to the lexical selection

process, highlighting the need for more reﬁned strate-

gies in phonological assessments.

Therefore, this study aimed to computationally

analyze the set of 84 words proposed by (Ceron et al.,

2020) as the basis for a pediatric phonological assess-

ment tool. Our proposal introduces a computational

logic capable of inferring whether a child can produce

a consonant cluster “XY” through words that contain

at least a part of it, i.e., words with other similar con-

sonant clusters like “XW”, “ZY”, “QX” etc. This idea

is based on the hypothesis that if a child can produce

at least parts of consonant clusters, they can produce

the consonant clusters formed by those parts. Addi-

tionally, we applied the reverse logic to identify the

structures that the child would not be able to produce.

To validate our proposal, we used a database

containing over 1200 phonological assessments con-

ducted on 1357 school-age children (3–7 years old) in

the southern region of Brazil. Each assessment used

the set of 84 words from (Ceron et al., 2020), and all

words were verbalized by the children and transcribed

in the database by the speech therapist.

Our results indicate that analyzing consonant clus-

ters separately reduces the minimum number of words

needed to validate the phonemes the child can articu-

late. Furthermore, the reduction of words does not im-

pact the indication of phonological disorder observed

in the assessment, meaning the same conclusions can

be reached using fewer words.

Finally, our inference logic was validated, and we

observed that 97% of the inferred consonant clusters

that the child would be able to produce were indeed

produced by the child. Conversely, we obtained 12%

false positives, indicating that the child would be able

to produce more phonemes than they actually did.

Additionally, the negative predictive value was 74%,

suggesting that the reverse logic to identify consonant

clusters the child would not be able to produce still

has room for improvement in future work.

The structure of the paper is as follows. In Sec-

tion 2, we present the main concepts related to this

work. In Section 3, we detail our proposal, and in

Section 4, we present how the proposal was validated.

Finally, in Section 5, we conclude the paper with our

ﬁnal considerations.

2 BACKGROUND AND CONTEXT

The study conducted by (Marques. et al., 2023)

revealed a notable overrepresentation of certain

phonemes in speciﬁc positions within the set pro-

posed by (Ceron et al., 2020). This implies that us-

ing this set in a phonological assessment tool may re-

sult in generating redundant data. In this scenario, the

speech therapist, when guiding the patient in produc-

ing a word, may at times be requesting the articula-

tion of phonemes that have already been previously

validated at least twice, a redundancy that could com-

promise the efﬁciency of the evaluation process (Mar-

ques. et al., 2023).

Through exploratory research, we found that there

is a collaborative effort between the ﬁelds of speech

therapy and information systems to develop tools that

assist in the identiﬁcation and diagnosis of phono-

logical disorders (Jothi and Mamatha, 2020; Attwell

et al., 2022). The interest in telemedicine solutions

in speech therapy has also increased, especially after

the global outbreak of COVID-19 (Uberti et al., 2022;

Bahar et al., 2022; Patel et al., 2022; Gallant et al.,

2023). The use of computational tools in this ﬁeld

is also attractive as they represent cost-effective solu-

tions compared to traditional methods of identifying

speech deﬁciencies and subsequent therapeutic mon-

itoring (Usha and Alex, 2023). However, the large

number of words in a comprehensive phonological as-

sessment complicates the screening process for cases

that truly require special attention.

Focusing on reducing the 84 words in the set pro-

posed by (Ceron et al., 2020), our work aims to pre-

dict the child’s ability to produce consonant clusters

based on their successes and errors. We introduce

a simpliﬁed method to verify which of these clus-

ters the child can or cannot produce, breaking down

An Alternative Way to Analyze and Predict Consonant Clusters Productions in Brazilian Portuguese Phonological Assessments

217

these complex phonetic structures into simpler units.

Then, by permuting these units to reconstruct com-

plex structures, we identify other consonant clusters

not yet assessed, suggesting the child’s ability to re-

produce them. Thus, it was possible to observe that in

97% of cases, if the child could produce the hypothet-

ical consonant clusters “XY” and “ZW”, they could

also reproduce the similar clusters “XW” and “ZY”.

This approach represents a perspective on evalu-

ating phonetic structures indirectly, substantially re-

ducing the need for a large number of words in a

phonological assessment. As advances in telepractice

in speech therapy continue to emerge (Chronopoulos

et al., 2021), similar approaches can even be used to

replace traditional methods of phoneme assessment,

since logical and statistical conclusions can be drawn

from the analysis of a limited set of data without com-

promising the quality of results.

2.1 Speech Therapy Concepts

In speech therapy, phonemes are analyzed in differ-

ent positions within the syllable and word (Armostis

et al., 2022). According to (Stoel-gammon, 1985), in

order to validate that a child is capable of articulating

a speciﬁc phoneme, it is necessary to assess them at

least twice. In the context of Brazilian Portuguese,

phonemes are evaluated in each of the following po-

sitions.

• (OI) Initial Onset: beginning of syllable, word be-

ginning - ca.sa [house];

• (OM) Medial Onset: beginning of syllable, mid-

dle of the word - ca.va.lo [horse];

• (CM) Medial Coda: end of syllable, middle of the

word - ca.dar.c¸o [shoelace];

• (CF) Final Coda: end of syllable, end of the word

- a.mor [love];

• (OCI) Initial Complex Onset: beginning of sylla-

ble, beginning of word - Bra.sil [Brasil];

• (OCM) Medial Complex Onset: beginning of syl-

lable, middle of the word - bi.blio.te.ca [library].

In this work, we are particularly interested in the

positions OCI and OCM as they indicate the presence

of consonant clusters. In Figure 1, it is emphasized

that the productions of the phonemes “br(OCI)” and

“br(OCM)” need to be differentiated, despite being

productions of the same cluster “br”. This is because

a child may ﬁnd it easy to produce the cluster at the

beginning of words but may have difﬁculty of pro-

ducing it in the middle of words. Such differentiation

helps the speech therapist better identify the phono-

logical articulation skills of children and possible re-

lated deviations. To validate the assessment, we need

two more words that stimulate the production of these

same consonant clusters, or in the best-case scenario,

a word that simultaneously contains the same cluster

in both OCI and OCM. Thus, there is a tendency for a

signiﬁcant increase in the number of words in an as-

sessment, stemming from the need to directly evaluate

consonant clusters.

a brirbra sil

OCI OCM

(OCM)

(OCI)

≠

Figure 1: Productions of the same consonant cluster in dif-

ferent positions are considered as different productions.

Thus, if we were able to apply a logic to infer an

individual’s ability to articulate a consonant cluster in

a speciﬁc position without the need for a direct assess-

ment, it would be possible to reduce the set of words

in the evaluation. This would lead to the exclusion

of words whose sole purpose is to encourage the pro-

duction of clusters that have already been indirectly

evaluated. Consequently, it would increase the efﬁ-

ciency of the assessment by using fewer resources to

reach the same conclusions. This is the focal point of

our study.

3 INFERRING CONSONANT

CLUSTERS PRODUCTIONS BY

SPLITTING

After analyzing the set from (Ceron et al., 2020) and

identifying all target phonemes, it was found that

there are 16 consonant clusters in the set. To ana-

lyze all of them at the word-initial position (OCI) and

word-medial position (OCM), we would need 2 oc-

currences for each phoneme, totaling 32 occurrences.

However, only 19 of these possibilities are evaluated

in the studied set, and of those, 14 are evaluated at

least 2 times, as shown in Figure 2. This indicates that

there would still be a shortage of words to adequately

evaluate all the consonant clusters in the set.

This work observed that the search for another

word with the same consonant cluster becomes eas-

ier when separating them into smaller phonetic struc-

tures, as shown in Figure 3. So, we start to break

the consonant clusters into smaller parts, instead of

analyzing them as they appear on the words, and we

observed that a inference logic could be applied.

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

218

OCI

pɾ pl bɾ bl

tɾ tl dɾ dl

kɾ kl gɾ gl

fɾ fl vɾ vl

OCM

pɾ pl bɾ bl

tɾ tl dɾ dl

kɾ kl gɾ gl

fɾ fl vɾ vl

evaluated in 2 words

evaluated in just 1 word

not evaluated

Figure 2: All consonant clusters found in the set proposed

by (Ceron et al., 2020).

bra

sil

OCI

…

OCI

No word with the exact br(OCI)

sil

2 words with b(OCI) + r(OCI)

OCI

…

Figure 3: Searching for words with the exact consonant

cluster vs searching for words with parts of the consonant

cluster.

3.1 Inferring Phonemes Capable of

Reproduce

Our approach introduces an inference logic capable of

predicting the consonant clusters that a child would be

able to produce based on the clusters that the child

has already produced at least once. This approach

analyzes each phoneme individually that constitutes

the consonant cluster, based on the premise that if the

child was able to articulate the clusters XY and ZW in

a speciﬁc position, then they are also capable of pro-

ducing XW and ZY in the same position. If this logic

holds true, it would be possible to test by inference

the other clusters highlighted in gray in Figure 2 with-

out the need to include more words in the assessment

solely for these clusters. Moreover, it would be pos-

sible to remove words from the assessment that serve

only to evaluate a consonant cluster that has already

been inferred that the child is capable of producing,

making the assessment shorter as it takes place.

We implemented a simple veriﬁcation method

presented in Figure 4. It is important to note that we

validated only the inferences of the colored phonemes

presented in Figure 2, which are present in the tar-

get words. This is because it would not be possible

to validate inferences of phonemes that the child was

not stimulated to produce, as we would not have the

necessary basis for validation.

pɾ (OCI)

bɾ (OCI)

pl (OCI)

bl (OCI)

…

Child’s

Assessment

Compare with

the validation set

Combine all the parts to form

known consonant clusters

Split all consonant clusters

in small parts

Validation

pl (OCI)

vɾ (OCI)

…

p (OCI)

ɾ (OCI)

b (OCI)

l (OCI)

…

compare

Inferred Consonant Clusters

Expected Produced

Target Words

Not

Produced

Phonemes

Non Repeated List

By looking only to this

list, we can’t know which

clusters was actually in

the entry data

pɾ (OCI)

bɾ (OCI)

bl (OCI)

pl (OCI)

vɾ (OCI)

…

Produced Phonemes

Entry

Set

Validation

Set

Figure 4: Algorithm to infer consonant clusters that the

child is able to reproduce.

3.2 Inferring Phonemes not Capable of

Reproduce

In addition, we employ reverse logic to infer the con-

sonant clusters that the child would not be able to pro-

duce. This approach is based on the consonant clus-

ters that the child was unable to produce in the assess-

An Alternative Way to Analyze and Predict Consonant Clusters Productions in Brazilian Portuguese Phonological Assessments

219

ment, using the same strategy shown in Figure 4.

From here, we have the following analysis: for

each word in the assessment, we check which con-

sonant clusters were expected to be produced by the

child and what was actually produced; then, we add to

a list the parts of the consonant clusters that were not

reproduced. For example, if it was expected in a word

that the child would reproduce the cluster “br” but

they produced “bl”, making a substitution, then only

the “r” should be added to the list of non-reproduced

parts.

4 VALIDATION AND RESULTS

We introduced a method that involves analyzing con-

sonant clusters in smaller parts, allowing the inference

of which clusters the child would or would not be able

to produce based on these parts.

To validate whether our approach is consistent

with real-world results, we executed the method on

our database to verify if the observed results align

with real-world data. In total, 1294 phonological as-

sessments were analyzed, conducted with 1357 chil-

dren aged 3-7 years. In each assessment, children re-

produced the 84 words from the set of (Ceron et al.,

2020), and their transcriptions were stored in our

database by a speech therapy specialist. Finally, each

transcription was analyzed using the method shown in

Figure 4.

It was observed that 76% of the inferences made

are of clusters that the child would be able to pro-

duce, while only 24% are from the reverse logic. This

is explained by the fact that our database consists

of 82% correct transcriptions, which represents an

imbalance regarding incorrect cases. Consequently,

this reﬂects in the quantity of inferences of consonant

clusters that the child would not be able to produce, as

such inferences are primarily based on cases contain-

ing phonemes that the child was unable to produce,

which are present mostly in incorrect transcriptions.

The complete result is shown in Figure 5.

As shown in Figure 6, our method was able to

cover all possibilities to evaluate consonant clus-

ters in the phonological assessment of (Ceron et al.,

2020). This indicates that, by employing an inference

method like ours, we can indirectly evaluate other

phonemes without the need to expose the child to ad-

ditional words in the assessment, providing greater ef-

ﬁciency to the process.

Currently, our research group is working on a

digital platform for phonological assessments, which

should include functionalities such as obtaining the

child’s phonetic inventory – a list of all the phonemes

Real Classification

True False

Total Population 24974 19994 4980

Predicted Classification

True 19511 581

False 1586 4605

Accuracy 0,97

Negative Predictive Value 0,74

Sensibility 0,98

False Positive Rate 0,12

False Negative Rate 0,08

Specificity 0,92

Figure 5: Confusion Matrix of our logic of inference.

OCI

pɾ pl bɾ bl

tɾ tl dɾ dl

kɾ kl gɾ gl

fɾ fl vɾ vl

OCM

pɾ pl bɾ bl

tɾ tl dɾ dl

kɾ kl gɾ gl

fɾ fl vɾ vl

evaluated in 2 words

evaluated in just 1 word

not evaluated directly (but inferred)

Figure 6: All possible consonant clusters in Brazilian Por-

tuguese with inferred phonemes by our method.

they are able to articulate satisfactorily. Additionally,

the platform will also have a contrastive analysis mod-

ule, which aims to identify the contrasts between the

child’s speech and that of an adult, providing a report

on which phonemes the child struggles with the most.

We developed the method proposed in this study

with the assistance of speech therapy specialists from

our group, and we already received a partial positive

validation from them. However, this method will need

to be validated in-depth when the platform is released

to our research colleagues, who will test it and pro-

vide their feedback. This way, our method can be

validated in ﬁeld and we can launch it to the gen-

eral public in the next months, after adjustments to

the platform containing the proposed method as part

of the system.

4.1 Analysis in Real Context

Finally, this subsection will present how the logic in-

troduced by this work can be applied in the real world

in phonological assessments.

In the algorithm proposed by (Marques. et al.,

2023), words were processed as they appeared in the

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

220

set, without any speciﬁc criteria. The present study

analyzed how many words would be necessary in

the set of (Ceron et al., 2020) to evaluate the same

phonemes in the same positions by varying two crite-

ria:

• The order of evaluation of the words;

• Whether or not to separate consonant clusters.

Regarding the order of evaluation of the words,

the analysis occurred in two ways: ﬁrst, analyzing

the easiest words to the most difﬁcult, and then in re-

verse order. The degree of difﬁculty of the words was

directly related to the number of errors recorded for

each word in our database. In short, words with more

errors would be the most difﬁcult words. Finally, we

applied a ﬂag that would determine whether conso-

nant clusters would be evaluated separately or not to

determine the impact of the logic introduced by our

study.

It was observed that, when not separating conso-

nant clusters, the number of words needed to validate

the child’s phonetic inventory was higher, as shown

in the graph in Figure 7. This is because it is more

difﬁcult to ﬁnd an exact consonant cluster separately

than the combination of simple phonemes that form

the cluster. Thus, we can replace a direct production

with an inference, eliminating the need to evaluate

consonant clusters entirely, only their parts.

Finally, as the number of words was reduced,

we also assessed the impact of this reduction on the

indication of speech disorder, obtained through the

PCC-R (Percent of Consonants Correct-Revised) by

(Shriberg et al., 1997) shown in Equation 1. This in-

dex is widely used in the ﬁeld of child speech therapy

as an indication of the level of phonological disorder

present in the child’s speech (McCabe et al., 2023;

Ceron et al., 2017), as shown in Table 1. The value

is calculated based on the number of correct produc-

tions of phonemes (PC) divided by the total number

of productions (TP), and them can be multiplied by

100 to have the percentage.

PCC-R =

T P

× 100 (1)

Table 1: Indication of speech disorder according with PCC-

R value (Shriberg et al., 1997).

PCC-R Value Indication of Disorder

Less than 50% High

Between 50% e 65% Moderate-High

Between 65% e 85% Low-Moderate

Greater than 85% Low

We measured the PCC-R for each of the assess-

ments observed in this study using all 84 words, and

associated an indication of speech disorder according

to Table 1. This indicator was taken as the validation

base, since it uses all words of the original set. The

next step was to perform the same measurements, but

now using subsets of words obtained in each scenario

addressed earlier.

The result is shown in Figure 7. It was observed

that, by using the logic introduced by this study and

separating consonant clusters to reduce the number

of words in the assessment, there were no signiﬁcant

changes in the accuracy of the indication of phono-

logical disorder compared to the original set. Further-

more, by not separating consonant clusters, the num-

ber of words needed in the subset slightly increases

without changing the accuracy, indicating an unnec-

essary amount in words.

100

Easiest First Hardest First

Qt Words [splitting] Precision [splitting]

Qt Words [no split] Precision [no split]

Figure 7: Number of words needed in the set and the re-

lationship with the accuracy in predicting the indication of

phonological disorder.

5 CONCLUSIONS

In this study, we introduced a method to infer a child’s

ability to produce consonant clusters based on their

ability to articulate similar clusters. In the literature,

computational methods have been applied in the ﬁeld

of pediatric speech therapy. An example of this is

the paper of (Franciscatto et al., 2019), where exist-

ing phonological processes in a child’s speech were

predicted through audio analysis and the application

of machine learning. However, while this study is be-

ing writing, no evidence has been found linking the

ability to articulate consonant clusters with the ability

to articulate similar clusters.

We employed a logic based on breaking down

these complex phonetic structures into elementary

structures. Our logic is grounded in the premise that:

An Alternative Way to Analyze and Predict Consonant Clusters Productions in Brazilian Portuguese Phonological Assessments

221

“if a child can produce the consonant clusters XY and

ZW in a speciﬁc position, then they can also produce

XW and ZY in the same position”. The phonological

assessment model studied in this work consists of 84

words, a considerably high number for a 3–7 years old

child to verbalize in a speech therapy session. With

our method, we could eliminate words from phono-

logical assessments by inferring consonant clusters

that the child can produce indirectly.

To validate our approach, we analyzed 1294

phonological assessments conducted in Southern

Brazil. Each assessment contained the 84 words pro-

posed by (Ceron et al., 2020), which were verbalized

by the children and transcribed in our database by ex-

perts in the ﬁeld. In each assessment, we obtained a

list of consonant clusters that the child produced and

that were present in the target words, which are the

clusters expected to be produced by the child during

the assessment. The list of clusters that the child did

not produce was also collected.

Finally, we reanalyzed each assessment with our

method, going through all transcriptions and breaking

down the consonant clusters that the child produced

and did not produce into smaller phonetic parts. At

the end of the analysis, our method was able to pre-

dict with 97% accuracy the articulation ability of the

inferred consonant clusters. However, our approach

had a false positive rate of 12% and a false negative

rate of 8%, indicating that the method predicts more

than ideal and still has room for improvements.

As future work, we will include adaptations to the

method to predict consonant clusters that the child

would not be able to produce, as the negative predic-

tive value was 74%, a value slightly low compared

with the general accuracy of the method. Addition-

ally, we believe that there is potential to apply the

same logic to predict other types of phonemes in the

language, based on their similarity and position in

words.

In conclusion, our approach introduces a fast way

to infer the child’s capability of producing consonant

clusters, eliminating the need for more words in the

phonological assessment tool. By proposing a new

set of target words for an assessment, our method can

be used to discard the need to directly evaluate cer-

tain consonant clusters in the language. This also re-

duces the assessment time and subsequent report ﬁll-

ing by the speech therapist, as fewer words would be

needed in the analysis. Thus, the assessment used

by speech therapists would become more efﬁcient,

achieving equally reliable results with less effort.

REFERENCES

Armostis, S., Petinou, K., and Kyprianou, A. (2022). The

implementation of the ‘testing of phonological skills

(tops)’ tool: a computer-based phonological analysis

algorithm. Clinical Linguistics & Phonetics, 0(0):1–

20.

Attwell, G. A., Bennin, K. E., and Tekinerdogan, B. (2022).

A systematic review of online speech therapy systems

for intervention in childhood speech communication

disorders. Sensors, 22(24).

Bahar, N., Namasivayam, A. K., and van Lieshout, P.

(2022). Telehealth intervention and childhood apraxia

of speech: a scoping review. Speech, Language and

Hearing, 25(4):450 – 462.

Ceron, M. I., Gubiani, M. B., Oliveira, C. R. d., and Keske-

Soares, M. (2017). Factors inﬂuencing consonant ac-

quisition in brazilian portuguese–speaking children.

Journal of Speech, Language, and Hearing Research,

60(4):759–771.

Ceron, M. I., Gubiani, M. B., Oliveira, C. R. d., and Keske-

Soares, M. (2020). Phonological assessment instru-

ment (infono): A pilot study. CoDAS, 32(4).

Chronopoulos, S. K., Kosma, E. I., Peppas, K. P., Taﬁadis,

D., Drosos, K., Ziavra, N., and Toki, E. I. (2021).

Exploring the speech language therapy through infor-

mation communication technologies, machine learn-

ing and neural networks. In 2021 5th International

Symposium on Multidisciplinary Studies and Innova-

tive Technologies (ISMSIT), pages 193–198.

Combiths, P., Amberg, R., Hedlund, G., Rose, Y., and Bar-

low, J. A. (2022). Automated phonological analysis

and treatment target selection using autopatt. Clini-

cal Linguistics & Phonetics, 36(2-3):203–218. PMID:

34085574.

Franciscatto, M. H., Lima, J. a. C. D., Trois, C., Maran,

V., Soares, M. K., and Rocha, C. C. d. (2019). A

case-based approach using phonological knowledge

for identifying error patterns in children’s speech. In

Proceedings of the 34th ACM/SIGAPP Symposium on

Applied Computing, SAC ’19, page 968–975, New

York, NY, USA. Association for Computing Machin-

ery.

Gallant, A., Watermeyer, J., and Sawasawa, C. (2023). Ex-

periences of south african speech-language therapists

providing telepractice during the covid-19 pandemic:

A qualitative survey. International journal of lan-

guage & communication disorders, 58.

odi, A. and T

oth, E. (2023). Exploring the opportuni-

ties for online assessment of phonological awareness

at the beginning of schooling. International Journal

of Early Childhood, pages 1–20.

Jothi, K. R. and Mamatha, V. L. (2020). A systematic re-

view of machine learning based automatic speech as-

sessment system to evaluate speech impairment. In

2020 3rd International Conference on Intelligent Sus-

tainable Systems (ICISS), pages 175–185.

Marques., J., Lima., J., Keske-Soares., M., Rocha., C., Ru-

bin., F., and Miollo., R. (2023). Algorithm for select-

ing words to compose phonological assessments. In

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

222

Proceedings of the 25th International Conference on

Enterprise Information Systems - Volume 1: ICEIS,,

pages 80–88. INSTICC, SciTePress.

McCabe, P., Preston, J. L., Evans, P., and Heard, R. (2023).

A pilot randomized control trial of motor-based treat-

ments for childhood apraxia of speech: Rapid sylla-

ble transition treatment and ultrasound biofeedback.

American Journal of Speech-Language Pathology,

32(2):629–644.

Patel, R., Loraine, E., and Gr

eaux, M. (2022). Impact of

covid-19 on digital practice in uk paediatric speech

and language therapy and implications for the future:

A national survey. International Journal of Language

& Communication Disorders, 57(5):1112–1129.

anen, O., Seshadri, S., Lavechin, M., Cristia, A., and

Casillas, M. (2021). Alice: An open-source tool

for automatic measurement of phoneme, syllable, and

word counts from child-centered daylong recordings.

Behavior Research Methods, 53:818–835.

Savoldi, A., Ceron, M. I., and Keske-Soares, M. (2013).

What are the best words to compose an evaluation

phonological instrument? Audiology-Communication

Research, 18(3):194–202.

Shriberg, L. D. et al. (1997). The speech disorders classi-

ﬁcation system (sdcs). Journal of Speech, Language

and Hearing Research, 40(4):723–740.

Sotero, L. K. B. and Pagliarin, K. C. (2018). The use of

software in cases of speech sound disorders. CoDAS,

30(6).

Stoel-gammon, C. (1985). Phonetic inventories, 15–24

months. Journal of Speech, Language, and Hearing

Research, 28(4):505–512.

Str

ombergsson, S., G

otze, J., Edlund, J., and Bj

orkenstam,

K. N. (2022). Simulating speech error patterns across

languages and different datasets. Language and

Speech, 65:105–142.

Uberti, L., Rodrigues Portalete, C., Pagliarin, K., and

Keske-Soares, M. (2020). Speech articulation assess-

ment tools: a systematic review. Journal of Speech

Sciences, 8(1):01–35.

Uberti, L. B., da Motta Forneck, L. L., Keske-Soares, M.,

and Pagliarin, K. C. (2022). How do speech-language

pathologists assess speech production through tele-

health? Audiology - Communication Research, 27.

Usha, G. P. and Alex, J. S. R. (2023). Speech assessment

tool methods for speech impaired children: a system-

atic literature review on the state-of-the-art in speech

impairment analysis.

Yavas, M., Hernandorena, C., and Lamprecht, R. (2001).

Avaliac¸

ao fonol

ogica da crianc¸a: reeducac¸

ao e ter-

apia. Biblioteca Artmed. Fonoaudiologia. Artmed Ed-

itora.

An Alternative Way to Analyze and Predict Consonant Clusters Productions in Brazilian Portuguese Phonological Assessments

223