Daniel Faustino Lacerda de Souza

Daniel Faustino Lacerda de Souza

Auto-generated profile, based on publicly available information

Daniel Faustino Lacerda de Souza

UFPB

Research Focus & Metrics

Academic Career Length

10Years (2017-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 January 2017 in Multimedia Tools and Applications
Verified
Journal Article
Closed access
PDF file
DSDaniel Souza
Daniel Souza
ABAquiles Burlamaqui
Aquiles Burlamaqui
GFGuido Souza Filho
Guido Souza Filho

26 January 2022 in Plasmonics
Verified
Journal Article
Closed access
PDF file
GAGleice M. de Aguiar
Gleice M. de Aguiar
LSLeandro de Souza
Leandro de Souza
DSDaniel F. L. de Souza
Daniel F. L. de Souza
LOLeiva Casemiro Oliveira
Leiva Casemiro Oliveira

26 May 2025 in Revista Principia
Verified
Journal Article
Open access
PDF file
DSDiego Ramon Bezerra da Silva
Diego Ramon Bezerra da Silva
SMSamuel de Moura Moreira
Samuel de Moura Moreira
DSDaniel Faustino Lacerda de Souza
Daniel Faustino Lacerda de Souza
RCRostand Edson Oliveira Costa
Rostand Edson Oliveira Costa
TATiago Maritan Ugulino de Araújo
Tiago Maritan Ugulino de Araújo
Os modelos recentes de Tradução Automática Neural (Neural Machine Translation – NMT) podem ser aplicados a idiomas e domínios de poucos recursos sem limitações significativas. Vários estudos investigam se novas técnicas de NMT também podem ser generalizadas para diferentes contextos, considerando a disponibilidade de dados e infraestrutura computacional. Nesse contexto, o objetivo principal deste estudo foi explorar métodos modernos de NMT e analisar a sua potencial aplicabilidade em cenários de poucos recursos, como os das línguas de sinais. Para uma melhor avaliação, foram adaptados e utilizados alguns modelos promissores identificados no componente de tradução automática da Suíte VLibras e os resultados foram comparados com aqueles fornecidos pela arquitetura LightConv atual, utilizando-se o mesmo corpus de treinamento e validação bilíngue Português-Libras, um dos maiores desse tipo disponíveis no mundo, constituído por mais de 70.000 frases geradas por linguistas. Os resultados indicam que a adoção de uma das duas arquiteturas de melhor desempenho – a Basic Transformer ou a ByT5 – ajudaria a melhorar a precisão e a qualidade da tradução da Suíte VLibras, com um aumento percentual de até 12,73% considerando a métrica BLEU.

24 July 2025 in Data in Brief
Verified
Journal Article
Open access
MLManuella Aschoff Lima
Manuella Aschoff Lima
DFDaniel Cruz de França
Daniel Cruz de França
DSDiego Ramon Silva
Diego Ramon Silva
DADilainne Daniel Albuquerque
Dilainne Daniel Albuquerque
DLDaniel Faustino Lacerda
Daniel Faustino Lacerda
TATiago Maritan de Araújo
Tiago Maritan de Araújo
VLibras-DB is a bilingual text corpus in Brazilian Portuguese (BP) and Brazilian Sign Language (Libras), designed and developed to support the creation of machine translation systems from BP-to-Libras. The corpus adopts a textual notation for Libras known as gloss, which serves as an interlingua between the source and target languages. To support this process, we initially defined a set of grammatical rules specific to Libras. Based on this notation, a bilingual textual database was built by a team of ten Libras interpreters, resulting in a corpus comprising 127,349 BP–Libras translation pairs. The dataset includes approximately 72,000 general-purpose sentences and around 55,000 sentences extracted from Brazilian federal government content and services.. The dataset was carefully constructed to include a wide variety of lexical and syntactic phenomena relevant to Libras translation, such as directional verbs, intensifiers, negation, and word-sense disambiguation. The resulting resource provides not only a substantial volume of parallel data but also a linguistically informed foundation for training and evaluating NMT models, contributing significantly to the advancement of accessible language technologies for the Deaf community. This comprehensive dataset is particularly significant for Neural Machine Translation (NMT) as it provides a much-needed, high-quality resource to train and evaluate NMT models for this low-resource language pair, facilitating advancements in BP-to-Libras translation systems. Beyond its direct application in NMT, VLibrasBD serves as a foundational linguistic resource for natural language processing, supporting tasks such as comparative linguistic analysis, bilingual embedding training, and the development of assistive technologies to enhance multilingual communication and information accessibility.

1 September 2020 in IEEE Latin America Transactions
Verified
Journal Article
Closed access
ABAntonio A. R. Beserra
Antonio A. R. Beserra
LSLeandro de Souza
Leandro de Souza
DSDaniel F. L. Souza
Daniel F. L. Souza
Compression algorithms have been proposed with the technology advance. However, there are not objective analysis procedures to guide a future choice of an algorithm directed for the type of data in the system they are intended for. This paper introduces a statistical framework, based on the bootstrap method, to execute the analysis of compression algorithms using an objective comparison parameter as a criterion. A case study using the compression ratio as the parameter and file samples of 4 different types was analyzed. The proposed scheme allowed us to infer which algorithm is better to be used for each data type. RLE has proven more suitable to image, audio and video files with Huffman obtaining comparable performance. For text files, LZW has remarkably outperformed all other algorithms.

3 September 2025 in Language Resources and Evaluation
Verified
Journal Article
Closed access
PDF file
DSDiego R. B. da Silva
Diego R. B. da Silva
MLManuella A. C. B. Lima
Manuella A. C. B. Lima
SMSamuel de M. Moreira
Samuel de M. Moreira
VCVirginia P. Campos
Virginia P. Campos
RCRenan P. O. Costa
Renan P. O. Costa
DADilainne D. de Albuquerque
Dilainne D. de Albuquerque