Emmanuel Agyei

Dr. Emmanuel Agyei

Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, China
orcid
The more we search, the greater our drive to continue search...

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
18 April 2025 in Applied Sciences
Verified
Journal Article
Open access
PDF file
Emmanuel AgyeiEmmanuel Agyei
Emmanuel Agyei
Xiaoling ZhangXiaoling Zhang
Xiaoling Zhang
AQAma Bonuah Quaye
Ama Bonuah Quaye
Victor Adeyi OdehVictor Adeyi Odeh
Victor Adeyi Odeh
JAJoseph Roger Arhin
Joseph Roger Arhin
Machine Translation (MT) for low-resource languages, such as Twi, remains a persistent challenge in natural language processing (NLP) due to the scarcity of extensive parallel datasets. Due to their heavy reliance on high-resource data, traditional methods frequently fall short, underserving low-resource languages. To address this, we propose a fine-tuned T5 model trained using a Cross-Lingual Optimization Framework (CLOF), a unique method that dynamically modifies gradient weights to balance low-resource (Twi) and high-resource (English) datasets. This cross-lingual learning framework leverages the strengths of federated training to improve translation performance while ensuring scalability for other low-resource languages. In order to maximize model input, the study makes use of a carefully selected parallel English-Twi corpus that has been aligned and tokenized. A thorough evaluation of translation quality is provided by the use of SPBLEU, ROUGE (ROUGE-1, ROUGE-2, and ROUGE-L) measures, and Word Error Rate (WER) metrics. A pretrained mT5 model is used to set baseline performance, which acts as a standard for the optimized model. The suggested method shows notable benefits, according to experimental results. The fine-tuned model achieves a remarkable increase in SPBLEU from 2.16% to 71.30%, a rise in ROUGE-1 from 15.23% to 65.24%, and a notable reduction in WER from 183.16% to 68.32%. These findings highlight the effectiveness of CLOF in addressing the challenges of low-resource MT and enhancing the quality of Twi translations. This work demonstrates the potential of combining cross-lingual learning and federated training to advance NLP for underrepresented languages, paving the way for more inclusive and scalable translation systems.

5 July 2024 in Discover Computing
Verified
Journal Article
Open access
PDF file
Emmanuel AgyeiEmmanuel Agyei
Emmanuel Agyei
XZXiaoling Zhang
Xiaoling Zhang
SBStephen Bannerman
Stephen Bannerman
AQAma Bonuah Quaye
Ama Bonuah Quaye
SYSophyani Banaamwini Yussi
Sophyani Banaamwini Yussi
VAVictor Kwaku Agbesi
Victor Kwaku Agbesi
Although Ghana does not have one unique language for its citizens, the Twi dialect stands a chance of fulfilling this purpose. Twi is among the low-resourced language categories, yet it is widely spoken beyond Ghana and in countries such as the Ivory Coast, Benin, Nigeria, and other places. However, it continues to be seen as the perfect resource for Twi Machine Translation (MT) of IS0 639-3. The issue with the Twi-English parallel corpus is eminent at the multiple domain dataset level, partly due to the complex design structure and scarcity of the digital Twi lexicon. This study introduced Twi-2-ENG, a large-scale multiple domain Twi to English parallel corpus, Twi digital Dictionary, and lexicon version of Twi. Also, it employed the Ghanaian Parliamentary Hansards, crowdsourcing, and digital Ghana News Portals to crawl all the English sentences. Our curled news portals accumulated 5,765 parallel corpus sentences, the Twi New Testament Bible, and social media platforms. The data-gathering method used means of translation, compilation, tokenization, and the final alignments with the Twi-English parallel sentences, including the technology employed in compiling and hosting the corpus, were duly discussed. The results reveal that the role of manually qualified linguistic professionals and Twi translation specialists across the media spectrum, academia, and well-wishers adds a considerable volume to the Twi-2-ENG parallel corpus. Finally, all the sentences were curated with the help of a corpus manager, sketch engine, linguistics, and professional translators to align and tokenize all texts, allowing the Twi professional linguists to evaluate the corpus.

17 February 2022 in Periodica Polytechnica Civil Engineering
Verified
Journal Article
Closed access
PDF file
IAIsaac Osei Agyemang
Isaac Osei Agyemang
XZXiaoling Zhang
Xiaoling Zhang
IAIsaac Adjei-Mensah
Isaac Adjei-Mensah
JAJoseph Roger Arhin
Joseph Roger Arhin
Emmanuel AgyeiEmmanuel Agyei
Emmanuel Agyei
Civil structural component detection plays an integral role in Structural Health Monitoring (SHM) pre and post-construction. Challenges including but not limited to labor-intensiveness, cost, and time constraints associated with traditional methods make it a less opti-mal approach in SHM. Despite the success of deep convolutional neural networks in diverse detection problems, the required computational resources are a challenge. This has led to rendering a chunk of resource-constrained edge nodes less applicable with deep convolutional neural networks. In this paper, a computational-efficient deep convolutional neural network is presented based on Gabor filters and a color Canny edge detector. Generic Gabor filters are generated and used as initializers in the computational-efficient deep convolutional neural network presented, afterward trained on building components data. Next, extensive offline and online experimentation with a resource-constrained edge node is conducted and evaluated using diverse metrics. The computational-efficient detection model demonstrates to be effective in detection and via NVIDIA GPU profiler, we observe conservation of around 30% of computational resources during training. The computational-efficient detection model adduces almost a 3% mean average precision higher than two state-of-the-art detectors and records a promising frame processing rate during the online experimentation.

17 December 2021
Verified
Conference Paper
Closed access
Emmanuel AgyeiEmmanuel Agyei
Emmanuel Agyei
XZXiaoling Zhang
Xiaoling Zhang
SYSophyani Banaamwini Yussif
Sophyani Banaamwini Yussif
BABless Lord Y. Agbley
Bless Lord Y. Agbley
Recent years, Neural Machine Translation outperforms the existing traditional-based models for language translation with the use of sequential architecture. However, low resource languages, such as Akan-Twi, lack structured language resources largely due to the limited amount of research geared towards their revival and resuscitation. In an effort to mitigate the issue of low resource language translation from Akan-Twi to English in machine learning, we utilize the low resource parallel corpus for the translation setting of Akan-English. Using our proposed training process and Transformer-based model, we generate augmented corpus which is used to improve the performance of our final translation model. We achieved best BLEU score of 12.96 and 17.57 for the Akan-English and English-Akan, respectively.

23 December 2024 in Preprints.org
Verified
Preprint
Open access
PDF file
Emmanuel AgyeiEmmanuel Agyei
Emmanuel Agyei
Xiaoling ZhangXiaoling Zhang
Xiaoling Zhang
AQAma Bonuah Quaye
Ama Bonuah Quaye
Victor Adeyi OdehVictor Adeyi Odeh
Victor Adeyi Odeh
JAJoseph Roger Arhin
Joseph Roger Arhin
Machine Translation (MT) for low-resource languages, such as Twi, remains a persistent challenge in Natural Language Processing (NLP) due to the scarcity of extensive parallel datasets. Due to their heavy reliance on high-resource data, traditional methods frequently fall short, underserving low-resource languages. To address this, we propose a fine-tuned T5 model trained using Cross-Lingual Optimization Framework (CLOF), a unique method that dynamically modifies gradient weights to balance low-resource (Twi) and high-resource (English) datasets. This cross-lingual learning framework leverages the strengths of federated training to improve translation performance while ensuring scalability for other low-resource languages. In order to maximize model input, the study makes use of a carefully selected parallel English-Twi corpus that has been aligned and tokenized. A thorough evaluation of translation quality is provided by the use of SPBLEU, ROUGE (ROUGE-1, ROUGE-2, and ROUGE-L) measures, and Word Error Rate (WER) metrics. A pretrained mT5 model is used to set baseline performance, which acts as a standard for the optimized model. The suggested method shows notable benefits, according to experimental results. The fine-tuned model achieves a remarkable increase in SPBLEU from 2.16% to 71.30%, a rise in ROUGE-1 from 15.23% to 65.24%, and a notable reduction in WER from 183.16% to 68.32%. These findings highlight the effectiveness of CLOF in addressing the challenges of low-resource MT and enhancing the quality of Twi translations. This work demonstrates the potential of combining cross-lingual learning and federated training to advance NLP for underrepresented languages, paving the way for more inclusive and scalable translation systems.