Mehul Ligade
Mehul Ligade has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Chandrakant D. Kokane
Department of Computer Science and Engineering (Artificial Intelligence), Vishwakarma Institute of Technology, Pune, Maharashtra, India
1 shared publication - Pranav Khandagale1 shared publication
- Shreeyash Garde1 shared publication
- Vilas Deotare1 shared publication
Latest Publications
- 17 April 2024
- Published by Taylor & Francis
- p. 55-76
Data-Centric AI Approaches for Machine Translation
Machine learning has transformed machine translation, a crucial component of multilingual communication. This chapter emphasizes its function, application, and effects. It starts with a system overview that highlights the statistical and rule-based constraints that led to the integration of machine learning.Parallel corpora and high-quality training data are necessary for effective integration. The chapter discusses data construction, focusing on bilingual datasets and hand-annotated data. To improve translation quality, preprocessing techniques such as tokenization, alignment, and feature engineering are investigated. We look at various neural network topologies, including transformers, LSTM, and RNNs. These do so by capturing context and improving accuracy through methods that balance source-target information. Model learning is accelerated by training and optimisation strategies like SGD and regularization. Evaluation is explained, covering both human and automatic metrics (BLEU, METEOR, TER). The use of cutting-edge methods like zero-shot, multilingual translation, and transfer learning is highlighted, highlighting the advancements made in neural machine translation with models like BERT. Future directions include contextual integration and improving translation of rare words. Challenges like low-resource languages and domain adaptation are faced. This chapter clarifies the influence of machine learning on translation, improving results and fostering multilingual dialogue. It is extremely helpful to academics, practitioners, and enthusiasts because it is based in scientific jargon.
- 17 April 2024
- Published by Taylor & Francis
- p. 55-76
Data-Centric AI Approaches for Machine Translation
Machine learning has transformed machine translation, a crucial component of multilingual communication. This chapter emphasizes its function, application, and effects. It starts with a system overview that highlights the statistical and rule-based constraints that led to the integration of machine learning.Parallel corpora and high-quality training data are necessary for effective integration. The chapter discusses data construction, focusing on bilingual datasets and hand-annotated data. To improve translation quality, preprocessing techniques such as tokenization, alignment, and feature engineering are investigated. We look at various neural network topologies, including transformers, LSTM, and RNNs. These do so by capturing context and improving accuracy through methods that balance source-target information. Model learning is accelerated by training and optimisation strategies like SGD and regularization. Evaluation is explained, covering both human and automatic metrics (BLEU, METEOR, TER). The use of cutting-edge methods like zero-shot, multilingual translation, and transfer learning is highlighted, highlighting the advancements made in neural machine translation with models like BERT. Future directions include contextual integration and improving translation of rare words. Challenges like low-resource languages and domain adaptation are faced. This chapter clarifies the influence of machine learning on translation, improving results and fostering multilingual dialogue. It is extremely helpful to academics, practitioners, and enthusiasts because it is based in scientific jargon.