Raj Nath Patel
Research Interests: I am interested in neural/statistical machine translation/transliteration, Neural/statistical language modeling, quality evaluation and estimation, part-of-speech tagging, and parsing. Current projects include ”Personalized Neural Machine Translation”, ”Translation Quality Estimation using Deep Learning”, and ”Recurrent Neural Network based Part-of-Speech tagging”.
Specialties/Areas of Interest: Natural Language Processing & Deep Learning
Skills: Analytical researcher with more than eight years of experience in Natural Language Processing. Good grasp in Neural Networks (Transformer, Convolution, and Recurrent Neural Network), Statistical/Probabilistic modeling and machine learning algorithms. Experience in developing/working with open source projects.
Research Contributions
-
Improving Robustness in Real-World Neural Machine Translation Engines, The work resulted in a user track long paper published in the Machine Translation Summit XVII, 2019.
-
English to Indian Language Statistical Machine Translation with language-specific preprocessing, The work resulted in a long journal paper published in the Journal of Intelligent Systems, 2019.
-
Personalized Machine Translation: Preserving Original Author Traits, which is in collaboration with University of Sheffield, United Kingdom; IBM Research-Haifa, Israel; University of Haifa, Israel; The work resulted in a long EACL paper, EACL (European Chapter of the Association for Computation Linguistics) is one of the most prestigious conferences in the field, with acceptance rates below 25\%.
-
Recurrent Neural Network based Translation Quality Estimation, which resulted in a long WMT (ACL 2016) paper, WMT (Workshop on Statistical Machine Translation) conference is known for organising different NLP shared tasks including machine translation, quality evaluation and estimation, automatic post-editing, and bilingual document alignment and produces state-of-the-art techniques for these tasks.
-
Recurrent Neural Network based Pasrts-Of-Speech Tagging for social media code-mixed text, this is the first attempt using deep learning in the Indian language code-mixed NLP, the work resulted in a short ICON paper, ICON (International Conference on Natural Language Processing) is a well known and the only Indian conference for natural language processing and computational linguistics.
-
Reordering rules for English to Indian languages Statistical Machine Translation (SMT) to reduce the syntactic divergence between source and target languages which resulted in a long HyTra (ACL 2013) paper, HyTra (Hybrid Approaches to Translation) is a prestigious workshop in the area of Hybrid machine translation.
-
Morphological analysis for Dravidian languages to Hindi SMT which includes Suffix Separation and Compound Word Splitting to reduce the morphological divergence between source and target languages. The work resulted in a long paper and two short papaers ICON papers.
Education
Master of Science in Computer Science, University of Allahabad, India, 2009-2011
MARKS:74%
Bachelor of Science in Computer Science, University of Allahabad, India, 2006-2009
MARKS:67%
Accomplishment & Awards
Best System out of 5 participating systems, using language specific preprocessing, in the workshop on ”Machine Translation in Indian Languages (MTIL) 2017”.
Best Systemout of 10 participating systems, using Deep Learning Models which includes Long Short-Term Memory (LSTM) and Gated Recurrent Unit(GRU), in the ”Shared Task on Translation Quality Estimation (phrase level)”, WMT16, Association for Computational Linguistics (ACL) 2016.
Best Systemout of 3 participating systems, using post-editing, syntactic reordering, and morphological word segmentation, in the ”NLP Tools Contest on Statistical Machine Translation in Indian Languages”, International Conference on Natural LanguageProcessing (ICON), 2015.
2nd Best Systemout of 5 participating systems, using syntactic reordering, and mor-phological word segmentation, in the”NLP Tools Contest on Statistical Machine Trans-lation in Indian Languages”, International Conference on Natural Language Processing(ICON), 2014.
2nd Best Systemout of 4 participating systems, using factored SMT, in the ”Workshopon Reordering for Statistical Machine Translation”, COLING-2012.