Multi-lingual code-mixed machine translation system
Amit Ramesh Khaparde, Akshat Kumar, Mahesh Kumar, Suyash Gupta · AIP conference proceedings · 2022
The Machine Translation System has been lately attracting considerable recognition due to its extensive use in multiple fields. This study focuses on abundant raw code-mixed textual data on microblogging sites, which can't be processed directly because it's multilingual. This raises a language barrier in communication and research. In order to use this immense data source, all unwanted symbols are removed to reduce noise in the input text. Also, the emoticon adding up the intention or emotion is permuted with the corresponding textual meaning. Recurrent Neural Network in one form or the other can be used for translating multilingual text from source to target language. A large neural network is used in the process of generating the expected noise-free output in the targeted language. The extensive experiments demonstrate that removing noise and transmuted emoticons improve the result and sustain the intention or semantics better. The proposed system can save anywhere from 35% to 98% of the cost of a human translator because of higher accuracy and desired output. It also finds a place as a subsystem to real-time multi-lingual translators, manuscript translators, other systems for providing input as required targeted language thus enlarging to input domain for that system.