Loading [MathJax]/extensions/MathMenu.js
Kannada to English Machine Translation using LSTM and GRU | IEEE Conference Publication | IEEE Xplore

Kannada to English Machine Translation using LSTM and GRU


Abstract:

This paper focuses on unidirectional translation of Kannada text to English text using Neural Machine Translation (NMT) and addresses the challenges posed by Kannada lang...Show More

Abstract:

This paper focuses on unidirectional translation of Kannada text to English text using Neural Machine Translation (NMT) and addresses the challenges posed by Kannada language such as its unique grammar, syntax, and vocabulary. Extensive studies have shown that Recurrent Neural Network (RNN) models have been highly effective for machine translation tasks. In this paper, the Sequence to Sequence (Seq2Seq) framework is employed, utilizing an Encoder-Decoder Mechanism with the Gated Recurrent Unit (GRU) and LSTM as RNN unit. The Seq 2 Seq architecture enables the model to learn the underlying patterns and dependencies within the input data, allowing it to generate coherent and meaningful translations. The model was trained and evaluated with two datasets. Dataset-1 consists of 2000 lines of normal sentences of Kannada to English translation pairs and Dataset-2 consisted of 2500 lines of text which is a mixture of normal sentences and sentences which are context dependent. The quality of machine translation was evaluated using standard translation quality metrics such as BLEU (bilingual evaluation understudy). Simulation results showed a BLEU score of 0.68 and 0.72 for Dataset-1 and BLEU score of 0.56 and 0.63 for Dataset-2 with LSTM and GRU model.
Date of Conference: 09-10 August 2024
Date Added to IEEE Xplore: 04 October 2024
ISBN Information:
Conference Location: Bengaluru, India

I. Introduction

Machine Translation(MT) is one type of natural language processing application. It is the task of converting text from one language, known as the source language, to another, known as the target language. In MT, exact word-for-word translation occurs, but the translated text may or may not have the same semantics as the source text. Translation in paraphrase is done at the sentence level rather than the word level. The semantics of the source text are preserved while translating into translated text. Even though MT-predicted transcriptions differ from human-looking translation, they are easy to understand and the conversion process is free from human interference. The translation approach’s effectiveness is demonstrated by its ability to generate semantically identical and grammatically valid target constructs. Prior to translation, an intellectual translation approach avoids word-for-word translation. The processing of natural language enables machines to understand languages spoken by humans. Natural language processing allows machines to understand human languages, and translation is critical in comprehension of information and ideas expressed in various languages. However, Kannada, a language spoken primarily in the Indian state of Karnataka, is getting little attention in comparison with other various languages.

Contact IEEE to Subscribe

References

References is not available for this document.