Corpus Based Machine Translation for Scientific Text

Abstract

From many years, machine translation and computational linguistic research community has given immense attention towards the development of machine translation techniques. In order to fulfill the goal of machine translation "translation without losing meaning", a lot of translation methods have been proposed. All of these translation methods differ in their theories and implementation strategies. Although some basic rules of translation are same but many of them vary with the selection of language pair. While concerning with the scientific text, every science domain has thousands of terminologies. Translation of these terminologies according to the domain boosts the performance of translation. Translation of scientific text is ignored in the literature, as it needs more effort and expertise of both domain and language are required. In this research, we have proposed an effective scientific text translator for English to Urdu to cope with the challenge of scientific text translation. This method tags and translate the terms according to the domain. We have introduced a term tagger for tagging terms. The system can work for any domain but for experimental purpose we have selected the domain of computer science. System is evaluated on selfgenerated corpus of computer science. It is also compared with the existing translators to demonstrate the dominance of proposed translator as compared to the competitor. The comparative results of proposed approach and existing are shown in the form of tables.

The role of non-English-language science in informing national biodiversity assessments

“…The quality of machine translation has improved drastically over the years 25 , and machine translation is increasingly being used in science communication, for example, to assist communication with patients in health settings 26 . However, understandably, concerns over the accuracy of machine translation, especially when applied to scientific terms 27 , still limit its broader implementation in science communication 26 . The inadequate quality of machine translation was also recognised by some of the report authors who participated in the survey (Extended Data Fig.…”

Opportunities and challenges of text mining in materials research

“…However, there are significant difficulties in implementing the parallel-translation approach tailored specifically to the peculiarities of the scientific text. The domain-specific vocabulary of scientific texts requires a significant bilingual corpora for training the parallel-translation model ( Tehseen et al., 2018 ). The latest development in unsupervised NMT models ( Lample et al, 2017 , 2018 ; Artetxe et al., 2017 ) utilizes monolingual corpora, escaping the need for parallel texts.…”