A Linguistic Evaluation of Google Translate Between Human and Machine Translation for English-Sorani Kurdish Languages Using Natural Language Processing (NLP)

Muhealddin, Dlnya Raouf; Mohammed Salih, Yusra Mohammed; Taher, Fatima Jalal

doi:10.24271/psr.2024.422619.1415

A Linguistic Evaluation of Google Translate Between Human and Machine Translation for English-Sorani Kurdish Languages Using Natural Language Processing (NLP)

Document Type : Original Article

Authors

Dlnya Raouf Muhealddin ¹

Yusra Mohammed Mohammed Salih ²

Fatima Jalal Taher ³

¹ International English Language Center, Presidency of Sulaimani Polytechnic University, Sulaimani Polytechnic University, Sulaymaniyah, Sulaymaniyah, Iraq, 46001

² Department of Computer Engineering, Faculty of Engineering, Tishk International University, Sulaymaniyah, Sulaymaniyah, Iraq, 46001

³ Department of Translation, College of Languages, University of Sulaimani, Sulaymaniyah, Sulaymaniyah, Iraq, 46001

10.24271/psr.2024.422619.1415

Abstract

Abstract
Online machine translation systems are widespread worldwide and free of charge. Among the most widely utilized multilingual machine translation services is Google Translate; however, Sorani Kurdish was one of the languages added to Google Translate in May 2022. The assessment of machine translation systems is crucial for understanding their performance. Several standard and efficient natural language processing metrics accurately approximate human assessments, including BLEU, NIST, and BERTScore. This study is interdisciplinary research between two intertwined fields: language and technology. In computational linguistics, Sorani Kurdish has received trivial attention; therefore, the current work targets to investigate and evaluate sets of English to Sorani Kurdish translations to determine whether machine translations differ from human translations. The machine translation was analyzed from a descriptive quantitative and qualitative perspective according to the investigators’ mother language intuitions (Sorani Kurdish) and their linguistic backgrounds of English and Kurdish regarding morphological, semantic, and syntactic features. Based on the results of these experiments, Google Translate performs at a lower level than the average performance of Kurdish native speakers, and further research is suggested to increase its quality. Despite several shortcomings in evaluating Google Translate utilizing NLP and the ongoing advances of Google Translate, Sorani Kurdish speakers can gain a lot from using this free online machine translation service.

Keywords

Google Translate (GT)

English-Kurdish Translation

Kurdish-Sorani Dialect

Linguistic Evaluation

Natural Language Processing (NLP)

Machine Translation (MT)

BLEU Score NIST BERTScore