''مشینی ترجمہ کاری کا ارتقا: نظریاتی بنیادیں، عملی اطلاقات اور عصری رجحانات''

Abstract

Machine Translation (MT) is a linguistic-technological phenomenon that emerged in the mid-20th century as a result of the convergence between computer science and linguistics. It aims to automatically convert one human language into another through computational processes. This paper explores the theoretical foundations, historical evolution, technological models, and contemporary applications of MT, with a particular focus on under-resourced languages like Urdu. From the philosophical visions of a universal language in the 17th century to the advent of neural machine translation systems like DeepL, Microsoft Translator, and Amazon Translate, this study traces the intellectual journey and technological advancements in the field. The paper examines key MT models including Rule-Based (RBMT), Corpus-Based (CBMT), Statistical (SMT), Neural (NMT), and Hybrid Machine Translation (HMT), emphasizing their strengths and limitations. It also investigates the challenges of applying MT to Urdu due to limited corpora, lack of annotated data, and cultural-linguistic complexities. Recent research on models like Fully Convolutional Neural Networks (FConv-NN), Long Short-Term Memory (LSTM), and Example-Based MT for English-Urdu translation is critically analyzed. The paper concludes by outlining future prospects and research directions for enhancing machine translation quality in low-resource languages.