Machine translation can be used as a language-based method, where words are translated into the most appropriate language where they will be replaced. Students are willing to automatically learn knowledge interpretation from large data instead of writing rules on human professionals. Although the end-to-end machine translation (MT) process has recently made considerable progress, the problem of low-resource language pairs and areas still suffers from data scarcity. In this paper, an architectural statistics source-based translation machine (ASS-TM) model has been introduced to deal with the data scarcity problem need to translate small body language. The discriminative learning process (DLP) is employed to enlarge the vocabulary of a system and set of syntactic structures by integrating the synonyms and paraphrases obtained in corpus training. The iteration pipelines for the integration and combination of various generation models using an effective decoding framework. Symmetric context-free grammar (SCG)is implemented to extract a translation memory that includes the conceptual relationships between the two component’s structures. The simulation analysis is performed based on accuracy and efficiency, proving the proposed framework’s reliability of 97.3%.