Hybrid neural machine translation with statistical and rule based approach for syntactics and semantics between Tolaki-Indonesian-English languages

Abstract

Machine Translation (MT) incorporates syntax lexical extraction and semantics to predict accurate results. Indonesian have many factors compared to English that related with syntax, especially morphophonemic factors in the language study. These factors are influenced by Lexical type and function while effected MT to frequently mistranslate sentences containing these factors. Meanwhile, semantic extraction is heavily reliant on syntaxis extraction results to predict accurate Lexical translations. In this study, we propose a hybrid statistical and rule-based for MT method that can solve syntaxis and semantic Indonesian problems that conducted the Local Languages in it, particularly Tolaki. First, we developed lexical extraction techniques in Statistical and Rule Based Approach to compile into hybrid MT. This lexical extraction technique is divided into three major tasks: morphophonemic extraction, Lexical Function, and Lexical type extraction. Then we forecast each output of forwards and backwards translations. We compare the predicted output to find accurate translations. Following that, we update the Lexical type based on the actual Lexical function for the translation updating process, which we mark as incorrect translation. Finally, we evaluated MT in both directions. As a result, the proposed method received significant evaluation results, with a percentage success of Indonesian-Tolaki to English translation achieved Precision 0.7231; Recall 0.7; F1-measure: 0.7114; Accuracy: 0.7417 and percentage of success English to Indonesian-Tolaki translation Precision: 0.7119; Recall: 0.7167; F1-measure: 0.7143; Accuracy: 0.7083.