Translation of Clefting Construction in Persian to English Apertium System

Since Persian and English languages differ in several ways including morphology, lexicon, and syntactic structures, this caused machine translation to suffer from numerous problems in translating between these two languages. This paper presents the syntactic challenges of Persian to English Apertium system, a shallow-transfer rule-based machine translation (RBMT) platform, confronted in translating clefted constituents from Persian to English. To this end, 100 sentences containing clefted constituents were gathered in two methods: 1) the contemporary Persian spoken data were collected from radio, television, and daily conversations, 2) the written data were from different Persian grammar books, newspapers, and novels. As the formal words were added to the Apertium dictionaries, it was necessary to pre-process the extracted structures before running them in the Apertium system. To improve the translation performance, new structural transfer rules at chunker level were added to the developed system. The Apertium system were evaluated via word error rate (WER) and position-independent error rate (PER). Besides, these sentences were translated using Google translate as a statistical machine translation system (SMT). The evaluation results showed that the Apertium system could generate closer translation to the reference text comparing that of Google translate due to adding new chunker rules.