Phrase-level English to Sinhala machine translation with multi-agent approach

Translating phrases including noun phrases, verb phrases, and preposition phrases has been identified as a key subtask of machine translation. Various techniques have been used to translate source language phrases into target language phrases including rule-based and machine-learning techniques. These phrase-level machine translation techniques have considerably increased the quality of the machine translation. However, human translation is considered as the perfect language translation so far. In view of this, we propose a novel approach to machine translation that is inspired by phrase-based and multi-agent approaches. The approach is stimulated by the fact that how humans translate a sentence with psycholinguistic parsing. The approach has been tested with a multi-agent system named EnSiMaS that translates an English sentence into Sinhala. The EnSiMaS has been implemented through the MaSMT framework. This paper presents how English phrases are translated to Sinhala through the multi-agent approach. Each English phrase in the input sentence is assigned as an agent. According to the availability of the Sinhala words, phrase agents are capable of generating multiple Sinhala phrases with support of the Sinhala morphological and semantic system. These agents consider the structure of the phrase, semantic features of the words, and thematic roles of the existing phrase to generate the Sinhala translation. Further, phrase agents are capable of classifying each generated Sinhala phrase according to its context and usage. Communicating with other related agents, phrase agents identify the most suitable Sinhala phrase form generated Sinhala phrases. The system has been tested with the different types of phrases and the successful results were obtained.