On Machine Translation for Persian

Abstract

This paper is an attempt to clarify the aims of MT and highlight some of the reasons why Persian MT is now a matter of necessity for Iranians in terms of its scientific, philosophical, social, and commercial benefits. After giving a short history of MT, I will illustrate that translation quality is not all that matters in the 21st century. I will show that there are some other factors involved that make MT an attraction for translators and non-translators the same. I will briefly review a few types of MT systems. From the rule-based approaches, transformer and linguistic knowledge (LK) architectures, and from the empirical approaches, example-based and statistical architectures are reviewed to give the reader an idea about how MT engines work. Linguistic and/or computational problems of each approach are discussed in short. It is argued that MT systems are better to use a combination of various architectures rather than sticking to a single approach. As an example of such architectures, Bond’s (2002) proposal for a Multi-Pass Machine Translation (MPMT) is reviewed. Wherever possible, technicality is avoided and references are made to the Persian language so that the material is more attractive and tangible for non-MT practitioners.