Machine translation for te reo Māori

Abstract

What is Machine Translation? Machine Translation (MT) is a process where computer software is used to translate texts from one natural (or spoken) language to another. Early research centred on two distinct approaches: Rules Based Machine Translation (RBMT) and Statistical Machine Translation (SMT). In simple terms RBMT makes use of large sets of linguistic rules that define languages whereas SMT uses statistical techniques to build language models from large language corpora. Increases in computing power and the amount of language corpora available has meant that SMT had become the preferred option with recent advances in neural networks also being applied to improve the accuracy of SMT. For commercial reasons, this is an area of research that has generated a lot of interest and funding support from some major international computer companies, including Google¹,Microsoft² and Facebook³. Why is Machine Translation important for te reo Māori? A te reo Māori purist may argue that it is not important to focus activities on having a machine undertake translations for te reo Māori; if people want to understand te reo Māori then they should put in the effort to learn the language. This line of reasoning is difficult to argue with. But, from a Māori language activist perspective the value of MT is not so much in the translation of te reo Māori to (say) English, but rather the translation of English to te reo Māori. If this translation can be done efficiently, with low costs, it will assist in the proliferation of te reo Māori into new contexts, new environments and will assist its normalisation in New Zealand's society. At this time, two of the major international companies, Google and Microsoft, have invested significantly in MT for te reo Māori. This paper summarises their endeavours and reports on the quality of translations they have been able to generate.