Abstract:
Undoubtedly, the internet enables easy communication between people with its
affordability and convenience. The demand for instant and accurate access to
information is growing day by day at the present speed of the world. However, language
differences continue to be a strong hindrance to the exchange of information on an
international level primarily due to the fact that human translation takes time and costs
money. Thus, the demand for proper translation from one language to another continues
to grow consistently. In a bid to address these issues, researchers have turned their
attention to machine translation as a way of making the easier transfer of content from
one language into many others possible. Machine translation has therefore become one
of the quickest fields of research, making it possible to bridge the language gap between
cultures.
In Natural Language Processing (NLP), machine translation is accountable for
an important role in web translation services, information extraction, document
retrieval, intelligence analysis and email processing among other applications. This
research presents the first systematic investigation into machine translation for Mon
language, a severely under-resourced language with no prior research in the field. The
lack of linguistic resources such as publicly accessible parallel corpora, monolingual
corpora and preprocessing toolkits is available for Mon, the absence of which has been
the single biggest hindrance to the development of Natural Language Processing (NLP)
applications for Mon. While machine translation has been tried in other Southeast Asia
languages, the Mon language remains relatively unworked on both academic literature
and practical implementation.
The primary objective of this research is to construct and evaluate machine
translation systems for English-to-Mon and Mon-to-English translation. One of the key
contributions of this work is the creation of 25k sentence pair Mon-English parallel
corpus as well as a Mon monolingual corpus. The parallel corpus was built from two
primary sources: the Book of “4-Languages: English, Mon, Burmese and Pali” from
which only the English and Mon sentences were utilized and “English Grammar
Explained in Mon” by Nai Sac Lun from which suitable English-Mon sentence pairs
were extracted. In addition, other English sentences were gathered from the UCSYcorpus to increase the dataset. Data were heavily cleaned and preprocessed for quality
and consistency.