Abstract
In Uganda, Luganda is the most dominant local language spoken all over the country with over 6 million speakers thus, learning Luganda would enable English speaking foreign traders, tourists and Non- Governmental Organisations (NGOs) to get the best out of their dealings with the Luganda speakers. Leading Neural Machine Translation (NMT) Systems such as Google Translate (GT), Microsoft Translator and those based on Statistical Machine Translation (SMT) cannot satisfactorily support Noun Phrase translations between major and minor languages such as the English language and Luganda pair because of inadequate digital resources for minority languages. Machine Translation (MT) tools which are potentially affordable and nvenient anguage learning options cannot help majority language speakers to learn minority languages due to diversities in the syntax and semantic structures of the languages but cheaper and more effective MT approach between major and minor languages is Rule- based Machine Translation (RBMT) which involves harvesting a language pair??s linguistic knowledge. Design Science Research Methodology (DSRM) allows for continuous refinement of a translation model and its implementation through rototyping techniques, Document Analysis and Focus Group Discussions to incorporate new translation rules into the model and use Holdout validation with Human Evaluation to test the model output. Deterministic Finite Automaton (DFA) automates the Noun Phrase Translation model, Java Formal Languages and Automata Package (JFLAP) designs the DFA while Python is the programming language. The bilingual dictionaries are implemented as Coma Separated Values (CSV) files, the Natural Language Tool Kit (NLTK) supports Natural Language Processing (NLP) tasks such as Part-Of-Speech (POS) tagging, parsing and tkinker develops a Graphical User Interface (GUI) for the MT application. Py2Exe creates an executable file from the python codes and Nullsoft Scriptable Install System (NSIS) builds the window installer for the application. The translation model does not cover complex noun phrases consisting of other phrases such as prepositional phrases and focuses on the commonest noun phrase pattern of the Pre-modifier + Head structure. Luganda as a noun centric language with 10 noun classes does not have standalone articles but exist as prefixes and noun modifiers are determined by noun??s grammatical number and its class. Evaluation results offers clear and unambiguous translation of English to Luganda noun phrase which invariably facilitate tutoring and teaching of Luganda. This research work aids the protection of both the language and the culture from a possible extinction. As a matter of fact, the dominancy of English in African societies has its advantages but it has damaged African local languages values and virtues that is language is culture therefore, if a language dies, culture dies and when a culture dies, its people die.
Keywords