Integrating Closed Domain Amharic into Multi-Lingual Translation Grammatical Framework

No Thumbnail Available

Date

2014-06

Journal Title

Journal ISSN

Volume Title

Publisher

Addis Ababa University

Abstract

One alternative approach to multi-lingual translation systems is Grammar-based approach in Grammatical Framework (GF). GF is a grammar formalism which is based on constructive type theory (2). GF mainly targets quality rather than coverage, when applied to build closed domain languages multi-lingual translation applications, it works well and with better quality. The absence of Amharic language from the 72 languages in Google-Translate™, the world number one widely used translation system is one indicator to reveal a broad research has to be made on Amharic NLP. In this study, we have made simple and a bit complex closed domain Amharic multi-lingual translation experiment using a rule based approach in GF. The methodology applied in the experiment involves designing English and Amharic foods grammars, generating active voice sentences as a data source and evaluating the translation accuracy based on sample data from the data source. Our experiment on food domain had passed two levels of experimentation involving simple and a bit complex design, the first covered only sentences of simple active voice singular present tense forms, while in the second level we have introduced paradigms and case handling functions onto the food grammar that expanded the foods grammar translation system capacity to grow by threefold. The advanced experiment involved designing an Amharic and English closed domain foods grammar that generates a set of 32,460 unique plural and singular active voice sentences for each language. Performing a manual translation accuracy test on a randomly chosen 3,000 records accounting 10% of the entire data source, our system achieved a 81% accuracy. The result of our experiment demonstrated the possibility of developing an effective small scale closed domain Amharic multi-lingual translation. However, our research have also discovered developing a robust Amharic multi-lingual translation application in GF requires the full Amharic resource grammar library and its API, thus we recommended further research to be made on improving and completing the partially implemented Amharic resource grammar(1) that currently exists in GF.

Description

Keywords

Amharic into Multi-Lingual Translation Grammatical

Citation