SALAMA -  Swahili Language Manager

Main | General info | For users | Dictionary compilation  | Machine translation | Language learning | Tone marking | Two phase method | Was used for | Future_plans | Technical reports 

Swahili to English Machine Translation

Supported by the Academy of Finland, we have been working with text-to-text machine translation from Swahili to English. We use a linguistic approach and make full use of linguistic rules.

Currently we have two different implementations of the morphological description. In the traditional implementation we use a finite state method with two-level description. Recently we have developed a new method for morphological description. In the two-phase method, as we call it, we first perform a meta-level description, and in the second phase we expand it to the final description.  With this new method we have been able to implement also the very complex verb reduplication, which has not been possible with finite state methods.

In disambiguation and syntactic analysis we make use of  Constraint Grammar.

Also multi-word expressions of Swahili are handled with the  Constraint Grammar parser.

Because the word order in Swahili and English is very different, a number of constituent re-ordering rules are needed, as well as rules for handling the presence of some pronouns and articles in English.

The production of correct word-forms in the target language is carried out with transfer rules, which, making use of the linguistic information of the source language, transform the English lexical form into the needed surface form.

English to Swahili Machine Translation

SALAMA is currently expanded to include also English to Swahili machine translation. The English text is first analysed with the Functional Dependency parser of Connexor, applied to English, and then processed into Swahili through several phases. The system makes use of various development environments, including Constraint Grammar rules and regular expressions.

Bible Translation

SALAMA was also adapted to Bible translation. The system was developed and tested using Swahili as source language and Luganda as target language.

More details pdf