Multiword Expressions and Machine Translation
Arvi Hurskainen · 2008
In this paper we approach the multiword expressions (MWE) from the viewpoint of rule-based machine translation (MT). Rather than trying to find a theoretical definition to the concept, we look at it as a problem of translation result. Because rule-based MT relies on a detailed description of the source language, each translation mistake can be traced and eventually corrected. A large part of translation mistakes can be attributed to inadequate handling of MWEs. The discussion is carried out in the context of SALAMA (Swahili Language manager), which, inter alia, translates text from Swahili to English. Various types of MWEs are discussed and solutions for handling them are introduced. The performance of SALAMA in identifying MWEs was tested. Also the distribution of various types of MWEs in text is presented. Key Words: multiword expressions, rule-based machine translation, Constraint Grammar, Swahili language, automatic dictionary compilation