This paper first presents a theoretical interpretation of the translation process. It then analyzes existing machine-translation research strategies and points out that some of the generally accepted principles of these strategies are not optimal. Finally, an alternative strategy is proposed, based on the author's theoretical position and research results.