Survey of Current Paradigms in Machine Translation

摘要

This paper is a survey of machine translation (MT) research from the United States, Europe, and Japan. A short history of machine translation is presented, followed by an overview of current research and representative examples of a wide range of different approaches adopted by machine translation researchers. These examples are described in detail along with a discussion of the practicalities of scaling up such approaches for operational environments. In support of this discussion, issues in, and techniques for, evaluating machine translation systems are discussed. While a number of MT surveys have been published, this one discusses a wide range of current research issues in light of results obtained from a survey and evaluation project conducted by Mitre. During this project, Mitre evaluated 16 MT systems and also studied 7 U.S. MT systems. Because a number of innovative MT approaches have surfaced since the completion of the Mitre study, the authors also include discussions of more recent research paradigms. Section 2 provides a brief description of the history of MT. Section 3 discusses the types of challenges (both linguistic and operational) that one must consider in developing a MT system. Section 4 describes three architectural designs that are used for MT. Following this is a comparison of translation systems along the axis of research paradigms (section 5); these include linguistic, nonlinguistic, and hybrid approaches. Section 6 is a discussion of the challenges of evaluating a MT system, and some approaches to doing so. A 233-item bibliography is included.