An incremental translation system for spoken language, which can produce the target language synchronously with a source language input, has been proposed to far. Producing so much source language structures in the CFG-based parsing stage, however, the system has two fatal problems: 1) difficult to work in real-time, and 2) difficult to generate correct results. This paper introduces a special form of CFG, called M normal-form, and describes a method of converting any CFG to M normal-form. Since it produces less structures in the incremental parsing, the system using M normal-from of CFG can generate translation results with higher quality. Experiments with a prototype system have shown the effectiveness of the conversion for fine-grained incremental machine interpreting.
展开▼