SL-TRANS is an experimental spoken language translation system from Japanese to English. It can recognize Japanese speech, translate it to English, and output a synthesized English voice. In this system, a new method of integrating speech recognition and language processing is adopted. This method is composed of three stages, predictive speech recognition using intra-phase grammar, candidate filtering using inter-phrase Kakariuke dependency, and, a unification based language analyzer to which sentence preference mechanism is added. In each stage, appropriate linguistic information is used in a step by step fashion in order to get the most plausible sentence. Some preliminary experiment results for the system are also reported.