This paper describes recent efforts to improve the HMM-LR speech recognition system for continuously spoken sentences. The HMM-LR system has been applied to Japanese phrase recognition and has attained high recognition performance. However, up to now, the system has not been applied to continuously spoken sentence recognition. In this work, several improvements have been made on the system. The first improvement is HMM training with continuous utterances as well as word utterances. In previous implementation, HMMs have been trained with only word utterances. Continuous utterances are included in HMM training data because coarticulation effects are much stronger in continuous utterances. The second improvement is the development of a sentential grammar for Japanese. The sentential grammar was created by combining inter- and intra-phrase grammars, which were developed separately. The third improvement is the incorporation of stochastic linguistic knowledge, which includes stochastic CFG and a bigram model of production rules. The system was evaluated using continuously spoken sentences from a conference registration task that include approximately 750 words. A sentence accuracy of 83.9% was attained in the speaker-dependent condition.