Refinements in BTG-based Statistical Machine Translation

Bracketing Transduction Grammar (BTG) has been well studied and used in statistical machine translation (SMT) with promising results. However, there are two major issues for BTG-based SMT. First, there is no effective mechanism available for predicting orders between neighboring blocks in the original BTG. Second, the computational cost is high. In this paper, we introduce two refinements for BTG-based SMT to achieve better reordering and higher-speed decoding, which include (1) reordering heuristics to prevent incorrect swapping and reduce search space, and (2) special phrases with tags to indicate sentence beginning and ending. The two refinements are integrated into a well-established BTG-based Chinese-to- English SMT system that is trained on largescale parallel data. Experimental results on the NIST MT-05 task show that the proposed refinements contribute significant improvement of 2% in BLEU score over the baseline system.