Incorporating Bigram Constraints into an LR Table

In this paper, we propose a method to construct a bigram LR table to incorporate bigram constraints into an LR table. An LR table which incorporates bigram constraints is called a bigram LR table. Using the bigram LR table, it is possible for a GLR parser to make use of both bigram and CFG constraints in natural language processing. A method for constructing bigram LR tables is proposed. Applying the resultant bigram LR table to our GLR method has the following advantages: 1. A language model utilizing a bigram LR table has lower perplexity than a bigram language model, since local constraints (bigram) and global constraints (CFG) are combined in the single bigram LR table at the same time. 2. Bigram constraints are easily acquired from a given corpus. Therefore data sparseness is not likely to arise. The former advantage leads to a reduction in complexity, and as the result, produces better performance for GLR parsing. Our experiments demonstrate the effectiveness of our method.