Abstract
Sequence alignment, which involves aligning elements of two given sequences, occurs in many natural language processing (NLP) tasks such as sentence alignment. Previous approaches for solving sequence alignment problems in NLP can be categorized into two groups. The first group assumes monotonicity of alignments; the second group does not assume monotonicity or consider the continuity of alignments. However, for example, in aligning sentences of parallel legal documents, it is desirable to use a sentence alignment method that does not assume monotonicity but can consider continuity. Herein, we present a method to align sequences where block-wise changes in the order of sequence elements exist. Our method formalizes a sequence alignment problem as a set partitioning problem, a type of combinatorial optimization problem, and solves the problem to obtain an alignment. We also propose an efficient algorithm to solve the optimization problem by applying column generation.
