Substring Alignment using Suffix Trees

Abstract. Alignment of the sentences of an original text and a translation is considerably better understood than alignment of smaller units such as words and phrases. This paper makes some preliminary proposals for solving the problem of aligning substrings that should be treated as basic translation unites even though they may not begin and end at word boundaries. The proposals make crucial use of suffix trees as a way of identifying repeated substrings of the texts that occur significantly often. It is fitting that one should take advantage of the few occasions on which one is invited to address an important and prestigious professional meeting like this to depart from the standard practice of reporting results and instead to seek adherents to a new enterprise, even if it is one the details of which one can only partially discern. In this case, I intend to take that opportunity to propose a new direction for a line of work that I first became involved in in the early 1990’s [3]. It had to do with the automatic alignment of the sentences of a text with those in its translation into another language. The problem is nontrivial because translators frequently translate one sentence with two, or two with one