| Resumen |
In this paper, we present an optimization algorithm for finding the best text alignment based on the lexical similarity and the results of its evaluation as compared with baseline methods (Gale and Church, relative position). For evaluation, we use fiction texts that represent non-trivial cases of alignment. Also, we present a new method for evaluation of the algorithms of parallel texts alignment, which consists in restoration of the structure of the text in one of the languages using the units of the lower level and the available structure of the text in the other language. For example, in case of paragraph level alignment, the sentences are used to constitute the restored paragraphs. The advantage of this method is that it does not depend on corpus data. |
