A filter for syntactically incomparable parallel sentences

Abstract

Massive automatic comparison of languages in parallel corpora will greatly speed up and enhance the development of a theoretical model of the syntactic properties all languages have in common and the range and limits of syntactic variation. When extracting syntactic differences from parallel corpora it is essential only to compare sentence pairs that are sufficiently similar syntactically. We explore four measures to automatically filter out syntactically incomparable sentence pairs: the Damerau-Levenshtein distance between POS- tags, the sentence-length ratio, the graph-edit distance between dependency parses, and a combination of the three in a logistic regression model. For all approaches we experiment with ignoring specific functional material and other parameters. We evaluate the filters on three labelled datasets consisting of sentence pairs that are and that are not syntactically comparable. Results suggest that the dependency-parse filter is the most stable throughout language pairs, while the combination filter achieves the best results.