Curve-fitting of frequency distributions of dependency distances in a multi-lingual parallel corpus