Creating a multilingual collocations dictionary from large text corpora