Tokens will be sorted by their last character first, then by the last but one character etc. This is useful for corpora without POS tags where such sorting helps to sort lines by word ending and group the same parts of speech or the same gramamtical categories together.