Polish tea is Czech coffee : advantages and pitfalls in using a parallel corpus in linguistic research