Computational linguistics

A candidate sentence receives higher probability from a language model than another candidate. What can this comparison establish on its own?