Averaging truth values in fuzzy linguistic logic