On avoiding paradoxes in assessing inter rater agreement