A linguistically-motivated evaluation methodology for unraveling model's abilities in reading comprehension tasks