We have looked at the evidence we use to justify postulating particular features and their values in particular languages. That evidence, especially from agreement and government, is what is available to the child learner and to the linguist. Given the genuine difficulties of analysis we have found, we might wonder what more general claims we could make when we turn to comparison across languages. A simple suggestion (Zwicky 1986a: 988) is that ‘universal grammar should permit only a finite number of attributes and values – indeed […] universal grammar should provide finite lists of the attributes and values available for service in a particular grammar’. This idea has since been discussed within Minimalism (thus Chomsky 2001: 10 appears to take a similar view). Zwicky points out the difficulty with the approach, as put to him by Gerald Gazdar:

Zwicky's suggestion, then, is that morphosyntactic features (‘properties’ in his terms) always have a semantic core, and it is this core which allows comparison across languages. They are often partly arbitrary, but never fully arbitrary. We return to this issue in §5.1.