Unit testing syntaxes / "paradigms" and their benefits/drawbacks?

I want to develop a unit testing framework for a new programming language and am thinking about how to design its interfaces.

The two most popular schemes seem to be JUnit-like (...Unit) and RSpec-like (e.g. RSpec, Jasmine).

What are the benefits and drawbacks? What other concepts would you recommend, and why?

Do you know of articles discussing the different styles and interfaces in general (as opposed to discussing specific implementations and their shortcomings)?

It is a "script" language, prototypal, single-inheritance with mix-ins, dynamically-typed, optionally parameter-/return-type-hinted, interpreted.