I'd definitely argue that no language ever will "feel natural" unless it fundamentally changes the metaphors of computation. Humans just don't think in sufficiently "hard-boundary" terms for programming as currently phrased to come naturally. Granted, culture has made up for a lot of that - in the UK, children are expected to be able to perform arithmetic, which presumably helps a lot in instilling the appropriate style of yes-no exact thinking - but I don't think the programming mindset aligns well at all to what was in the ancestral environment. Our minds are classifying machines built to do Just Well Enough in a world where everything is shades of grey; programming is the art of making the world concretely align to a well-specified model.
When we tell a person to do something, we always give them some freedom in how to do it. When we tell a computer to do something, we don't. If one were to instruct a human being to go to the grocery store in absolutely full micromanaged detail, it would look equally unnatural ("Lift your right knee. Swing your right foot forward. Put down your right foot...") Sort of like that videogame QWOP.
This is because reality is not intuitive, too.
Take natural numbers. They are utterly intuitive. Three-year-olds can grasp them. But as you try to calculate more and more complex things in the real world, you discover the need in counter-intuitive concepts like zero, negative numbers, fractions, even the aptly named irrational numbers. If you try to calculate √3 using the intuitive natural numbers, your solution will be rather approximate.
Same thing with many "intuitive" languages, from Basic-80 to ES6. You can solve a number of problems with them, but many solutions end up inexact and full of holes. In a lot of cases, this is acceptable, or seen as acceptable.
But when you need tools for a precise solution, you have to study some "advanced concepts" (another word for math) and take a language like Haskell which support them. Sometimes you end up with something as unwieldy as your average three-story analytic formula, but this is often not a shortcoming of the language but the nature of your problem domain, described correctly.