I think you are right, but: how about tools like Excel, which is very non-linear, and yet it seems to be easier to pick up, and less intimidating, for complete newcomers compared to traditional programming?
Actually, I think that this line of thinking holds a lot of promise: building specific tools well-tailored to solving specific types of problems. I think we could easily find models and alternate modes of thinking that outperform text for given domains.
I guess my argument is that as humans language is our most fluent medium for communicating abstractions. In the general case I don't think it will be outperformed.
For one, constructs like loops that we have mostly internalized as natural, can look a bit intimidating to non-programmers (they are not found in regular language either though!). Even among fairly experienced coders, for example in statistical programming, people who come from general-purpose coding background sometimes find loops easier to read and write compared to "vectorized" code, and vice versa.
Thinking about it, the way say much Java code is written is not what I would call a linear language either, although that I suspect might be more a result of path dependence and optimizing for "large teams of replaceable programmers".