Considering the Towards the Anti-Mac article:
Radagast made a follow up and adressed ncm about a possible pseudo-english parsing shell. There is a fatal inherent flaw to such a shell, but it also made an old idea of mine resurface. Thus:
The name says it all: "pseudo-english". Obviously you'd want pseudo-german as well. Even considering a pseudo-japanese one is daunting to say the least. Porting such a shell to the different semantic structures (or tendencies towards such structures) of different natural languages is not practical.
Lets imagine some kind of scratchboard where you could mix representations of various acts (picked from tables divided according to the nature of the various tasks). Just like using redirection or variables in a shell script, the user would connect the representations using arrows. Thus forming a directed graph -- possibly a cyclic one with conditions for termination.
The acts would require the user to fill in parameters (in popups or whatever) just like a shell command does.
When a graph is constructed it could be saved as a new act, thus adding to the table of pickable acts.
Acts (or graphs of acts) could then be used as traditional scripts that the user executes manually or assigns to cron-like or event-driven agents. The acts could however also be used interactively by simply executing some act and adding subsequent ones as necessary. Backtracking would be possible since a session with history would exist naturally in form of the directed graph formed by the acts.
Finally, only the program managing the sets of acts (and the popups for various acts) would require translation between different natural languages. The acts themselves or the machinery underneath would require no knowledge of natural(-like) language to function.
The construction of original acts would still require skill in putting together shell scripts, but it wouldn't be a prerequisite for using existing ones.
How about that?