Java makes me cringe. But my understanding is Java programmers use IDE's with autocomplete. They don't do much typing.
As for your pipe comment, are you a whiz with passing file descriptors? What do you think about chaining programs together in this way? Would that solve your pipe needs?
Maybe you could accomplish what you want using lex? The general paradigm is pattern-action, just like awk. Lots of flexibility in what you can create. You could create your own interpreter.
I like writing simple filters. I guess I have succombed to the UNIX voodoo. I think of everything as a stream of text.
I'd be interested to take a crack at your "capture group" problem, if you can give me an example of what it looks like.
/the number is ([0-9]+)/ { print $1; }
perl -pe 'print "$1\n" if /the number is ([0-9]+)/'
Not sure what you mean about lex; if I decide to go ahead and write my ideal shell I could use it, but that's A Project, a large part of which is designing the syntax and deciding what commands to implement. And I'm lazy. :)
Dunno about sed - once I wrote a Brainfuck interpreter in it for fun, but I see it as a Turing tarpit - for things more complicated than "/a/s/a/b/" I prefer awk, though I wish awk had the succinct "s/a/b/", and I wish either of them had a quick way to extract a capture group from a match. (vi was a teenager when I was born, so I personally don't care about adhering to ed.)
I've never tried any APL-like languages and am therefore somewhat scared of them... but most of the programs I write outside of the interactive shell are non-throwaway and/or the kind of glue code that uses lots of random system functions, which decreases the value of extreme terseness. (But I prefer anything over Java and dislike the inflexibility of languages without macros, so perhaps it's only logical for me to go ahead and learn an array language.)