This usually isn't done because integers and decimals are fundamentally very similar, so for addition there is a platonic correct result when you add an integer and a float (the floating point result), and the operation returns that. Then, if you wanted an int, you can cast your result.
That doesn't work if you want to consider string concatenation to be a form of numeric addition. It isn't. There is no ideal result when adding a string to a number, and if you do one when you meant to do the other, there is no way of producing the data you wanted ("11") from the data you have (2).
Perl doesn't have much of a type system. But it solves the common problem of manipulating strings when you meant to manipulate numbers, or vice versa, by not assigning the same operators to these radically different operations, and I think that was the correct choice, even for languages with stricter type systems.
After further consideration, I believe this is incorrect. This usually isn't done because integers are closed under addition, subtraction, and multiplication, so the "promote according to argument type" C behavior doesn't lose accuracy in those cases. There obviously is a platonic correct result when you divide two integers, but it is not in general an integer, so the C behavior introduces bugs (explaining the popularity of divide_and_return_float() -- I don't care what types the arguments were, as long as I get the result I'm looking for).
I mostly agree with your point here, but in this specific example I actually think Perl did the right thing, where
are all 2, and are all 11. Letting you specify the result you want by the function you call basically solves that whole problem.