• You don't say so, but your script is hard-coded to handle text that uses CP936 encoding for Chinese. It would probably work with other GB-based encodings as well as Big5, which all use the same basic strategy, but it would go wrong if the input text turned out to be any sort of unicode.
• All the encodings for Chinese (including unicode) have a section of code points for "wide" versions of the ASCII characters: in addition to the single-byte ASCII digits, alphabet, punctuation marks and brackets, there are two-byte renderings for these characters also -- but your code treats all 2-byte characters as "Chinese". (It looks like there's a two-byte comma in the last line of your DATA.)
• The code could be written more simply, especially if you have Perl 5.8.x and convert the text to internal utf8 before applying regexes; depending on what version of Perl you're using, the unicode might slow it down noticeably (probably only a problem with 5.8.0 and 5.8.1), but you gain a lot in clarity and maintainability.

Here's how the code could look if the data is converted to utf8 internally -- I'm also using simpler logic: split the input strings into chunks of ideographic and non-ideographic characters, then re-join the chunks, adding spaces where necessary.

This will produce slightly different output than the code you posted, especially where the input text contains "fullwidth" (2-byte) versions of ASCII characters, but it might be easier to tweak in order to make the spacing come out the way you want.

#!/usr/bin/perl -w

use strict;

# NOTE: use a pipe or redirection to feed input data to this script

binmode( STDIN, ":encoding(cp936)" );
binmode( STDOUT, ":encoding(cp936)" );

# (you could add a command-line option to select
# a different input/output character encoding)

while (<>)
{
# first, convert any "fullwidth" ascii characters to normal ascii
# (ff01-ff5e is the unicode range for "fullwidth ascii", and it
# can be transferred directly to the ascii range 0x21-0x7e):

    tr/\x{ff01}-\x{ff5e}/!-~/;  

# now split into chunks: ideographic vs. non-ideographic
# note that we put capturing parens around the split regex):

    my @chunks = split /(\p{Ideographic}+)/;

# put the chunks back together, adding spaces to non-ideographics as n
+eeded

    my $out = '';
    if ( @chunks == 1 ) {
        $out = shift @chunks;
    } else {
        for ( my $i=0; $i <= $#chunks; $i++ ) {
            $chunks[$i] =~ s/([!-~])$/$1 / unless $i == $#chunks;
            $chunks[$i] =~ s/^([!-~])/ $1/ unless $i == 0;
            $out .= $chunks[$i];
        }
    }
    print $out;
}
[download]