Using sed, how can a regular expression match Chinese characters?

Question

I decided to post a question, after spending quite some time and still not figuring out the problem. Also read a bunch of seemingly related posts, none really fit my simple (?) problem.

So I have a possibly large text file (>1000 lines) that contains Mandarin Chinese chars, with a sample line like:

"ref#2-5-1.jpg#2#一些 <variable> 内容#pic##" (the Chinese just means "some content"). 

All that needs to be modified is that a space should be inserted between each character, if there is not one already:

"ref#2-5-1.jpg#2#一 些 <variable> 内 容#pic##".

I started naively with straightforward stuff like the following, but there is no match at all:

sed -e 's/\([\u4E00-\u9fff]\)/\1 /g' <test_utf_sed.txt > test_out.txt

where 4E00-9fff are supposed to be the code range for Mandarin Chinese. Unamazingly, this has not worked, so I also had wanted to try

sed -e 's/\([一-龻]\)/hello/g' <test_utf_sed.txt > test_out.txt

This failed because my bash cannot display (?) the "一" character.

Then I did some basic test, which failed as well:

sed -e 's/\(\u4E00\)/hello/g' <test_utf_sed.txt > test_out.txt //一
sed -e 's/\(\u4E9B\)/hello/g' <test_utf_sed.txt > test_out.txt //些

Same with another notation for utf encoding (found here on stackoverflow):

sed -e 's/\(\u'U+4E00\)/hello/g' <test_utf_sed.txt > test_out.txt

1) As tool for dealing with double byte chars, is sed the right choice at all?

2) Is sed able to handle unicode at all, or do I need a special switch?

3) I am not looking for a workaround solution like this:

step1: insert space after each character 
  //like 's/\(.\)/\1 /g')
step2: remove space after each chacter which is not a Chinese character 
  //like 's/\([a-zA-Z0-9]\) /\1/g')

I know how to do this but it is unelegant and error-prone. This must be possible using utf-8 in regex in sed.

4) My environment is bash-3.2 on a MacOS 10.6.8 (oldish OS).

5) If you know of any pointers to some open regEx-onliners as library dealing with Chinese text or language processing, it would be great to share.

Thanks a lot in advance, your help is much appreciated!

Answer 1

Perl has pretty good support for dealing with Unicode. That might be a better bet for your task than sed. This one-liner works like your first sed example:

perl -CIOED -p -e 's/\p{Script_Extensions=Han}/$& /g' filename

The -CIOED tells perl to do its I/O in utf8. -p runs the given code once for each line of the input file, then prints the result. -e specifies a line of Perl code to run. See the documentation on command-line arguments for more.

The regular expression uses named ranges to identify the characters to match.

You might also want to read the Perl Unicode documentation.

Answer 2

sed doesn't understand \u escape sequences (apparently). I don't know if bash-3.2 does either, but I think it does; if so, you could write

sed $'s/\u4E9B/hello/g'

but you still wouldn't be able to do the range specification.

However, by translating to UTF-8 by hand, you could arrive at the following extended regular expression which will, I believe, match any UTF-8 sequence for a character in the range U+4E00...U+9FFF:

(\xe4[\xb8-\xbf][\x80-\xbf]|[\xe5-\xe9][\x80-\xbf][\x80-\xbf])

(But the character ranges will only work if you invoke sed in a single-byte locale, preferably the C locale.)

With GNU sed, you get extended regular expressions if you provide the -r flag. With MacOSX I believe you need the -E flag. So you could try:

LANG=C sed -E \
       $'s/(\xe4[\xb8-\xbf][\x80-\xbf]|[\xe5-\xe9][\x80-\xbf][\x80-\xbf])/\\1 /g' \
       <test_utf_sed.txt >test_out.txt

(The above lets bash handle the \x escapes. If you leave out the $, then sed will handle the \x escapes, but you'll have to change the substitution from \\1 to \1. I don't have a Mac, nor do have the old version of bash, so I really don't know whether your sed does hex escapes or not; I'm pretty sure that your bash will, but I can't guarantee it.)

---

By the way, it's not that difficult to get the utf-8 encodings for those characters; I did it with a little copy-and-paste from the original post. Eg.:

$ hd <<<"一些"
00000000  e4 b8 80 e4 ba 9b 0a                              |.......|

It helps to know that the entire range of plane 0 ideographs (U+4E00...U+9FFF) have three-byte codes, so that 一 is E4 B8 80 and 些 is E4 BA 9B. (The 0A is, of course, a line-end.)