selecting text in multilanguage document

1. Press Ctrl+H to open the Replace dialog.
2. Select the Find tab of the dialog.
3. Click More>> to expand the dialog.
4. Click Format, then Language.
5. Select the language of the text you want to copy/paste and click OK.
6. Click Find In and select Main Document (you can do other layers later as necessary). This will select all the text in the document in the selected language.
7. Click the title bar of the document to return focus to the document without deselecting the text.
8. Press Ctrl+C to copy (or Ctrl+X to cut) the selected text.
9. In a new document, press Ctrl+V to paste. Each run of text will be pasted as a separate paragraph.
10. Repeat as required for other languages.

If your content uses Arabic and English proofing languages, or at least has one lot of text with a specified proofing language and the remainder set to 'No proofing', you can use Find Replace, to do this.

For example, suppose you have Arabic as a proofing language. In that case, you can extract the English text by using Find/Replace to Find all Arabic text and delete it (by not specifying anything, or perhaps just a paragraph break as the replacement). Conversely, to you can extract the Arabic text the same way (if you have English as a proofing language) or, if you only have Arabic as a proofing language, by first hiding all text, then using Find/Replace to Find all hidden Arabic text and make it visible. Having done that you can use Find/Replace to Find all remaining hidden text and delete it.

Failing either of the above, you could use a wildcard Find/Replace to:

find & delete all text with an ASCII value of 33-255 (which is the character range within which English text will be found), thus leaving you with the Arabic text; or

by first hiding all text, then using Find/Replace to Find all hidden text with an ASCII value of 33-255 and make it visible. Having done that you can use Find/Replace to Find all remaining hidden text and delete it. this will leave you with the English text.

This latter approach would probably require some fine-tuning to minimize false matching & non-matching.