WPML xliffs are an old problem that can only be fully solved in Studio with workarounds like creating custom XML filetypes.
Since it is possible to process the CDATA sections by enabling embedded HTML processing, the mess of putting HTML and text in a big CDATA container can easily be avoided.
This just leaves the problem of the segmentation that is "hardcoded" in the xliff files.
Other CAT tools seem to be able to overcome this easily but in Studio we can't seem to break up the huge "trans-units" by "sentence" (I think it may have something to do with it being converted to SDLXLIFF first).
This generates huge TUs that are hard to work with and not very useful in TMs.
What I suggest is either:
1- Creating a new filetype based on the existing XLIFF that has CDATA embedded processing enabled by default and that allows for breaking up segments per-sentence;
2- Adding the ability to customize the segmentation rules for the existing XLIFF filetype (instead of it only being available at the TM settings level, where it doesn't seem to be able to do anything) so that we can segment these anoying files by sentence.
Wordpress is a very popular CMS system and WPML the most popular translation plugin, so I should think this to be a concern for the major CAT tool out there, especially since others seem to already be focusing on it.