Unfortunately this could happen in some edge cases, and I can’t write a script to automatically fix it. You’ll need to manually review your exp_publisher_phrase_data table and remove the duplicate rows. You can run this query to identify the duplicates, then determine which rows to delete.

select distinct phrase_id, site_id, lang_id, `status`, is_auto, edit_by
  from exp_publisher_phrase_data as t1
 where exists (select * 
                 from exp_publisher_phrase_data as t2 
                where t2.id != t1.id
                  and t2.phrase_id = t1.phrase_id
                  and t2.site_id = t1.site_id
                  and t2.lang_id = t1.lang_id
                  and t2.`status` = t1.`status`
                  and t2.is_auto = t1.is_auto
                  and t2.edit_by = t1.edit_by
               )

I wish I had a better answer for you but I can’t create a query to fix it automatically because I don’t know which row contains the correct translation for your site.