Currently, when the artificial alignments are created for a termbase entry, they undergo a different tokenization process compared to when the entries are added to the tb model. This causes discrepancies with alignment length, whenever there are unicode white spaces in the text.