Currently, the spec requires declaring full source–target pairs. For 100 languages, that blows up to ~10,000 pairs, which feels a bit excessive. I'm considering listing languages individually instead—both source & target by default, with opt-out flags: ``` <item> <engine jid='translation.shakespeare.lit' name='SYSTRANS 2005 Release 2' source_detection='1'> <!-- By default, each language is both a source and a target. Same-language pairs are excluded. --> <lang xml:lang='en' name='English'/> <lang xml:lang='fr' name='French'/> <lang xml:lang='zh-Hans' source='false'/> <lang xml:lang='ko' target='false'/> </engine> </item> ``` Also, I’m undecided on how best to handle source auto-detection. Which approach makes more sense to you? An engine attribute: source_detection='1' A special entry: <lang xml:lang='und' target='false'/>