Google找到了

Google found this, the gist of which is: it's hard for malformed documents because parsing requires synthesizing valid tokens that don't have any corresponding input. It's possible for valid documents, but most parsing libraries don't support it.