Probability values

[RobH123]

My experience in dealing with human language shows that sometimes we are pretty sure that that's what a person meant. So if the text says, Jesus answered and said, "That." then we can assign "That" to Jesus with pretty-much 100% confidence. But I'm sure that in producing the Glyssen and FCBH data sets, that there's also ambiguities. If the text says, "He answered and said, "This." maybe we're 90% sure that it's Peter or maybe there's cases where we're not really sure at all??? I know there's a case where Matthew (IIRC) wrote "Isaiah said..." but it's in Jeremiah or something.

So just asking: if we added probability/reliability data to the data files, could a Bible translation assistant software package later use that to say highlight passages that need a bit more manual attention?

[tombogle]

Currently in Glyssen there are two ways to get to ~100% confidence:

1. If a verse has only one known speaking character, any quoted text in that verse is assumed to be that character.
2. In a verse with two or more known speaking characters, if the order of the speaking parts and narrator parts correspond 1-for-1 to the reference text being used, then the matchup is assumed to be correct.

Obviously, neither of these is guaranteed to be perfect because we could be missing a needed entry for a verse and there's always a slight chance that in some language we'll happen to get an exact match even though the characters spoke in a different order.

For passages where we know that things are a bit more ambiguous, we have a special character ID "Needs Review" that will automatically force the passage to require a user to look at it. Unfortunately, if the person doing the disambiguation does not know the target language (as is almost always the case with FCBH projects), the scripter can still just guess without really knowing if they got it right. The hope is that the mistake will be caught during the recording process.

For the most part (with the possible exception of your Isaiah/Jeremiah example), if the person doing the markup speaks the language, they should be able to say with near 100% confidence who is speaking. But, yeah, there are a handful of exceptions to that. They don't usually affect the ability to put together a recording script, but they could affect what we'd want to store in the metadata.