Re:I'm a Judge and I Think Criminal Court Is Horrifying
by tgflynn on Sunday August 14, 2016 @12:30AM
(#12284271)
What form does the "not directly intelligible" transcript take ? If it's possible to obtain that maybe there could be a way to generate an intelligible transcript automatically.
Re:I'm a Judge and I Think Criminal Court Is Horrifying
by Spooky23 on Sunday August 14, 2016 @01:32AM
(#12284390)
It has to be processed and formatted. The reporter uses a shorthand system that needs to dealt with, and you need to be able to figure out which party is which. Also, I'm pretty sure the reporter just blasts on through, so it would be difficult to distinguish between cases without some human intervention. You probably have hundreds of people getting arraigned in one session in the Bronx.
Digital audio can be more difficult. Automated transcription sucks, and the costs for human annoyed machine transcription are very high.
Re:I'm a Judge and I Think Criminal Court Is Horrifying
by dredmorbius on Sunday August 14, 2016 @12:46AM
(#12284299)
I don't know. I do know that preparing a relatively short transcript (in terms of court proceedings) can be a nontrivial request -- e.g., it's not simply pressing "generate" on some system.
That said, the records I've seen have been accurate.
Re:I'm a Judge and I Think Criminal Court Is Horrifying
by mLuby on Sunday August 14, 2016 @12:34AM
(#12284278)
Could potentially use cases for which both proceeding notes _and_ transcripts are available to train a sentiment analysis system to interpret the proceeding notes directly.
Re:I'm a Judge and I Think Criminal Court Is Horrifying
by dredmorbius on Sunday August 14, 2016 @12:44AM
(#12284293)
A further complication -- my understanding, this may be incorrect: court reporters are in at least some cases independent contractors. They hold rights to their transcripts. You're not actually dealing with the court in requesting them.
Your AI solution would 1) tremendously reduce consts and 2) almost certainly remove a large share of the income of these independent reporters.
Mind: the problem could be resolved by making reporters employees of the court.
Straight off the top, one problem is that transcripts aren't routinely generated, AFAIU. Yes, there's a court reporter taking down the proceedings, but what's generated isn't itself directly intelligible. Preparing an actual transcript from the proceedings for a small (2-day) case can run close to a thousand dollars. Doing this at scale would likely be highly expensive.
You bring to mind an alternative though: an automatic, random, appeal of a select number of cases to a review system of some sort. This might initially be a completely random sampling across all courts, but could then be subject to some sort of stratified sampling, with more cases subject to review from areas highlighted as problematic, and periodic reweighting.