4 ms·

I have been quietly advocating for version controlled legislation for a long time. Here in Massachusetts, bills describe what they would change in the text of a

I have been quietly advocating for version controlled legislation for a long time. Here in Massachusetts, bills describe what they would change in the text of a statute directly - a bit like an ed script. Here's an actual example: >SECTION 2. Said section 35 of said chapter 123, as so appearing, is hereby further amended by striking out the words “is an alcoholic or substance abuser”, in lines 17 and 18, and inserting in place thereof the following words:- has an alcohol or substance use disorder. >SECTION 3. Said section 35 of said chapter 123, as so appearing, is hereby further amended by inserting after the word “a”, in line 36, the third time it appears, the following word:- qualified. >SECTION 4. Said section 35 of said chapter 123, as so appearing, is hereby further amended by striking out the fourth and fifth paragraphs and inserting in place thereof the following 3 paragraphs:- As someone who attempts to keep informed about changes to the law, this style is a huge obstacle. It often necessitates a lot of manual piecing-together in order to form a complete view of the final change. A simple diff view would make it much easier to understand. I have considered tracking changes to the law in git, including representing bills as branches, as a side project idea, but I determined it would require far more effort than I am willing to put in.

sova 10y ago
Wow sir. That is really a great list of examples! It actually seems [very very] feasible to make a simple system that could automate this based on the language used. It would be a worthwhile endeavor, but like you say, would take a lot of time/effort investment. Perhaps an open-source effort that does this (tracks and updates current laws and shows diffs) could be a worthwhile beginning? I think every senator and representative that has ever had to amend legislation would delight at the thought.
- oftenwrong 10y ago
  I was only considering doing it manually. I don't know how feasible it would be to automate the conversion process. The formatting and language used in these "edit script"-style bills varies considerably, as they are written for humans by humans with no standardisation.
  - sova 10y ago
    Honestly, this seems like one of the more realistic problems NLP could actually solve. Yes there may be many variants, say 100 or even 1000 different structures and vocabularies for updating versions, but a differential neural network where you have inputs (like the pre-amended law) and outputs (like the laws after the "amendments" or version bumps) would actually be perfect for learning what means what and when to do it. It would be the perfect grad project for someone interesting in bridging the gap between computation/machine learning and legislation. Of course, it would be a little tedious setting up the learning (thousands of sets of input cases and output cases) but in the end the findings could be used across the board.

sova 10y ago

Wow sir. That is really a great list of examples! It actually seems [very very] feasible to make a simple system that could automate this based on the language used. It would be a worthwhile endeavor, but like you say, would take a lot of time/effort investment. Perhaps an open-source effort that does this (tracks and updates current laws and shows diffs) could be a worthwhile beginning? I think every senator and representative that has ever had to amend legislation would delight at the thought.

oftenwrong 10y ago
I was only considering doing it manually. I don't know how feasible it would be to automate the conversion process. The formatting and language used in these "edit script"-style bills varies considerably, as they are written for humans by humans with no standardisation.
- sova 10y ago
  Honestly, this seems like one of the more realistic problems NLP could actually solve. Yes there may be many variants, say 100 or even 1000 different structures and vocabularies for updating versions, but a differential neural network where you have inputs (like the pre-amended law) and outputs (like the laws after the "amendments" or version bumps) would actually be perfect for learning what means what and when to do it. It would be the perfect grad project for someone interesting in bridging the gap between computation/machine learning and legislation. Of course, it would be a little tedious setting up the learning (thousands of sets of input cases and output cases) but in the end the findings could be used across the board.