words-to-data 0.3.0

Convert Legal Documents Into Diffable Data Structures
Documentation

Words To Data - Convert Legal Documents Into Diffable Data Structures

CI

Overview

words_to_data parses US Code titles and Public Laws (bills) from USLM XML format, providing structured access to legislative text, the ability to track changes between document versions, and tools for annotating how bills amend existing law.

Available for both Rust and Python with high-performance Rust core and ergonomic Python bindings via PyO3.

Features

  • Dataset-centric workflow - Manage versioned legal documents, bills, and annotations in a single structure
  • Parse USC and Public Law documents - Extract hierarchical structure from USLM XML files
  • Rich text content - Capture heading, chapeau, proviso, content, and continuation fields
  • Bill amendment extraction - Identify USC references and amending actions from bills
  • Hierarchical diffing - Compute word-level differences between document versions
  • Congress data integration - Fetch bill metadata and text from Congress.gov API
  • Python bindings - Full API access from Python with PyO3

Installation

Rust

Add to your Cargo.toml:

[dependencies]
words-to-data = "0.3.0"

Python

pip install words-to-data

Note: Pre-built wheels are available for Linux x86_64. Other platforms will build from source (requires Rust toolchain).

Getting Data

Quick Start

Dataset Workflow

The Dataset is the primary abstraction for working with versioned legal documents. It holds document versions, bills, and annotations together.

Rust:

use words_to_data::dataset::{Dataset, DatasetMetadata};
use words_to_data::uslm::bill_parser::parse_bill_amendments;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let metadata = DatasetMetadata {
        name: "Tax Code Changes".to_string(),
        description: "Tracking Title 26 changes".to_string(),
        author: "Author".to_string(),
        source_urls: vec![],
        license: "MIT".to_string(),
        version: "1.0.0".to_string(),
    };
    let mut dataset = Dataset::new(metadata);

    // Add document versions
    dataset.add_uslm_xml("path/to/old.xml", "2025-07-18", Some("Before".into()))?;
    dataset.add_uslm_xml("path/to/new.xml", "2025-07-30", Some("After".into()))?;

    // Add bill
    let bill = parse_bill_amendments("119-21", "path/to/bill.xml")?;
    dataset.add_bill(bill);

    // Compute diff
    let diff = dataset.compute_diff("2025-07-18", "2025-07-30")?;

    // Navigate to specific section
    if let Some(s174a) = diff.find("uscode/title_26/subtitle_A/chapter_1/subchapter_B/part_VI/section_174/subsection_a") {
        for change in &s174a.changes {
            println!("{:?}: {} → {}", change.field_name, change.old_value, change.new_value);
        }
    }

    // Save dataset
    dataset.save("my_dataset.json")?;
    Ok(())
}

Python:

from words_to_data import Dataset, DatasetMetadata, parse_bill_amendments

metadata = DatasetMetadata(
    name="Tax Code Changes",
    description="Tracking Title 26 changes",
    author="Author",
    source_urls=[],
    license="MIT",
    version="1.0.0",
)
dataset = Dataset(metadata)

# Add document versions
dataset.add_uslm_xml("path/to/old.xml", "2025-07-18", label="Before")
dataset.add_uslm_xml("path/to/new.xml", "2025-07-30", label="After")

# Add bill
bill = parse_bill_amendments("119-21", "path/to/bill.xml")
dataset.add_bill(bill)

# Compute diff
diff = dataset.compute_diff("2025-07-18", "2025-07-30")

# Navigate to specific section
s174a = diff.find("uscode/title_26/subtitle_A/chapter_1/subchapter_B/part_VI/section_174/subsection_a")
if s174a:
    for change in s174a.changes:
        print(f"{change.field_name}: {change.old_value} → {change.new_value}")

# Save dataset
dataset.save("my_dataset.json")

Download from Congress API

Bills can be automatically fetched with additional metadata from the congress.gov API

Rust:

use words_to_data::congress::CongressClient;
use words_to_data::dataset::{Dataset, DatasetMetadata};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create client with API key (get from https://api.congress.gov/sign-up/)
    let client = CongressClient::new(std::env::var("CONGRESS_API_KEY")?);

    // Download bill data (XML + sponsors + members)
    let download = client.download_bill("119-hr-1")?;

    // Create dataset and load bill
    let metadata = DatasetMetadata {
        name: "HR 1 Analysis".to_string(),
        description: "Tracking HR 1 amendments".to_string(),
        author: "Author".to_string(),
        source_urls: vec![],
        license: "MIT".to_string(),
        version: "1.0.0".to_string(),
    };
    let mut dataset = Dataset::new(metadata);

    // Load bill into dataset (parses XML, stores sponsors/members)
    let bill_id = dataset.load_bill_download(&download)?;
    println!("Loaded bill: {}", bill_id);

    // Access bill data
    let bill = dataset.get_bill(&bill_id).unwrap();
    println!("Amendments: {}", bill.amendments.len());

    Ok(())
}

Python:

import os
from words_to_data import CongressClient, Dataset, DatasetMetadata

# Create client with API key
client = CongressClient(os.environ["CONGRESS_API_KEY"])

# Download bill data
download = client.download_bill("119-hr-1")

# Create dataset and load bill
metadata = DatasetMetadata(
    name="HR 1 Analysis",
    description="Tracking HR 1 amendments",
    author="Author",
    source_urls=[],
    license="MIT",
    version="1.0.0",
)
dataset = Dataset(metadata)

# Load bill into dataset
bill_id = dataset.load_bill_download(download)
print(f"Loaded bill: {bill_id}")

# Access bill data
bill = dataset.get_bill(bill_id)
print(f"Amendments: {len(bill.amendments)}")

Core Concepts

Dataset

The Dataset is the primary abstraction for working with versioned legal documents:

  • DatasetMetadata: Name, description, author, license, version
  • VersionSnapshot: A document tree at a specific point in time
  • Bills: Parsed bill data with extracted amendments
  • Annotations: Links diff paths to bill amendments with verification status

Use Dataset to load documents, compute diffs, and build training data for ML models.

USLM Elements

Documents are represented as trees of USLMElement structures. Each element contains:

  • ElementData: Metadata, text content, and identification
  • Children: Nested child elements forming the document hierarchy

The library uses two types of paths:

  1. Structural Path: Full hierarchy including all elements Example: uscode/title_26/subtitle_A/chapter_1/section_174

  2. USLM ID: Official USLM identifier (excludes structural-only elements) Example: /us/usc/t26/s174/a/1

Text Content Fields

Each element can contain up to five distinct text fields:

  • Heading: Section or subsection title
  • Chapeau: Opening text before enumerated items
  • Proviso: Conditional or qualifying clauses
  • Content: Main body text
  • Continuation: Text appearing after child elements

Diffs

The TreeDiff structure mirrors the element hierarchy and tracks:

  • Field changes: Word-level differences in text content fields
  • Added elements: New child elements in the newer version
  • Removed elements: Elements that existed in the older version
  • Child diffs: Recursive diffs for matching child elements

Diffs are computed using word-level granularity via the similar crate.

Amending Actions

Bills can perform these operations on existing code:

Amend, Add, Delete, Insert, Redesignate, Repeal, Move, Strike, StrikeAndInsert

API Documentation

Rust

Generate and view the full API documentation:

cargo doc --open

Development

# Run Rust tests
cargo test

# Build and install Python bindings locally
maturin develop

# Run Python tests
python -m pytest