Peng/run natural reasoning

[PengLin-Zilv]

Summary

Per Kaustuv's feedback, add a wrapper script that takes a user-input paragraph, runs the full natural language → PyReason → natural language pipeline, and prints the reasoning results split into Input and Output sections.

Changes and next steps

• New file: examples/run_natural_reasoning.py
• will remove old natural_language_to_pyreason_ex.py if needed
• will revise the tutorial file after this script gets approval and finalize

Pipeline

1. Step 1 — LLM extracts facts and rules in plain English from the paragraph
2. Step 2 — LLM converts English facts/rules into PyReason syntax
3. Step 3 — Parser splits the LLM output into a facts list and a rules list
4. Relax body bounds — auto-revise rule bodies so probabilistic conclusions can chain across rules (PyReason's default [1,1] body bound blocks [0.8,1] derivations from chaining)
5. Step 4 — Build a NetworkX graph from the entities, load facts/rules into PyReason, run pr.reason()
6. Step 5 — Split results into Input (relevant facts) and Output (derived conclusions). Entities for which no rule fired are filtered out
7. Step 6 — LLM translates each section back into natural English with proper grammar and percentage formatting

Example Output

Lawyer paragraph:

Nurse paragraph: