This diagram shows how the current Reverse Analyzer processes an
Arabic or Latin token. The current engine is not a simple one-way “strip everything”
pipeline. Instead, it preserves context, protects exact/non-derived items, routes
the token through competing analysis families, and then ranks the resulting candidates.
Key idea: the current Reverse Analyzer is not just reversing the forward
pipeline step by step. Instead, it uses a router-based, evidence-weighted
approach: preserve context, protect exact items, test family-specific analyses,
then rank the resulting candidates.
Current summary:
normalize → context → proclitic handling → exact protection →
family routing → family-aware suffix testing → root/form analysis →
enrichment → ranking → final display