Proper Noun Translation Drift Simulator
NMT / LLM Diagnostic
Diagnose sub-token embedding confusion ('Ao' → 'Aoyama Toshihide') and evaluate NER preservation constraints
Reset Defaults
Export Report (JSON)
Standard Unconstrained NMT
Drift Detected (0.82)
Source Sentence (English)
Sub-token Drift
0.82
Hallucination Risk
HIGH
Entity Confidence
34%
Constrained NER Engine
Entity Anchored (0.03)
Target Language Output (Japanese)
Sub-token Drift
0.03
Hallucination Risk
LOW
Entity Precision
98%
Sub-Token Attention & Alignment Bipartite Graph
Source Sub-Token
Drift / Hallucination Candidate
Target Exact Entity
Entity Preservation Constraints
Target Dictionary Anchoring Strength
0.85
Soft-Max Attention Penalty (Out-of-dict drift)
0.70
Enable Hard Entity Boundary Masking (Lock "Ao Tanaka" tokens)
Custom Proper Noun Anchoring Dictionary
Source Entity
Pinned Target
Embedding Distance
0.012 (Strong)
0.005 (Strong)
Live Diagnostic Telemetry & JSON Proof Surface
Focus Diagnostic Proof
Enjoy this tool? Build your own with Super