Cross-dialect Text-to-SQL Fidelity on BIRD
100CoverageLLM-as-a-Judge (No Exe)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLM-as-a-Judge (No Exe)Class=LLM, Requires Migration (DBs)=No, Requires Migration (Queries)=No2026.05 | 100 | 0.38 | 0.66 | 0.78 | |
| Gold → Target (LLM)Class=LLM, Requires Migration (DBs)=Yes, Requires Migration (Queries)=Yes2026.05 | 100 | 0.66 | 0.68 | 0.84 | |
| PolySQLClass=Exec (Ours), Requires Migration (DBs)=Yes, Requires Migration (Queries)=No2026.05 | 100 | 0.72 | 0.75 | 0.85 | |
| Pred → Source (SQLGlot)Class=Static, Requires Migration (DBs)=No, Requires Migration (Queries)=Yes2026.05 | 46.1 | 0.39 | 0.56 | 0.55 | |
| Gold → Target (SQLGlot)Class=Static, Requires Migration (DBs)=Yes, Requires Migration (Queries)=Yes2026.05 | 40.9 | 0.48 | 0.69 | 0.65 |