Synchromesh
Pairs target-similarity retrieval with Constrained Semantic Decoding (CSD), using an incremental parser and type/scope engine to enforce valid code.
The loop
- Proposes
A code language model predicts next-token distributions for domain-specific languages (SQL, Vega-Lite, SMCalFlow).
- Judged by
An incremental Context-Free Grammar parser and dynamic semantic validator (table/column scope, type compatibility) compute the regex of valid next tokens.
- Steers
Invalid tokens are masked at every decoding step so the sampler only explores syntactically and semantically valid AST completions.
Headline result
Eliminated syntax, type, and scope errors across SQL, Vega-Lite visualization specs, and calendar DSLs while boosting execution accuracy.
What this does not prove
Requires authoring an incremental grammar and completion-engine specification for each target domain language.
How it generalizes the harness
An incremental grammar and type-checking completion engine constrains autoregressive token generation at every step.
Classification
| What it proposes | Program |
|---|---|
| Generative model | Codex / GPT-3 with constrained token masking |
| Algorithmic authority | Incremental CFG parser and static type/scope completion engine |
| Feedback | Verifier errors |
| What gets updated | Search tree |
| When it runs | At inference time |
| Validity | Formal (machine-checked) |
| Evidence | Code released |
| Lineage | Original system |