Weighted Attribute Grammars

Weighted Attribute Grammars (WAG) are grammars which production rules can have weights or probabilities, while nonterminals can have inherited, synthesised or local attributes. In WAG, weights coexist with attributes, and either can refer to the other.

It is an extension of both weighted grammars as well as attribute grammars, in terms of both notation and expressiveness.

Research on WAG is ongoing and is mostly done in collaboration with Marcus Gerhold.
If you're interested in collaboration, get in touch!

Completed constituent projects (latest rightmost)

Jan Douwe Beekman (2021)Procedural Location Generation with WAG

Berke Güdücü (2022)WASTs for Program Comprehension in Java

Vadim Zaytsev (2022)Speak Well or Be Still: Solving Conversational AI with WAG

Andrei Popa (2023)Gotta adjust them all! DDA of RPG through PCG of NPC

Suzanne van der Veen (2024)NoWa: A CNF Converter for WAG

Rafael Dulfer (2024)WAGon: WAG Oriented Notation

Lotte Steenmeijer (2025)WAG for HMI in IoT

Nick Wolters (2025)Generating Samples for WAG

Tools and artefacts

• https://github.com/helixaril/Generative-Grammar-for-NPCs (MIT)
• https://github.com/SvdVeen/NoWa
• https://github.com/Rafaeltheraven/wagon (Apache) + documentation
• https://github.com/NickWolters/wagenator (BSD)

---

The page is maintained by Dr. Vadim Zaytsev a.k.a. @grammarware. Last updated: February 2026.