Weaviate v1.37 Makes Tokenization Observable with Accent Folding and Stopword Presets
Below validation threshold — auto-passed without scoring
Weaviate released v1.37 on April 23, 2026, exposing new REST endpoints /v1/tokenize and per-property tokenization for debugging. André Mourão and Ivan Despot at Weaviate documented how asciiFold normalizes accented characters during indexing and querying for 20+ languages. The release added named stopword presets at the collection level and per-property overrides. Weaviate v1.37 supports language-specific tokenizers including kagome_ja for Japanese. The company enabled per-property configuration without reindexing existing data.