NL2SHACL-Bench: A Benchmark Suite for Natural Language to SHACL Translation
The episode discusses a paper introducing NL2SHACL-Bench, a benchmark suite for translating natural language descriptions into SHACL rules for validating knowledge graphs. The hosts review findings from testing four state-of-the-art LLMs. While the models produce syntactically valid SHACL, they struggle with complex logical and structural patterns, particularly in certain domains.