New benchmark evaluates LLMs on natural-language-to-Scratch programming

Researchers have introduced NL2Scratch, a new benchmark designed to evaluate natural-language-to-code generation specifically for block-based programming environments like Scratch. This benchmark includes over 300,000 program-description pairs, with a focus on semantic alignment rather than just lexical similarity. Experiments with large language models show a significant gap between their performance on traditional metrics and their ability to achieve semantic consistency, particularly with longer or more complex programming tasks. AI

IMPACT This benchmark could improve LLM capabilities in translating natural language instructions into structured code for educational and creative programming tools.

RANK_REASON The item describes a new academic paper introducing a benchmark and evaluation metric for natural-language-to-code generation in a specific programming environment. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• NL2Scratch
• ScienceCast
• Scratch
• Semantic Alignment Consistency

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →