MultiLang Code Parser Dataset (MLCPD)
收藏资源简介:
MLCPD是一个大规模的多语言代码解析数据集,它统一了十种主要编程语言的语法和结构表示。该数据集包含超过七百万个解析的源代码文件,按照我们提出的通用抽象语法树(AST)模式进行规范化,从而能够在不同语言之间进行一致的推理、结构学习和多语言软件分析。MLCPD为每个文件提供层次树表示和丰富的元数据,确保了无损失的语法覆盖和结构统一。每个条目都包括一个规范化的模式、语言级别元数据和存储在Parquet格式中的抽象节点语义,以便于可扩展的检索。经验分析揭示了强烈的跨语言结构规律性,证明了从Python、Java和Go等不同语言中提取的语法图可以在共享模式下对齐。我们在Hugging Face上公开发布了数据集,并在GitHub上发布了配套的代码库,其中包括用于数据集再现、语法编译和跨语言统一AST可视化工具的完整管道。这些资源共同建立了MLCPD作为未来跨语言表示学习和程序理解研究的一个开放、可重复的基础。
MLCPD is a large-scale multilingual code parsing dataset that unifies the syntax and structural representations of ten major programming languages. This dataset contains over seven million parsed source code files, normalized according to our proposed Generalized Abstract Syntax Tree (AST) schema, enabling consistent reasoning, structural learning, and multilingual software analysis across different languages. MLCPD provides hierarchical tree representations and rich metadata for each file, ensuring lossless syntax coverage and structural unification. Each entry includes a normalized schema, language-level metadata, and abstract node semantics stored in Parquet format to facilitate scalable retrieval. Empirical analyses reveal strong cross-language structural regularities, demonstrating that syntax graphs extracted from diverse languages such as Python, Java, and Go can be aligned under a shared schema. We have publicly released the dataset on Hugging Face and published an accompanying code repository on GitHub, which includes a full pipeline for dataset reproduction, syntax compilation, and cross-language unified AST visualization tools. Together, these resources establish MLCPD as an open, reproducible foundation for future cross-language representation learning and program understanding research.

- 1通过乔治华盛顿大学计算机科学系 · 2025年



