AutoDSL: Automated domain-specific language design for structural representation of procedures with constraints

Yu-Zhe Shi,Haofei Hou,Zhangqian Bi,Fanxu Meng,Xiang Wei,Lecheng Ruan,Qining Wang

2024-06-18

Abstract:Accurate representation of procedures in restricted scenarios, such as non-standardized scientific experiments, requires precise depiction of constraints. Unfortunately, Domain-specific Language (DSL), as an effective tool to express constraints structurally, often requires case-by-case hand-crafting, necessitating customized, labor-intensive efforts. To overcome this challenge, we introduce the AutoDSL framework to automate DSL-based constraint design across various domains. Utilizing domain specified experimental protocol corpora, AutoDSL optimizes syntactic constraints and abstracts semantic constraints. Quantitative and qualitative analyses of the DSLs designed by AutoDSL across five distinct domains highlight its potential as an auxiliary module for language models, aiming to improve procedural planning and execution.

Robotics

What problem does this paper attempt to address?

The paper attempts to address the problem of accurately representing program requirements in specific scenarios (such as non-standardized scientific experiments), especially when precise description of constraints is needed. Existing domain-specific languages (DSLs) can effectively structure these constraints, but they often require manual customization for each case, which is both time-consuming and labor-intensive. Therefore, the paper proposes a framework called AutoDSL, aimed at automating the DSL design process across different domains. By leveraging a corpus of experimental protocols from specific domains, AutoDSL optimizes syntactic constraints and abstracts semantic constraints, thereby reducing the cost of DSL design and enhancing its potential in assisting language models, improving program planning and execution.

Specifically, the paper focuses on how to generate DSLs suitable for specific domains through automated methods to overcome the customization and high-cost issues in current DSL design. This not only helps improve the accuracy of program representation in complex scenarios such as scientific experiments but also promotes the performance of language models in handling new protocols, especially in terms of syntactic complexity and semantic errors.