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

TL;DR

Quantitative and qualitative analyses of the DSLs designed by AutoDSL highlight its potential as an auxiliary module for language models, aiming to improve procedural planning and execution.

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.

Figures (6)

• Figure 1: Representative constraints in protocols.(A) Parameter omission: This refers to the absence of essential parameter values within a predefined set, e.g., the lack of temperature specification during the denaturation step in Protein Gel Electrophoresis. (B) Parameter under-specification: This occurs when a quantitative parameter is described using qualitative terms, leading to ambiguity, e.g., unclear mixture configurations in DNA Extraction. (C) Action undefinition: This involves the description of procedural steps at a high level without grounding to the specific, executable actions required, e.g., the vague change operation in Cell Preparation. (D) Iterative control logic: Loops that operate iteratively to satisfy a final condition rather than straightforwardly, as seen in PCR Optimization. (E) Memory management: Drawing a parallel with computer memory mechanisms, laboratory procedures also face constraints on the availability of storage for intermediates, necessitating explicit reallocation of containers and devices, such as managing buffers in Protein Digestion. (F) Concurrent management: The synchronization of actions without dependencies to maximize time efficiency and resource utilization, e.g., reagent splitting in RNA Extraction.
• Figure 2: Protocols in different structural representations
• Figure 3: The AutoDSL framework and the resulting dsl-based procedure constraints.(Top) Given domain specified corpus, AutoDSL conducts bidirectional syntax optimization and non-parametric semantic reduction, resulting in syntactic constraints and semantic constraints. (Bottom) A dsl-based constraint takes novel procedures as input, handles the nonlinear syntax structures like loop by syntactic constraints, and handles the semantic errors like missing by semantic constraints.
• Figure 4: Illustration on syntactic constraint optimization and semantics constraint reduction.(A) Resulting syntactic constraints derived from the CFG prior model. (B) Resulting semantic constraints. (C) Convergence curve of syntactic constraint optimization. (D) Convergence curve of semantic constraint reduction. (E) Frequency profile of the semantic constraints of Genetics-DSL and Medical-DSL. (F) Acquisition of different syntactic constraints on Ecology and Medical domain corpora. (G) Syntactic constraint Function-procedure is differently acquired by the five distinct domains. (H) Confusion matrix indicating the overlapped semantic constraints between the five distinct domains.
• Figure 5: Constraint design results on 5 experimental science domains.(A) Quantitative evaluation on the 5 dsl-based constraints and the BioCoder baseline. (B) Empirical evaluation on the 5 dsl-based constraints and the BioCoder baseline.