发表机构
University of California, Berkeley（加州大学伯克利分校）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对资源受限语言模型在自然语言组合调度中可行性不足的问题，提出神经符号框架SDDL，其在300实例调度子集上大幅提升小模型的可行调度准确率，逼近大模型性能
AI 中文摘要
组合调度对语言模型构成重大挑战，要求其在指数级大的搜索空间内识别满足复杂约束的可行解。在资源受限场景中，这一挑战尤为突出：更大规模的语言模型不具实用性，仅能选择较小模型，而这类模型在直接从自然语言调度时往往无法保证可行性。为解决这些局限，我们提出SDDL，一种神经符号框架，可将自然语言调度问题转换为任务、资源、约束及目标的紧凑且与求解器对齐的表示，同时将低层建模与搜索任务交由确定性编译器和外部求解器处理。在包含300个实例的多系列调度问题子集上，SDDL使所有受测资源受限模型的独立验证可行性均得到提升。两种最强的SDDL配置分别达到55.3%和28.3%的可行性，而直接生成基线的可行性仅为23.7%和1.3%，求解器代码基线的可行性为21.7%和7.0%，且可行调度的中位数最优性差距为0.0%。SDDL通过表达问题结构而非生成解或求解器代码，使较小模型能够接近所评估的最强直接配置和求解器代码配置，包括规模大得多的前沿模型。
英文摘要
Combinatorial scheduling poses a significant challenge for language models, requiring them to identify feasible solutions within exponentially large search spaces while satisfying complex constraints. This challenge is especially pronounced in resource-constrained settings, where larger language models are impractical and selection is limited to smaller models which often fail to preserve feasibility when scheduling directly from natural language. To address these limitations, we introduce SDDL, a neuro-symbolic framework that translates natural-language scheduling problems into compact, solver-aligned representations of tasks, resources, constraints, and objectives, while delegating low-level modeling and search to a deterministic compiler and external solver. On a 300-instance, multi-family subset of scheduling problems, SDDL improves independently verified feasibility for every resource-constrained model tested. The two strongest SDDL configurations reach 55.3% and 28.3%, up from direct-generation baselines of 23.7% and 1.3% and solver-code baselines of 21.7% and 7.0%, with a 0.0% median optimality gap among feasible schedules. By expressing problem structure rather than generating solutions or solver code, SDDL enables smaller models to approach the strongest evaluated direct- and solver-code configurations, including substantially larger frontier models.
CommentsTo appear, The 4th Annual Workshop on Mathematical Natural Language Processing (MathNLP2026) @EMNLP2026
