Semantic Parsing

Semantic parsing is described in 139. 【Agent的综述】和苏煜聊Agent技术史、OpenClaw Moment、边界的消弭和社会的辐射 as the NLP route that translates natural language into machine-readable semantic representations. Those representations can connect human intent to databases, knowledge graphs, websites, or executable tools.

Su Yu / 苏煜 frames semantic parsing as complementary to neural agents: neural agents went deep in constrained environments, while semantic parsing broadened the possible action space by letting language address structured systems. His own PhD work in the area is presented as a precursor to his later Language Agent and Computer Use Agent work.

Key Claims

• Semantic parsing tries to let machines understand human instructions without forcing people to think like software systems.
• It expands agent action space by connecting language to structured back ends and executable environments.
• It prefigures current Agent-Facing Interfaces work because both ask how human intent becomes reliable machine action.

Connections

• Su Yu / 苏煜 — researcher whose path moved from semantic parsing toward language agents.
• Language Agent — later paradigm that absorbs language-to-action capabilities into LLM-based systems.
• Agent-Facing Interfaces and Agent Harness — modern infrastructure layers that continue the language-to-action problem.
• Universal Digital Agent — convergence target that requires broad action spaces across software environments.