Construction grammars are commonly written by hand. This is not only time and resource intensive, but also limits the potential for applications that need to extract accurate and deep semantics from real-world language. In the present paper we explore novel ways to tap into existing resources to expand Embodied Construction Grammar (ECG) semi-automatically from FrameNet, an approach motivated by the shared theoretical underpinnings of ECG and FrameNet. We show how FrameNet data can be readily translated into ECG constructions and schemas, and discuss ways to identify the kinds of general patterns that are crucial to construction grammar approaches. The results achieved thus far indicate how a data-driven approach to construction grammar development is not only desirable but feasible.
展开▼
