Design and Implementation of a Natural Language Interface for Controlling the Web of Things Devices

We present a systematic workflow for developing a Natural Language Interface (NLI) using Knowledge Graphs (KGs) and the Web of Things (WoT) specification as an abstraction layer. We first synthesize a data corpus from a given device interface description using manually created templates and parameter sheets. Then we paraphrase the synthesized data corpus with pre-trained Large Language Models (LLMs). The resulting corpus serves as training data for a text-to-code encoder-decoder neural network model, enabling the mapping of diverse natural language commands into an executable code format. To the best of our knowledge, no existing data corpus or NLI system has been tailored for WoT device interactions. Our work provides a baseline and can be extended to a broader range of real world WoT use cases.