Smart Contract Generation Assisted by AI-Based Word Segmentation

• Published in: Applied Sciences
• Main Authors: Yu Tong, Weiming Tan, Jingzhi Guo, Bingqing Shen, Peng Qin, Shuaihe Zhuo
• Format: Article
• Language: English
• Published: MDPI AG 2022-05-01
• Subjects: smart contract; collaborative drafting; semantic understanding; automatic word segmentation
• Online Access: https://www.mdpi.com/2076-3417/12/9/4773

In the last decade, blockchain smart contracts emerged as an automated, decentralized, traceable, and immutable medium of value exchange. Nevertheless, existing blockchain smart contracts are not compatible with legal contracts. The automatic execution of a legal contract written in natural language is an open research question that can extend the blockchain ecosystem and inspire next-era business paradigms. In this paper, we propose an AI-assisted Smart Contract Generation (AIASCG) framework that allows contracting parties in heterogeneous contexts and different languages to collaboratively negotiate and draft the contract clauses. AIASCG provides a universal representation of contracts through the machine natural language (MNL) as the common understanding of the contract obligations. We compare the design of AIASCG with existing smart contract generation approaches to present its novelty. The main contribution of AIASCG is to address the issue in our previous proposed smart contract generation framework. For sentences written in natural language, existing framework requires editors to manually split sentences into words with semantic meaning. We propose an AI-based automatic word segmentation technique called Separation Inference (SpIn) to fulfill automatic split of the sentence. SpIn serves as the core component in AIASCG that accurately recommends the intermediate MNL outputs from a natural language sentence, tremendously reducing the manual effort in contract generation. SpIn is evaluated from a robustness and human satisfaction point of view to demonstrate its effectiveness. In the robustness evaluation, SpIn achieves state-of-the-art F1 scores and Recall of Out-of-Vocabulary (R_OOV) words on multiple word segmentation tasks. In addition, in the human evaluation, participants believe that 88.67% of sentences can be saved 80–100% of the time through automatic word segmentation.