The Uncanny Valley of Computable Contracts

Assessing Controlled Natural Languages for Computable Contracts

Abstract

Automated legal relationships of one form or another are the future. One does not have to listen to legal tech enthusiasts to come to this conclusion. In some limited settings, such automated legal relations are already here. These embedded, automated legal relations can be as seemingly innocuous as a social media app that implicitly encodes its privacy policy within its computer code. One avenue to make these relations more accessible is the combination of controlled natural languages (CNLs, languages based on existing languages such as English but more restrictive) and programming languages, creating a CNL that reads like (for example) English but can be used to create executable computable contracts. These work by reducing the complexity of the natural language to a manageable size by only allowing a limited set of syntax and semantics. While CNLs are more precise and more suitable for such a task than machine-learning-based natural language processing, they also have limitations, especially in their inherently limited complexity and when learning and writing them. This contribution sketches the historical development of controlled natural languages and their relation to programming languages and then assesses how valuable CNLs are for computable contracts. In this process, I describe the specific property of CNLs, namely, that they are often easy to read but hard to master, referring to the "Uncanny Valley of Computable Contracts" by way of analogy to the hypothesised phenomenon described for human reactions to humanoid androids.

Reply by Emma Tosch, Northeastern University