Towards improved collaborative text editing CRDTs by using Natural Language Processing

Abstract

Collaborative text editing systems are used in a variety of cloud-based products. To ensure that documents remain consistent between users, these systems often rely on CRDTs, operational transformation, or other techniques for achieving (strong) eventual consistency. CRDT-based approaches are appealing as they incorporate strategies to ensure that concurrent updates cannot conflict. However, these strategies do not necessarily take into account program semantics and may result in unexpected behaviour from the end-user's perspective.For example, conflict resolution strategies in collaborative text editors may lead to duplicate words and incorrectly merged sentences. This position paper investigates the use of deterministic natural language processing (NLP) algorithms to improve the concurrency semantics of collaborative text editing systems that rely on CRDTs, aiming to provide a better end-user experience. We explore what is needed to ensure convergence, and highlight potential difficulties with the approach.