We're working on a new way to check the grammatical correctness of your natural-language text. It'll be a part of our upcoming suite of intelligent tools—powered by innovative machine learning—that'll work offline and that'll respect your privacy. Privacy-friendly machine learning is a bold, new frontier, and we're just getting started.
To help you understand the significance of MLGrammar, we've outlined our thoughts in a clear, delineated manner.
The existing tools for checking grammatical correctness generally fall into two categories: primitive rules and intelligent server programs. Both categories have significant compromises.
They're static and unreliable, and they don't incorporate semantic context.
Example: The squiggly green lines in Microsoft Word
They compromise your privacy by sending your text to a server that you don't control.
The reason why most intelligent services run on remote servers that you don't control is that they were built at a time when the devices that most people used every day weren't powerful enough to compute the complex algorithms that power those services. Soon, the companies that ran those services realized that they could make money off of your private data, so they stuck with the remote-server model.
More recently, everyday devices like iPhone and iPad gained special processors that enable powerful machine learning directly on those devices themselves, eliminating the need for a remote server. However, the server-based companies didn't want to give up the money that they make off of your private data, so they didn't embrace this new capability. This is where Gerzer comes in: we're challenging the status quo by introducing privacy-friendly intelligent tools—powered by innovative machine learning—that don't need any remote servers. As an added benefit, these tools will work completely offline.
MLGrammar will be our first entry in a suite of intelligent tools that respect your privacy. We're training powerful machine learning models with advanced tokenization and lemmatization, accurate non-binary classification, and deep sequence-to-sequence neural networks, all running locally on the Neural Engine in your personal device. Unless you work in the field, you don't need to understand what any of that means; just know that it all results in useful, accurate grammar checks and suggestions that can outperform the primitive rules that you'll find in programs like Microsoft Word without the privacy risks that come with services like Grammarly.
Even after its initial release, we'll continue to improve MLGrammar, incorporating feedback from our users. Later on, we'll also introduce other intelligent tools—powered by innovative machine learning—that respect your privacy. These future tools will be built on top of the same core technology that we're pioneering with MLGrammar. We're just getting started.