EPO affirms that AI cannot invent
Den Rise / Shutterstock.com
The popular text correction software must navigate the nuances around patenting language processing systems, explain Parminder Lally and Simon Ambroz of Appleyard Lees.
“It looks like you’re writing a letter. Would you like help?”
Some of you may read that sentence and immediately remember Clippy, Microsoft’s Office assistant. Clippy would pop up while you were typing in Microsoft Word and offer you advice on how to improve your writing. Clippy eventually retired in the early 2000s, but since then, blog writing, social media posts, and content creation have become incredibly popular. Even in the workplace, most of us communicate using emails now rather than phone calls. This is where Grammarly comes in.
In this article, we look at this AI-based business, and provide advice on whether its AI technology is patentable.
Grammarly identified a gap in the market in 2009 and built a successful company based on Artificial Intelligence (AI) technology. Grammarly (which is also a registered trademark) is a cloud-based writing assistant that checks text written in English dialects.
The writing assistant reviews spelling, grammar, punctuation, and clarity in real-time, and can also check whether your writing is engaging and strikes the right tone. The AI-powered writing assistant not only identifies mistakes and other errors within your writing, but also searches for an appropriate replacement for the mistakes and errors, providing a detailed explanation of the mistake or error.
Grammarly requires you to be connected to the internet to work, because the AI writing assistant checks your writing against a database of mistakes and errors, as well as data collected from users.
Grammarly currently has a multi-billion-dollar valuation but has a small patent portfolio. However, when it comes to a patent portfolio, size certainly isn’t everything. Grammarly’s portfolio focuses on its key innovations, ie the core technology that underpins its business. This is an important lesson, particularly for smaller companies and start-ups with limited budgets: use your IP budget to ensure you protect the core technology.
Patents provide start-ups with options for revenue generation (such as licensing, or acquisition) and collaboration (more negotiating power). If you’d like to know more about why patents can be useful for start-ups and how to make the most of a limited budget, take a look at this article.
Grammarly’s patent portfolio
Grammarly was founded in 2009, and its earliest patent application dates back to 2010. It is great to see that the company understood the importance of protecting its technology at an early stage. We explain the importance of filing patent applications sooner rather than later in this article.
The company’s patent portfolio also appears to cover various iterations of its technology. For example, the earliest patent application which has a priority date of 2010 describes using a computer-based grammar checking tool and a human proof-reader. This reliance on humans to perform part of the task is unsurprising, as other companies were doing similar things at this time. An example is Amazon’s Mechanical Turk, which used humans to perform tasks, such as query response, that computers were either unable to do or which humans could do faster.
However, it appears that Grammarly has moved away from partly relying on human intelligence, and its patent portfolio reflects this move. This is important and something innovating technology businesses should ensure, as otherwise a patent portfolio might not cover what the business is actually doing.
In 2020, Grammarly filed two patent applications that do not mention using humans to perform part of the mistake identification and correction. One of these discloses a machine learning model that can suggest synonyms to replace a word in a text sequence that also fits in the context of the text sequence.
The other describes a machine learning model that has been trained to analyse digital input text sequences and produce digital output text sequences that are grammatically corrected and fluency-adjusted versions of the corresponding digital input sequences, taking into account the user's particular native language and proficiency level.
Currently, it seems that Grammarly has only filed patent applications and obtained granted patents in the US. There may be many reasons why it has focussed on the US: the company is based in the US and the US is a big market for this technology.
Also, since Grammarly is cloud-based, if the servers it uses to perform the text analysis are based in the US, it may have filed patent applications in the US to cover the jurisdiction in which its technology is running. Another reason may be due to differences between what patent offices around the world consider patentable subject matter.
Is Grammarly’s technology patentable in Europe?
Generally, the chance of obtaining a software patent in the US and Europe is very similar. That said, the US Patent and Trademark Office and European Patent Office (EPO) have different assessment approaches. We explain the different approaches in this article.
Some examples of patentable AI inventions are specified in the EPO’s Guidelines for Examination: “For example, the use of a neural network in a heart-monitoring apparatus for the purpose of identifying irregular heartbeats makes a technical contribution. The classification of digital images, videos, audio or speech signals based on low-level features (eg, edges or pixel attributes for images) are further typical technical applications of classification algorithms.”
It is important to note that all the above-mentioned examples of software/AI have “further technical effects” which go beyond the "normal" physical interactions between the program (software) and the computer (hardware) on which it is run.
Clearly, the EPO generally considers AI-based inventions to be patentable, but what is the EPO’s view on AI inventions that relate to language processing? The EPO’s Guidelines specify that “Classifying text documents solely in respect of their textual content is however not regarded to be per se a technical purpose but a linguistic one (T 1358/09)”.
This statement suggests that natural language processing inventions may not be patentable at the EPO. Accordingly, this may be one of the reasons that Grammarly has not pursued any European patents and mainly focuses on the US.
However, this does not mean that obtaining patent protection for Grammarly would be impossible in Europe. Indeed, the number of patent applications being filed for natural language processing at the EPO continues to grow.
Firstly, the patent application could focus on a technical use of the natural language processing invention, rather than just a linguistic one.
As an example, granted European patent EP3230896B1 uses a linguistic-based model to predict a complexity score of an input text in a first language, where the complexity score represents a difficulty to translate the input text from the first language into a second language.
This score enables a user to modify the input text to improve the complexity score. EP3378059B1 relates to a similar scoring method, which can be used to generate audio data based on the input text. EP2387031B1 describes analysing statements to determine how confusing they would be if they were spoken and processed by a speech recognition system. In all these cases, the text processing is used for a technical purpose (machine translation, or text to speech conversion).
Another technical use of text processing is drug discovery—see our article about patenting such techniques.
Secondly, the patent application could focus on a wider system which includes technical features as well as the natural language processing.
For instance, EP3583747B1 describes an intelligent digital assistant system that comprises a microphone and a speaker. The system receives audio input, recognises speech of multiple users, and processes the speech to determine what the users are saying and to provide an appropriate response.
Speech analysis is considered to be patentable by the EPO, even if the analysis uses linguistic models. It may be that speech processing using grammar models becomes more prevalent in the near future as we increasingly use our voices to control our devices or to obtain information, and as we communicate verbally with colleagues, friends and family in different languages.
Thirdly, even if the invention is purely linguistic, it can be advantageous to use a European patent attorney to prepare your patent application and assist you with the filing and prosecution in other countries.
Similarly, it may be worth first-filing your text processing patent application at the EPO. This is because it is possible to obtain a quick search and written opinion from the EPO within around four months of filing, which can be useful to determine your subsequent patent filing strategy and to identify potentially relevant prior art. Similarly, using the EPO as the International Searching Authority when filing a PCT application may be useful, as outlined here.
Parminder Lally is a senior associate at Appleyard Lees. She can be contacted at: firstname.lastname@example.org
Simon Ambroz is a trainee patent attorney at Appleyard Lees. He can be contacted at: email@example.com
Today’s top stories
Start-ups, SMEs, AI, Universities, software, patents, USPTO, EPO