The Surprising Reason Why Europe Denied Google’s Machine Translation Patent

European Patent Office headquarters in Munich, Germany

Google’s Neural Machine Translation (MT) obtained a patent in the United States in 2017. On the European side, however, the European Patent Office (EPO) rejected (download) the application on October 30, 2023, eight years and seven days after first receiving it and reviewing multiple appeals, saying it is not an “inventive step.”

“Neural machine translation systems with rare word processing” is the title given by Google to its invention in the patent application for the technology, submitted to the EPO on October 23, 2015.

The application originally submitted by researchers Quoc V. Le, Minh-Thang Luong, Oriol Vinyals, Wojciech Zaremba, and Ilya Sutskever (now a board member of OpenAI) was first rejected on July 6, 2020.

In communicating the first rejection, the EPO legal brief did not specify a reason, but cited five documents as grounds for the decision and opened the matter to appeal, which is what Google legal representatives did.

The EPO’s examiners reviewing the latest appeal refused the application definitively because it deemed that improvements to machine translation (MT) were not an inventive step. The “date of legal effect,” was November 9, 2023, when the EPO’s Board of Appeal made the decision final.

The patent appeal reviewers determined that using artificial intelligence, machine learning, and training for a translation is a matter of linguistics, not a technical purpose. “Merely finding a computer algorithm to implement an automated translation process does not render the resulting computer program technical,” reads the EPO conclusive statement. 

Google scientists submitted claims about the technology to the EPO examiners describing it as a neural network translation model, a “computer-implemented translation system for translating natural language text from a source sentence in a source language to a target sentence in a target language.”

“Merely finding a computer algorithm to implement an automated translation process does not render the resulting computer program technical”.

On appeal, Google representatives cited as a precedent the patent granted to Systran (a process taking place from application in 1987 through appeals and then a final decision in 2002), alleging, among other things, that “[t]he implementation of the information and methods related to linguistics as a computerized translation process similarly requires technical considerations and thus provides a technical aspect to per se non-technical things such as dictionaries, word matching or to translating compound expressions into a corresponding meaning.”

The EPO Board of Appeals rejected specifically the argument that a computerized translation process is a technical application and confers “technical character to non-technical aspects.”

The conclusions drawn by the EPO examiners were that, since the translation system could be implemented as a computer program, “features contribute to [an] inventive step only to the extent that they interact with the technical subject-matter of the claim to provide a technical effect,” but that “the board had been unable to identify a technical effect achieved by the distinguishing features … even if the computer program includes algorithmic aspects which are not directly based on linguistic concepts.”