Grammarly, known for its English grammar editing software refined over 16 years by a team of linguists, has expanded its offerings to include five additional languages: Spanish, French, Portuguese, German, and Italian. This move addresses the most requested feature since the company’s inception, according to Luke Behnke, VP of enterprise product at Grammarly.
The AI-driven update allows the app to not only catch spelling mistakes but also rework sentences and paragraphs to better match native tone and improve overall clarity in all six languages. Grammarly also facilitates translation between these six core languages and 19 other languages directly within the tool, eliminating the need for external translation services, according to Behnke.
The market for AI-based multilingual software is becoming increasingly competitive. Other tech companies are making similar inroads, with Google Search’s AI Mode recently adding five languages and Apple’s AirPods introducing real-time language translation capabilities.
Grammarly has signaled its broader ambition to evolve into an AI productivity application. Key steps in this direction include the July acquisition of the email application Superhuman and the launch of nine AI agents designed for students and educators.
According to the company, machine learning has been integral to Grammarly’s software architecture from the beginning. Behnke noted that the company referred to its proofreading product as an “agent” even before the widespread adoption of large language models (LLMs) in 2023. The software utilizes open-source LLMs fine-tuned by analytical linguists.
The expansion to new languages did not require a proportional increase in the number of linguist teams, said Behnke. Grammarly leveraged a smaller group of experts, alongside internal feedback and thorough evaluations, to ensure quality across the new languages.
These in-house models generate suggestions and paragraph rewrites. According to Behnke, the models are “hosted on our infrastructure with our own security controls and very tight training rules.”
During a beta rollout involving approximately one million users, native speakers of the five new languages showed similar acceptance rates for Grammarly’s suggestions compared to English speakers. Many beta users were surprised to see the familiar red underlines appearing in their native languages, Behnke noted.
Grammarly also integrates third-party LLMs for advanced functionalities, allowing users to opt for external LLMs from providers such as OpenAI. While these third-party LLMs cannot train on Grammarly user data, Grammarly does train its own models on user data. Data training is automatically disabled for enterprise and education customers, while other users have the option to opt out.
While Grammarly has not disclosed specific details about future language additions, Behnke mentioned that customers in the customer support sector have expressed interest in expanding language support to regions with global offshore call centers.
