Grammarly, a leader in grammar editing software, is set to broaden its offerings beyond English after 16 years of refinement. The company has announced plans to leverage artificial intelligence (AI) to introduce its services in five additional languages, namely Spanish, French, Portuguese, German, and Italian. This expansion comes in response to a consistent demand, described as the “number one feature request” by Luke Behnke, Grammarly’s VP of enterprise product.
The enhanced app not only identifies spelling errors but also restructures sentences and paragraphs for better tone and clarity, now supporting six languages in total. Furthermore, Grammarly’s software provides translation services for these six languages across 19 different languages, allowing users to remain within the tool without needing external translation services, according to Behnke.
The landscape for AI-driven language tools is intensifying, particularly with Google expanding its AI capabilities to five new languages in its search features and Apple introducing live language translation with its latest AirPods.
Grammarly’s ambitions extend further into the AI productivity realm, evidenced by its acquisition of the email platform Superhuman in July and the introduction of nine new AI agents tailored for students and educators.
The foundation of Grammarly’s software relies on machine learning from its inception. “We refer to it as an agent, perhaps an overused term in the AI context, but consider this the first agent that users have interacted with,” Behnke remarked, highlighting the tool’s legacy in proofreading. In 2023, large language models were integrated into their platform, built on open-source models fine-tuned by expert linguists.
Behnke noted that the expansion in linguist teams did not proportionally align with the increase in languages incorporated. Instead, the company utilized a smaller group alongside internal testing and feedback processes to optimize the new language functionalities. Suggestions and rewrites are generated from specialized models hosted under Grammarly’s secure infrastructure.
The initial beta version has reached around a million users, where native speakers of the newly added languages are responding positively to Grammarly’s suggestions, similar to its English-speaking user base. Many were taken aback to see the familiar red underlines appear while writing in their respective languages, noted Behnke.
In addition, Grammarly utilizes external large language models (LLMs) for advanced functionalities. Users have the flexibility to opt for LLMs from prominent companies like OpenAI, although these third-party models are not permitted to learn from Grammarly’s user data. Grammarly’s in-house models, which employ user data for training, have additional safeguards in place for enterprise and educational users who have their training feature turned off by default.
While Grammarly has not disclosed specific languages that may be added in the future, Behnke expressed that clients in customer support sectors have shown interest in incorporating more languages that are prevalent in global offshore call centers.
0 Comments
