For 16 years, the Linguist team carefully performed and refined the grammar editing software to match natural English language patterns. Now the company receives a immense assist of artificial intelligence to expand similar offers to five consecutive languages: Spanish, French, Portuguese, German and Italian. It is the expansion that satisfies the “request number one function” from the foundation of the company, according to the Vice President for Grammarly, Luke Behnke.
In addition to catching spelling errors, the application transfers sentences and paragraphs to match the tone of native speech and improve overall transparency – currently in six languages. Grammarly also explains six basic languages on 19 different languages ”without having to leave the tool and transition to another translation provider,” says Behnke.
Grammarly indicated a wider ambition to become an AI product application, starting from taking over the e -mail superhuman buzz in July, and then launched nine AI agents designed for students and teachers.
The company claims that its software has been built machine learning from the very beginning. “We call it an agent. … I know that it may be the abused word of AI at the moment, but this is a kind of first agent from which people have ever used,” says Behnke about the product correction of grammar. Enormous language models have introduced a photo 2023. The software is based on immense models of immense languages that are refined by analytical linguists.
Behnke said that the number of modern linguistic bands was not “linear” with an increased number of languages offered by Grammarly. Behnke says that the company used a “smaller group of people” in addition to internal feedback and modern languages assessments. Suggestions and rewriting paragraphs emerge from these internal models “hosted in our infrastructure with our own safety controls and very strict training rules,” says Behnke.
In the implementation of beta for users, “about a million” native speakers from five modern languages adopted suggestions for grammar according to rates similar to English. The implementation surprised some users when known red lines suddenly appeared under the text written in their native language, says Behnke.
Grammarly also uses other companies for advanced functions. Behnke says that users can choose the inclusion of external LLM from the main companies such as OpenAI. While LLM of third pages cannot train grammar users, Grammarly train their own models of user data. Entrepreneurship and education clients automatically exclude training, while other users must give up training.
Although the company has not disclosed which languages are in preparation, Behnke claims that grammar clients specializing in customer service demanded more languages in global areas in which there are an offshore call center.