Survey finds most lexicographers use AI but value dictionaries’ reliability

Most lexicographers who responded to an international survey use artificial intelligence in their work, but the researchers say AI-generated answers are not as reliable as dictionaries. The preliminary findings were reported by ERR in an article by Lydia Risberg, Jelena Kallas and Eleri Aedmaa of the Estonian Language Institute.

The institute is part of ELEXAI, an international project running from June 2026 to May 2029. The project aims to advance European lexicography by exploring how AI can help compile dictionaries and how high-quality dictionary information can help large language models better understand European languages.

The survey, which the institute helped develop, collected 141 responses from around the world between late September and early November. Sixty-three percent of respondents said they use AI at work, while 18 percent plan to start using it and 19 percent do not use it. Most respondents were experienced lexicographers: 65 percent had worked in the field for more than 10 years, including 38 percent with more than 20 years of experience. Sixteen percent had less than three years’ experience.

Respondents most often use AI to draft definitions and explanations of meaning, and to find synonyms and translations. For these tasks, 51 percent rated the results good or very good, while nearly 10 percent rated them poor or very poor. Translation suggestions received the strongest ratings: 59 percent called them good, 35 percent average and 6 percent poor.

Lexicographers said AI was used less for creating multimedia, such as images and videos, reflecting the largely text-based nature of dictionary work. Respondents also noted that dictionary data is increasingly used to assess language models.

Views on barriers to using AI varied. Some respondents said they faced none and used it routinely; others said they did not trust its output and preferred to rely on their own judgment. They also cited weaker results in some languages, including Macedonian and Maltese, and institutions that have not yet adopted AI. Concerns included hallucinations, ethical issues and possible risks to the reliability and principles of dictionary work.

Respondents said dictionaries’ strength lies in providing reliable, checked language information. They also warned that if AI-generated material is mistaken for human-written language in corpora used for dictionary work, dictionary entries could become less grounded in actual human language use.

The findings suggest that AI is already a tool for many lexicographers, while the authors’ account of the survey highlights the continuing importance of verified dictionary information.