Speakers
Description
In the presentation we will introduce the research on applying LLMs to help create definitions for learners of Estonian as a second language. For Estonian, there have been experiments conducted to create general language dictionary definitions with LLMs (see Tuulik et al., 2025). For English, LLM-s are tested for generating learner’s definitions as well (see for example Ide et al., 2026; Rees & Lew, 2024). Our research is based on a survey we conducted with 55 learners representing different proficiency levels (A1–C1). The aim of the survey was to investigate how learners of Estonian comprehend dictionary definitions in a learner-oriented digital resource and identify linguistic features that support or hinder pedagogical effectiveness. Participants evaluated two alternative definitions – an original and a pedagogically adapted version – for 20 headwords from a dictionary portal Learner’s Sõnaveeb. The analysis focused on lexical, morphological, and syntactic parameters influencing comprehension.
The results showed that learners consistently prefer pedagogically simplified definitions that are compact, clearly structured, and free of semantically empty pronouns, excessive conjunctions, and abstract modifiers. At lower proficiency levels (A2–B1), learners tend to favour familiar and internationally transparent loanwords, while this preference diminishes at higher levels. Morphological analysis revealed that the impersonal voice does not significantly impede comprehension, whereas non-finite constructions present notable difficulties. From a syntactic perspective, comprehension is enhanced when the headword appears at the beginning of the definition and when sufficient, relevant information is provided, even at the expense of brevity.
We are applying the results of the survey to adjust the existing definitions in Learner’s Sõnaveeb and to guide LLMs to create definitions for new B2-level headwords. Based on expert evaluations, the best performing model is selected for the task. We prompt LLMs using new principles and exemplary definitions to explore how to automatically get best quality draft definitions for lexicographers to edit. Using drafts provided by LLMs would significantly speed up the compilation process and thus allow for a faster expansion of the headword set in Learner’s Sõnaveeb.