Exploring Automated Keyword Mnemonics Generation with Large Language Models via Overgenerate-and-Rank

Sep 14, 2026

Uploaded by Aldword

This paper studies keyword mnemonics for language and vocabulary learning, focusing on the automated creation of verbal cues that are traditionally time-consuming and labor-intensive. The proposed overgenerate-and-rank approach uses prompting with large language models to generate candidate verbal cues, then ranks them using psycholinguistic measures and evidence from a pilot user study. Automated imageability and coherence evaluation, complemented by human judgments from English teachers and learners, shows LLM-generated mnemonics match human ones while still leaving room to improve.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Abstract

1 Introduction

Keyword mnemonics and two-step cue linking

Challenges in automated KM generation

Related work on automated KM generation

Evaluation challenges and metrics

Contributions and proposed method

Exploring Automated Keyword Mnemonics Generation with Large Language Models via Overgenerate-and-Rank Jaewook Lee, Hunter McNichols, Andrew Lan University of Massachusetts Amherst {jaewooklee,wmcnichols,[andrewlan}@cs.umass.edu](andrewlan}@cs.umass.edu) Abstract In this paper, we study an under-explored area of language and vocabulary learning: keyword mnemonics, a technique for memorizing vocabulary through memorable associations with a target word via a verbal cue. Typically, creating verbal cues requires extensive human effort and is quite time-consuming, necessitating an automated method that is more scalable. We propose a novel overgenerate-andrank method via prompting large language models (LLMs) to generate verbal cues and then ranking them according to psycholinguistic measures and takeaways from a pilot user study. To assess cue quality, we conduct both an automated evaluation of imageability and coherence, as well as a human evaluation involving English teachers and learners. Results show that LLM-generated mnemonics are comparable to human-generated ones in terms of imageability, coherence, and perceived usefulness, but there remains plenty of room for improvement due to the diversity in background and preference among language learners. 1 Introduction Recent advances in natural language processing have expanded the exploration of developing automated methods for applications in language learning, including second language acquisition (Zhang et al., 2021 ; Yeung and Lee, 2021 ; Okano et al., 2023), linguistic skill modeling (Zylich and Lan, 2021), language assessment and correction (Katinskaia and Yangarber, 2021), and language practice through conversational chatbots (Tyen et al., 2022 ; Liang et al., 2023) . In this work, we investigate an intriguing but relatively under-explored vocabulary learning application: keyword mnemonics (KM) (Atkinson and Raugh, 1975) . KM is a widely utilized technique employed by language learners to memorize vocabulary effectively, through the creation of memorable associations with keywords that they already know, aided Target Word (Keywords) Verbal Cue Visual Cue | alleviate (a, leaf, he, ate) | | --- | | On his plate, there was a leaf he ate to alleviate his hunger. | | | Figure 1: An example of human-authored verbal cue (Geer and Geer, 2018) for the target word “alleviate” and visual cue generated by Stable Diffusion XL. We study automated verbal cue generation in this work and leave visual cue generation for future work. by verbal and visual cues. This technique is used in many language learning resources, from traditional books (Burchers et al., 2000 ; Heisig, 2011 ; Geerand Geer, 2018) to modern community-based learning platforms (Mnemonic Dictionary, 2024 ; Koohii Kanji, 2024), where users contribute mnemonicsand the community vote on their effectiveness. KM involves a two-step process that creates an acoustic and imagery link for the learner. For instance, consider a learner learning the target word“alleviate”, as shown in Figure 1. First, an acoustic link is established to an already-known set of keywords with similar pronunciation, e.g., “a, leaf, he, ate.” Second, a imagery link is established to these keywords with a verbal cue that evokes a mental image, e.g.,“On his plate, there was a leaf to alleviate his hunger.” These links can be further reinforced using a visual cue. Despite its effectiveness, KM generation is challenging since it demands significant human effort to meticulously find keywords that resemble the target word and create memorable verbal cues. For instance, the word “alleviate” has a large number of possible keyword combinations, such as “a, leaf, he, ate” or “a, levy, it”; sorting through them takes considerable effort from teachers and learners. Moreover, the task of creating verbal cues using the keywords that evoke vivid mental images addsan extra layer of complexity to the process. Existing work on automated KM gen

FAQs

How is the quality of generated mnemonics evaluated?

The paper uses automated evaluation of imageability and coherence and also includes human evaluation with English teachers and learners to assess perceived usefulness and overall cue quality.

Sep 14, 2026

Uploaded by Aldword