Problem
- Lack of mnemonic generation models for typologically distinct language pairs. (e.g. Korean and English)
- Previous work only supports English-only. It does not address the cross-lingual IPA variation.
Solution
- Design and Implementation
- Leveraging previous work with English-only support
- Applying our own transliteration methodology to retrieve keywords.
- Validation: Automated evaluation and human experiments.
Contributions (Co-first author)
- Designed and implemented the complete pipeline from English words to verbal cue.
- Developed automated evaluation methods and designed/implemented ablation studies.
- Participated in human experiment design.