Jinju (Pearl) Kim

Authors

Jinju Kim, Seungho Kwon, Jong Hwan Ko

Abstract

to be updated

Keywords

to be updated

Authors

Jinju Kim, Haeji Jung, Youjeong Roh, Jong Hwan Ko, and David R. Mortensen

Abstract

Despite the fact that much communication takes place in them, low-resource language varieties used in specific regions or by specific groups remain neglected in the development of Multilingual Language Models. A great deal of cross-lingual research focuses on inter-lingual language transfer which strives to align allied varieties and minimize differences between them. However, for low-resource varieties, linguistic dissimilarity is also an important cue allowing generalization to unseen varieties. Unlike prior approaches, we propose a two-stage Language Generalization framework that focuses on capturing variety-specific cues while also exploiting rich overlap offered by high-resource source variety. First, we propose TOPPing, a source-selection method specifically designed for low-resource varieties. Second, we suggest a lightweight VAÇAÍ-Bowl architecture that learns variety-specific attributes with one branch while a parallel branch captures variety-invariant attributes using adversarial training. We evaluate our framework on structural prediction tasks as proxy for performance on other downstream tasks. Using VAÇAÍ-Bowl with TOPPing yields an average 54.62% improvement in the dependency parsing task, which serves as a proxy for performance on other downstream tasks across 10 low-resource varieties.

Keywords

to be updated

Links

OpenReview (AAAI LM4UC Oral) |

Authors

Haeji Jung, Jinju Kim, Kyungjin Kim, Youjeong Roh, and David R. Mortensen

Abstract

Transliteration has emerged as a promising means to bridge the gap between various languages in multilingual NLP, showing promising results especially for languages using nonLatin scripts. We investigate the degree to which shared script, overlapping token vocabularies, and shared phonology contribute to performance of multilingual models. To this end, we conduct controlled experiments using three kinds of transliteration (romanization, phonemic transcription, and substitution ciphers) as well as orthography. We evaluate each model on two downstream tasks—named entity recognition (NER) and natural language inference (NLI)—and find that romanization significantly outperforms other input types in 7 out of 8 evaluation settings, largely consistent with our hypothesis that it is the most effective approach. We further analyze how each factor contributed to the success, and suggest that having longer (subword) tokens shared with pre-trained languages leads to better utilization of the model

Keywords

Transliteration, Tokens, Phonemics

Links

arxiv |

Authors

Jinju Kim*, Taesoo Kim*, Dong Chan Kim, Jong Hwan Ko, and Gyeong-Moon Park

Abstract

The rapid advancement of Zero-Shot Text-to-Speech (ZS-TTS) technology has enabled high-fidelity voice synthesis from minimal audio cues, raising significant privacy and ethical concerns. Despite the threats to voice privacy, research to selectively remove the knowledge to replicate unwanted individual voices from pre-trained model parameters has not been explored. In this paper, we address the new challenge of speaker identity unlearning for ZS-TTS systems. To meet this goal, we propose the first machine unlearning frameworks for ZS-TTS, especially Teacher-Guided Unlearning (TGU), designed to ensure the model forgets designated speaker identities while retaining its ability to generate accurate speech for other speakers. Our proposed methods incorporate randomness to prevent consistent replication of forget speakers' voices, assuring unlearned identities remain untraceable. Additionally, we propose a new evaluation metric, speaker-Zero Retrain Forgetting (spk-ZRF). This assesses the model's ability to disregard prompts associated with forgotten speakers, effectively neutralizing its knowledge of these voices. The experiments conducted on the state-of-the-art model demonstrate that TGU prevents the model from replicating forget speakers' voices while maintaining high quality for other speakers.The demo is available at https://speechunlearn.github.io/ .

Keywords

Machine Unlearning, Generative Unlearning, Voice Privacy, Zero-Shot Text-to-Speech

Links

Openreview (ICML Main) | Openreview (ICML X MuGen) | conference | poster | github

Authors

Jinju Kim, Taehan Kim, Abdul Waheed, Jong Hwan Ko, and Rita Singh

Abstract

AI music generation is rapidly emerging in the creative industries, enabling intuitive music generation from textual descriptions. However, these systems pose risks in exploitation of copyrighted creations, raising ethical and legal concerns. In this paper, we present preliminary results on the first application of machine unlearning techniques from an ongoing research to prevent inadvertent usage of creative content. Particularly, we explore existing methods in machine unlearning to a pre-trained Text-to-Music (TTM) baseline and analyze their efficacy in unlearning pre-trained datasets without harming model performance. Through our experiments, we provide insights into the challenges of applying unlearning in music generation, offering a foundational analysis for future works on application of unlearning for music generative AI models.

Keywords

Machine Unlearning, Generative Unlearning, Music Generation, Text-to-Music

Links

preprint

Authors

Kyungjin Kim, Jinju Kim, Haeji Jung, David R Mortensen, Jongmo Seo

Abstract

Medical natural language processing (NLP) has greatly benefited from the increasing availability of electronic health records (EHRs) across multiple languages. However, most available resources are heavily biased toward English, restricting the development of non-English medical NLP models and limiting their broader adoption in healthcare AI. In general domain NLP, monolingual models trained on high-resource European languages often outperform multilingual models. However, this paper takes a domain-specific perspective, highlighting the unique linguistic characteristics of medical terminology. Using a diverse corpus of medical texts in English, Italian, Spanish, and French, we pretrain RoBERTa-based models in both monolingual and multilingual settings. Contrary to conventional NLP trends, our results suggest that multilingual training enhances cross-lingual knowledge transfer and can even outperform monolingual models in medical domains. Through a comparative analysis of orthographic and phonemic representations, we find evidence suggesting that the strong Latin roots of medical terminology may facilitate effective knowledge sharing among languages with similar scripts. These findings indicate that orthographically focused multilingual strategies could provide a more robust framework for modeling specialized medical terminology. In particular, multilingual training that capitalizes on script similarities appears to enable richer information utilization, which may contribute to improved medical NLP performance across languages.

Keywords

Phonemic Representations, NLP, Multilingual Models, Medical AI

Links

Paper