[论文] [论文] Preserve to Expand: Source-Aware Adaptation and Model Merging for...

论文概要

研究领域: NLP
作者: Yuchen Zhang, Haralambos Mouratidis, Ravi Shekhar
发布时间: 2026-10-26
arXiv: 2610.35846

中文摘要

多语言 ASR 系统扩展到新语言时，若适配必须保持已有语言性能、又不能回放源语言训练数据，挑战很大。我们在基于连接器的 Speech LLM 中研究该问题：可训练连接器把语音表征映射进冻结 LLM。我们评估适配连接器的目标更新构造与合并策略，以在获取新语言的同时保住源语言。单目标扩展中，仅目标微调与标准 TIES 合并能学会新语言但造成严重源语言遗忘，SSU-Wanda+TIES 给出最佳保源折中。多目标扩展中，更新合并稳定保持源语言；递归扩展在顺序有利时可改进目标适配，但对顺序高度敏感。结果表明，多语言 ASR 扩展需要在新语言获取、源语言保持以及各目标语言更新之间的交互上取得平衡。

原文摘要

Expanding multilingual ASR systems to new languages is challenging when adaptation must preserve previously supported languages without replaying source-language training data. We study this problem in connector-based Speech LLMs, where a trainable connector maps speech representations into a frozen large language model. We evaluate target-update construction and merging strategies for adapting the connector while preserving source-language performance. In single-target expansion, target-only fine-tuning and standard TIES merging learn new languages but cause severe source-language forgetting, while SSU-Wanda+TIES gives the best source-preserving trade-off. In multi-target expansion, update merging provides stable source retention, whereas recursive expansion can improve target adaptation under favorable orders but is highly order-sensitive. These results show that multilingual ASR expansion requires balancing target acquisition, source retention, and interactions among target-language updates.

自动采集于 2026-10-28

#论文 #arXiv #NLP #小凯