用大模型零样本检测多语言认知障碍，发现结构化语言特征更可靠
Multilingual Cognitive Impairment Detection in the Era of Foundation Models
- 用语音转录本和语言特征融合做认知障碍分类
- 结合人工特征与嵌入的监督模型表现最佳
- 小样本下语言差异影响标注数据价值，适合临床辅助诊断
我们评估了在英语、斯洛文尼亚语和韩语三种语言中，基于语音转录本进行认知障碍（CI）分类的效果。比较了零样本大语言模型（LLM）在三种输入设置下的直接分类性能——仅转录本、仅语言特征、两者结合——并与在留一法协议下训练的监督表格式方法进行对比。表格式模型基于手工设计的语言特征、转录本嵌入以及早期或晚期融合两种模态。跨语言实验表明，零样本LLM提供了有竞争力的无训练基线，但监督表格式模型整体表现更优，尤其当加入手工语言特征并与其嵌入融合时。少量样本实验显示，有限监督的价值具有语言依赖性：某些语言从额外标注样本中显著获益，而另一些语言则因缺乏丰富特征表示仍受限制。总体而言，小数据场景下，结构化语言信号与简单融合分类器仍是强而可靠的判别依据。
原文摘要 · Abstract (English)
We evaluate cognitive impairment (CI) classification from transcripts of speech in English, Slovene, and Korean. We compare zero-shot large language models (LLMs) used as direct classifiers under three input settings -- transcript-only, linguistic-features-only, and combined -- with supervised tabular approaches trained under a leave-one-out protocol. The tabular models operate on engineered linguistic features, transcript embeddings, and early or late fusion of both modalities. Across languages, zero-shot LLMs provide competitive no-training baselines, but supervised tabular models generally perform better, particularly when engineered linguistic features are included and combined with embeddings. Few-shot experiments focusing on embeddings indicate that the value of limited supervision is language-dependent, with some languages benefiting substantially from additional labelled examples while others remain constrained without richer feature representations. Overall, the results suggest that, in small-data CI detection, structured linguistic signals and simple fusion-based classifiers remain strong and reliable signals.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。