Text Classification by Contrastive Learning and Cross-lingual Data Augmentation for Alzheimer’s Disease Detection

Proceedings of the 28th International Conference on Computational Linguistics, pages 6161–6171 Barcelona, Spain (Online), December 8-13, 2020 6161 Text Classification by Contrastive Learning and Cross-lingual Data Augmentation for Alzheimer’s Disease Detection Zhiqiang Guo1, Zhaoci Liu1, Zhen-Hua Ling1∗, Shijin Wang2, Lingjing Jin3, Yunxia Li3 1National Engineering Laboratory for Speech and Language Information Processing, University of Science and Technology of China, Hefei, China 2iFLYTEK Research, iFLYTEK Co., Ltd., Hefei, China 3Shanghai Tongji Hospital, Tongji University School of Medicine, Shanghai, China {gzq,zcliu8}@mail.ustc.edu.cn, zhling@ustc.edu.cn, sjwang3@iflytek.com , {lingjingjin,doctorliyunxia}@163.com Abstract Data scarcity is always a constraint on analyzing speech transcriptions for automatic Alzheimer’s disease (AD) detection, especially when the subjects are non-English speakers. To deal with this issue, this paper first proposes a contrastive learning method to obtain effective representations for text classification based on monolingual embeddings of BERT. Furthermore, a cross-lingual data augmentation method is designed by building autoencoders to learn the text representations shared by both languages. Experiments on a Mandarin AD corpus show that the contrastive learning method can achieve better detection accuracy than conventional CNN-based and BERT- based methods. Our cross-lingual data augmentation method also outperforms other compared methods when using another English AD corpus for augmentation. Finally, a best detection accuracy of 81.6% is obtained by our proposed methods on the Mandarin AD corpus. 1 Introduction Alzheimer’s disease (AD) is a worldwide popular neurodegenerative disease and the major cause of de- mentia. It is estimated that 32.8 million people worldwide living with AD in 2015, resulting in economic costs of 604 billion US dollars, and the number of patients is expected to almost double every 20 years (Prince et al., 2016). With no cure has been found yet, in order to allow effective intervention in early stages to delay onset of the disease, early detection of AD is crucial (Dubois et al., 2016). Nonethe- less, nowadays common methods of AD diagnosis, including cognitive tests and biomarker detection, are too time-consuming and expensive to cover all potential patients. Thus, developing an effective and convenient method for early AD detection becomes a valuable research topic. Language impairment generally appears at the early stages of AD (Morris, 1996). Suffering the dis- ease, patients have significant worse performance than controls on a standard aphasia test, specifically in the areas of verbal expression, auditory comprehension, repetition, reading, and writing (Murdoch et al., 1987). The performances of AD patients on some language tasks, such as picture description and sentence repetition, are distinctive from healthy people, which give us a hint to utilize these tasks to detect AD. Other advantages of these language tasks include their simple procedure and quick process. One usually completes the picture description task within 5 minutes and needs few instructions. A lot of researches have applied machine learning approaches to automatic AD detection based on language tasks. Fraser et al. (2016) extracted total 370 features considering part-of-speech, syntax, acoustics, as well as other aspects of linguistics, and obtained a best average accuracy of 82% for binary classification using the 35 top-ranked features on …

Content truncated to 500 words