Detecting dementia in Mandarin Chinese using transfer learning from a parallel corpus

Proceedings of NAACL-HLT 2019, pages 1991–1997 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 1991 Detecting Dementia in Mandarin Chinese using Transfer Learning from a Parallel Corpus Bai Li1,2, Yi-Te Hsu1,2,4, Frank Rudzicz1,2,3 1 University of Toronto, Toronto, Canada 2 Vector Institute, Toronto, Canada 3 Toronto Rehabilitation Institute, Toronto, Canada 4 Academia Sinica, Taipei, Taiwan {bai, eeder, frank}@cs.toronto.edu Abstract Machine learning has shown promise for au- tomatic detection of Alzheimer’s disease (AD) through speech; however, efforts are hampered by a scarcity of data, especially in languages other than English. We propose a method to learn a correspondence between independently engineered lexicosyntactic features in two lan- guages, using a large parallel corpus of out- of-domain movie dialogue data. We apply it to dementia detection in Mandarin Chinese, and demonstrate that our method outperforms both unilingual and machine translation-based baselines. This appears to be the first study that transfers feature domains in detecting cog- nitive decline. 1 Introduction Alzheimer’s disease (AD) is a neurodegenerative disease affecting 5.7 million people in the US (As- sociation et al., 2018), and is the most common cause of dementia. Although no cure yet ex- ists, early detection of AD is crucial for an ef- fective treatment to delay or prepare for its ef- fects (Dubois et al., 2016). One of the earliest symptoms of AD is speech impairment, includ- ing a difficulty in finding words and changes to grammatical structure (Taler and Phillips, 2008). These early signs can be detected by having the patient perform a picture description task, such as the Cookie Theft task from the Boston Diagnos- tic Aphasia Examination (Goodglass and Kaplan, 1983). Previous models have applied machine learn- ing to automatic detection of AD, for example, Fraser et al. (2016) extracted a wide variety of lex- icosyntactic and acoustic features to classify AD and obtained 82% accuracy on the DementiaBank (DB) dataset. However, clinical studies of AD are expensive, so datasets of patient data are often scarce. Noorian et al. (2017) augmented DB with more a much larger corpus of normative data and improved the classification accuracy to 93% on DB. Similar linguistic differences between healthy and AD speech have been observed in Mandarin Chinese (Lai et al., 2009), but machine learning has not yet been applied to detecting AD in Man- darin. Daume III (2007) proposed a simple way of combining features in different domains, assum- ing that the same features are extracted in each domain. In our case, ensuring consistency of fea- tures across domains is challenging because of the grammatical differences between Mandarin and English. For example, Mandarin doesn’t have determiners or verb tenses, and has classifiers, which don’t exist in English (Chao, 1965). An- other method trains a classifier jointly on multiple domains with different features on each domain, by learning a projection to a common subspace (Duan et al., 2012). However, this method only accepts labelled samples in each domain, and can- not make use of unlabelled, out-of-domain data. Other work from …

Content truncated to 500 words