A Morphological Approach to the Longitudinal Detection of Dementia Najoung Kim School of Computing, Korea Advanced Institute of Science and Technology 291 Daehak-ro, Daejeon, Republic of Korea njkim@nlp.kaist.ac.kr Jong C. Park* School of Computing, Korea Advanced Institute of Science and Technology 291 Daehak-ro, Daejeon, Republic of Korea park@nlp.kaist.ac.kr ABSTRACT The impact of cognitive impairment on linguistic abilities has been a topic of continuous interest in dementia studies. However, there is a lack of systematic agreement on the longitudinal association between dementia progression and the patients' morphological capacity, and the role of morphological phenomena other than inflection has been relatively underreported. We present a longitudinal study of writings by Iris Murdoch (diagnosed of Alzheimer's Disease after her death) and Arthur Conan Doyle (no known record of dementia diagnosis), using two novel measures to account for the usage of complex morphology and lexical innovation. The results imply an association between lexical innovation and cognitive decline caused by dementia, as observed in Murdoch's works beginning from her mid-fifties, in contrast to a milder tendency in Doyle's works. Our findings contribute to a potential for facilitating early diagnosis of dementia through automated language processing approaches. ∗ Author Keywords Lexical Innovation; Morphology; Dementia; Alzheimer’s Disease; Mild Cognitive Impairment, Natural Language Processing; Automatic Text Analysis. ACM Classification Keywords I.2.7 Natural Language Processing: Text analysis. INTRODUCTION The goals of this study are twofold. First, we investigate whether we could establish a statistically significant connection between the progression of dementia and the subjects' command of morphology. Whether the trends found in previous dialog-based experiments are similarly observed in text analysis or not, and whether including non- inflectional morphology yields a different result would be the focal points of this research. Second, if it is indeed the case that a significant correlation can be established between the subjects’ command of morphology and dementia progression, we would like to propose an automation guideline in order to facilitate future language processing applications for the detection of early symptoms of dementia or signs of a mild cognitive impairment (MCI) phase. ∗ Corresponding Author Motivation Dementia, such as Alzheimer's Disease (AD), is known to have a long preclinical phase; for example, a decade or longer for AD [22]. Prior research suggests that the patients go through a transitional stage of mild cognitive impairment (MCI) from healthy aging to dementia, in which the symptoms are often non-obvious or mild [16, 17, 20]. For this reason, identifying early signs of cognitive decline in an individual has been an important issue for clinical dementia research [23]. Studies have been conducted to apply a text processing approach to this issue, examining dementia patients' linguistic characteristics manifested in writings in a longitudinal manner [5, 6, 9, 12, 15, 18]. However, the role of morphology has not been investigated in such studies. The effect of dementia on morphology has been explored mostly in oral, task-based experiments involving a real- time speech production of the subjects. As noted in [1], there is a lack of research on the association between morphological competence 1 and diachronic progression of the disease. Moreover, most existing studies have exclusively considered inflectional morphology [1]. Building on from these previous studies, we attempt a longitudinal examination of potential declines in morphological capacity reflected in the subjects' writings. We furthermore expand the types of morphological phenomena accounted for by capturing non-inflectional paradigms through novel quantification methods. We also investigate whether these methods can be developed or integrated into an automated language processing system that detects preclinical stages of dementia. Related Work A similar longitudinal study has been conducted in [9] and [15] using the same dementia-diagnosed subject – Iris Murdoch. [9] examined novels by Murdoch in terms of diachronic lexical and syntactic changes. However, as pointed out by [15], this analysis was limited to 3 of her novels and showed problems in methodology. [15] presented a reinforced, larger-scale study of Murdoch's novels and additionally analyzed the works of P.D. James 1 The term “competence” in this article is not used in the context of any particular linguistic theory where the term bears a contrastive meaning to “performance.” Proceedings of HCI Korea 2016 - 8 - © 2016 Hanbit Media & HCI Korea