Enriching Complex Networks with Word Embeddings for Detecting Mild Cognitive Impairment from Speech Transcripts

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1284–1296 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1118 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1284–1296 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1118 Enriching Complex Networks with Word Embeddings for Detecting Mild Cognitive Impairment from Speech Transcripts Leandro B. dos Santos1, Edilson A. Corrˆea Jr1, Osvaldo N. Oliveira Jr2, Diego R. Amancio1, Let´ıcia L. Mansur3, Sandra M. Alu´ısio1 1 Institute of Mathematics and Computer Science, University of S˜ao Paulo, S˜ao Carlos, S˜ao Paulo, Brazil 2 S˜ao Carlos Institute of Physics, University of S˜ao Paulo, S˜ao Carlos, S˜ao Paulo, Brazil 3 Department of Physiotherapy, Speech Pathology and Occupational Therapy, University of S˜ao Paulo, S˜ao Paulo, S˜ao Paulo, Brazil {leandrobs,edilsonacjr,lamansur}@usp.br, chu@ifsc.usp.br {diego,sandra}@icmc.usp.br Abstract Mild Cognitive Impairment (MCI) is a mental disorder difficult to diagnose. Lin- guistic features, mainly from parsers, have been used to detect MCI, but this is not suitable for large-scale assess- ments. MCI disfluencies produce non- grammatical speech that requires man- ual or high precision automatic correction of transcripts. In this paper, we mod- eled transcripts into complex networks and enriched them with word embedding (CNE) to better represent short texts pro- duced in neuropsychological assessments. The network measurements were applied with well-known classifiers to automati- cally identify MCI in transcripts, in a bi- nary classification task. A comparison was made with the performance of tra- ditional approaches using Bag of Words (BoW) and linguistic features for three datasets: DementiaBank in English, and Cinderella and Arizona-Battery in Por- tuguese. Overall, CNE provided higher accuracy than using only complex net- works, while Support Vector Machine was superior to other classifiers. CNE pro- vided the highest accuracies for Dementia- Bank and Cinderella, but BoW was more efficient for the Arizona-Battery dataset probably owing to its short narratives. The approach using linguistic features yielded higher accuracy if the transcriptions of the Cinderella dataset were manually revised. Taken together, the results indicate that complex networks enriched with embed- ding is promising for detecting MCI in large-scale assessments. 1 Introduction Mild Cognitive Impairment (MCI) can affect one or multiple cognitive domains (e.g. memory, language, visuospatial skills and executive func- tions), and may represent a pre-clinical stage of Alzheimer’s disease (AD). The impairment that affects memory, referred to as amnestic MCI, is the most frequent, with the highest conversion rate for AD, at 15% per year versus 1 to 2% for the general population. Since dementias are chronic and progressive diseases, their early diagnosis en- sures a greater chance of success to engage pa- tients in non-pharmacological treatment strategies such as cognitive training, physical activity and socialization (Teixeira et al., 2012). Language is one of the most efficient in- formation sources to assess cognitive functions. Changes in language usage are frequent in patients with dementia and are normally first recognized by the patients themselves or their family mem- bers. Therefore, the automatic …

Content truncated to 500 words