Idea density for predicting Alzheimer’s disease from transcribed speech

Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 322–332, Vancouver, Canada, August 3 - August 4, 2017. c⃝2017 Association for Computational Linguistics Idea density for predicting Alzheimer’s disease from transcribed speech Kairit Sirts1, Olivier Piguet2,3 and Mark Johnson4 1Institute of Computer Science, University of Tartu 2School of Psychology and Brain & Mind Centre, The University of Sydney 3Neuroscience Research Australia, The University of New South Wales 4Department of Computing, Macquarie University kairit.sirts@ut.ee, olivier.piguet@sydney.edu.au mark.johnson@mq.edu.au Abstract Idea Density (ID) measures the rate at which ideas or elementary predications are expressed in an utterance or in a text. Lower ID is found to be associated with an increased risk of developing Alzheimer’s disease (AD) (Snowdon et al., 1996; Engel- man et al., 2010). ID has been used in two different versions: propositional idea den- sity (PID) counts the expressed ideas and can be applied to any text while semantic idea density (SID) counts pre-defined infor- mation content units and is naturally more applicable to normative domains, such as picture description tasks. In this paper, we develop DEPID, a novel dependency-based method for computing PID, and its version DEPID-R that enables to exclude repeat- ing ideas—a feature characteristic to AD speech. We conduct the first comparison of automatically extracted PID and SID in the diagnostic classification task on two different AD datasets covering both closed- topic and free-recall domains. While SID performs better on the normative dataset, adding PID leads to a small but significant improvement (+1.7 F-score). On the free- topic dataset, PID performs better than SID as expected (77.6 vs 72.3 in F-score) but adding the features derived from the word embedding clustering underlying the auto- matic SID increases the results consider- ably, leading to an F-score of 84.8. 1 Introduction Idea density (ID) measures the rate of propositions or ideas expressed per word in a text and it is con- nected to some very interesting results from neu- roscience related to Alzheimer’s disease (AD). In The old gray [MARE] has a very large [NOSE]. Dependencies Propositions det(The, mare) amod(old, mare) (OLD, MARE) amod(gray, mare) (GRAY, MARE) nsubj(mare, has) (HAS, MARE, NOSE) det(a, nose) advmod(very, large) (VERY, (LARGE, NOSE)) amod(large, nose) (LARGE, NOSE) dobj(nose, has) (HAS, MARE, NOSE) punct(., has) Table 1: The alignment of the dependency and propositional structures. The example sentence is due to Brown et al. (2008). The predicative proposition (HAS, MARE, NOSE) is represented by two dependency arcs. particular, two longitudinal studies—the Nun Study (Snowdon et al., 1996) and the Precursors Study (Engelman et al., 2010)—suggest that lower ID, as measured from the essays written in young age, is associated with the higher probability of develop- ing AD in later life. Two alternative definitions of idea density have been used in relation to AD. Propositional idea density (PID) counts the number of any ideas ex- pressed in the text, setting no restriction to the topic (Turner and Greene, 1977; Chand et al., 2010). An example sentence with its ideas or propositions is …

Content truncated to 500 words