Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Assessment of Social Engagement and Cognitive Function for Studying Aging Izhak Shafran Center for Spoken Language Understanding (CSLU) Oregon Health & Science University Portland, OR CMU LTI Colloquium
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Number of Social Ties Vs. Cognitive Decline 1 • 2812 adults, 65 yrs or older, 1982-94 • 0 vs. 5-6 ties: Twice more likely to decline!! 1 S. S. Bassuk et al. “Social disengagement and incident cognitive decline in community-dwelling elderly persons.” In: Ann Intern Med 131.3 (1999).
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Social Engagement and Health Deleterios Affects of Social Disengagement • Cognitive decline 2 • Higher depression 3 • Slower recovery from health incidents Understanding Social Engagement • What aspects of social engagement matter? • Can we detect unhealthy levels of disengagement? • Can we intervene and promote engagement? How? 2 S. S. Bassuk et al. “Social disengagement and incident cognitive decline in community-dwelling elderly persons.” In: Ann Intern Med 131.3 (1999). 3 T. A. Glass et al. “Social engagement and depressive symptoms in late life: longitudinal findings”. In: J Aging Health 18.4 (2006), pp. 604–628.
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Measuring Social Function: Subject’s Perspective • Questionnaires • E.g. “How many friends do you have?” • Relies on memory, hence confounding • Experience sampling • E.g. Beep: “Were you alone or with someone?” • No easy trade-off: frequent sampling vs perturbing behavior In Summary, • Easy to administer • Subject’s perspective, has inherent value, but need more • Need fine-grained information
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Measure’s from Observer’s Perspective Follow a subject and record their everyday life • One Boy’s Day 4 • The lived day of an individual 5 • Intrusive, measurement perturbs behavior • Labor-intensive 4 R. G. Barker et al. “One boy’s day”. In: (1951). 5 K. H. Craik. “The lived day of an individual”. In: (2000).
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion An Acoustic Window into Social Behavior Electronically Activated Recorder (EAR) 6 • Record ambient conversations throughout the day • Annotators listen to recording and annotate • Annotations include transcripts, social context, affect • For privacy-protection, recording not continuous 6 M. R Mehl and J. W. Pennebaker. “The sounds of social life: a psychometric analysis of students’ daily social environments and natural conversations.” In: J Pers Soc Psychol 84.4 (2003), pp. 857–870.
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion An Acoustic Window into Social Behavior • Unobtrusive, doesn’t perturb behavior • Samples subjects’ naturalistic conversations • Layers of information • Interaction: in-person, on the phone, alone • Talking to: male(s), female(s), mixed group • Location: at home, in transit, dining/bar, recreation • Activity: radio/tv, work, chores, sports, entertainment • Mood: laugh, sing, cry, mad, sigh • Health: cough/sneeze Many successful social psychology studies 7 7 M. R Mehl. “The lay assessment of subclinical depression in daily life”. In: Psychol Assess 18.3 (2006), pp. 340–5.
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion An Acoustic Window into Social Behavior In Summary • No effort by the subject, doesn’t peturb behavior • Observer’s perspective, consistency can be controlled • Easy to record observations • But, need to listen and annotate, labor intensive! • And too noisy for automation with current technology
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Social Engagement via Telephony Premise • Older adults are less mobile, rely on telephones heavily 8 • Entire interaction occurs through voice – no gestures, facial expressions, . . . • Many forms of dementia directly effect language • We can recognize the content automatically, can scale !! 8 P . Taylor et al. Growing old in America: Expectation vs. Reality . Tech. rep. Pew Research Center, 2009.
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion From Call Logs: Social Networks
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion From Call Content: Social Relationships
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Social Engagement via Telephony • Unobtrusive, doesn’t perturb behavior • Samples subjects’ naturalistic conversations • Layers of information • Talking to: male(s), female(s), mixed group • Affect: happy, sad, angry, . . . • Health: cough/sneeze
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Subject Pool: Fairly Active! Activity Daily Weekly Monthly Yearly Rarely Read a newspaper Listen to radio/TV news Use a computer Listen to music Watch TV Watch movies Follow finances/investments Have visitors Visit others at their homes Eat out Take a class Read a book Attend a club meeting Travel out of town Care for pet
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Subjects and Corpus • 10 subjects, 79 years or older • Social questionnaire • Unique Corpus • Call logs, includings numbers called to/from, time, duration • ALL incoming/outgoing telephone conversations recorded • Enrollment and exit interviews, picture description task • Ongoing collection: 45 residences more, 2500 hours so far Valuable Orthogonal Data • Cognitive (neuropsychological) tests, MRI, activity reports • Sensor data, including doors, motion, medicine, . . . • Longitudinal analysis: backtrack future health outcomes
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion ORCATECH’s Living Lab Secure Internet
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Security and Privacy of the Data Annonymized Speech & speaker IDs Speaker Encrypted Detection Encryption Speech using Computation of Advanced the Markers Encryption Standard Automatic It was good Data Storage Encrypted transcript talking to you! Speech Encrypted Recognition w23 w56 w24 Lexicon w46 w59 w45! OHSU Subject’s Residence
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Our Tasks Task 1: residential vs. business Task 2: family vs. non-family Task 3: familiar vs. unfamiliar Task 4: family vs. other residential • Subset of data was labeled for training and testing • For example, business vs. residential • ≈ 8.3k conversations, after trimming short ones • labels for ≈ 4.3k (2.7k residential, 1k business) • no labels for ≈ 4k • balanced training (1.8k) and test (328k) sets
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Can Duration Distinguish Calls? No! 0.8 Global duration Estimated probability Res. call duration 0.6 Biz. call duration 0.4 0.2 0 1.5 2 2.5 3 3.5 4 LOG10 [word count]
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Can Days of the Week Distinguish Calls? No! Biz. 0.3 Res. Probability of call 0.25 0.2 0.15 0.1 0.05 0 Mon Tue Wed Thu Fri Sat Sun Day
Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion Can Hours of the Day Distinguish Calls? No! Biz. 0.3 Res. Probability of call 0.25 0.2 0.15 0.1 0.05 0 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Hour In Summary • Simple features are not sufficient! • Need to examine the content of the conversations
Recommend
More recommend
