Assessment of Social Engagement and Cognitive Function for Studying - - PowerPoint PPT Presentation

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 Decline1

• 2812 adults, 65 yrs or older, 1982-94
• 0 vs. 5-6 ties: Twice more likely to decline!!
• 1S. 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 decline2
• Higher depression3
• 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?
• 2S. S. Bassuk et al. “Social disengagement and incident cognitive decline

in community-dwelling elderly persons.” In: Ann Intern Med 131.3 (1999).

• 3T. 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 Day4
• The lived day of an individual5
• Intrusive, measurement perturbs behavior
• Labor-intensive
• 4R. G. Barker et al. “One boy’s day”. In: (1951).
• 5K. 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
• 6M. 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 studies7

• 7M. 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 heavily8
• Entire interaction occurs through voice

– no gestures, facial expressions, . . .

• Many forms of dementia directly effect language
• We can recognize the content automatically, can scale !!

8P

. 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

Encrypted transcript Automatic Recognition Speech Encryption Encryption using Standard Advanced Encrypted Lexicon Speech & Speaker Detection Data Storage OHSU Speech Encrypted the Markers Computation of speaker IDs Annonymized Subject’s Residence talking to you! It was good w23 w56 w24 w46 w59 w45!

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.

• ther 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!

1.5 2 2.5 3 3.5 4 0.2 0.4 0.6 0.8 LOG10 [word count] Estimated probability Global duration

• Res. call duration
• Biz. call duration

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Can Days of the Week Distinguish Calls? No!

Mon Tue Wed Thu Fri Sat Sun 0.05 0.1 0.15 0.2 0.25 0.3 Day Probability of call Biz. Res.

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Can Hours of the Day Distinguish Calls? No!

4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0.05 0.1 0.15 0.2 0.25 0.3 Hour Probability of call Biz. Res.

In Summary

• Simple features are not sufficient!
• Need to examine the content of the conversations

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Content via State-of-the-Art Speech Recognizer

• Acoustic Models
• Trained on 2000 hour of speech
• 8000 pentaphone clustered states
• 150K Gaussians, w/ semi-tied covariance
• Language Models
• 47k vocabulary, 10M parameters
• 10M n-grams, trigrams
• Three Stage Decoding
• Speaker-independent models
• Vocal-tract length normalized models
• Speaker-adaptation
• Speaker-adapted models
• Maximum likelihood linear regression models
• 24% word error rate on 2004 NIST RT Task

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Simple Supervised Classification

• Audio =

⇒ transcript = ⇒ features = ⇒ classifier = ⇒ labels

• Transcripts: errorful

E.g., hello, this is mark is

• Features: simple word counts or lexical unigrams

E.g., c[hello] = 1, c[this] = 1, c[is] = 2, c[mark] = 1

• Classifier: support vector machines, linear, radial basis

functions

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Simple Supervised Classification: Results

Task 1: residential vs business, Task 2: family vs non-family, Task 3: familiar vs non-familiar, and Task 4: family vs other residential.

Features Task 1 Task 2 Task 3 Task 4 Unigram 87.2 76.6 72.9 78.0 Bigram 85.1 77.8 73.5 77.2 Trigram 83.2 74.0 71.4 76.3 Surface 69.6 72.0 62.1 75.7 Unigram + Surface 86.9 81.2 74.4 77.2

• High accuracies, 74-87%, in spite of ASR errors
• Fully automated classification of social relationships!

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

What Features Matter?

• Linguistic Inquiry and Word Count9
• 32 psychological constructs (affect, cognition, biological)
• 22 linguistic dimensions (POS)
• 7 personal categories (work, home, leisure activities)
• 3 paralinguistic dimensions (assents, fillers, nonfluencies)
• 9J. W. Pennebaker. “Linguistic inquiry and word count (LIWC)”. In: (2001).

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

What Features Matter?

Task 1: residential vs business, Task 2: family vs non-family, Task 3: familiar vs non-familiar, and Task 4: family vs other residential.

Features Task 1 Task 2 Task 3 Task 4 Unigram 87.2 76.6 72.9 78.0 Unigram-stem 87.8 76.0 74.3 76.0 LIWC 77.1 74.6 64.8 69.1 POS-unigram 78.4 66.8 59.8 67.1 POS-bigram 77.7 70.8 63.9 70.5 Unigram × POS 84.2 76.3 72.5 79.8 Unigram + POS 86.9 76.0 72.6 77.5 Stemming and Unigram × POS helps, LIWC not so much

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Feature Selection via Mutual Information

10 10 10 10

2

10

3

10

4

80 82 84 86 88 Verification accuracy Dictionary size Frequency truncation MI truncation

• More effective than POS
• Optimal performance with 1000+ words

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Mutual Information: Top 30 Words Business oriented Social oriented Press, thank, calling, infor- mation, service, customer, number, quality, please, pressed, representative, account, zero, seven, moni- tored, transferred, nine, six, transfer, services. Hi, dinner, she’s, high, home, dad, night, everybody, do- ing, tonight, later, hello, mom, anyway, bad, nice, sleep, to- morrow, house.

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Alternative: Classifying using Conversation Topics

• Fortunately, we can learn topics automatically, using

Latent Diriclet Allocations

• Utilize 4k unlabeled conversations to learn topics
• Each conversation may contain multiple topics
• Estimate the proportion of each topic in a conversation
• Then, use that to classify conversations

spoken words = ⇒ posterior over topics (θ) = ⇒ classifier = ⇒ labels

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Topic Features from Latent Dirichlet Allocation

2 5 10 20 30 50 75 100 76 78 80 82 84 86 88 Number of topics Accuracy Cross validation Verification

• No loss in performance, all the way down to 30 topics
• With 2-topics, naturally clusters into biz vs. social calls

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

LDA Two-Topic Word Distribution Topic 1 Topic 2 Invalid, helpline, eligibility, transactions, promotional, representative, mastercard, touchtone, activation, nom- inating, receiver, voicemail, digit, representatives, ballots, refills, classics, metro, ad- minister, transfers, reselling, exclusive, submit. Adorable, aeroplanes, Ar- lene, Astoria, baked, bis- cuits, bitches, blisters, blue- grass, bracelet, brains, Char- lene, cheeses, chit, Chris, clam, clientele, cock, crab, Davenport, debating, demen- tia, dime, Disneyland, Eileen, follies, gained

• For biz, probability mass is concentrated on few words
• For social, probability mass is more widely distributed

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Error Distribution: Across Homes Home Records Accuracy 1 8 87.5 2 103 84.5 3 42 81.0 4 6 100.0 5 27 77.0 6 74 94.6 7 25 88.0 8 43 90.7

• Accuracy uniformly better than 77%

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Error Distribution: Conversation Length Word Counts Chance Accuracy Percentile Range 0-20 30-87 62.12 75.76 20-40 88-167 51.52 83.33 40-60 168-295 60.61 90.91 60-80 296-740 59.09 93.94 80-100 741+ 59.38 93.75

• Accuracy degrades for shorter conversations
• Accuracy is stable > 300 words (2-3 minutes)

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Calls from Certain Number Always Correctly Classified?

• Upto 300 conversations from some numbers

10 20 30 40 50 60 70 80 90 100 20 40 60 Number of telephone contacts Classification accuracy (%)

• 50 / 125 correct all the time
• 5 consistently wrong (e.g., 65 calls to a lighting store)

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Nature of Everyday Telephone Conversations

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Openings & Closings

• Do different parts of the conversations contribute equally?
• Schegloff & Sacks: Openings and closings are distinct

30 50 100 250 500 1000 5 10 15 20 25 30 Number of words sampled Res/biz classification errror (%) Word sample from start Word sample from end Word sample randomly taken

• Openings are good, but closings are not

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Nature of Short vs. Long Calls

• Just saw, first 30 words are sufficient to classify
• But, accuracy degrades for short conversations
• Sparsity or intrinsic nature of short conversations?

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Sparsity or Intrinsic Nature: Why Are Short Calls Difficult? Truncate all calls to 30-words, then comparing accuracy Original Length (# Words) Split Accuracy Percentile Range Res / Biz 0-20 30-87 62.1 / 37.9 78.6 20-40 88-167 48.5 / 51.5 82.8 40-60 168-295 39.4 / 60.6 91.4 60-80 296-740 40.9 / 59.1 87.8 80-100 741+ 59.4 / 40.6 93.4

• Original longer calls are still easier to classify
• Degradation is not due to sparsity, but inherent ambiguity

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

What Length of Observation is Sufficient? Jensen-Shannon divergence

• 12-month estimate vs shorter windows
• Averaged over all windows and residences

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

What Length of Observation is Sufficient?

2 4 6 8 10 12 0.02 0.04 0.06 0.08 JS−Div Duration (months)

(a) Business vs. residential

2 4 6 8 10 12 0.005 0.01 0.015 0.02 0.025 JS−Div Duration (months)

(b) Family vs. non-family

2 4 6 8 10 12 0.002 0.004 0.006 0.008 0.01 0.012 JS−Div Duration (months)

(c) Familiar vs. non-familiar

2 4 6 8 10 12 0.01 0.02 0.03 0.04 JS−Div Duration (months)

(d) Family vs. res. non-family

• Reference labels, discard more calls, need longer obs.
• Automatic labels, use all calls, stable w/ shorter obs.

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Summary and Implications

• Framework for measuring social engagement
• Infer types of social interaction automatically
• Accuracies of 74-88%, with 30 topics or first 30 words
• Can be improved by collating information across calls
• Content more useful than the medium specific features;

applicable to emails, chats, . . . ; cover other demographies

• More importantly, our framework allows deeper analysis
• Now, expanding to 50 subjects, cross-sectional analysis
• Additionally, include affect, health topics, who spoke what

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

Cognitive Function

• Digit Span (forward, reverse)
• Stroop test
• . . .
• Narrative retelling task

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Weschler Logical Memory Test

Reference text

Anna Thompson of South Boston employed as a cook in a school cafeteria reported at the police station that she had been held up on State Street the night before and robbed . . . police touched by the woman’s story took up a collection for her.

An example retelling

Ann Taylor worked in Boston as a cook. And she was robbed of sixty-seven dollars. Is that right? And she had four children and reported at the some kind of

• station. The fellow was sympathetic and made a

collection for her so that she can feed the children.

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Clinical Scoring of WLM

Reference text, chunked into story elements.

Anna / Thompson / of South / Boston / employed / as a cook / in a school / cafeteria / reported / at the police / station / that she had been held up / on State Street / the night before / and robbed / . . . / police / touched by the woman’s story / took up a collection / for her.

An example retelling with 12 recalled story elements.

Ann Taylor worked in Boston as a cook. And she was robbed of sixty-seven dollars. Is that right? And she had four children and reported at the some kind of

• station. The fellow sympathetic and made a collection

for her so that she can feed the children.

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Our Task: Emulate Clinical Scoring Challenges

• Diverse lexical variants
• Paraphrasings
• Disfluencies
• ASR errors

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

One Approach: ASR + MT

• Compute best hypothesis from the ASR
• Align the hypothesis with reference text
• Use MT word-alignment model for aligning

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Detecting Story Elements

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Alternate Approach: Tagging Problem

!"#####$%%$##&'!()*!%###&'$&+*####',-#####%$(,# (./0.1#23./453672# &.84942:3;#<1<=</24># ?###########@################A##############################?####### $*-#@BC<42# $*-#DE%# ',F####$%%$#&'!()*!%###&'$&+*#####$%#####$%GH,-# ',F# '$F# $%%$# $%I# &'!()*!%# &'$&+*# !J-# ',-# $%# %$(,#

• KL&#

$%GH,-#

L<.203<#M<52:34#9NBO6=># P:3#2Q<#E-L# $R?# $%R?# S# '$FR@# ',FR@# S# $R?# $%R?# $%IR@# $%%$R@# S# $R?# $%R?# S# &'!()*!%R@# S# $R?# $%R?# S# &'$&+*R@# S# S# $R?# $%R@# S# $%GH,-R@# S#

• KG'&R@#

$R?# $%R?# S#

S#

S#

• KG'&R@#

%$(,R@#

P(y|x) = exp(Σd

i=1λiφi(x, y))

Z(x)

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Experiments

• Training: retellings from 144 subjects
• Testing: retellings from 70 subjects

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

ASR System Baseline: Broadcast News ASR System

• 4000 clustered pentaphones, 150K Gaussians
• 84K vocab, 3M language model ngrams
• Multistage discriminative decoding
• Performance: 13.1% on RT04

System WER (%) Baseline 47.2 AM adaptation 38.0 LM adaptation 28.3 AM+LM adaptation 25.6

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Experiments: Configurations

• Two tagging schemes

Tagging anna rent was due UO-tags U1 U19 U19 U19 BIO-tags B1 B19 I19 I19

• Two types of ASR systems: baseline, adapted
• Two types of ASR outputs: 1-best, confusion nets (WCN)

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Experiments: Results 1-best WCN Manual BL AM+LM BL AM+LM N/A Context Independent Features UO 79.3 89.3 80.8 88.1 91.0 BIO 78.9 89.0 79.3 87.7 91.1 Context Dependent Features UO 78.4 90.0 79.7 87.7 91.6 BIO 78.2 89.3 80.5 88.3 91.9

• WCN > 1-best, when ASR errors are high
• F-score from ASR close to manual

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

MCI Experiments: Results 1-best WCN Manual BL AM+LM BL AM+LM N/A Context Independent Features UO + SVM 0.65 0.72 0.67 0.75 0.78 BIO + SVM 0.65 0.72 0.70 0.76 0.77 Context Dependent Features UO + SVM 0.66 0.73 0.67 0.73 0.79 BIO + SVM 0.67 0.73 0.69 0.73 0.79

• Surprisingly high AUC, considering this is only one test!
• Best results with WCN, again close to that from manual

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Summary

• Fully automate scoring of a cognitive task
• Easy to include reverse digit recall, animal recall, etc
• Applicable for evaluating fidelity of any narrative retellings

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Assessing Health & Well-Being: Wish List In-Clinic − → Real-World Episodic − → Continual Subjective − → Objective Intrusive − → Non-intrusive Labor-Intensive − → Automated

• Technology is begining to transform assessments
• Physcial Domain: AGPS, accelerometer, in-home sensors
• Social and Cognitive Domain: Speech & language

technology!

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Acknowledgements

• Post-doctoral researchers: Anthony Stark
• Doctoral students: Alireza Bayesteh, Meysam Asgari,

Maider Lehr, Emily Prud’hommeaux

• Collaborators: Jeffrey Kaye, Kathy Wild, Brian Roark

Social Engagement, Cognitive Decline and Measurements Assessing Social Engagement Assessing Cognitive Function Conclusion

Questions? Comments? Suggestions? For publications, see http://www.csee.ogi.edu/~zak