Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Voice and Pictographs for Disaster Communication


Published on

Talk at the 5th International Conference on Translating Voice Translating Regions:

Published in: Mobile
  • Be the first to comment

  • Be the first to like this

Voice and Pictographs for Disaster Communication

  1. 1. Voice and Pictographs for disaster communication with marginalized communities in Sri Lanka and the Philippines 5th International Conference Translating Voices, Translating Regions ‘Minority languages, risks, disasters and regional crises’ Europe House, London, UK 13 December 2017 Nuwan Waidyanatha Kunming, China
  2. 2. Outline ❏ About Sahana ❏ Who we are and what we offer ❏ Free and open source software ❏ Scope of the work ❏ Mobile Pictographs for Linguistically Challenged ❏ Research question, realization, objectives, & design ❏ Pictographs appropriateness and comprehension ❏ Voice-enabled ICTs for Disaster Management ❏ Research question, realization, objectives, & design ❏ Voice quality and interpretation ❏ m-Health for Real-Time Disease Surveillance & Notification ❏ Research question, realization, objectives, & design ❏ Data quality and disease predictions 2
  3. 3. About Sahana 3 ● Sahana EDEN is open source (MIT license); no licensing fees or restrictions. ● You have “ the rights to study, change, and distribute the software to anyone and for any purpose.” ● For over a decade, Sahana has been offering high quality information management systems for emergency preparedness, response, recovery and resilience-building accessible to all. Sahana is an Open Source Community with a mission to save lives by providing free and open source information management systems that improves the effectiveness of organizations and communities throughout the disaster cycle. Driving adoption to over 25 countries with 20 different disaster management modules:
  4. 4. Sahana Eden architecture 4 Host locally in the command center on a Laptop Host in the Cloud Synchronize these 2 together...
  5. 5. Focus on two disaster communication workflows 5 1. Incident Reporting Public asking for help from the Emergency Services indicating hazard (threat) and Immediate Needs HELP ME! DANGER! 2. Alerting / Warning Alerting Authority warning the public, in the area, indicating hazard and required response actions Emergency Services Receive Reports Send Alerts
  6. 6. Mobile Pictographs for Disaster Communication in support of Linguistically Challenged 6
  7. 7. Why Pictographs? 7 05 Reasons you cannot read: Illiterate, functionally illiterate, don’t know the language (tourist) sub Assume “linguistically challenged
  8. 8. Present day disaster symbols 8 ● Usually designed for use on maps and/or signage, not for alerting or reporting ● Symbols often concentrate on threat and omit response actions. Not representing complex situations. ● Designed and evaluated with professionals in mind ● Not explicitly addressing illiterates / linguistically challenged
  9. 9. Research Question 9 “How do we design symbols, workflows and technology to communicate disaster information through pictographs such that it can be used by linguistically challenged populations?”
  10. 10. Followup questions 1. How should these pictographs look? 2. How can they be entirely understood? 3. Are they universal enough to reach across cultural borders? 4. How can the pictures indicate the severity of an incident? 5. How can we communicate about actions to be taken? 6. How does this all go along with mobile devices? 7. ... 8. 10
  11. 11. Insights from Literature review 1. Pictographs generally add to understanding 2. Pictographs must be specific according to cultural background 3. Alerting pictographs are a demand of disaster agencies 4. Response actions rarely included in disaster symbol collections 5. Symbol usage mostly restricts to indicators on maps 11
  12. 12. Exercise with Experts at the CAP Workshop Bangkok 2016 12
  13. 13. Findings from the experiment with Experts ● Details were often omitted, especially numbers. If at all, they were indicated through depictions of context (e.g., flood height on a house) ● Arrows served for both location markers (“going downhill”) as well as to indicate time sequences. Both was well understood ● Time was represented by clocks. 13
  14. 14. Universal symbols 14 “Chemical Spill! “Help, Pirates!”
  15. 15. User-centered design and field-work 15 Philippines Sri Lanka
  16. 16. First survey research questions 1. What are the disaster information needs? 2. How would they categorize icon groups? 3. What are the choices of icons for illustrating various scenarios? 16 UN-OCHA Humanitarian Icons Guemil icon set FDGC Emergency Symbols
  17. 17. Disaster information needs 17 They all have information needs - Hazard specific information needs (average)
  18. 18. Current weather bulletins 18
  19. 19. Comprehension of text-based disaster information 19 In Sri Lanka, almost two thirds of the participants could understand official warnings when they had been read to them. Less than 20% said they did not understand them at all. Two thirds of deaf participants in the Philippines did not understand the warnings. This is also caused by the need to translate warnings in sign language.
  20. 20. User selection of icons for various categories 20 Flood (3, 1) Landslide (8, 10) Storm (11, 14)Heavy Rain (17) Shelter (23) Evacuate (36, 40) Observe (42, 43) Food (46, 50) Prepare (51) Rescue (58, 60) All Clear (63, 62) Injured (68, 70) People (72, 73) Direction (77, 78)
  21. 21. Second survey research questions ● Do people understand more complex pictographs? ● Does context/reference objects have an effect on comprehension? ● Do people recognize response actions as response actions? ● Do people understand indications of time? ● Does indication of response actions/time have negative/positive effects on comprehension? ● Do certain aspects affect impression of severity? 21
  22. 22. Alert pictograph layout 22 Response Action (e.g. Observe) Response Time (e.g. 24:00 hrs) Event (e.g. Rain) Incident (e.g. Flood) Severity (e.g. Extreme) Certainty (e.g. Likely) Urgency (e.g. Expected) EDXL-CAP
  23. 23. Comprehension counts 23 Group A Group B Group C Group D The event was understood. Response not understood when not indicated A B C D Activity - Each group was presented with a pictographs and asked to tell use what illustration was.
  24. 24. Presence and absence of context 24 ● Adding contextual entities (house and car) adds to comprehension. ● Only 52% understood the meaning without context. 67% 53% 70% Comprehension
  25. 25. Counts of “what will you do?” 25 They will tell others and ask for help; deaf will always ask for help Icons with reference and response will instigate to run/evacuate Group A Group B Group C Group D A B C D Activity - Participants were asked to indicate what they would do, with respect to the pictographs presented?
  26. 26. Relative severity rating 26 A B C D E Activity - Participants were asked to pick an icon to represent the immediate need for each situation: river is rising, reached a danger level and is flooding Reference is absolutely important Depictions with response actions regarded more severe
  27. 27. Response action 27 ● Additionally depicting a response action “evacuate” (right) caused confusion – comprehension rate drops ● However, understanding of what to do (correct response action) slightly increased – even if overall comprehension was lower
  28. 28. Response time 28 ● Time representations are generally not correctly understood ● Several representations of “for six hours” were tested ● >95% understood this is “about time”, but: ● Comprehension rate: 7.5% (for “in/for 6 hours”)
  29. 29. Reporting pictograph layout 29 Required Resources (e.g. boat, wheelchair, ambulance) Severity (e.g. water level reached river bank) Victim (e.g. Family) Number (e.g. 10) EDXL-SITREP
  30. 30. Choice of immediate need when reporting 30 ● Rising - mostly observe and evacuate ● Danger level - likelihood of evacuation increases ● Flooding - mostly evacuate but some said observe (why) Participants were asked to pick an icon to represent the immediate need for each situation: river is rising, reached a danger level and is flooding Do nothing Observe Evacuate
  31. 31. Choice symbol for reporting the number of victims 31 We asked the participants whether the symbol they chose represented FAMILY or PEOPLE? We asked the participants which of the three symbols would they chose to represent the number of VICTIMS?
  32. 32. Local and Cultural context 32 Cebuano (Filipino people) living in boathouses perceived the symbol as a rainy day and and strong waves Clocks — Bad Luck. In Chinese, saying 'giving a clock' (送钟 sòng zhōng /song jong/) sounds exactly like the Chinese words for 'attending a funeral ritual' (送终 sòng zhōng) and thus it is bad luck to gift clocks or watches. Evil eye is a kind of black magic that is transmitted on victim overlooking. More precisely, by gazing negative energy and negative thoughts are transferred or transmitted to the targeted person. Symbol for OK is the same as Money in sign language
  33. 33. Findings ● Local Design: ○ Pictographs must be designed locally with target audiences to address cultural differences ● User-centered Symbology ○ Usual recommendations for icon design might not hold for illiterates due to deficiencies in abstraction and categorization ● Level of abstraction ○ Choosing level of abstraction is crucial due to different cultural background and experiences ● Time and Numbers: ○ Abstract concepts like time and numbers must be handled with care – important, but hard to communicate clearly ● Response actions ○ Usually not considered, but crucial part of the information ● Limited success ○ Evaluations do not show required comprehension rates, even with professionals Literature survey available: 33
  34. 34. 34 Voice-enabled ICTs for Disaster Communication
  35. 35. Sarvodaya disaster management 35 Voice for Alerting and Response >> Full Story ● Community members (victims or associates) ○ call the District or Head Office ○ report of incidents. ● Collect ground truth ○ Dispatched youth with cameras, laptops, and dongles ● Blog situation on ● Get word to media and donors ● Secure response resources
  36. 36. Freedom Fone (IVR) and Sahana (IMS) 36 Integrate Freedom Fone IVR with Sahana disaster information management system to offer voice for alerting and reporting YOUTUBE VIDEO
  37. 37. Alerting (TTS) 37 What we did - human Interpreters EDXL-CAP What we wanted - Procedure for automation
  38. 38. Reporting (ASR) 38 EDXL-SITREP What we did - human Interpreters What we wanted - Procedure for automation
  39. 39. Voice-enabled disaster comm research design 39 Lanka Jathika Sarvodaya Shramadana Sangamaya Hazard Information Hub @ Community Disaster Management Center, Moratuwa, HIH Manager, 3 HIH Operators Four Districts: Colombo, Matara, Nuwara-eliya, Ratnapura, ~ 10 - 13 CERT members from each district: Divisional/District Coordinators, Staff
  40. 40. 1. Mean Opinion Score scale: 5=Excellent, 4=Good, 3=Average, 2=Poor, 1=Bad 2. Percent Difficult (or Difficulty Score) scale: 4=Clear, 3=Partial), 2=Unclear, 1=Missing 3. Comparison Categorical Ratio scale: 7=much better, 6=better, 5=slightly better, 4=same, 3=slightly worse, 2=worse, 1=much worse 40 Research question – “What are the design strategies for facilitating an interoperable emergency data exchange platform for voice enabled alerting and situational reporting?” ITU-T Voice Quality Tests
  41. 41. Mean Opinion Score 41 Speaker-dependent 63% was good and excellent Speaker-independent 53% was good and excellent
  42. 42. MOS as a diagnostic test for voice-enabled reliable Em. Comm 42 ● Patterns of urban (Matara, Colombo) rural (Ratnapura, Nuwara-eliya) divide ● Except Colombo all others agree ~80% with classifier ● Colombo behavior could pure chance?
  43. 43. Percent difficult ● Partial, Unclear, or Missing emergency info can lead to false predictions/actions ● Speaker-dependent: survey with set words to speak ● Evaluators could predict (trained) ● Speaker-independent: gave reporting attributes but free to speak ● Evaluators unaware (untrained) ● User variability → speaker-dependent unfavorable (i.e. impractical to train each user) ● Perhaps a hybrid 43 Speaker-dependent ~13% was difficult in comprehending Speaker-independent ~30% was difficult in comprehending
  44. 44. Mean Opinion Score (SNR) ❏ MOS cut-off = 4.0 is a reliable choice for setting a cap for emergency comm IVR applications ❏ Circuit (mechanical) and environmental noise prevalent ❏ Approx. 50% bad, poor, and fair ❏ Calling CERT member to confirm info would be inefficiencies 44
  45. 45. 45 m-Health Real-Time Disease Surveillance and Notification
  46. 46. Syndromic Surveillance 46 ❏ Activated passive surveillance ❏ Data limited to 25 infectious diseases (avg LK=70 TN=600 monthly health records per district) ❏ Only 20% of the diagnosis are confirmed, rest are probable and suspected, likely to miss signs of emerging outbreak ❏ No data on other-communicable or chronic diseases ❏ Trend analysis is based on < 0.05% of patient population ❏ Planning and resource allocation is based on expert opinion, departments are not data driven Current Practicem-Health Introduced
  47. 47. 47
  48. 48. Pilot project in India and Sri Lanka 48
  49. 49. Quality of the digitized data 49 Cumulative 23% noise reduced to 4% after informing the consequences of false detections (SNR for sub intervals: 0.18, 0.40. 0.31. 0.04, 0.07) Cumulative 45% noise persistent - Assistants have no formal health training, not affiliated to hospitals/clinics, no incentive to correct the (SNR for sub intervals: 0.58, 0.30, 0.53, 0.57, 0.17) 1 Low quantities of data received from Health Sub Centers 2 Volume of records were better after including Primary Health Centers 3 Holiday effect: no records received 4 Learning curve getting medical officers to adopt to the new procedures of writing the diagnosis 5 Release of mHealthSurvey v1.3 with better predictive text
  50. 50. Digitizing problems that affected the data 50 Use of synonyms Goal fever =jail fever = typhus fever; Dementia = memory loss; Enteric fever = typhoid fever; Encephalitis = meningitis Insertion of symbols body ache = body-ache, body pain = body pain Changing the order of words muscle weakness = weakness in muscle stomach pain = pain in the stomach Inclusion and exclusion of adjectives 'Severe' memory loss vs memory loss Using local language when terms are unknown Leg vettuthal = 'broken leg' in tamil Using proposition and conjunctions between different terms Nasal stuffiness 'or' sneezing Over bleeding 'with' abdominal pain Long sentences Not able to identify color white and shining patches without any sense Mistaking Treatment for diagnosis Oral pils, remove catheter, vaccination Instructions Please specify details specify symptoms UK vs USA spelling Diarrhoea = diarrhea Vomiting = vommitting Test results as symptoms or signs BP 140/90, BP 120/100 Singular vs plural Fit / fits , cut / cuts Inconsistencies in the verb tense Faint, fainted, fainting
  51. 51. Observations of the data digitizing uncertainties 51 Fever greater than 7 days concentrated in February and March of 2010, mainly from a single location, during the non rainy season is legitimate The likelihood of a measles outbreak emerging only in a single location without spreading to other areas, given that it is a viral disease, is highly unlikely (miscoding muscle pain). The assistant entering the data had submitted data for “Toxoid vaccine” as Tetanus, resulting in unusual case Uniformity of geographic distribution of disease cases (low: concentrated in a few locations, high: spread over)
  52. 52. Investigate the use of ontologies and semantics 52
  53. 53. THANK YOU 53