BabyTalk-Clan
First Year Presentation
Wendy Moncur
Universities of Aberdeen & Dundee
Supervisors: Ehud Reiter & Judith Masthoff
BabyTalk-Clan
Introduction Summary of Research Areas Current Research Research Direction
Who is this research about?
BabyTalk: Tailoring reports about babies in Neonatal Intensive Care
Sensor readings
Event records
Data about the baby
Data Interpretation
& NLG processing
Heart rate, breathing,…
Medication, lab results, equipment settings.
Tailored reports
Medical staff
Parents, friends & family
BabyTalk: Tailoring reports about babies in Neonatal Intensive Care
Sensor readings
Event records
Data about the baby
Heart rate, breathing,…
Medication, lab results, equipment settings.
Tailored reports
Medical staff
Parents, friends & family
Web/ mob phone text
Data Interpretation
& NLG processing
BabyTalk-Clan (BT-Clan) Objectives Part of the BabyTalk project
Communicate tailored updates about the baby to family & friends- Mobilise emotional support for parents
- Parents may be too busy /upset to communicate info themselves
NLG application• Adaptation to recipient
• Adaptation to parents’ requirements
• Deliver web/ mobile phone texts to friends & family
The Problem: Mobilising Emotional Support in a Crisis
The newborn baby is the parents’ priority.
Parents benefit from the emotional support of friends and family.
If they are to give support, friends and family first need to know that something is wrong.
BT-Clan can mobilise support.
BabyTalk-Clan
Introduction Summary of Research Areas
• Social networks
• Affective NLG
• Personalisation for eHealth Current Research Research Direction
Research Area: Social Networks
Describe relationships between people or groups (Churchill et al, 2005)
Applications • Social - Friendster, FlickR
• Business – LinkedIn
• Psychology – social support
Research Area: Social Networks
Describe relationships between people or groups
Applications • Social - Friendster, FlickR
• Business – LinkedIn
• Psychology – social support
But ... • Don’t map social closeness.
• Don’t know how to acquire this data
Node Tie
Hi Sam, I’ll meet you at 8, ok?Flow
Research Area: Social Networks Social Networks studied by psychologists
“…provision of psychological and material resources intended to benefit an individual’s ability to cope with stress.” (Cohen 2004).
Social support can be :• Instrumental – provision of material aid• Informational – advice, guidance• Emotional – empathy, caring, chance to express feelings
Emotional support reduces stress
Most effective when provided by friends and family, rather than clinicians or support groups.
BT-Clan aims to mobilise emotional support in the social network
Research Area: Affective NLG
NLG (Natural Language Generation) • Tailored, comprehensible text from underlying
knowledge base
Affective NLG:• Portrays emotional state of interlocutor
• Induces emotional effect on hearer
• Goal = believability
• Early days as a research area
Research Area: Affective NLGResearch Directions: Detect the emotional state of the user(de Rosis, Mellish 2007)
• Physiological tests – cortisol, electromyography, heart rate, …• Validated questionnaires – Big 5, PANAS, HADS, EPDS…
• Only detect relevant emotions/ traits
• We will use questionnaires - EPDS & HADS - to establish degree of depression of parents: this affects how they communicate.
Research Area: Affective NLG
Research Directions: Develop goals involving emotional state (de Rosis, Mellish 2007)
• Advice-giving. E.g. - STOP tailored letters for smoking cessation (Reiter, Robertson et al. 2003)
• Intelligent tutoring (Lester et al, 2000).
• Adaptation to anxiety and belief state (Hudlicka, McNeese M.D. 2002; Fleischman, Hovy 2002).
• Social interaction and amusement (Binsted, Bergen et al. 2006; Sundström 2005)
• BT-Clan: Tailor information appropriately for recipients according to social closeness.
Research Area: Affective NLG
George has been changed from nasal CPAP to ventilation. He has been in oxygen between 25 and 40%. George has been given a dose of replacement surfactant. He has been started on morphine to keep him comfortable.
He has been started on vecuronium to keep him from moving and fighting the ventilator.
He has been started on dopamine to help his blood pressure.
Medical summary report on ‘test’ baby George, supplied by current system.
(Links to glossary underlined).
Research Directions: use language in a directed way to achieve goals involving emotional state (de Rosis, Mellish 2007)
Research Area: Personalisation for eHealth Patients prefer personalised information (Cawsey, Jones
et al. 2000)
Personalised information seen as more relevant (Bental 1998)
Affective NLG can be used to tailor info in the health domain (Grasso,Cawsey et al. 2005).
No work on patients’ supporters yet, but ….“…the caregiver is no longer a bystander in ….care, but is
an actual or potential ‘co-user’ or 'co-client’ of services.” (O'Mara 2005)
BT-Clan considers personalisation for patient supporters
BabyTalk-Clan
Introduction Summary of Research Areas Current Research Research Direction
Initial Research
Prototyped a simple modelling tool to capture social closenesss.
Tested with parents who have had babies in NICU in the past. N=7 (2 male, 5 female).
Simplified for usability
Button for baby at centre
Explored how these dynamics influenced information transmission from parents.
A mother’s social network
* Note: Photos are not of the real individuals concerned.
Husband. He’s been keeping everyone up-to-date about Emma.
Our beautif ul baby, Emma , born 10 weeks premature.
Friends across the country. They’ve all been phoning, asking how me & Emma are doing.
Me, J ane. Anxious. Too tired to phone anyone right now.
Brother. Single, lives in Australia.
My mum. I tell her everything.
My mother-in-law. Housebound. Worries a lot.
Colleagues. I went into labour at work. They are worried about me & the baby.
Friend. She doesn’t have kids, and is always busy.
Figure 1: “My social network” by Jane, mother of Baby Emma
Husband’s best f riend.
Husband. He’s been keeping everyone up-to-date about Emma.
Our beautif ul baby, Emma , born 10 weeks premature.
Friends across the country. They’ve all been phoning, asking how me & Emma are doing.
Me, J ane. Anxious. Too tired to phone anyone right now.
Brother. Single, lives in Australia.
My mum. I tell her everything.
My mother-in-law. Housebound. Worries a lot.
Colleagues. I went into labour at work. They are worried about me & the baby.
Friend. She doesn’t have kids, and is always busy.
Figure 1: “My social network” by Jane, mother of Baby Emma
Husband’s best f riend.
Possible report tailoring by band of social networkBand Report content Recipient
A Detailed report
Current photos of baby.
B Mildly summarized version of report, avoiding medical terminology.
Current photos of baby.
NICU visiting times.
C Extensively summarized version of report, avoiding medical terminology.
NICU visiting times.
D New developments, but no serious bad news.
NICU visiting times.
E One-off communication to say that the baby has been born & is in NICU.
Option to get Band D information on request.
Results: Factors that affected information dissemination.
Social closeness affected by:• Circumstances
• Ability to cope
“My husband’s mum is elderly & house bound. We didn’t tell her too much. We didn’t want to worry her.” Mother
of twins
Results: Factors that affected information dissemination.
“She’s his godmother, but she wasn’t interested.” Mother of boy born with operative condition
• Demand for news
• Time
Results: Effect of gender
Small study More work needed Men’s maps simpler –
consistent with psychology findings
“You know how it is with men… they don’t tell you anything.”Mother of baby born
at 36 weeks gestation
Results: ‘Information brokers’
“…she phoned everyone for us.” Father of baby born at 36 weeks
Nominated by the mother. Responsible for communication. Know her social network intimately. Make judgements on what information to give.
Discussion of Research
Uniqueness of individual social network
Autonomy for parents
Encourage contact - Information push & pull
Relevance
Consistency of findings • Education, employment, etc
• People set boundaries according to closeness
• Correlation between social closeness and information given
BT-Clan will use this model in further research – focus groups and diaries.
BabyTalk-Clan
IntroductionSummary of Research AreasCurrent ResearchResearch Direction
• Mobilising emotional support through affective NLG.
• Generalisability of approach
Research Direction
Data gathering - Focus groups,
Diaries
NLG Prototyping
Evaluate accuracy with clinicians
User testing – parents, clan.
Sample data about babies
Generalisation We all edit information according to the recipient.
Possible applications:
• Health: an adult Intensive care patient struggles to communicate for themself.
• Industry: Dissemination of corporate information to management hierarchy & staff
• Social: University students updating family and friends on personal news.
Thank-you