The production of a reminiscing conversational agent and the

implementation of an ontology of reminiscence

Collette Curry, James O’Shea, Keeley Crockett and Laura BrownBackground

In 2008, 1.3 million people in the United Kingdom were

aged 85 and over; this number is projected to reach 3.3

million by 2033 [1]. Aging memory impairment problems

will become more acute as the population profile

changes [1]. Improvement of memory impairment

reduces distress and enhances an individual’s wellbeing

and independence [2:3]. Quality of life in old age can

also be improved by increased subjective well-being,

which is concerned with how people experience their

lives, and includes both emotional reactions and

cognitive judgments [4].

Why Reminiscence?

Reminiscence concerns telling stories of the past,

personal histories, individual perceptions of social

worlds inhabited, and events experienced personally or

at a distance [5]. Bender [9] provides a wide range of

twenty possible purposes and benefits that can derive

from reminiscence. Using a three-Cs model, these

include

1) benefits for clients, such as interacting, socializing,

learning and engaging in therapeutic activities;

2) benefits for carers to aid communication and

improve staff skills;

3) benefits for the work context or culture of the unit [9].

Since publication of the Life Review paper by Butler [5]

there has been an exponential growth in literature

concerning reminiscence and life review, making the

importance of reminiscence and life review in the caring

services clear.

Betty speaks to the

user and responds to

enquiries with

seeming intelligence.

Objectives

There are many different algorithms available to produce

conversational agents, using a variety of computer

languages, databases and flat text files. The study

explored these different algorithms and proposed a

method that utilised a database as well as an ontology of

reminiscence to provide faster and more realistic

conversation in the reminiscence domain.

Results

Usability of the system was checked with a small sample of

participants and the system modified accordingly. This

research conducted a comparative usability test to explore

that a CA effectively contributed to reminiscence in terms of

its functionality and interface and did not create more

problems than it solved. This was done by separate use of a

web based questionnaire compared with the same questions

during a conversation with ‘Betty’.

Future evaluation of the CA ‘Betty’ will be by the use of a

general anxiety and depression scale which will test well-

being of the person both before and after application of the

CA. In addition, the use of standard instruments such as the

Everyday Memory Questionnaire (EMQ) [10] both before and

after the application of the CA will inform whether there is a

noticeable difference in cognitive ability after use of the CA.

Standard instruments can be used to screen for cognitive

impairment. They can estimate the severity of cognitive

impairment at a specific time and to follow the course of

cognitive changes in an individual over time, thus making

these instruments an effective way to document an

individual's response to any intervention. The EMQ is used

as a subjective measure of memory failure in everyday life.

This more direct assessment of the errors experienced by

older adults during their daily activities may be more useful

for directing the research into developing an intervention that

will have a practical impact. Well-being can be tested with a

range of mood assessment techniques including self-

reporting measures. These could be collected to show levels

of satisfaction with the system.

Methodology

The CA was given a voice using a Text to Speech

system. The web interface was built in Flash using

Action-script 3. It was implemented using HTML5 and

tested on Linux, Windows and Apple systems.

Contact:

Collette Curry collette.curry@mmu.ac.uk

Manchester Metropolitan University, Room E113, John Dalton Building, Chester Street. Manchester M1 5GD

Adaptive narrative

The ability to generate narrative is of importance to computer

systems that wish to use reminiscence effectively for a wide

range of contexts ranging from entertainment to training and

education. The typical approach for incorporating narrative

into a computer system is for system builders to script the

narrative features at design time. This CA uses an ontology

to propagate the content and learns from the conversation

logs using keywords.

An ontology of

reminiscence was drawn

up and used with Betty.

Ontology of ReminiscenceThe CA system consists of data imported from WordNet with a

reminiscence ontology. The definition of meaning in WordNet is words

that are synonyms in some particular context. Such a collection in

WordNet is called a synset. Since words can have multiple meanings

(and be multiple parts of speech), the flags of a word are a summary of

all of the properties it might have and it has a list of entries called

"meanings". Each entry is a meaning and points to the circular list, one of

which marks the word you land at as the synset head. This is referred to

as the "master" meaning and has the gloss (definition) of the meaning.

The meaning list of a master node points back to all the real words which

comprise it.

Since WordNet has an ontology, its synsets are hooked to other synsets

in various relations, particular that of parent and child. The CA

represents these as facts. The hierarchical relationship uses the verb "is"

and has the child as subject and the parent as object.

Conclusions

Using the ‘Betty’ package, older adults took part in

reminiscence themed conversations. Conversations were

logged and used to create personal dictionaries and themes

for further future conversation. The CA was able to acquire

new knowledge in this way. The ontology grew as the

participant related more information during the conversation

Aims

To produce an ontology of reminiscence that can be

used to inform the knowledgebase of a conversational

agent (CA) called ‘Betty’. This will then be used as a

reminiscence aid for people with aging memory loss as

part of normal aging.

The ontology is

linked to WordNet.

Ontology

hierarchy References1 Stockport Metropolitan Borough Council (2010), Dementia: Strategy Document.

[Online] [Accessed on 12th January 2012]

2 Dorin, M. (2007) ‘Online education of older adults and its relation to life

satisfaction’. Educational Gerontology, 33(2), 127-143.

3 Wagner, N., Hassanein, K. and Head, M. (2010) ‘Computer use by older adults. A

multi-disciplinary review’. Computers in Human Behaviour, 26, 870-882

4 George, L. K (2010) Still happy after all these years: Research frontiers on

subjective well-being in later life. The Journal of Gerontology Series B,

Psychological Sciences and Social Sciences. 65B(3):331-9. doi:

10.1093/geronb/gbq006.

5 Butler, R.N.(1963) The Life Review: An interpretation of reminiscence in the aged.

Psychiatry, 26: 65-76.

6 Trueman, I. and Parker, J. (2004) Life review in palliative care. European journal of

palliative care, 11(6): 249-53.

7 Parker, J. (2003) Positive communication with people who have dementia. In:

Adams, T. and Manthorpe, J. (eds.). Dementia Care. London: Arnold, pp.148-63.

8 WordNet: An electronic lexical database available from Princeton University

[online] http://wordnet.princeton.edu/wordnet/ [Accessed 20th December 2012]

9 Bender, M., Bauckham, P. & Norris, A., 1999. The therapeutic purposes of

reminiscence. London: Sage.

10 Sunderland, A., Harris, J.E., & Baddeley, A, (1983) The Everyday Memory

Questionnaire

The ontology began as a list of facets and developed through several iterations. In the example it can be seen that

classes have properties and relationships.

The CA has a standard response mechanism to deal with off-topic

user utterances. This is known as the ELIZA layer.

The CA began life as a text block which displayed the CA

response as well as the user’s input. The CA remembered

past visits and conversations

Future work

The CA ‘Betty’ could be contained in a multi-activity environment.

Games and other challenges could be provided as well as email and

personal photo album access.

Improvement of mood can result from speaking with the CA

NLDB-2013

Salford University

Media City

The CA evolved into a speaking avatar, displaying the user utterance and CA response as text on the screen. This helped

with reinforcement of the conversation.