Laurent, M., Bretier, P. and Manquillet, C. (2010).
Ad-hoc evaluations along the lifecycle of industrial spoken dialogue systems: heading to harmonisation?. In LREC 2010. Malta
Ad hoc evaluation of SDS along project lifecylcle in Orange
1. Ad Hoc evaluations along the lifecycle of industrial spoken
dialogues systems: heading to harmonisation?
Marianne Laurent, Philippe Bretier, Carole Manquillet
pla Har Hard /
Interactive systems
implying a human-
amchine dialogue, A technology based
e
om
atin
aut man
g
What is a
rvic e hu rs low
Spoken
a QoS & an service tailored to
tfo d/s software individualised Se th to lue
a srm doftw Automate routing platform access the various business a
per ed-va
the er eli are service satisfying to information customers' needs o d
cu vice veri ad tasks
Dialogue
Ha
sto to ng
me
the customer soft rd /
r
relationship plat ware
management form
System?
Ergonomics Hosting
View
Contracting Technical Experts Business Providers
Métier Developers Marketing Managers
Owners Experts
points
Goals
Enhance the Deliver a service to the
customer with respect to Tailor the development Enhance the Responsible of the
customer service with Develop the service Answer the users'
the customer with a user-centric expectations. organisation service hosting and
automated vocal development approach
solutions relationship policy maintenance.
Evaluations
Ensure the respect of Evaluate the service Compare the adequation Compare the adequation Monitor the good
Control the service Monitor the service of the service
the predefined QoS quality and its adequacy performance, monitor
of the service functionning of the live
commitments and the perceived quality performance with the performance with the
to the organisation and correct anomalies. services they host
end-user satisfaction. users needs organisation needs
Indicators
ROI (project financing Measure of the service
Track the end-user Indicators linked to the Expert evaluation Measure of the service Measure of the service performance, monitor
decisions), QoS, campaign (field tests performance and the performance and the
dialogue performance, experience service design live services and
and real users) users' perception. users' perception technical support
user satisfaction
Ho
pro sti
vid ng
ers
Difficult
communication
be tween actors eed : ion
N at
evalu ator
One valu
s
tandard for
rts ic
No s
pe om
tion sharing, er e tuation
p i
ex n
informa
go
per s Local Contraints:
Er
cooperation Budget,
Ma
ex rket
and collaboration ! data collection,
pe ing
rts
etc.
Nomadism
of ev aluation
m ethodol ogies !! al ts
oc er
ctions
formance
ymynonymy
V p
ex
mon Sy
ctions
Ho and
a
orms
nter
int e
nt n the
nc
era
fere ng on
l per
Dif di
ers i
rie
Service
pe
epen munities
Technica
Platform
d
SDS-Us
U s er e x
Com ractice
of P
C , low: , exit
all-f path %
12
ntry
58
e 36
46
3
,2 14
Need to refine
indicators Traceabilit
+ collected data y
Caller Reference
history Call values
motive
Exploitation
Generalisation
Evaluation
The service is maintained
by both the hosting
Pilot providers and the initial
Contexts
Access is granted to the developers (for
overall flow of traffic. corrections).
Experiments .
Collect the users experience feedback in live
Service tested under real conditions.
conditions, with a The service keeps constantky monitored.
Development fragment of the real flow
of traffic.
Dashboards containing high-level indicators are
Test the service so as to sent daily to the various interested teams (number
identify design issues
The evaluation constitute
Framing Development of a
and refine the
development to the users a comprehensive
prototype integrating the practices. Alignment of validation of the service
initial design, linguistic, observation with a priori before its access is
Analysis Specification of project
usability and technical
considerations.
projections. extended to all users. It
includes the relevance of
scope, specifications and Tests organised on a the technical
project plan. Test within the panel of recruited users. architecture.so as to
development team to detect and fix the last
What is done? Analysis of end-users' beta-solution issues.
needs and practices. Feasibility evaluation. detect major bugs and
Evaluation context? Solutions benchmark. enrich the vocal
recongition model,
Marianne.laurent@orange-ftgroup.com, Philippe.Bretier@orange-ftgroup.com, Carole.Manquillet@orange-ftgroup.com
Orange Labs, 2 avenue Pierre Marzin, 22307 Lannion, France