Now far ahead the Roadmap has gone,

unknownlippsΤεχνίτη Νοημοσύνη και Ρομποτική

16 Οκτ 2013 (πριν από 4 χρόνια και 22 μέρες)

87 εμφανίσεις


The Roadmap goes ever on and on

Down from the institute where it began.

Now far ahead the Roadmap has gone,

And we must follow, if we can,

Pursuing it with eager feet,

Until it joins some larger way

Where many paths and errands meet.

And whither then? I cannot say.


Frodo’s song, free after J.R.R. Tolkien


Language technology


Machine Learning


Tilburg University



Speech technology


Twente

University


Telecats


HLT roadmaps in the Low Countries & EU


Then, now, plans for the future


HLT: broadly or narrowly speaking?


Overlap with HCI, AI


Application areas: Health, security, gaming


Software engineering


Whither fundamental research?


New roadmaps


In the African context

Fundamental
HLT

HLT

Pilots

HLT

Applications

Machine
Learning

Emotion
detection

Image
processing

Database
Technology

Information
Retrieval

Workflow
systems

PABX

Education
control

Serious

gaming

Fundamental
HLT

HLT

Pilots

HLT

Applications

Users


Strong focus on academic research.


Collaboration of academia with the research
departments of some (big) industries

Resources

Celex

(1986


2001)

Batavo

(2000)

Corpus Spoken Dutch (1998
-
2004)

Twente News Corpus (1998
-


)

Infrastructure & Networking

Elsnet

(1991


2004)

NOTaS (2000
-


)



Fundamental LST
-
research

SPIN
(NL 1988
-
1994)

TST
(NL 1995
-
2000)

Korte
Termijn

Programma’s

(FL)


Signal processing

Speech processing

Phonetic research

G2P

Parsing

Semantic web


Technology becomes mature and useful


Applications move from the academic labs to
industrial labs, (commercial) pilots and real world
applications.


Because basic HLT starts to work, “adjacent”
technologies become interested too!


Scientific programs reflect these developments:
industrial partners are moving in

Programme

Description

Sponsors

IOP
-
MMI

(1999
-
2009)

Man
-
Machine Interaction

IMIX

(2002
-
2008)

Interactive Multimodal Information
eXtraction

CATCH

(2004
-
2008)


Continuous Access to Cultural Heritage

CATCH

Plus

(2008
-
2010)

Transfer from CATCH
-
research



䍁T䍈
-
慰plica瑩潮s

䍁T䍈C䥉I

⠲〰0
-
㈰ㄲ2

䵯癩湧 晲潭⁃ l瑵t慬a䡥物H慧攠



䡵H慮i瑩es

䵵M瑩浥mi慎

⠲〰0
-
㈰〸2

卯l癩湧 晵nd慭ent慬ⁱaes瑩潮s 潦⁍ l瑩
-
浥di慎

S呅噉T


⠲〰0
-
㈰ㄱ2

䑵D捨
-
䙬敭ish Rese慲捨c偲潧r慭浥 f潲o
䑵D捨c䱡ngu慧攠慮d⁓ ee捨cTe捨c潬潧o


Because HLT is “working” (although not perfect),
less money is available for fundamental HLT
-
research


Tendency to stress what already works is
strengthened by valorisation demands


HLT is used as an
enabling technology


In various application areas (education,
cure&care
,
humanities, governmental structures, safety, etc)


In most new (National AND International)
programmes, HLT is part of the research: not the
main part of it!


Again, tendency to stress what already works, to be a
useful partner



As an HLT community, be prepared to step into
other large initiatives


In other areas, such as


Occasional national funding possibilities


Multidisciplinary opportunities


With social sciences, life sciences, exact sciences, humanities


E.g. NWO: three areas, large
-
scale


Resource / infrastructure funds


NWO Groot /
Middelgroot


Hercules


ESFRI

There is a substantial gap
between


Academic hacking
and


Real software development


Software architecture design


Choices of frameworks,
standards, languages


Programming


Unit testing and quality control


Release


Bug reporting


Maintenance


HLT shares fundaments with


AI, HCI, cognitive science


[
applied|theoretical|socio|psycho
] linguistics


We share some deep, fundamental challenges


If we want to continue being innovative, we
NEED
to tie in with


Reasoning, planning, & search


Common sense & world knowledge


Human factors in HCI


Else, we stop being a field of innovation


For South Africa


Learning from past: Low Countries / EU


Some fundamental choices:


Alliances with other fields


Software, resources


Fundamental research goals


Utilizing current initiatives


NHN (
http://www.meraka.org.za/nhn
)


AfLaT (
http://www.aflat.org
)


ALT
-
I (
http://www.alt
-
i.org
)


Bisharat (
http://www.bisharat.net
)


Anloc (
http://www.africanlocalisation.net
)




Language technology researchers for African
languages over the continent face typical, similar
problems:


Complex morphology


Noun classes


Encoding


Tonality


Resource
-
scarceness


But no central point to access information

(e.g. links, forums, publications)


1 central resource point might improve visibility of
African Language Technology



Community
-
driven portal dedicated to
language (and speech) technology for African
languages


Share information (own work, interesting
links to other people’s works)


Backbone = drupal content management


Spoken Command&Control

Dictation in controlled environments

Traintable information

When both hands are needed

Spelling control

Lip reading

Educational software

Automatic summarization

Translation software

Searching in large audio
-
data

Zorg
, care en cure, open democracy, education, heritage,


Bio food,


Road map coming from
stevin

demo


Buchenwald

digitalized

(historic) collections

H.M. Koningin

Wilhelmina

Digital recorded collections

Second Feministic Wave

Memories of Indonesia

Getting your daily newspaper
if it was
not

delivered,

by identifying yourself over
the telephone (Speech
Recog
)

In the evening the
“newspaper” is calling you to
ask if everything is ok.

If not, a human is taking over.

What

is the
reason

you

are of
calling

us
?

Analyzing 40K calls/per day in
call centres,

to tell the management why
people are calling.


GemeenteConnect
: asking questions to your
local authorities


Q
-
A on the web: asking questions in a “natural
way”


HLT for dyslectic children (6


12 year)


Monitoring elderly patients by Q
-
A over the
phone


48 other RW application recently described


The economic potential and societal importance of
language and speech technologies in Services and ICT


In information and communication, people are naturally
drawn to the means that nature gave them to express
messages in natural language, and to understand the
messages of others.


While having embraced the computer and new ICT
technologies, and having acquired the ability to use the
mouse and the desktop metaphor, people still have
complex desires and questions that remain best expressed
in the form of words.


In order to offer their services in a maximally effective way,
it is vital that computers step up from their current “silent”
state into being talking and understanding “intelligent
systems”.


Why collaborate Low Countries and
South Africa?


What’s in for me/you


Roadmap, but to what?


Economic
efficiency, social issues,
educational issues?