HUMAN LANGUAGE TECHNOLOGIES FOR E-GOV
Mário Rodrigues
1
, Gonçalo Paiva Dias
2
ESTGA
1,2
/ IEETA
1
/ GOVCOPP
2
, University of Aveiro, Aveiro, Portugal
António Teixeira
DETI / IEETA, University of Aveiro, Aveiro, Portugal
Keywords:
e-gov, Life-event, Human language technologies, Natural language interface, Information extraction.
Abstract:
Effective provision of government services implies that, besides being provided online, services become avail-
able through other channels, are organized according to citizen’s expectations, are accessible to everyone,
anytime and anywhere, and include information from unstructured sources. It is also essential to provide
the tools that allow citizens to correctly identify the services they need. In this paper we will discuss how
it is possible to improve e-gov service delivery by using human language technologies. We argue that these
technologies can contribute to: deliver services in more inclusive manners; provide human centered and mul-
tilingual service and support; and include non-structured information scattered across different sources.
1 INTRODUCTION
eGov “refers to the use by government agencies of in-
formation technologies (such as the Internet and mo-
bile computing) that have the ability to transform re-
lations with citizens, businesses, and other arms of
government. These technologies can serve a variety
of different ends: better delivery of government ser-
vices to citizens, improved interactions with business
and industry, citizen empowerment through access to
information, or more efficient government manage-
ment. The resulting benefits can be less corruption,
increased transparency, greater convenience, revenue
growth, and/or cost reductions” (World Bank, 2009).
Human language technologies (HLT) refer to
speech and natural language processing (NLP). Re-
search and development in speech include automatic
speech recognition (ASR) to get a textual representa-
tion of a speech sound wave, and text to speech (TTS)
to generate a speech sound wave representing a given
text. NLP refers to computer systems that analyze,
attempt to understand, or produce one or more hu-
man languages, such as English, Japanese, Italian, or
Russian. The input might be text, spoken language,
or keyboard input. The task might be to translate to
another language, to comprehend and represent the
content of text...” (Allen, 2003).
Paper structure: in this paper we will discuss how it
is possible to improve e-gov services by using HLT.
After a brief overview of needs regarding better ser-
vice deliveryin e-gov (section 2), followsan overview
of HLT (section 3) and the potentials of its application
to e-gov (section 4). The last sections of the paper ad-
dress some recent related work, our ongoing work and
the relevant conclusions.
2 CITIZEN NEEDS IN E-GOV
eGov is a dynamic concept of varying meaning and
significance” (Relyea, 2002). It is commonly used
to refer to several alternative or complementary con-
cepts, including the use of the Internet in the interac-
tion between governmentand citizens or businesses, a
reengineering of government processes catalyzed by
Information and communication technologies (ICT),
and a symbol of ICT usage in increasing the efficiency
and effectiveness of government.
Electronic service provision is probably the most
common approach to e-gov. In effect, ICT allows tra-
ditional services to become online, with clear bene-
fits for clients (citizens and businesses) and govern-
ment, avoiding travel, speeding up processes and di-
minishing costs. However, besides being delivered
online, government services should also be integrated
in such a way that they fit the citizens and business
400
Rodrigues M., Paiva Dias G. and Teixeira A.
HUMAN LANGUAGE TECHNOLOGIES FOR E-GOV.
DOI: 10.5220/0002857804000403
In Proceedings of the 6th International Conference on Web Information Systems and Technology (WEBIST 2010), page
ISBN: 978-989-674-025-2
Copyright
c
2010 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
concrete needs (Dias and Rafael, 2007). One-stop e-
gov corresponds to this perspective: services should
be provided through a single entry point and be in-
tegrated across agencies from the client’s point of
view (Wimmer et al., 2001). They can be organized
into life-events (as birth of a child, marriage, etc) tar-
geted at specific costumers at particular times in their
life (Oteniya et al., 2006).
Digital divide refers to the gap between people
with effective access and capacity to use ICT and
those with very limited or no access or capacity at
all. If not properly addressed, online service provi-
sion can contribute to increase this problem. Before
using an e-gov service users sometimes cannot iden-
tify which public administration institutions provide
the services they need and what inputs are required
to execute the service (Sroga, 2008). It is necessary
to have tools to help every client to find and use the
appropriate service regardless of its level of expertise
in government services or ICT, because government
must serve 100% of its clients (Trochidis et al., 2008).
Finally, the development of a customer oriented
approach implies that government agencies learn how
to communicate with each other, interoperation be-
coming a crucial issue. To this respect, it is important
to note that a very relevant set of government infor-
mation is not stored in searchable databases. Many
documents although digitally stored, remain in their
original format. eGov services should be able to in-
clude and combine these sources of information.
3 HUMAN LANGUAGE
TECHNOLOGIES
3.1 Interface Technologies
A natural language interface allows people to inter-
act using a human language, such as English, as op-
posed to a computer language, command line inter-
face, or graphical user interface. Spoken language in-
terfaces are capable of handling spoken human lan-
guage. Advances in ASR, language understanding,
language generation, and speech synthesis (see Juraf-
sky and Martin, (2000) for technical details) enabled
the emergence of complex conversational spoken lan-
guage interfaces (Bohus and Rudnicky, 2009).
These systems are typically connected in a
pipeline architecture (Bohus and Rudnicky, 2009).
The audio signal from the user is captured and passed
through a ASR module that produces a sequence of
words. This recognition hypothesis is forwarded to
a language understanding component to create a cor-
responding semantic representation which is passed
to the dialog manager that, using also the discourse
context, produces the next system action, usually in
the form of a semantic output. A language genera-
tion module produces the corresponding textual form,
subsequently passed to a TTS module to produce syn-
thetic speech. It is possible to create an automated
chat by adding text input alongside with the ASR
module and a text renderer alongside TTS or to con-
nect a software agent (e.g. a talking head). Exam-
ples of conversational spoken language interfaces in-
clude (Bohus and Rudnicky, 2009): Jupiter, AdApt,
and TRIPS.
3.2 Natural Language Queries
A natural language interface requires algorithms to
query information using natural language. Also, in
dealing with large amounts of information as in e-
gov, a central problem is the formulation of queries
that are communicable to the system. Database query
languages can be intimidating to the non-expert, lead-
ing to the popularity for keyword based search in spite
of its significant limitations (Hendrix et al., 1978).
Two major obstacles lie in the way of support-
ing arbitrary natural language queries: automatically
understanding natural language is itself still an open
research problem; to translate the understood natural
language query into a formal query requires mapping
the understanding of intent into a specific database
schema. Regarding the first problem, state-of-the-art
techniques can reach acceptable results but it is still
far from being totally resolved. The second prob-
lem has been tackled with good results by works like
NaLIX (Li et al., 2005), Panto (Wang et al., 2007),
and ESTER (Bast et al., 2007).
3.3 Unstructured Information
As described above, HLT allows users to perform
queries in an intuitive way and also allows to feed the
system knowledge base with information from more
sources. The knowledge base is accessed by the dia-
log manager or the query system.
Information extraction (IE) takes texts as input
and produces fixed-format, unambiguous data, as out-
put. It involves processing text to identify relevant in-
formation, such as named entities (NE) or relations
between them (Appelt, 1999). NE include people,
organizations, locations and so on, while relations
include physical relations (located, near, partwhole,
etc.), personal or social relations (business, fam-
ily, etc.), and membership (employ-staff, member-of-
group, etc.) (Bontcheva et al., 2008). State-of-the-art
HUMAN LANGUAGE TECHNOLOGIES FOR E-GOV
401
systems perform NE recognition over the web con-
tents with good results (Whitelaw et al., 2008).
To find relationships between named entities is a
task related to syntactic parsing, which is the pro-
cess of analyzing a text to determine its grammatical
structure with respect to a given formal grammar. Re-
cent works achievedgood results in any context by in-
cluding machine learning algorithms to find relation-
ships (Suchanek et al., 2007).
4 HLT POTENTIAL IN E-GOV
eGov services should be provided in a way that meets
customers natural communication paradigm. There-
fore, whenever possible, e-gov services should use
natural language interfaces: accepting written and
speech inputs (over the web and telephone) and allow-
ing people to experience a level of interaction compa-
rable to traditional, face-to-face services.
In brief, the great potential of speech and natural
language technologies for e-gov can be supported by
the following advantages:
More intuitive user interface - profiting from the re-
cent advances of natural language interfaces in other
areas, we argue that it is possible to develop e-gov
support systems that are able to solve user queries for-
mulated in natural language, instead of forcing users
to look for instructions or to find information in the
set of rules that regulate the service. The linguistic
coverage of natural language interfaces is not always
obvious but some strategies allow dialog systems to
intuitively drive the user to use the right vocabulary
set (Gorin et al., 1997).
Dialog becomes an option - dialog processing is, by
its very nature, incremental. An incremental system
can work with units smaller than utterances, allow-
ing the creation of a more reactive system capable of
taking the initiative in the conversation to clarify, ask
for missing information, or suggest. By having a con-
versation, individuals would be able to explain their
problem to discover a solution, instead of having to
make a thorough search to find how to proceed.
Take advantage of voice channels - speech is the
most natural and easy existing interface, not only for
people with special needs, but for people in general,
as Nass & Brave (Nass and Brave, 2005) state: Ubiq-
uitous computing - access to all information for any-
one, anywhere, at any time - relies on speech for those
whose eyes or hands are directed to other tasks or for
those who cannot read or type (such as children, the
blind, or the disabled)”. Another advantage is that
telephonesare more popular than computers and more
people feel comfortable with them. Services provided
by voice are already available to the general public
(e.g. Google Voice).
Multilingual - machine translation is not adequate for
online documents because its performance is not per-
fect. The same is not true when using natural lan-
guage interfaces with dialog support because the use
of clarification can resolve translation errors and the
communication can be effective and more efficient
than having the user speaking a foreign language.
Access to unstructured information - important
sources of information relevant for e-gov are origi-
nally created in natural language documents, such as
forms, laws, regulations, etc. Those documents are
usually available to the public in pdf and html for-
mats. A computer system that does not feature NLP
capacities cannot derive meaningful information from
it, being limited to store and display. Systems able to
understand the information of this type of documents
could assist users more efficiently by selecting the in-
formation that is relevant in the client context.
Access for all - the above advantages contribute to
reduce the digital divide. Written and spoken natural
language interfaces facilitate the access of minorities
as the visual impaired, people with speech disabilities,
and people not familiar with ICT. The usage of ubiq-
uitous telephone increases the potential to reach more
users, including those who live in remote areas, are
homebound, etc. Finally, multilingual support and the
ability to search and contextualize unstructured docu-
ments can serve other minorities, such as immigrants,
tourists and people with low literacy levels in gen-
eral. Written interfaces are robust and speech tech-
nology is becoming mature as demonstrated by sev-
eral commercial products in different languages (TTS
systems: Microsoft Reader, Nuance RealSpeak, Lo-
quendoTTS; ASR systems: Nuance Dragon Naturally
Speaking, Microsoft SpeechFX).
5 RELATED WORK
Although a lot of research activity has been done
in e-gov, so far, to our knowledge, just one project
was dedicated to the study of using HLT in e-gov:
HOPS, funded by the European Commission under
Framework-Programme 6 (FP6) (Gatius et al., 2006).
Other e-gov projects use semantic technologies to
specify, develop, deploy services, and are rather cen-
tered in solving problems as interoperability and ser-
vice integration, which are very important problems
and still need to be further addressed. See for example
WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies
402
OneStopGov (Chatzidimitriou and Koumpis, 2008),
also FP6 funded, and Access-eGov (Sroga, 2008).
The authors are particularly interested in applying
the expressed position in the scenario of municipal
service delivery because it is often the closest point of
service for citizens and enterprises. The implementa-
tion will integrate a web and speech interfaces, a ques-
tion and answering (Q&A) module, a dialog manager,
a knowledge manager, and an IE module. For TTS
and ASR off-the-shelf software will be used.
The research has started by addressing the IE and
the knowledge management problems by extending
a semantic knowledge base ideas and public code
(YAGO) to cope with Portuguese texts (Suchanek
et al., 2007). Preliminary tests with a first prototype
for the Q&A module, working with the collected data,
are scheduled for near future. The first version of
the dialog system will be developed using the CMU
Olympus open-source framework with the Ravenclaw
dialog manager (Bohus and Rudnicky, 2009).
6 CONCLUSIONS
The authors strongly believein the potential of HLT to
improve e-gov services and are working on the appli-
cation of these ideas/technologies to new e-gov ser-
vices for Portuguese municipalities. Other projects,
such as HOPS, show that this is a challenging prob-
lem, which has been seldom addressed and requires
more research. Although good results have been
achieved in some HLT tasks, the integration of these
technologies in complex systems as e-gov services
needs to be further studied in order to have systems
ready to serve any client.
Using HLT in e-gov is still a challenge because it
requires research and adaptation to the specific area.
The benefits are clear: reduction of the digital di-
vide; more intuitive human interfaces; communica-
tion comparable to traditional face-to-face dialogs;
more channels available; and more knowledgeable
systems. These benefits are particularly relevant in
e-gov because government must serve 100% of the
population and because e-gov success also depends
on how easy it is to use its services.
There is still a lot of work to be done to have liable
applications in Portuguese language - our goal. We
believe that our research can be useful to other lan-
guages in the same way we found fruitful to study, un-
derstand, and adapt to Portuguese the developments
of HLT for other languages, such as English.
REFERENCES
Allen, J. F. (2003). Natural Language Processing. In Ency-
clopedia of Computer Science, pages 1218–1222.
Appelt, D. E. (1999). Introduction to Information Extrac-
tion. AI Communications, 12(3):161–172.
Bast, H., Chitea, A., Suchanek, F., and Weber, I. (2007).
ESTER: Efficient Search on Text, Entities, and Rela-
tions. In Proc. 30th ACM SIGIR, pages 679–686.
Bohus, D. and Rudnicky, A. I. (2009). The ravenclaw di-
alog management framework: Architecture and sys-
tems. Computer Speech and Language, 23.
Bontcheva, K., Davis, B., Funk, A., Li, Y., and Wang, T.
(2008). Human Language Technologies. Semantic
Knowledge Management.
Chatzidimitriou, M. and Koumpis, A. (2008). Marketing
One-stop e-Government Solutions: the European On-
eStopGov Project. IAENG.
Dias, G. P. and Rafael, J. A. (2007). A simple model
and a distributed architecture for realizing one-stop e-
government. ECRA, 6(1):81 – 90.
Gatius, M., Mangham, A., and Lesmo, L. (2006). HOPS:
Developing Transformational Government Services.
Gorin, A. L., Riccardi, G., and Wright, J. H. (1997). How
May I Help You? Speech Communication.
Hendrix, G., Sacerdoti, E., Sagalowicz, D., and Slocum,
J. (1978). Developing a natural language interface to
complex data. ACM TODS, 3(2):105–147.
Li, Y., Yang, H., and Jagadish, H. V. (2005). NaLIX:
an interactive natural language interface for querying
XML. In Proc. ACM SIGMOD, page 902.
Nass, C. and Brave, S. (2005). Wired for Speech: How
Voice Activates and Advances the Human-Computer
Relationship. MIT Press.
Oteniya, O., Janowski, T., and Ojo, A. (2006). Government-
Wide Workflow Infrastructure-Enabling Virtual Gov-
ernment Organizations. Springer.
Relyea, H. (2002). E-gov: Introduction and overview. Gov-
ernment Information Quarterly, 19(1):9–36.
Sroga, M. (2008). Access-eGov-Personal Assistant of Pub-
lic Services. In IMCSIT, pages 421–427.
Suchanek, F. M., Kasneci, G., and Weikum, G. (2007).
Yago: a core of semantic knowledge. In WWW ’07.
Trochidis, I., Tambouris, E., and Tarabanis, K. (2008). One-
Stop Government: Literature Review. Prague. 6th
Eastern European eGov Days April 2008.
Wang, C., Xiong, M., Zhou, Q., and Yu, Y. (2007). Panto:
A portable natural language interface to ontologies.
LNCS, 4519:473.
Whitelaw, C., Kehlenbeck, A., Petrovic, N., and Ungar, L.
(2008). Web-scale named entity recognition. In CIKM
’08, pages 123–132, New York, NY, USA. ACM.
Wimmer, M., Traunmüller, R., and Lenk, K. (2001). Elec-
tronic Business Invading the Public Sector: Consider-
ations on Change and Design. HICSS, 5:5006.
World Bank (2009). http://www.worldbank.org/egov.
HUMAN LANGUAGE TECHNOLOGIES FOR E-GOV
403