WHAT HAPPENS IF WE SWITCH THE
DEFAULT LANGUAGE OF A WEBSITE?
Te Taka Keegan, Sally Jo Cunningham
Computer Science Department, University of Waikato,Hamilton, New Zealand
Keywords: User Interfaces, Usability, Digital Libraries, Multi-lingual web sites
Abstract: In this paper we investigate the effect of the default interface language setting on a bilingual website. Log
file analysis is undertaken to determine usage patterns of the Niupepa digital library (a collection of historic
Māori language newspapers) when the default interface language is switched between Māori and English in
alternate weeks. Activity is grouped into active user sessions, which are further analysed to determine
methods of access and searching patterns. The results clearly show that changing the default language of a
website will affect the ways in which users access information.
1 INTRODUCTION
The bilingual website that we chose to investigate is
called the Niupepa Collection. It is being served by
the Greenstone software (Witten and Bainbridge
2002) of the New Zealand Digital Library (NZDL)
at: www.nzdl.org/niupepa. Niupepa is a collection of
historic Māori newspapers published between 1842
and 1933. It is a large source of historic texts, almost
18,000 newspapers pages. The newspapers are
available in a full text
format and in two facsimile forms: a low resolution
image that downloads quickly for previewing, and a
high resolution image that takes longer to download
but is readable on screen (Apperley et al 2002).
70% of the documents are written in Māori, 27%
are written bilingually in both Māori and English,
and 3% written in English only. The collection is a
rich source of Māori language texts in an
environment where there is a dearth of Māori
language resources. The default language of the
collection is normally set to Māori.
Figure 1: Home Page of the Niupepa Website
263
Taka Keegan T. and Jo Cunningham S. (2005).
WHAT HAPPENS IF WE SWITCH THE DEFAULT LANGUAGE OF A WEBSITE?.
In Proceedings of the First International Conference on Web Information Systems and Technologies, pages 263-269
DOI: 10.5220/0001231602630269
Copyright
c
 SciTePress

Approximately 14% of New Zealand’s total
resident population are Māori and about 1 in 4 of
these are able to converse in Māori (Te Puni Kōkiri
2003).
When we consider that 65% of Māori have
never accessed the Internet, the potential users of a
Māori language interface is perhaps 1-2% (40,000-
80,000) of New Zealand’s population. In contrast,
potential English-speaking New Zealand users of the
website are approximately 51% (2,040,000) (Te
Puni Kōkiri 2001). By setting the default language
of the Niupepa Collection to Māori we are clearly
going against the preferences of the majority of
potential users. Research undertaken by Jones (Jones
et al 2000) suggests that users of a digital library
system rarely amend the default settings for options
with regard to query types and result displays.
Through the use of log file analysis this research
seeks to find answers to what differences will occur
in user behaviour if the default language of an
interface is alternated between Māori and English.
2 LIMITATIONS OF LOG FILE
ANALYSIS
As a method of gathering data on user actions log
file analysis has its shortcomings– the primary one
being the effect of web caches. Web caches sit
between web servers and clients and will serve
repeated requests to the client without having to
bother the original server. This saves time and
reduces network traffic. However the issue is that
the original server does not receive the request and
consequently it is not be recorded in its web log file.
There are two main types of web caches: a
browser cache which is handled by a user’s browser
software and a proxy cache which is configured
within a network. Both types of caches will prevent
repeated requests from a single user appearing in
web log files. However a network cache will also
mask repeated requests from different users within
its network.
There are other limitations of log file analysis
including false hits due to web robot activity, false
hits due to server upgrades and maintenance; and the
inability to accurately delimit user sessions.
3 GATHERING THE DATA
The NZDL website (www.nzdl.org) makes available
over 40 different collections in various formats. All
user activity is logged. Every request or ‘hit’ is
recorded along with information such as the page
requested, the language used in the interface, the
time of the request, the type of request, the previous
action, the IP address of the requestor and the
various preferences that are set.
The NZDL site is mirrored with a site located at
the University of Lethbridge in Alberta, Canada. The
New Zealand site, located at the University of
Waikato, serves the collection to Web requests from
within New Zealand. The Lethbridge mirror site is
responsible for serving the collection to Web
requests from outside of New Zealand. The data
collected in this analysis is from the University of
Waikato site only, and thus only reflects usage
within New Zealand.
We chose to analyse a four week period running
from 8.40am Monday 5 July 2004 to 8:40am
Monday 2 August 2004. In the first week we
changed the Niupepa default language setting to
English. The second week we changed it back to
Māori. The third week it was in English and the
fourth week we changed it back to Māori again. The
raw NZDL log file was collected for this time period
and the hits relating to the Niupepa collection were
extracted.
The raw Niupepa data was then further filtered to
remove hits of unwanted origin. These included
incorrect language argument (27), undefined IP
address of requestor (213), web crawler and web
robot hits (33) and hits from the local research team
(18). This left a total of 14,416 hits, of which 7724
were in the weeks that the default language was set
to English and 6692 were in the weeks when the
default language was set to Māori (Graph 1).
Clearly there are significantly more hits (15.4%
more) when the default language of the website is
set to the more commonly spoken English language.
It is also clear to see, and perhaps unsurprising to
note, that the number of hits in English increases
when the default language is set to English and the
number of hits in Māori increases when the default
language is set to Māori.
4 DEFINING THE SESSIONS
To further analyse what the users of the web site
were doing, the hits recorded in the log files were
grouped into sessions. Cookies were used to define
these sessions. When users connect to the website a
cookie is created on their machines which holds
information that includes the IP address of the
machine connecting and the time that the cookie was
created. This information is recorded with each hit
as the z argument of a hit and so a sessions is simply
a group of hits with the same z argument within a
given time frame. Web analysis software usually
defines a session as a series of hits such that the time
WEBIST 2005 - WEB INTERFACES AND APPLICATIONS
264

gap between any two successive hits is less than 30
minutes. However because Niupepa users may spend
long periods reading single newspaper pages is was
more appropriate to extend the minimum time length
between hits to 60 minutes.
Once the log file of hits was grouped into
sessions, the types of sessions were then arranged
based on the length of the session and what the user
accessed. This gave three types of sessions:
o Single hit: these could be a request where the
user has clicked on the site and decided not
to look any further, or who could be users
that have not enabled cookies.
o Exploratory only sessions: multi-hit sessions
where the user only accessed the home page,
the help page, and/or the preferences page.
No documents in the collection were
accessed and no searches were undertaken.
o Extended sessions: multi-hit sessions where
documents of the Niupepa collection where
accessed and/or searches were undertaken.
From Graph 2 we can see there is an increase
(23.4%) from the number of single hits in the EN
weeks (367) to the number of hits in the MI weeks
(453). This can be expected because the majority of
users logging onto the website are not Māori literate,
so when they first log onto the site and see that it is
in Māori they are perhaps not prepared to continue.
By contrast the number of extended sessions and
exploratory sessions were quite similar for both
English and Māori default language settings of the
website. This suggests that active users of the
website were intent on using the website no matter
which default language the website is set to.
Graph 1 showed that the total number of hits was
lower in the Māori weeks than in the English weeks.
However Graph 2 showed that the number of
sessions in the Māori weeks was higher than in the
English weeks. This implies that the difference rests
in the activity that the users are undertaking when
they are accessing the web site. The single hit and
exploratory sessions are not as important as they do
Session Types
367
24
273
30
277
453
0
100
200
300
400
500
single hit exploratory session extended session
EN weeks MI weeks
Graph 2: Sessions Types for the weeks under analysis
Graph 1: Total fi
ltered hits for the weeks under
analysis
Filtered Niupepa Hits
6795
4975
929
1717
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
EN weeks MI weeks
English Hits Māori Hits
WHAT HAPPENS IF WE SWITCH THE DEFAULT LANGUAGE OF A WEBSITE?
265

not represent activity from users who are actively
using the website. Although steps have been taken to
remove hits from known web crawlers and known
web robots we still cannot be sure that hits generated
by non human activity have been completely
removed from the statistics. However by definition
an extended session involves undertaking searches
and/or browsing the collection; consequently we can
be confident that these hits are a result of human
activity. The next stage of the analysis involves
having a closer look at the extended sessions.
5 EXTENDED SESSION ANALYSIS
The extended sessions were further subdivided into
language categories. Initially we decided that there
were just two categories: those sessions that were
conducted in English, and those sessions that were
conducted in Māori. However when we looked at the
data closely we realized that there was a third type
of session: the bilingual session, in which users
conducted their interaction with both English and
Māori interfaces.
A Māori language session was defined as a
session that was conducted in Māori at least 80% of
the time and did not involve more than two user
interface language switches. An English language
session was defined as a session that was conducted
in English at least 80% of the time and did not
involve more than two user interface language
switches. A bilingual language session was a session
that involved three or more user interface language
switches and/or sessions that spent at least 20% of
the activity in both languages.
Table 1: Summary Statistics for English Sessions.
English Extended Sessions
EN wks MI wks
total sessions: 210 172
min hits/session: 2 2
max hits/session: 286 136
mean hits/session: 29.4 25.3
median hits/session: 12.5 16
Std Dev hits: 41.5 26.2
shortest (min): <1 <1
longest (min): 233 191
mean (min): 22.3 22.3
median (min): 7.5 8.5
Std Dev (min): 36.2 31.5
When we consider the English Extended
Sessions and compare the weeks that the default
language is set to English (EN weeks) with the
weeks that the default language is set to Māori (MI
weeks), (Table 1) we can see a number of
differences. There are more sessions, 210 as opposed
to 172, when the default language is set to English.
The mean number of hits in the English default
weeks is larger, the standard deviation is larger, and
as the median is lower in the English default weeks
which indicates large sessions is at the upper
extreme of the data. Overall these statistics suggest
that sessions in English involve more activity over a
similar time period when the default language of the
interface is set to English.
Table 2: Summary Statistics for Māori Sessions
Māori Extended Sessions
EN wks MI wks
total sessions: 48 76
min hits/session: 2 2
max hits/session: 76 132
mean hits/session: 13.5 16.7
median hits/session: 8 11.5
Std Dev hits: 14.9 19.3
shortest (min): <1 <1
longest (min): 103 116
mean (min): 12.4 14.4
median (min): 5.5 5.5
Std Dev (min): 20.0 23.2
A similar pattern occurs for Māori language
sessions when the default language is set to Māori
(Table 2): there are more Māori sessions, sessions
contain more activity (hits), and sessions are
conducted over a longer time period.
Table 3: Summary Statistics for Bilingual Sessions
Bilingual Extended Sessions
EN wks MI wks
total sessions: 15 29
min hits/session: 2 3
max hits/session: 182 62
mean hits/session: 31.7 19.2
median hits/session: 8 8
Std Dev hits: 47.5 16.7
shortest (min): <1 <1
longest (min): 241 117
mean (min): 32.5 14.3
median (min): 8 8
Std Dev (min): 64.0 23.4
WEBIST 2005 - WEB INTERFACES AND APPLICATIONS
266

The statistics for extended bilingual sessions, as
shown in Table 3, also show differences between the
English default weeks and the Māori default weeks.
Twice as many sessions are defined as bilingual in
the Māori default weeks perhaps because a larger
number of users are switching more often when the
default language is set to Māori. Generally, the
extended bilingual session averages in the English
default weeks are similar to the English sessions
averages in the English default weeks and the
bilingual session averages in the Māori default
weeks are similar to the Māori sessions averages in
the Māori weeks. However as the standard
deviations are high and the number of sessions that
we are dealing with is low it would be premature to
deduct any firm conclusions.
6 ACCESSING DOCUMENTS
The NZDL web log records if the pages accessed
were the result of a search, a browse by newspaper
publication (series) or a browse by date. The results
are grouped by language of the extended session. In
Table 4 we can see the document types accessed in
the English Extended Sessions. It is apparent that
switching the default language doesn’t make any
difference in how users access documents in the
English sessions. However Table 5 shows quite a
large difference with Māori Extended Sessions,
indicating a large preference (90.9%) to access
pages by using a search when the default language is
set to English but a much lower preference (69.9%)
when the default language is set to Māori. It appears
that when the default language is set to Māori, the
Māori sessions have a higher possibility of browsing
the documents by series or by date. It should be
noted however that the number of documents viewed
in the Māori sessions is much smaller than the other
two sessions. The bilingual extended session
preference for accessing pages (Table 6) is similar to
the English sessions, with no significant differences
when the default language is switched.
Table 4: Types of Documents Accessed in English
Sessions
English Extended Sessions
EN wks MI wks
total documents viewed: 4265 3070
pages viewed from search: 81.6% 81.9%
pages viewed from series: 12.0% 12.7%
pages viewed from date: 3.5% 3.0%
other: 2.9% 2.4%
Table 5: Types of Documents Accessed in Māori Sessions
Māori Extended Sessions
EN wks MI wks
total documents viewed: 495 901
pages viewed from search: 90.9% 69.9%
pages viewed from series: 7.1% 13.0%
pages viewed from date: 1.6% 14.9%
other: 0.4% 2.2%
Table 6: Types of Documents Accessed in Bilingual
Sessions
Bilingual Extended Sessions
EN wks MI wks
total documents viewed: 4760 3971
pages viewed from search: 82.6% 79.1%
pages viewed from series: 11.5% 12.7%
pages viewed from date: 3.3% 5.7%
other: 2.6% 2.4%
7 SEARCHING
The final analysis undertaken for this paper was an
examination of the searching characteristics with
each of the three types of sessions to see if there
were any differences when the default language was
changed. The searching summary for English
Extended Sessions is shown in Table 7. It can be
seen that more searches are submitted when the
default language is set to English; a consequence of
having a higher number of English extended
sessions in these weeks. The average number of
searches per session is slightly higher (5.0 versus
4.4) in the English default weeks, and there is a
higher percentage of sessions without any searches
(36.2% versus 19.2%) in the English default weeks
as well. The average number of terms submitted per
search is very similar (1.9 versus 2.2) for both
default language settings.
WHAT HAPPENS IF WE SWITCH THE DEFAULT LANGUAGE OF A WEBSITE?
267

Table 7: Searching Summary for English Sessions
English Extended Sessions
EN wks MI wks
number of searches: 1046 749
average per session: 5.0 4.4
maximum searches: 80 43
minimum searches: 0 0
standard deviation: 3.8 2.2
0 searches: 36.2% 19.2%
1 search: 11.9% 14.0%
2-3 searches: 14.8% 27.9%
4+ searches: 37.1% 39.0%
average search terms: 1.9 2.0
Table 8 presents the searching summary for
Māori extended sessions. More searches are
submitted when the default language is set to Māori;
a consequence of having a higher number of Māori
extended sessions in these weeks. The average
number of searches per sessions is slightly higher
(3.1 versus 2.6) in the Māori default weeks, and
there is a slightly higher percentage of sessions
without any searches (28.9% versus 27.1%). Again,
the average number of terms submitted per search is
similar (1.9 – 2.2) for both default language settings.
Table 8: Searching Summary for Māori Sessions
Māori Extended Sessions
EN wks MI wks
number of searches: 126 238
average per session: 2.6 3.1
maximum searches: 19 18
minimum searches: 0 0
standard deviation: 2.0 1.8
0 searches: 27.1% 28.9%
1 search: 25.0% 19.7%
2-3 searches: 29.2% 22.4%
4+ searches: 18.8% 28.9%
average search terms: 1.9 2.0
The searching summary for bilingual extended
sessions is displayed in Table 9. It can be seen that
there are more than twice as many searches
submitted in the Māori default weeks (115) than in
the English default weeks (51). This is surprising as
Table 6 indicates that there are fewer documents
accessed in the Māori default weeks (3971) than in
the English default weeks (4760). This, and the fact
that the English default weeks have a higher
percentage of 0 searches per session (40.0%
compared with 27.6%), suggests that the bilingual
sessions use more browsing when the default
language is set to English and have very effective
searches, but when the default language is set to
Māori more searches are undertaken and the
searches are not as effective.
Table 9: Searching Summary for Bilingual Sessions
Bilingual Extended Sessions
EN wks MI wks
number of searches: 51 115
average per session: 3.4 4.0
maximum searches: 18 14
minimum searches: 0 0
standard deviation: 2.7 1.7
0 searches: 40.0% 27.6%
1 search: 26.7% 17.2%
2-3 searches: 13.3% 20.7%
4+ searches: 20.0% 34.5%
average search terms: 1.5 1.8
8 CONCLUSION
Designers of bilingual websites should be aware that
setting the default language of a website strongly
favours usage of the website in that language. This
will occur despite users having the ability to easily
switch the interface language.
An initial look at the data in Table 2 suggests
that there is little difference in the two default
language settings. However when the actual user
data is analysed, the extended session analysis, we
can see that the number of sessions in the default
language is higher, the sessions are longer and
include more user activity, involve the accessing of
more pages, and there is less of a reliance on
searching to access pages.
We have also discovered a new user type, the
bilingual user, who conducts a significant amount of
activity in both languages and whose usage
characteristics seem to alter depending on the default
language setting.
One consideration with these results is the data is
drawn from just two 2 week time periods. Analysis
over a longer period will produce more conclusive
results.
WEBIST 2005 - WEB INTERFACES AND APPLICATIONS
268

REFERENCES
Apperley M. D., Keegan T. T., Cunningham S. J., Witten,
I. H., 2002. Delivering The Māori Newspapers on the
Internet in Curnow J, Hopa N, McRae J (ed.s), Rere
Atu Taku Manu! Discovering History Language and
Politics In The Māori Language Newspapers.
Auckland University Press. Pages 211-36
Jones S., Cunningham S. J., McNab R. J. and Boddie S.,
2000. A transaction log analysis of a digital library. In
International Journal on Digital Libraries 3(2) 152-
169.
Te Puni Kōkiri, 2001. Māori Access to Information
Technology. Te Puni Kōkiri, Wellington, New Zealand.
Te Puni Kōkiri, 2003. Speakers of Māori within the Māori
Population. Te Puni Kōkiri, Wellington, New Zealand.
Witten, I. H., Bainbridge, D., 2002. How to Build a Digital
Library. Morgan Kaufmann. San Francisco, CA.
WHAT HAPPENS IF WE SWITCH THE DEFAULT LANGUAGE OF A WEBSITE?
269