See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262310186

Amharic-English bilingual web search engine

Conference Paper · October 2012

DOI: 10.1145/2457276.2457284

CITATIONS READS
2 3,084

2 authors:

Mequannint Munye Solomon Atnafu

University of Oslo Addis Ababa University
1 PUBLICATION   2 CITATIONS    57 PUBLICATIONS   279 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Similarity-Based Algebra for Image Database Systems View project

Modélisation et traitement de requêtes images complexes View project

All content following this page was uploaded by Solomon Atnafu on 05 January 2016.

The user has requested enhancement of the downloaded file.

 Amharic-English Bilingual Web Search Engine
Mequannint Munye
Department of Computer Science
College of Engineering and Technology
Jijiga University
P.O.BOX 1020, Jijiga, Ethiopia
Tel: (+251) 91 211 99 15
email: mokemz24@yahoo.com
ABSTRACT
As non-English languages are growing exponentially on the
Web, the number of online non-English speakers who realizes the
importance of finding information in different languages is
enormously growing. However, the major general purpose search
engines such as Google, Yahoo, etc have been lagging behind in
providing indexes and search features to handle non-English
languages. Amharic, which is the family of Semitic languages
and the official working language of the federal government of
Ethiopia, is one of these languages with a rapidly growing content
on the Web. As a result, the need to develop bilingual search
engine that handles the specific characteristics of the users‟ native
language query (Amharic) and retrieves documents in both
Amharic and English languages becomes more apparent.
In this research work, we designed a model for an Amharic-
English Search Engine and developed a bilingual Web search
engine based on the model that enables Web users for finding the
information they need in Amharic and English languages. In
doing so, we identified different language dependent query
preprocessing components for query translation. We have also
developed a bidirectional dictionary-based translation system
which incorporates a transliteration component to handle proper
names which are often missing in bilingual lexicons. We have
used an Amharic search engine and an open source English search
engine (Nutch) as our underlying search engines for Web
document crawling, indexing, searching, ranking and retrieving.
To evaluate the effectiveness of our Amharic-English
bilingual search engine, precision measures were conducted on
the top 10 retrieved Web documents. The experimental results
showed that the Amharic-English cross-lingual retrieval engine
performed 74.12% of its corresponding English monolingual
retrieval engine and the English-Amharic cross-lingual retrieval
engine performed 78.82% of its corresponding Amharic
monolingual retrieval engine. The bilingualism advantage of the
system is also evaluated by comparing its results with general
purpose search engines. The overall evaluation results of the
system are found to be promising.
Permission to make digital or hard copies of all or part of
this work for personal or classroom use is granted without
fee provided that copies are not made or distributed for profit
or commercial advantage and that copies bear this notice
and the full citation on the first page. To copy otherwise, or
republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
MEDES‟12, October 28-31, 2012, Addis Ababa, Ethiopia.
Copyright © 2012 ACM 978-1-4503-1755-9/10/10...$10.00.
Solomon Atnafu
Department of Computer Science
College of Natural Sciences
Addis Ababa University
P.O.Box 1176, Addis Ababa, Ethiopia
Tel: (+251) 91 140 69 46
email: solomon.atnafu@aau.edu.et
Categories and Subject Descriptors:
H.3.3 [INFORMATION STORAGE AND RETRIEVAL]:
Information Search and Retrieval-Search process
General terms: Design, Algorithms and Language,
Keywords: Bilingual search engines, cross-lingual information
retrieval, query preprocessing, query translation, transliteration.
1. INTRODUCTION
As non-English language Web contents which are written in
different languages and encoding schemes are available online on
the Web, the number of non-English speakers that use the Web as
their major source of information and a means of communication
channel is rapid growing. However, general purpose search
engines, such as Google, Yahoo, etc, often ignore the special
characteristics of non-English languages although information
search and retrieval needs language specific treatment [2, 7]. The
non-English speaking users use these search engines that do not
take into account the structure and the special characteristics of
the specific language they use, because they may not have
alternatives. This has led to the need to develop language specific
search engines instead of general purpose once. However,
developing search engines that support only a specific language
and writing scheme do not allow the users to access all relevant
documents available on the Web. This is because; there may be
relevant documents in response to the user‟s query that are not in
the same language and script of the query language.
In recent years, multilingual information retrieval (MLIR) or
bilingual IR, for the case of two languages, has got considerable
attention because it searches and presents the results in second or
third language of which the user is aware of [6]. MLIR system
involves providing a query in one language and searching
document collections in one or more different languages. Since
the query and the document are written in different languages, any
Bilingual Information Retrieval (BLIR) system integrates a
machine translation system as its core component to translate
either the query or the document. According to [5], many BLIR
systems have used a query translation approach since translating
the document needs large bilingual corpus particularly in
morphologically rich languages such as Amharic.
Although machine translation (MT) systems use different
approaches, dictionary based approach is preferable for under
resourced language bilingual web search engines that use query
translation approach for several reasons. We can see these reasons
from two perspectives: The first one is from the advantages of lingual information retrieval by focusing on the translation
machine readable dictionary translation approach perspective[13]: approaches used, the information retrieval model or tool used, the
Machine-readable dictionaries that are used in dictionary-based language pairs used, and whether they are web applications or
translation approach are more widely available and easier to use not.
than the parallel corpora required by the corpus-based A work of Mohammed et al. [14], evaluated the use of a
approach. Machine Translation based approach for query translation in an
The limited availability of existing parallel corpora cannot meet Arabic-English CLIR system. The work is experimented using the
the requirements of practical retrieval systems in today‟s three query types to determine the effects of query length on the
diverse and fast-growing Web environment. performance of the machine translation based method for CLIR.
The dictionary-based Cross-Lingual Information Retrieval The results showed that the machine translation achieved 61.8%,
(CLIR) approach is more flexible, easier to develop, and easier 64.7%, and 60.2% for title, description, and narrative fields,
to control when compared with Machine Translation based respectively. As Arabic is a relatively widely researched Semitic
CLIR which has little space for users to modify it for their language and has a number of common properties that share with
specific purposes. Amharic, the tools developed for this language can be customized
and used for Amharic.
Secondly, from the current IR models perspective [17]:
Another work on Cross-Lingual Information Retrieval is the
Most IR models are based on bag-of-words models. Therefore,
work of P. L. Nikesh et al. [15], the paper described about an
they don‟t take the syntactic structure and word order of queries
English-Malayalam Cross-Lingual Information Retrieval system.
into account and hence they are easy to develop.
The system retrieves Malayalam documents in response to query
Queries submitted to IR systems are usually short. Therefore,
given in English or Malayalam language. The authors used a
they don‟t describe the user‟s information need in an
bilingual dictionary developed in house for translation. For
unambiguous and precise way.
document ranking and retrieval, a system developed based on the
This means that a simpler translation approach may suffice to
vector space model (VSM) was used. As Malayalam is under
implement the translation process. Having these ideas and the
resourced language like Amharic, the query translation
scarcity of linguistic resources for Amharic in mind, we believe
approaches that the authors follow and the Information Retrieval
that developing the query translation system based on a word-by-
(IR) tool used are the important points that we have learnt to
word translation method by looking up the general-purpose
develop our system.
Amharic-English bilingual dictionary can perform better.
Joanne Capstick et al. [12], developed MULINEX which is a
However, dictionary based approach has one serious
fully implemented multilingual search engine and it is available in
problem, that is, word coverage limitations of dictionaries because
German, French and English languages. MULINEX is a
of the appearance of Out-Of-Vocabulary (OOV) words in the
multilingual Internet search engine that supports selective
query. This often occurs because most of the queries contain
information access, navigation and browsing in a multilingual
proper names and borrowed words that do not often present in the
environment. MULINEX incorporates Web spiders, concept-
bilingual lexicons [6]. As a result, the translation component
based indexing, relevance feedback, translation disambiguation,
should integrate a transliteration component to alleviate the
document categorization, and summarization functionalities. In
problem of OOVs and to translate them into the target language
this system, queries are morphologically analyzed and then
without using dictionary.
translated by making use of multilingual dictionaries.
Amharic is the second most-spoken Semitic language in the
Another work on multilingual search engine is the work of
world, next to Arabic, and the official working language of the
Wen-hui Zhang et al. [11]. This work is a multilingual Chinese-
Federal Democratic Republic of Ethiopia (A country with more
English search engine developed by a project of Chinese
than 80 million populations). As a result, there are significant
Academy of Sciences. The work was conducted with the intention
amount of Amharic documents on the Web. Although the
to develop systems that can efficiently search, index and retrieve
percentage of Amharic content on the Web is very much less
multilingual information for Chinese (mother tongue) and English
compared to English, the Amharic contents could be different
information. The system has uniform query interfaces for both
than contents that are published in English language, as these
Chinese and English languages to allow the user to conduct
contents are related mostly to Ethiopian events. Thus, using only
searching.
one of the languages do not allow for accessing all the available
Another work on Multilingual Web Retrieval with an
relevant documents. There are a number of works about Amharic-
Experiment in English–Chinese Business Intelligence is
English information retrieval [1, 3, 4] and a little about Amharic
developed by Jialun Qin et al. [13]. This work dealt with
search engine [2]. However, to our knowledge, there are no
developing and evaluating a multilingual English–Chinese Web
attempted works about Amharic-English information retrieval on
portal that incorporates various CLIR techniques for use in the
the Web that accepts user queries according to the user‟s language
business domain. According to the paper [13], the dictionary
preferences and returns the result in both languages.
based approach of query translation is the most promising for
This research work is thus to design and develop a generic
Web applications. The authors strongly recommended to use
model for a bilingual search engine that integrates a query
dictionary based approaches for Web based applications
translation system for Amharic and English languages.
especially for under resourced languages like Amharic. It has also
shown the effectiveness of the approach as queries are usually
short and IR models are based on bag-of-words.
2. RELATED WORK Other related works are on Amharic-English Information
We reviewed and discussed some related works on CLIR, Retrieval. Atelech Alemu, in her three consecutive research works
multilingual/bilingual search engines, and Amharic-English cross- on Dictionary-based Amharic - English Information Retrieval [4],
Amharic-English Information Retrieval [3], and Amharic-English
Information Retrieval with Pseudo Relevance Feedback [1]
discussed cross lingual information retrieval between Amharic
and English languages. For all the research works the authors
used Amharic-English machine readable dictionaries and an
online Amharic-English dictionary for the query terms translation
with some additional enhancements for query translation,
indexing and searching from one research to the next. The
Amharic topic set used in all the experiments was constructed by
manually translating the English topics. As the experimental
results showed, progressive performance achievements were
observed form the first research to the next one and the challenges
related to the issues were discussed. The Amharic-English cross
lingual information retrieval researches done by the researcher
were not information retrieval on the Web and is not used to
retrieve both Amharic and English language documents. However,
cross lingual information retrieval on the Web has additional
overheads that should be considered compared to traditional
CLIR due to several factors exist on the Web documents.
Multilingual Web retrieval differs from traditional CLIR because
of the factors that [13]:
 Web pages are more unstructured and are very diverse in terms
of document content and document format (such as HTML or
ASP).
 Traditional CLIR usually focuses on effectiveness, measured in
recall and precision. However, Web retrieval is also concerned
with efficiency to end users (i.e. response time and query
length).
3. ANALYSIS OF THE AMHARIC
LANGUAGE
Amharic is the second most-spoken Semitic language in the
world, next to Arabic, and the official working language of the
Federal Democratic Republic of Ethiopia. Amharic uses a unique
script, which has originated from ancient language, the Ge‟ez
alphabet, which is the liturgical language of the Ethiopian
Orthodox Church. It has been the working language of the
government, the military, and of the Ethiopian Orthodox
Tewahedo Church throughout modern times. Outside Ethiopia,
Amharic is the language of some 2.7 million emigrants (primarily
in Egypt, Israel, Sweden, Eritrea, and the United States) [8, 9, 10].
Thus, it has official status and is spoken by many people as their
native and second language. In addition, it is a language with
many literatures. Of these who speak Amharic, a significant
number of them (usually the educated class) can understand and
speak English as well.
Amharic has a complex morphology which combines
consonantal roots and vowel intercalation. Amharic and English
differ substantially in their morphology, syntax and the writing
system they use. Therefore, the search engines which are mainly
developed for English cannot efficiently be used to retrieve
Amharic documents. Amharic language has the following
characteristics that should be considered in information retrieval
and translation systems.
 Morphological variations
As Amharic language is morphologically complex, words are
inflected with prefixes, suffixes and infixes. For example, the
Amharic word “በሊ” has morphological variations such as:
“በለ” ፣“በሊን”፣“በሊች” ፣“በሊሁ” ፣“በሊችሁ”፣ “አስበሊ”, etc.
 Character Redundancy
Amharic has some redundant symbols with the same sound
in its alphabet. For example, አ, ኣ, ዏ and ዓ, ሰ and ሠ, and ጸ
and ፀ have the same sound. These characters can be used
interchangeably without any meaning difference in the
language.
 Short Words
In Amharic, it is common to shorten some words using the
forward slash („/‟) and the English full stop (ነጥብ(.)). For
example the Amharic words ወታደር, ዓመተ ምህረት, ወልደ ስሊሴ
can be shortened as ወ/ር, ዓ.ም, ወ/ስሊሴ, respectively. These
shortened words will be expanded to their normal forms for
dictionary lookup or IR systems.
 Variations due to Pronunciations
Some of the Amharic words have multiple spelling variants
like ጧት፣ጥዋት፣ጡዋት፣ጠዋት፣ etc. The main reason for such
problems is due to the multiple regional dialects of the
spoken language. In addition, most of the words which are
adopted from English or other languages are written in
different formats like ሚሉዮን፣ ሚልዮን፣ ሚሉየን, etc. There is no
standardization in spelling for such words which in turn
result in huge variation in transliteration. These variations in
spelling are difficult for Web information retrieval
applications such as Google, which retrieves documents
based on exact term matching.
 Stop-words
Most search engines and query translation systems do not
consider extremely common words in order to save disk
space or to speed up search results. Although Amharic does
not have standard stop word lists, ሁለ, ሁኔታ, ሆነ, በኋሊ, በጣም,
ብቻ, ወደ, ናቸው, etc are considered to be stop words.
 Amharic Punctuation Marks
There are different punctuation marks in Amharic language
and among these punctuation marks ።(አራት ነጥብ), ፣(ነጠሊ
ሰረዝ), ፤(ድርብ ሰረዝ),?(የጥያቄ ምልክት) are the most commonly
used ones. They are used in query or document tokenization
as word delimiters in addition to white spaces.
4. THE ARCHITECTURE OF AMHARIC-
ENGLISH BILINGUAL SEARCH
ENGINE
As it can be surmised from its name, our bilingual search
engine is designed to pull up results in not one, but two languages
by accepting queries in either of the languages (Amharic or
English). As most of the bilingual/multilingual search engines,
our proposed Amharic-English bilingual search engine has basic
components such as Query Preprocessing, Query Translation,
Amharic Search Engine, and English Search Engine as shown in
Figure 1.
Query Preprocessing
The query posed by the users must pass through the query
preprocessing stage before it has been translated for cross
language retrieval and at the same time it can be directly
submitted to the appropriate search engine for monolingual
retrieval. Since query preprocessing is a language specific task,
the query preprocessing process is done in two different query Query Translation
preprocessing modules. The query translation process also uses two components
(Amharic Query translation and English Query translation) which
are responsible for lexical transfer or dictionary lookup in the
bilingual Amharic-English dictionary and transliterating the out of
dictionary words assuming that they are proper names or
borrowed words as shown in Figure 2.
The Amharic Query translation component in turn has two
subcomponents: a lexical transfer and a transliterator. The Lexical
transfer receives Amharic words from the query preprocessing
component and automatically translates them into English words
only by using the bilingual dictionary.

Transliteration
All the words that are not found in the bilingual dictionary will
pass through the transliterator module. As shown in the algorithm
in Figure 3, this module works by segmenting each stemmed
Amharic words into Amharic phoneme (or characters),
transforming each character into its corresponding English
characters based on the convention for the transcription of
Ethiopic script for ASCII, and concatenating each translated
English phoneme (or character) into a single English word. In the
Figure 1: The Architecture of the Amharic-English Bilingual transliteration convention a single Amharic character is replaced
Search Engine by one, two or three Latin characters. For instance, መ: me, ሙ:mu:
ሚ: mi, ማ:ma, ሜ:mie, and ሞ:mo. However, sadises such as “ም” will
The first is Amharic query preprocessing module which in vary according to their positions. It will be mapped to “mi” when
turn consists of subcomponents such as tokenizer, normalizer, it comes first or simply “m” when it comes at the end of the word.
stop-words remover, and stemmer. This module is responsible for It is these constants and vowels are called “Xi” in the algorithm.
tokenizing the Amharic queries into words using white spaces and
Amharic punctuation marks, normalization of Amharic redundant
symbols which have the same sound (such as: አ, ኣ, ዏ and ዓ, ሰ and
ሠ, and ጸ and ፀ) and expanding short form of Amharic words such
as ወ/ሮ፣ ዓ.ም፣ ወ/ሚካኤል፣ etc, eliminating stop-words (less
informative words) such as: ነው፣ሆነ፣ወደ፣ናቸው፣ውስጥ፣etc, and
stemming inflectional and some derivational Amharic morphemes.
For example, the Amharic words “በለ” ፣“በሊን”፣“በሊች” ፣“በሊሁ”
፣“በሊችሁ”፣ “አስበሊ”, etc can be reduced to their citation word “በሊ”
using the stemming subcomponent. This helps the query
translation component to handle morphological variations and to
find matches in the dictionaries for as many of the query words as
possible. The output of Amharic query preprocessing module is a
set of Amharic preprocessed bag of words which is used as an
input for the Amharic query translation component.
The second is English query preprocessing module, like
Amharic query preprocessing module it has subcomponents like
tokenizer, stop words remover, and stemmer. This module is
responsible for tokenizing the English queries into words using
white spaces and some English punctuation marks as word
delimiters, eliminating English stop-words (such as: “a”, “an”,
“are”, “be”, “for”, etc), and stemming inflectional and some
derivational English morphemes. For example, the English words
such as: “charger”, “charging”, “charged”, “charges” can be Figure 2: Amharic Query Translation Component
reduced to their base word “charge”. In general, these
subcomponents have similar functionalities to their corresponding
Amharic query preprocessing subcomponents except they are used
for English language queries.
1. Input: Amharic bags of words
2. Output: English bags of words
3. For each Amharic stemmed word not found in the dictionary
4. Split the word into its Amharic character
5. For all the characters in the word
6. If (the character is at the beginning of the word and the
character is sadis)
7. Replace with its Latin conventional „Xi‟
8. Put the characters at the beginning of the word
9. else if (the character is sadis and the character before it is
also sadis)
10. Replace with its Latin conventional „Xi‟
11. Append the character in its appropriate place
12. else if (the character is sadis and it is before the last
character)
13. Replace with its Latin conventional „Xi‟ except for
some exceptions
14. Append the character in its appropriate place
15. else
16. Transliterate the character based on the transliteration
convention
17. Concatenate the characters
18. End if
19. End for
20. End for
Figure 3: Algorithm for Amharic-English transliteration
The English Query Translation component has also two
subcomponents and works like the Amharic query translation
component except the transliterator subcomponent uses the
transliteration database as its knowledge base instead of direct
character mappings.
Like the algorithm in Figure 3, some other algorithms such
as Amharic-English translator, English-Amharic translator,
English-Amharic transliterator, etc have been developed for the
purpose of Amharic-English and vice versa translation and
transliteration.
Amharic and English Search Engines
The results of our bilingual search engine are highly
dependent on the underlying English and Amharic search engines
used to collect Web documents from the Web. Each search engine
performs the crawling, indexing and ranking according to its own
mechanism. The Amharic search engine component is
responsible for crawling, indexing, and ranking of the Amharic
Web documents [2]. The English search engine component is
responsible for crawling, indexing, and ranking of English Web
documents [16].
5. EXPERIMENTAL RESULTS
In order to evaluate the effectiveness of our search engine, an
experiment which measures the precision of the system was
designed and conducted using both Amharic and English queries
and on both Amharic and English language web document
collections. We also conducted an experiment to visualize the
significance of our bilingual search engine. This is done by
comparing the relevant bilingual retrieval results of our system
(Amharic and English documents) with the relevant results of
Google monolingual search results. We also conducted
experiments to evaluate the performance of some individual
subcomponents which have direct impact on the effectiveness of
our system.
Results of Translation Component
To visualize the effectiveness of the Amharic-English
translation component, an in house developed bilingual dictionary
with 5000 Amharic words and their corresponding translations is
used, and 15 Amharic queries, which have a total number of 36
words, are selected for the cross-lingual evaluation. This
component properly translates 10 queries out of the 15 queries.
When we see it word wise, out of the 36 words, 31(88.57%)
words are properly translated. To test the effectiveness of the
English-Amharic query translation component, 15 English queries
which are the manual translation results of the 15 Amharic queries
are used. Out of the 15 English queries, 14 (93.33%) queries are
properly translated.
Results of the Transliteration Component
The transliteration component is also independently
evaluated for its performance. To do this, we have collected 1100
(one thousand one hundred) proper names of which most of them
are person names and some are the names of countries, places,
hotels, well known cities, etc. The manual transliteration of the
gazetteer is made with the help of linguists. Out of the 1100
proper names, 887(80.64% of the names) names are properly
transliterated.
Bilingual Retrieval Performance Evaluation
Since our system is bidirectional, within a single query and
its corresponding manual translation four different search results
will be retrieved. For the Amharic query “የአማርኛ ቋንቋ”, the
Amharic-English search engine retrieved both Amharic and
English web documents in the same browser window separated
byframes as shown in Figure 4. In the same manner, the English-
Amharic results are as shown in Figure 5 for the English query
“Election 2002”.
The Amharic-English cross-lingual retrieval engine is
evaluated using the 15 Amharic queries and their manual
translations. Since it was impossible to read all Web page
documents collected during crawling to judge the relevancy of the
documents, we emphasized on precision only for the top 10
retrieved Web pages for each query. Based on the precisions
calculated for the English monolingual and Amharic-English
cross-lingual retrieval for each query, the precision of the English
monolingual retrieval engine is 85% and that of the Amharic-
English retrieval engine is 63%. From this, we can observe that
the Amharic-English cross-lingual retrieval system out performed
74.12% of its corresponding monolingual retrieval engine.
The English-Amharic cross-lingual retrieval engine is also
evaluated using the 15 English queries and their manual
translations. The Amharic monolingual retrieval engine has an
average precision of 85% and its corresponding English-Amharic
cross-lingual retrieval engine has an average precision of 67% as
calculated for only the top 10 retrieved English web pages. This
experimental result also showed us the English-Amharic cross-
lingual retrieval engine out performed 78.82% of its
corresponding monolingual retrieval engine.
 Figure 4: A Snapshoot of the Search Result for the Amharic query “የአማርኛ ቋንቋ”

Figure 5: A Snapshoot of the Search Result for the English query “Election 2002”
 The Significance of Our System Compared to As shown in Table 1, our system has two advantages over
general purpose search engines. The first one is our system
General Purpose Search Engines
considered the characteristics of Amharic language and retrieved
To evaluate the significance of our system when compared to
the same results for different variations of the words in the query
general purpose search engines, some Amharic queries which
but not Google. Secondly, our system retrieved more number of
have morphological variations, character variations, and short
relevant documents by simultaneously searching both Amharic
words are given to our system and to „Google‟ search engine. For
and English Web documents for the given Amharic query but
the experiment, our search engine has worked on seven hand-
Google retrieved only Amharic Web documents.
picked seed URLs provided to the crawler. Domain experts are
also involved to judge the relevance of the documents retrieved to
the query.

Table 1: Google versus our system results for queries that show different characteristics of Amharic language.
Our System Results
Google Results
Language Amharic Documents No. of Rel. Total No. of
Query
Characteristics No. of Rel. Documents No. of Rel. Documents English Relevant
Documents Similarity Documents Similarity Documents Documents

የኢትዮጵያ ፓርቲዎች 10 7 5 12
0 7
Morphological ኢትዮጵያ ፓርቲ 7 7 5 12
Variations የኢትዮጵያ ምርጫዎች 9 8 7 15
0 8
ኢትዮጵያ ምርጫ 6 8 7 15
አማርኛ ዜና 9 10 7 17
0 10
Character ዏማርኛ ዜና 2 10 7 17
Variations መሇስ ዜናዊ 10 6 4 10
0 6
መሇሥ ዜናዊ 3 6 4 10
የትምህርት ጥራት 10 8 5 13
Short words 1 8
የት/ት ጥራት 6 8 5 13

6. CONCLUSION AND FUTURE WORKS
In this work, we designed a generic model and developed a
bilingual Web search engine for Amharic and English languages.
The search engine has different components which address the
basic language specific issues for the two languages in query
translation and information retrieval. These major components
are: Amharic query preprocessing, English query preprocessing,
Amharic query translation, English query translation, Amharic
and English search engines.
To evaluate the effectiveness of our Amharic-English
bilingual search engine, precision measures were conducted on
the top 10 retrieved Web documents. The precision was calculated
for each query and the average precision measure showed
promising results for both the Amharic-English and English-
Amharic cross-lingual evaluations.
With this work, several novel ideas and designs are proposed as
contributions. Among these contributions, we:
 Identified the major components of Amharic-English
bilingual search engine which should be considered in
developing Amharic-English bilingual search engine.
 Identified the critical issues that need to be dealt in cross-
lingual search engines and proposed appropriate techniques.
 Proposed a general model for Amharic-English bilingual
search engine.
 Paved the way for further large scale research projects on
Web based bilingual search engines by identifying language
dependent components.
 Developed algorithms for query translation and
transliteration. These algorithms can also be adopted for
other cross-lingual information retrieval.
 Designed and created a bilingual Amharic-English dictionary
for query translation purpose and a repository reference table
for transliteration of words in queries.
 Customized the Amharic stemmer developed for information
retrieval to fit for translation systems.
 Designed and developed an Amharic-English bilingual
search engine based on the proposed model.
 Evaluated the prototype developed for its effectiveness.
 7. REFERENCES
[1] Atelach Alemu Argaw. "Amharic-English Information
Retrieval with Pseudo Relevance Feedback". In: Peters, C.,
et al. (eds.) Advances in Multilingual and Multimodal
Information Retrieval: 8th Workshop of the Cross
Language Evaluation Forum, CLEF 2007, Budapest,
Hungary, September 19-21, 2007.
[2] Tessema Mindaye, Hassen Redwan, Solomon Atnafu.
“Searching the Web for Amharic Content”. On the
International Journal of Multimedia Processing and
Technologies (JMPT), 2010. ISSN Print ISSN: 0976-
4127.”
[3] Atelach Alemu Argaw and Lars Asker. "Amharic-English
Information Retrieval". Working Notes of CLEF 2006,
Alicante, Spain. September 2006.
[4] Atelach Alemu Argaw, Lars Asker, Rickard Cöster and
Jussi Karlgren. "Dictionary-based Amharic - English
Information Retrieval". In Proceedings of Cross Language
Evaluation Forum (CLEF 2004), Bath, UK. September
2004.
[5] Kristen Parton, Kathleen R. McKeown, James Allan, and
Enrique Henestroza. “Simultaneous Multilingual Search for
Translingual Information Retrieval”. In Proceeding of the
17th ACM conference on Information and knowledge
management, Napa Valley, California, USA. ACM New
York, NY, USA, 2008.
[6] Karunesh Arora, Ankur Garg, Gour Mohan, Somiram
Singla, and Chander Mohan. “Cross Lingual Information
Retrieval Efficiency Improvement through Transliteration”.
In Proceedings of ASCNT – 2009, CDAC, Noida, India,
pp. 65 – 71, 2009.
[7] Judit Bar-Ilan and Tatyana Gutman: “How do search
engines handle non-English queries? - A case study”.
WWW (Alternate Paper Tracks), Budapest, Hungary, 2003.
[8] Amharic-WIKIPIDIA, the free encyclopaedia.
Available at: http://en.wikipedia.org/wiki/Amharic,
Accessed on 21 September 2009.
View publication stats
[9] Amharic Ethiopia Language. Available at: http://www.free-
press-release.com/news/200907/1248234344.html,
Accessed on July 21, 2009.
[10] Amharic Language. Available at:
http://multilingualbooks.com/amharic.html, Accessed on 11
August 2010.
[11] Wen-hui Zhang, Hua-lin Qian, Wei Mao and Guo-nian Sun.
“A Multilingual (Chinese, English) Indexing, Retrieval,
Searching Search Engine”. Available at
http://www.isoc.org/inet99/proceedings/posters/210/index.h
tm, Accessed on August 10, 2009.
[12] Joanne Capstick, Abdel Kader Diagne, Gregor Erbach and
Hans Uszkoreit. “MULINEX: Multilingual Web Search and
Navigation”, Accessed on August 25, 2009, Available at
http://eprints.kfupm.edu.sa/52030/1/52030.pdf, Published
on 08.02.99.
[13] Jialun Qin, Yilu Zhou, Michael Chau and Hsinchun Chen.
“Multilingual Web retrieval: An experiment in English–
Chinese business intelligence”. John Wiley & Sons, Inc.
New York, NY, USA, 2006
[14] Mohammed Aljlayl, Ophir Frieder, and David Grossman.
“On Arabic-English Cross-Language Information Retrieval:
A Machine Translation Approach”. IEEE Computer
Society Washington, DC, USA, 2002
[15] P. L. Nikesh, Sumam Mary Idicula, S. David Peter.
“English-Malayalam Cross-Lingual Information Retrieval–
an Experience‟, In Proceedings of IEEE International
Conference on Electro/Information Technology, Ames,
Iowa State University, May 2008
[16] A. Arasu, J. Cho, H. Garcia-Molina, A. Paepcke, and S.
Raghavan. “Searching the Web”. ACM Transactions on
Internet Technology (TOIT), 2001.
[17] Wessel Kraaij, Jian-Yun Nie, and Michel Simard.
“Embedding Web-based statistical translation models in
cross-language information retrieval.” MIT Press
Cambridge, MA, USA, 2003.