515.txt

The 9th International Conference on Informatics and Systems (INFOS2014) – 15-17 December
Natural Language Processing and Knowledge Mining Track
Extracting N-gram Terms Collocation from
Tagged Arabic Corpus
Waseem Alromima
Faculty of Computer and
Information Science
Ain Shams University, Egypt
waseem.2020@yahoo.com
Rania Elgohary
Faculty of Computer and
Information Science
Ain Shams University, Egypt
rania_elgohary@yahoo.com
Abstract—Information Extraction (IE) is one of the most
important Natural Language Processing (NLP) applications,
which extracts information such as Named-Entities (NE) and
collocation of terms from the corpus. Collocation is a sequence of
terms that co-occur together in the corpus. In Arabic
Information Extraction, there are many problems because of the
complex of Arabic’s grammar and ambiguity. In general, in
linguistics research, the more efficient corpus is the one
annotated by Part of Speech Tagging (POST). In this paper, we
propose a prototype that extracts collocation of N-gram words
(from 2- 6 gram) based on the sequence of POST from Arabic
Quran corpus. This approach extracts the collocation of N-gram
words by matching the input structured pattern of Arabic
language versus the Part of Speech Tagging of Quran corpus.
The system enables users to select a sequence of tags (2-6 gram)
and scope of the corpus source (whole Quran Corpus or specific
Surah). To show how the system is beneficial for linguistic
research, a set of experiments has been conducted in different
scenarios.
Keywords— Information Extraction; Arabic Phrases;
Computational linguistics; n-gram; Part-of-Speech Tagging
(POST)
I. I NTRODUCTION
Information extraction (IE) is one of the most important
Natural Language Processing (NLP) applications, which
extracts structured information such as Named-Entities (NE) or
collocation of terms [14] from unstructured or semi-structured
text. Generally, the term extraction (TE) process consists of
two steps: (1) identifying term (either single or multi–word
terms) from text, and (2) filtering through the candidates to sort
out terms from non-terms [11]. Example for using TE is to
extract the most important words from the text and using these
words to summarize the content based on them. Moreover
Arabic Natural Language Processing (ANLP) has gained an
importance field and several state of the art systems have
developed for a wide range of applications, including text
summarization, Proper Noun Extraction [17]. ANLP still in its
open research compared to the works have done in the Latin
language, because there are some issues that slow down
progress in ANLP [18] [19].
These issues illustrate as the following:
a. Arabic is highly inflectional and derivational, which
makes morphological analysis a very complex task.
Ibrahim F. Moawad
Mostafa Aref
Faculty of Computer and
Faculty of Computer and
Information Science
Information Science
Ain Shams University, Egypt Ain Shams University, Egypt
ibrahim_moawad@cis.asu.edu.eg
Aref_99@yahoo.com
b.
The absence of diacritics in the written text creates
ambiguity and, therefore, complex morphological rules
are required in order to identify the tokens and parse the
text.
c. Diacritics are not consistent or predictably marked with
Arabic letters.
d. The writing direction is from right-to-left and some of the
characters change their shapes depending on their
location in the word.
e. Capitalization is not used in the Arabic language, which
makes it hard to identify proper nouns and abbreviations.
The collocation term is defined as a sequence of terms that co-
occur together in the corpus. Other definition, it is the two or
more words, which appear together and always seem as
associated [9]. The collocations are important in many
applications, such as, machine translation, information
retrieval, information extraction, word sense disambiguation
and lexicography.
In general, in linguistic research, the more efficient corpus
is the one annotated by Part of Speech Tagging (POST). The
main usages of the corpus are statistical analysis, occurrences
checking, and linguistic rules validating on a specific universal
[13]. The tagged is assigned every word in the corpus by tag
such as (noun, verb, adjective, etc.), so that the tag of the word
gives important information about the word to understand the
context. Moreover stop Words eliminations are improved the
collocation while it’s improved in the Arabic text classification
[21].
In Arabic Information Extraction, there are many
problems because of the complex of Arabic’s grammar and
ambiguity, and hence this constitutes the difficulties of
identifying the collocations. The Arabic Quran Corpus is
annotated linguistic resource consisting of 77,430 words.
According to the Quran project developed by Leeds University
[3], it aims to provide morphological and syntactic annotations
for researchers who want to study the language of the Quran.
In this paper, we propose a prototype that extracts
collocation of N-gram words (from 2- 6 gram) based on the
pattern sequence of tags from the Arabic Quran corpus. This
approach extracts the collocation of N-gram words by
matching the input structured pattern of Arabic language versus
the Part of Speech Tagging of Quran corpus. The system
Copyright© 2014 by Faculty of Computers and Information–Cairo University
NLP-10The 9th International Conference on Informatics and Systems (INFOS2014) – 15-17 December
Natural Language Processing and Knowledge Mining Track
enables the users to select the scope of corpus data source and
the pattern sequence of tags, and then the system extracts and
displays the collocations. To show how the system is beneficial
for linguistic research, a set of experiments has been conducted
in different scenarios.
The rest of this paper is organized as follow: Section 2
presents the Related Work briefly. Section 3 describes the
System Architecture. Section 4 describes the System
Methodology while section 5 presents the Implementation of
the system with case studies. Section 6 shows the experiments
conducted to extract set of n-gram collocations and their
results. The paper is concluded and presented in section 7.
II. R ELATED WORK
In this section, we overview the related researches
concerning to extract Single-Word Term (SWT), Multi-Word
Term (MWT), and collection terms.
There is little research available for single-word term
extraction, which relies on frequency-related data. According
to [10], the authors have used the TF-IDF method to extract
single-word terms from three domains: financial management
succession, stock market, and crime drug domain. They have
mentioned that; the TF-IDF approach is really appealing for
extracting single-word terms, yet though it failed to deliver
particular data, relating to its efficiency.
In another research work at [5], the authors have been
coming up with a method to assign the weight to each
candidate word according to a vector of four features (Term
Frequency-Inverse Document Frequency (TF-IDF), Part-of-
speech tagging (POST), relative position of first occurrence,
and Square chine ( ) statistics). The method, starts an
automatic sorting of the candidates according to their scores in
order to select the characters of the highest rank Authors
showed the experimental results of the effectiveness of the
approach.
Alkhateb K. et al. at [1] presented a hybrid approach to
extract Multi-Word Terms (MWT) from Arabic corpus. The
authors compound nouns as an important type of MWT and
select a bi-gram terms. The approach depends on two main
filtering, starts to linguistic filter (preprocessing, extract of
noun sequence as well as noun connected by prepositional, and
extract of bi-gram) and then statistical filter: rank the bi-gram
based on Log-Likelihood Ratio (LLR) and C-value method.
The dataset used in a specific domain for the environment. The
combined two evaluating methods, Log-Likelihood Ratio
(LLR) and C-Value in the ranking process, the results with the
top 150 terms are 89%.
The research work in [7] presented an approach to extract
collocations terms from Arabic corpora. Depend on using a
Gate tool, which is a software tool kit written in Java and
widely used worldwide for NLP and IE especially those of
Named Entities. Extract binary or ternary additional
information about morphology. The POS by using a tagged
corpus developed at Leeds University and wrote new Jape
(Java Annotation Pattern Engine) rules. The authors evaluated
the system text with a reference (hand-annotated) text. The
testing for Noun-Adjective annotation, annotated manually the
same documents that are annotated using Jape rules with Gate,
then used AnnotationDiff to calculate F.Measure, it gives 0.66.
We think it’s a good approach for extraction collocation, but
the authors presented few forms of extraction with low result,
but the authors stated that they are working on improving it.
Ibrahim Bounhas et al. at [2] presented an approach to
build a semi-automatic linguistic method for Multi-Word
Terms (MWT) from a specialized domain in Arabic corpus.
By using two filtering methods: linguistic rules and statistical.
The linguistic is based on the morphological features and POS
(Part of Speech) tags to parse documents and retrieve candidate
terms. The statistical is a measure used to rank candidate terms
according to the relevancy. The dataset used in a specific
domain for the environment. This approach is relevant to
building ontology more than to extraction multi words, because
the terms in the domain too small to calculate MWT.
The research work in [12] designed a multi-word term
extraction program for Arabic language. They used a hybrid
method to extract multi-word terminology from Arabic corpus.
From linguistic respective, they used some linguistic
information to extract and filter the candidates of multiword
terminology. Their method uses the part-of-speech tagging of
the corpus that has been assigned by the Diab et al [16], to use
in the linguistic filtering. The linguistic filter is to identify the
Arabic MWT patterns, such as, N ADJ, N N and N PREP N.
Finally, the tagging terms in the Quran corpus is an important
part in the extraction terms based on the tagging sequence.
According to [3], the researcher developed a new project of the
Holy Quran, which can be freely downloaded. The linguistic
data has been manually verified by multiple annotated and
machine of readable. The project is annotated linguistic
resource depends on the Arabic grammar, syntax; morphology
for each word in the Holy Quran. The corpus provides three
levels of analysis: morphological, syntactical tree bank, and
semantic ontology.
The research work at [6] developed the Arabic part-of-
speech tags and employ to analysis and annotation of
traditional Arabic text especially in the Holy Quran text. The
Hidden Markov Models (HMMs) based-on the Arabic sentence
structure used to develop the approach. The system represents
each tag in the state of the HMM and transitions between tags
governed by the syntax of the sentence. The trial is to extract
text from a corpus of book ALJAHEZ's book titled by
"Albayan-WA-Tobin" (255 Hijri) and It is obtained from
"Ashamila" library. The experiments not used complete the
Holy Quran in this approach, but the experiment gives a good
accuracy for SURAH name ALFatiha and show the accuracy
rate. They are using a training set and testing set to give
recognition rate 96%.
III. S YSTEM A RCHITECTURE
As shown in Figure (1), the system architecture consists of
two phases. The first phase is the offline phase, which
Copyright© 2014 by Faculty of Computers and Information–Cairo University
NLP-11The 9th International Conference on Informatics and Systems (INFOS2014) – 15-17 December
Natural Language Processing and Knowledge Mining Track
preprocesses the corpus. The second phase is the online phase,
which consists of two main modules: the User Interface
Module and the Matching Engine Module.
The offline phase objective is to preprocess the corpus, the
Quran corpus stored in a semi-structured format (XML
language) obtained online from Leeds University [3]. This
phase transforms the semi-structured format into structured
format (relational database). The corpus (dataset) used in our
approach is the scripts of the Holy Quran dataset, which
consists of (114) chapter (SURA). Also, the Quran contains a
total of (6,236) verse (AYA), a total of (77,430) word. All
words in the corpus annotated by part-of-speech tagging
(POST). The tag set used is the tag set developed by Leeds
University [3].
On the other hand, the online phase consists of two main
modules: User Interface Module and Matching Engine Module.
The user interface module interacts with the end user to input
the collocation pattern features such as (scope of data source,
type of gram, and pattern tags). The Matching Engine module
matches the pattern features to the Arabic Quran Corpus
(Structured Tagged Corpus), and then extracts the collocation
terms based on the boundary of the AYAH in the SURA.
Finally, the user interface module displays the results of the
collocation terms to the end user.
Fig,1 System Architecture
IV. N-GRAM COLLOCATION METHODOLOGY
The methodology of the n-gram collocation extraction process
is illustrated by the flowchart shown in figure (2).The main
flowchart steps are:
1. The user starts to select the source of data (Specific
SURA, Whole Quran Corpus).
2. The user selects the type of N-gram such as bi-gram, tri-
gram, n-gram where n from 4-6
3. The user selects the pattern sequence of tags depending on
the gram type selected in step 2.
4. The matching engine starts to match the input data versus
the tagged Quran corpus, and then displays the results. The
pseudo code of matching engine algorithm is shown in
figure (3).
Fig. 2 flowchart of the N-gram collocation methodology
Input: scope_of_data, type_of_gram, Seq-tags ,
Output: collocation _of_words
Begin {
Phrase_collocation =‖ ‖, i=0, count =0
For i=0 to scope_of_data do
If Type_ of_ gram is ―bi-gram‖ then
If Seq-tag= tag (i) and Seq-tag1= tag (i+1)
Count_match_collocation +=1
Phrase =Pharse + Phrase_collocation
End if
Return Phrase
Else if Type_ of_ gram is ‖tri-gram‖ then
If Seq-tag =tag (i) and Seq-tag1= tag (i+1) and Seq-tag2= tag (i+2)
Count_match_collocation +=1
Phrase =Pharse + Phrase_collocation
End if
Return Phrase
Else if Type_ of_ gram is ‖n-gram‖ then
Input: n
// 4 <= n <= 6
Phrase = Function n-gram (n, scope-data, Seq-tags)
End for
Function n-gram (n,Seq-tage, scope_of_data)
Switch (n)
Case (n=4)
If Seq-tag=tag (i) & Seq-tag1=tag (i+1) & Seq-tag2=tag(i+2)& Seq- tag3=tag
(i+3)
Count_match_collocation +=1
Phrase =Pharse + Phrase_collocation
Return phrase
Else Return 0
Case (n=5)
If Seq-tag=tag (i) & Seq-tag1=tag (i+1) & Seq-tag2=tag (i+2) & Seq-tag3=tag
(i+3) & Seq-tag4=tag (i+4)
Count_match_collocation +=1
Phrase =Pharse + Phrase_collocation
Return phrase
Else Return 0
Case (n=6 :)
If Seq-tag=tag (i) & Seq-tag1=tag (i+1) & Seq-tag2=tag (i+2) & Seq-tag3=tag
(i+3)& Seq-tag4=tag (i+4) & Seq-tag5=tag (i+5)
Count_match_collocation +=1
Phrase =Pharse + Phrase_collocation
Return phrase
Else Return 0
End if
Default (―sorry, entered n value not equal from 4- 6”)
End function
Fig.3 Pseudo Code of the Matching Engine Algorithm
Copyright© 2014 by Faculty of Computers and Information–Cairo University
NLP-12The 9th International Conference on Informatics and Systems (INFOS2014) – 15-17 December
Natural Language Processing and Knowledge Mining Track
Figure (3) shows the pseudo code algorithm of the Matching
Engine process, which matches the pattern features input from
the end user (Scope of data source, type of gram, and sequence
tags) to the Arabic Quran Corpus (Structured Tagged Corpus).
It then validates the matching collocation terms, if exist, and
displays the results. If the matching process does not succeed,
it enables the user to insert another pattern features.
V. IMPLEMENTATION AND C ASE STUDY
We have designed and implemented the prototype of
extraction n-gram collocation terms. The system has been
implemented using the Visual Basic.Net and the database has
been developed using SQL server. The system has been tested
on a CORE i5 machine running Windows 7. The following two
cases are illustrated briefly. The two cases are collocation
extracting from specific SURA and from the whole Quran
Corpus.
 Case 1: Collocation Extraction from specific SURA
In this case, the user would like to perform a collocation
pattern tags such as N+ADJ (Noun +Adjective) from the
specific SURA under the name ―ALKhahf‖ as shown in
figure (4). The number of collocation terms is 28.
Fig.5 Results of Collocation N+ADJ Pattern Sequence in the whole Quran
Corpus
VI. E XPERIMENTS
By analyzing the Quran corpus by retrieving the number of
frequency of different part of speech (POS), we found the
Noun (N), Verb (V), Adjective (ADJ), Proper-Noun (PN), and
Preposition (P) are the most frequently used in the Quran
corpus. Table (1) shows the number frequency of those POS
tags.
Table (1) Frequency of most used tags in the Holly Quran corpus
Fig.4 Results of Collocation Noun +Verb (N+V) Pattern Sequence from
specific SURA ―ALKhahf‖
 Case 2: Collocation Extraction from the whole Quran
Corpus
In this case, the user would like to perform a collocation
pattern tags such as N+ADJ (Noun +Adjective) from the
whole Quran Corpus. As shown in figure (5), the number
of collocation terms is 1486.
Tag Description Frequency
N
PN
ADJ
V
P Noun
Proper-Noun
Noun
Adjective
Verb
Preposition 24769
3900
1977
19360
7358
In this section we have conducted three experiments to extract
collocation terms based on select type of N-gram from the
whole Quran corpus. The first experiment is bi-gram
collocation, which may be composed of N, V, ADJ, and PN.
The second experiment is the tri-gram collocation, which may
be composed of N, V, and P. Final experiment is the n-gram
collocation, which may be composed of the simple noun phase
and verb phrase structures. In the following these experiments
are illustrated in details.
 Experiment 1: Bi-gram Collocation
Table (2) shows a different sequence of parts of speech
obtained from this experiment. We selected the probability of
the tag sequences of N (noun), V (verb), ADJ (adjective), PN
(proper noun), which is 24= 16 pattern sequences. The results
of sequences are shown in figure (6).
Table (2) Pattern Sequence of tags by the bi-gram
Tag
N
V
ADJ N
N+N
V+N
ADJ+N V
N+V
V+V
ADJ+V ADJ
N+ADJ
V+ADJ
ADJ+ADJ PN
N+PN
V+PN
ADJ+PN
PN PN+N PN+V PN+ADJ PN+PN
Copyright© 2014 by Faculty of Computers and Information–Cairo University
NLP-13The 9th International Conference on Informatics and Systems (INFOS2014) – 15-17 December
Natural Language Processing and Knowledge Mining Track
As shown in figure (7), we conclude the results of V+P+N
pattern sequence tags is the highest collocation value and most
frequently in the Quran corpus, followed by N+N+N pattern
sequence tags from the other tri-gram collocation. On the other
hand, the collocation pattern sequence tags such as P+V+N and
P+V+P as shown in the figure (7) have low frequency times
results, because don’t widely use pattern of Arabic language.
-
-
‫ٱن‬ ٌَ‫ى‬ ً ُ ‫يُمِي‬ ) (Verb+Noun)
( َ‫ة‬ ‫ى‬ َٰ َ‫ه‬ ‫ص‬
‫ق‬
( ِ ‫ٱّلل‬
‫ز‬ ْ ‫ر‬ ِّ ) (Noun+Proper-Noun)
As shown in figure (6), we conclude the results of N+N pattern
sequence tags is the highest collocation value and most
frequently in the Quran corpus, followed by V+N pattern
sequence tags from the other bi-gram collocation. On the other
hand, the collocation pattern sequence tags such as ADJ+ V
and V+ADJ as shown in the figure (6) have low frequency time
results, because don’t widely use pattern of Arabic language.
7000
6500
6000
5500
5000
4500
4000
3500
3000
2500
2000
1500
1000
500
0
5000
4500
4000
Tag-Sequences
3500
3000
2500
2000
1500
1000
500
0
Tag Sequence
Fig.7 Frequency of Tri-gram Collocation in the whole Quran Corpus
Bi-gram Experiment Results
Tri-gram Experiment Results
Examples of these sequences are:
- ( ‫ى‬ َ ‫خَمِي‬ ‫س‬ ْ ً ُ ‫ٱن‬ ْ ‫ط‬ َ ‫ر‬ َ َٰ ‫ص‬
‫)ٱن‬ (Noun+Adjective)
- ( ‫فَج‬ ‫س‬ ِ َُ ‫ل‬ ُ ‫بَا‬ ‫ج‬ ِ ‫ٱن‬ ) (Noun+Verb)
Fig.6 Frequency of Bi-gram Collocation in the whole Quran Corpus
 Experiment 2:Tri-gram Collocation
Table (3) shows the different pattern sequence of the part of
speech obtained from this experiment. We selected the
probability of the tag sequences of N (noun), V (verb), and P
(Proposition), which is 33= 27 pattern sequences. The results
of sequences show in figure (7), where the tri-gram having a
result of sequence less than 60 was ignored.
Table (3) Pattern Sequence of tags by the tri-gram
Sequence Pattern of Tags
NNN NNV NNP NVN NVV NVP NPN NPV NPP
VNN VNV VNP VVN VVV VVP VPN VPV VPP
PNN PNV PNP PVN PVV PVP PPN PPV PPP
Examples of these sequences are:
- ( ٍ ِ ‫ٱندِّي‬ ‫و‬ ِ ْ‫يَى‬ ‫ك‬ ِ ِ‫ه‬ ‫ي‬ َ َٰ ) (Noun+Noun+Noun)
- ( ٌ ِ ‫دَا‬ ‫ج‬ ُ ‫س‬ ْ َ‫ي‬ ‫ر‬ ُ ‫ج‬ َ ‫ش‬ َّ ‫ٱن‬ ‫و‬ َ ‫ى‬ ُ ْ‫ٱنَُّج‬ ‫و‬ َ )(Noun+Noun+Verb)
- ( َ‫ت‬ ً َ ‫ك‬ ْ ‫ح‬ ِ ‫ٱن‬ ْ ‫و‬ َ ‫ب‬ َ َ‫خ‬ َٰ ‫ك‬ ِ ‫ٱن‬ ْ ‫ى‬ ُ ُ‫ه‬ ً ُ ِّ‫ه‬ ‫ع‬ َ ُ‫ي‬ ‫و‬ َ )(Verb+Noun+Noun)
 Experiment 3: N-gram Phrases
Table (4) shows a different sequence of parts of speech
obtained from this experiment. We have selected N-gram
where n=4 grams (4 tags). The Pattern Sequence of tags chosen
is based on the Arabic noun phrase and verb phrase structures.
Table (4) Results of Extracting Phrase by n-gram sequences n=4
N-gram Sequence N-P-N-ADJ N-N-V-N V+P+N+ADJ
Number Phrases 88 369 114
Examples of these sequences are:
- ( ‫ى‬ ِ ‫ي‬ ‫ح‬ ِ ‫ر‬ َّ ‫ٱن‬ ٍ ِ ً َ َٰ ْ‫ٱنرَّح‬ ٍَ ‫ي‬ ِّ ‫م‬ ٌ ‫ي‬ ‫ُس‬
‫ح‬ ) ( Noun+P+Noun+ADJ)
‫ار‬
- (‫ا‬ ً َ ُ‫يَه‬ ‫د‬ ِ ‫ي‬ ْ َ‫أ‬ ‫ا‬ ۟ ‫ُى‬ ٓ ‫ع‬ ‫ط‬ َ ‫ل‬ ْ ‫فَٱ‬ ُ‫لَت‬ ‫ار‬
‫س‬
‫ٱن‬ ‫و‬ َ ‫ق‬
‫ٱنس‬ ‫و‬ َ ) (NounNounVerbNoun)
- ( ‫ج‬ َُِّ‫بَي‬ ‫ج‬ ٍۭ َ‫ي‬ َٰ ‫ا‬ ‫ء‬ َ ‫ك‬ َ ‫ي‬ ْ َ‫إِن‬ ٓ‫َُا‬ ‫َن‬ ْ ‫أََس‬ ) (Verb+P+Noun+ADJ)
The structure of the phrase in the Arabic language depends on
the type of phrases such as Noun Phrase (NP), Verb Phrase
(VP), Preposition Phrase (PP), etc. The Noun Phrase (NP) is a
phrase that has a noun in its head. NP can be composed of
either a proper noun (PN) followed by one or more noun. The
verb phrase can be located by checking the head verb of the
phrase. The following examples show the collocation terms
based NP and VP structure by n-gram where n=4 gram.

Noun phrase structure with example:
‫ي‬ َ َٰ ‫هُى‬ ‫ع‬ ْ ‫ي‬ َّ ‫ر‬ ٌ ُ‫ه‬ ‫ش‬ ْ َ‫أ‬ ُّ‫ج‬ ‫ح‬ َ ‫‖ٱن‬
- (N+N+ADJ) ― ‫ج‬
- (N+P+N+N) ― ‫ة‬ ِ ‫ر‬ َ ‫خ‬ ِ ‫ا‬ ‫ء‬ َ ْ‫ٱل‬ ‫و‬ َ َٰ‫ُونَى‬ ‫ٱْل‬ ْ ‫فِى‬ ‫د‬ ُ ً ْ ‫ح‬ َ ‫‖ٱن‬
Copyright© 2014 by Faculty of Computers and Information–Cairo University
NLP-14The 9th International Conference on Informatics and Systems (INFOS2014) – 15-17 December
Natural Language Processing and Knowledge Mining Track

Verb phrase structure:
- (V+N+ADJ): ‫ى‬ َ ‫خَمِي‬ ‫س‬ ْ ً ُ ْ‫ٱن‬ َ‫ط‬ ‫ر‬ َ َٰ ‫ص‬
‫ٱن‬ ‫ََا‬ ‫د‬ ِ ‫ه‬ ْ ‫ٱ‬
- (V+P+N+N): ‫ك‬ ِّ ‫ح‬ َ ‫بِٱن‬ ‫ب‬
‫ي‬ ْ َ‫ه‬ ‫ع‬ َ ‫ل‬ َ ‫س‬ َّ ََ
َ‫خ‬ ‫ك‬ ِ ‫ٱن‬ ‫ك‬
VII. CONCLUSIONS
In this paper, we have presented a prototype of extracting N-
gram collocation terms from a tagged Arabic Corpus (the Holly
Quran). The prototype enables the user to select the scope of
data source (whole Quran Corpus or specific SURA) and the
sequence of tags based on the type of gram (from 2 to 6 gram).
This approach extracts the n-gram collocation words by
matching the input structured pattern of Arabic language versus
the tags of words in the Quran corpus. These collocations will
be used for the automatic extend the Query and construction of
domain ontology. A set of experiments has been conducted in
different scenarios to find different possible pattern sequences
of the collocation terms in the Quran corpus.
REFERENCE
[1] K., AlKhateeb and A. Badarneh, "Automatic extraction of
Arabic multi-word terms," in Computer Science and
Information Technology (IMCSIT), Proceedings of the
2010 International Multiconference on, Wisla, 2010.
[2] I., Bounhas and Y., Slimani, "A hybrid approach for
Arabic multi-word term extraction," in Natural Language
Processing and Knowledge Engineering, 2009. NLP-KE
2009. International Conference on, Dalian, 2009.
[3] K. Dukes, E., Atwell, and AB., Sharaf, "Syntactic
Annotation Guidelines for the Quranic Arabic
Dependency Treebank," in Proceedings of the Seventh
International Conference on Language Resources and
Evaluation (LREC'10), Malta, 2010.
[4] M., Sawalha and E., Atwell, "Linguistically Informed and
Corpus Informed Morphological Analysis of Arabic," in
Proceedings of the 5th International Corpus Linguistics
Conference CL2009, Liverpool, 2009.
[5] W. Liu and W. Li, "To Determine the Weight in a
Weighted Sum Method for Domain-Specific Keyword
Extraction," in Computer Engineering and Technology,
2009. ICCET '09. International Conference on (Volume:1
), Singapore, 2009.
[6] Y.O. Elhadj, "Statistical Part-of-Speech Tagger for
Traditional Arabic Texts," Journal of Computer Science,
vol. 5, no. 11, pp. 794-800, 2009.
[7] Zaidi. S., Laskri M., and Abdelali, A., "Arabic
collocations extraction using Gate," in Machine and Web
Intelligence (ICMWI), 2010 International Conference on,
Algiers, 2010.
[8] Abdelali, A., Cowie J. and Soliman H., "Building a
modern standard Arabic corpus," in workshop on
computational modeling of lexical acquisition. The split
meeting. Croatia, 25th to 28th of july, Croatia, 2005.
[9] A. M. Saif and M. J. A. Aziz, "An Automatic Collocation
Extraction from Arabic Corpus," Journal of Computer
Science, vol. 7, no. 1, pp. 6 - 11, 2011.
[10] Xu, F. a. Kurz, D. a. Piskorski, Jakub, Schmeier and
Sven, "A Domain Adaptive Approach to Automatic
Acquisition of Domain Relevant Terms and their
Relations with Bootstrapping.," in LREC, Las Palmas,
2002.
[11] A. M. Abed, S. Tiun and M. Albared, "Arabic Term
Extraction Using Combined Approach On Islamic
Document," Journal of Theoretical & Applied
Information Technology, vol. 58, no. 3, pp. 601 - 608,
2013.
[12] B. S., B. Daille and D. Aboutajdine, "A multi-word term
extraction program for Arabic language," in Proceeding
of the 6th International Conference on Language
Resources and Evaluation, Morocco, 2008.
[13]CorpusLinguistics,http://en.wikipedia.org/wiki/Corpus_lin
guistics (Accessed : Jan. 2014)
[14]Informationextraction, http://en.wikipedia.org/wiki/Inform
ationextraction, (Accessed : June. 2014)
[15] http://www.ahewar.org/debat/show.art.asp?aid=109126
‫انمىاييس‬ ‫في‬ ‫ويعانجخها‬ ‫انعربيه‬ ‫انهغه‬ ‫في‬ ‫انًخالزياث‬ , ٌ‫انًخًد‬ ‫ر‬ ‫ا‬ ‫انحى‬ "
" ‫انهغه‬ ‫انثُائيه‬ (Accessed : June 2014)
[16] M., Diab, K. Hacioglu and D. Jurafsky,‖Automatic
tagging of Arabic text: From raw text to base phrase chunks‖.
Proceeding of the NAACLHLT, Boston, USA, pp: 149152,
2004
[17] R., Al-Shalabi, Kanaan G, ‖ Proper Noun Extraction
Algorithm for Arabic Language‖ international Conference on
IT to celebrate S. Charmonment’s 72nd Birthad. March 2009,
Thailand
[18] K., Al-Daimi, Abdel-Amir, M. "The Syntactic Analysis
of Arabic by Machine". Computers and Humanities, Vol. 28,
No. 1, pp. 29—37, 1994.
[19] A., Farghaly, and Shaalan K., "Arabic Natural Language
Processing: Challenges and Solutions", ACM Transactions on
Asian Language Information Processing (TALIP), vol. 8, no.
4, New York, NY, USA, ACM, pp. 1-22, 2009.
[20] S., Evert and K. Brigitte, Using small random samples for
the manual evaluation of statistical association measures.
Speech Language Spec. Issue Multiword Exp., 19: 450- 466.
DOI: 10.1016/j.csl.2005.02.005
[21] Al-Shargabi B., Olayah F., and Alromimah W., ‖An
Experimental
Study for the
Effect of Stop Words
Elimination for Arabic Text
Classification Algorithms‖,
International Journal of Information Technology and Web
Engineering 6(2), 68-75, April-June 2011.
Copyright© 2014 by Faculty of Computers and Information–Cairo University
NLP-15