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

net/publication/327753798

Challenges in Arabic Natural Language Processing

Chapter · November 2018

DOI: 10.1142/9789813229396_0003

CITATIONS READS

24 5,522

4 authors, including:

Khaled Shaalan Sanjeera Siddiqui

British University in Dubai British University in Dubai
365 PUBLICATIONS   10,214 CITATIONS    6 PUBLICATIONS   67 CITATIONS   

SEE PROFILE SEE PROFILE

Manar Alkhatib
British University in Dubai
27 PUBLICATIONS   216 CITATIONS   

SEE PROFILE

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

Two-class support vector machine with new kernel function based on paths of features for predicting chemical activity View project

Arabic natural langauge processing tools View project

All content following this page was uploaded by Khaled Shaalan on 26 October 2018.

The user has requested enhancement of the downloaded file.

 59

Chapter 3

Challenges in Arabic Natural Language Processing

Khaled Shaalan1, Sanjeera Siddiqui2, Manar Alkhatib3 and Azza Abdel Monem4
Faculty of Engineering & IT, The British University in Dubai123,
Block 11, Dubai International Academic City,
P.O. Box 345015, Dubai, UAE
School of Informatics, University of Edinburgh1, UK
Faculty of Computer and Information Sciences, Ain Shams University4,
Abbassia, 11566 Cairo, Egypt
[email protected], [email protected],
[email protected] and [email protected]

Natural Language Processing (NLP) has increased significance in

machine interpretation and different type of applications like discourse
combination and acknowledgment, limitation multilingual data
frameworks, and so forth. Arabic Named Entity Recognition,
Information Retrieval, Machine Translation and Sentiment Analysis are
a percentage of the Arabic apparatuses, which have indicated impressive
information in knowledge and security organizations. NLP assumes a
key part in the preparing stage in Sentiment Analysis, Information
Extraction and Retrieval, Automatic Summarization, Question
Answering, to name a few. Arabic is a Semitic language, which contrasts
from Indo-European lingos phonetically, morphologically, syntactically
and semantically. This paper discusses different challenges of NLP in
Arabic. In addition, it inspires scientists in this field and others to take
measures to handle Arabic dialect challenges.

1. Introduction

Natural language processing (NLP) is a domain of computer science that

aims at facilitating communication between machines (computers that
understand machine language or programming language) and human
60 Khaled Shaalan et al.

beings (who communicate and understand natural languages like English,

Arabic and Chinese etc.) NLP is very important as it makes a huge impact
on our daily lives. Many applications these days use concepts from NLP.
This paper discusses different challenges of NLP in Arabic.
Arabic is the sixth most spoken language in the world. Ref. 1 expressed
the significance of Arabic. Arabic connected with Islam and more than
200 million Muslims perform their petitions five times daily utilizing this
dialect. Moreover, Arabic is the first language of the Arab world countries,
which has significant importance worldwide. Arabic is an uncommonly
rich language which is related to another linguistic family, particularly
Semitic vernaculars, which is different from the Indo-European lingos
talked in the West. Arabic is interesting and any person with a slight
knowledge of Arabic can read and understand a text written fourteen
centuries ago.
Arabic, as a language or dialect is exceedingly derivational and
inflectional according to Refs. 2, 3 and 4, and there are no rules for
emphasis5,6,7,8. Truly, there are principles; however, there are no firm
rules1.
Arabic language has a rich and complex grammatical structure9,10. For
instance, a noun and its modifiers need to agree in number, gender, case,
and definiteness11. Moreover, in Arabic, there are advancements that really
mean, “Mother of” or “father of” to show ownership, a trademark, or a
property, and use gendered pronouns; it has no fair-minded pronouns12.
Arabic sentences can be nominal (subject–verb), or verbal (verb–
subject) with free order; however, English sentences are fundamentally in
the (subject–verb) order. The free order property of the Arabic language
presents a crucial challenge for some Arabic NLP applications13.
Three types characterize Arabic: Classical (Traditional or Quranic)
Arabic, Modern Standard Arabic and Dialect Arabic14,15. Arabic language
takes these forms in light of three key parameters including morphology,
syntax and lexical mixes1,16,17,18. Classical Arabic is primarily use in
Arabic speaking countries, as opposed to within the diaspora. Classical
Arabic is found in religious writings such as the Sunnah and Hadith, and
numerous historical documents19. Diacritic marks (also known as
“Tashkil” or short vowels) are commonly used within Classical Arabic as
phonetic guides to show the correct pronunciation. On the contrary,
 Challenges in Arabic Natural Language Processing 61

diacritics are considered optional in most other Arabic writing20. Modern

Standard Arabic (MSA) is used for TV, newspapers, poetry and in books.
Arabic Courses at the Arab Academy are also taught in the Modern
Standard form. The MSA can be transformed to adapt to new words that
need to be created because of science or technology. However, the written
Arabic script has seen no change in the alphabet, spelling or vocabulary in
at least four millenniums. Hardly any living language can claim such a
distinction.
Dialect Arabic or “colloquial Arabic” is casually utilized daily by
Arabs. It is found in various nations and districts of a nation19. It is grouped
into Mesopotamian Arabic, Arabian Peninsula Arabic, Syro Palestanian
Arabic, Egyptian and Maghrebi Arabic. Arabic dialect generally used,
mostly written, by Internet clients21 and social media19; varies from locale
to area is Dialect Arabic. In vernacular Arabic, portions of the words are
acquired from MSA22,23. Ref. 1 showed the significance of building local
devices to chip away both Modern Standard and Dialect Arabic. Ref. 22
presented a hybrid pre-processing approach that has the ability to convert
paraphrases of Egyptian dialectal input into MSA such that the available
NLP tools can be applied to the converted text. Ref. 24, as well as its
enhanced version25, worked on Sentiment Analysis on the data containing
different Arabic Dialects.
In this paper, illustrative examples are used for clarification. The
examples are given in MSA as it represents the bulk of written material
and formal TV shows, lectures, and papers. Besides, it is a universal
language that is understood by all Arabic speakers.

2. Challenges

The Arabic is an extremely bended tonguea, with unique sound, especially

when pronounce the letters “‫( ”ﺽ‬ḍād), “‫( ”ﻅ‬Tha'a), and “‫( ”ﻍ‬Ghain).
Arabic grammar has a rich morphology and intricate sentence structure
and grammarians have described it as the language of ḍād (“‫)”ﻟﻐﺔ ﺍﻟﻀﺎﺩ‬26,18.
Ref. 15 states that Arabic has a greatly rich morphology depicted by a mix
of templatic and affixational morphemes, complex morphological norms,

aThe top alveolar ridge is located on the roof of the mouth between the upper teeth and the
hard palate.
62 Khaled Shaalan et al.

and a rich part system. Arabic makes use of many inflections because of
the appendages, which incorporate relational words and pronouns. Arabic
morphology is perplexing because there are about 10,000 roots that are the
basis for nouns and verbs27. There are 120 patterns in Arabic morphology.
Ref. 28 highlighted the importance of 5000 roots for Arabic morphology.
The word order in Arabic is variant. We can have a free choice of the
word we want to emphasize and put it at the head of sentence. Generally,
the syntactic analyzer parses the input tokens produced by the lexical
analyzer and tries to identify the sentence structure using Arabic grammar
rules. The relatively free word order in an Arabic sentence causes syntactic
ambiguities which require investigating all the possible grammar rules as
well as the agreement between constituents13,24.
In this paper, we discuss the challenges of Arabic language with regard
to its characteristics and their related computational problems at
orthographic, morphological, and syntactic levels. In automating the
process of analyzing Arabic sentences, there is an overlap between these
levels, as they all help in making sense and meaning of words, and in
disambiguating the sentence.

2.1. Arabic orthography

Within the orthographic patterns of the written words, the shape of a letter
can be changed depending on whether it is connected with a former and
subsequent letter, or just connected with a former letter. For example, the
shapes of the letter “‫( ”ﻑ‬f), i.e. “‫”ﻓـ‬/“‫”ـﻔـ‬/“‫”ﻑ‬, changes depending on
whether it occurs in the beginning, middle, or end of a word, respectively.
Arabic orthography includes a set of orthographic symbols, called
diacritics, that carry the intended pronunciation of words. This helps
clarify the sense and meaning of the word.
As far as Qur’an is concerned, these vowel signs are absolutely
necessary in order for children and those who are not well versed in
classical Arabic language to pronounce religious text properly. It is worth
noting that written copies of the Qur’ān cannot be accredited by religious
institutes or authorities that review them unless the diacritics are included.
The absence of short vowels (e.g. inner diacritics) prompts diverse sorts
of equivocalness in Arabic writings (both basic and lexical) on the grounds
 Challenges in Arabic Natural Language Processing 63

that distinctive diacritics speak to distinctive implications9,20. These

ambiguities can be determined just by relevant data and a satisfactory
information of the language or dialect. Contextual features and an
adequate knowledge of the language can only resolve these ambiguities29.
Arabic orthography includes 28 letters20, all letters are consonants
except three long vowels: “‫( ”ﺃ‬alef), “‫( ”ﻭ‬Waw), and “‫( ”ﻱ‬yeh) and short
vowels are represented by diacritical signs. This specificity brings into
existence two forms of spelling: with or without vocalisation.
The vowels added through a consonantal skeleton by means of
diacritical marks produce a shallow orthography whereas vocalisation is
missing. Orthography is deep and the word behaves as homograph that is
semantically and phonologically ambiguous For instance, the unvoweled
word “‫( ”ﻛﺘﺐ‬ktb), supports several alternatives such as “‫َﺐ‬َ ‫( ” َﻛﺘ‬he wrote,
kataba), “‫ﺐ‬ ُ
َ ِ‫( ”ﻛﺘ‬it was written, kutiba), “‫ﺐ‬ ُ ُ
ُ ‫( ”ﻛﺘ‬books, kutubun), etc.
Voweled spelling is taught to novice readers, while unvoweled spelling
constitutes the standard form and is gradually imposed at later reading
literacy stages. Unfortunately, MSA is devoid of diacritical markings and
the restoration of these diacritics is an important task for other NLP
applications such as text to speech30.

2.1.1. Lack of consistency in orthography

Hamza Spelling
The most critical use of Hamza letter (“‫ ”ء“ )”ﺍﻟﻬﻤﺰﺓ‬brings in more
challenges. With the very significance of Hamza being an additional letter
seen at the top or bottom of the letters following the sounds of “‫”ﺍ‬, “‫”ﻭ‬, or
“‫”ﻯ‬, i.e. “‫”ﺃ‬, “‫”ﺅ‬, or “‫”ﺉ‬, respectively. As these rules are confusing even
for native speakers, Hamza is ignored most of the time while typing. NLP
based systems should handle this assumption.
There are many orthographical forms of the Hamza letter “the seat of
Al-Hamza”, which is decided by the diacritics (“Tashkeel”) of both the
“Hamza” itself as well as the letter preceding it, i.e. either “Fatha”,
“Dama”, “kasra” or “Sukun”. Exceptionally, when Hamza comes at the
beginning of the word, we always write it over an “Alef”, e.g. “‫( ”ﺃﻧﺎ‬I, 'ana),
or under it, e.g. “‫( ”ﺇﻳﻤﺎﻥ‬faith, 'iiman).
64 Khaled Shaalan et al.

According to its appearance and pronunciation, there are two types of

Hamza: “‫( ”ﻫﻤﺰﺓ ﻗﻄﻊ‬Hamza Al-Qata’) and “‫( ”ﻫﻤﺰﺓ ﻭﺻﻞ‬Hamza Al-Wasl).
Distinguishing each type is a challenge for both text and speech
processing. Hamza Al-Qata’ is the regular Hamza and is always written
and pronounced, e.g. “‫ ”ﺇﻳﻤﺎﻥ‬and “‫”ﺃﻧﺎ‬.
On the contrary, Hamza Al-Wasl is neither written nor pronounced
unless it is at the start of the utterance; a bare Aleh is used instead. A
simple rule to recognize Hamza Al-Wasl is to add “‫( ”ﻭ‬waw, and) before
it and see whether or not it is pronounced; hence, written.
For example, the Hamza in ‫( ﺍﻟﻜﺘﺎﺏ" "ﺇﻗﺮﺃ‬iq-ra' AL’Kitab, read the book),
is pronounced and written. However, if we add “‫( ”ﻭ‬waw, and) at the
beginning of sentence as in “‫( ”ﻭﺍﻗﺮﺃ ﺍﻟﻜﺘﺎﺏ‬waq-ra’ Al-Kitab, and read the
book) the Hamza is neither pronounced nor written. A more complicated
example is “‫( ”ﺃﺧﺬﺕ ﺍﺑﻨﻨﺎ‬a-khadh-tu ibnana, I grabbed our son). In the first
word “‫( ”ﺃﺧﺬﺕ‬grabbed), the Hamza is a glottal stop (pronounced strongly)
and should be pronounced, but in the second word “ ‫” ﺍﺑﻨﻨﺎ‬, it is neither
written nor pronounced.
When the diacritic mark of the “Hamza” is either “Fatha” or “Dama”,
the Hamza appears at the middle or the end of a word and is written over
the letter. Table 1 presents some examples with the addition of the Hamza
and the challenge it brings in causing orthographic confusion. The Hamza
is following a hierarchy of vowels in the language: The Kasra has the
highest priority, the Dama has the medium priority, and the Fatha has the
lowest priority.

Table 1. The Hamza diacritic is determined by its own diacritics and the preceding letter.

Example Tashkeel of Tashkeel of letter before Pronunciation Translation

Hamza Hamza
‫ﺳﺄَﻝ‬
َ Fatha Fatha Sa-ala Asked
‫ﺳﺌِﻞ‬ ُ Kasra Kasra So-ela Was asked
‫ﺳ َﺆﺍﻝ‬ ُ Fatha Dama So-aal Question

If the diacritic is Kasra/Fatha/Dama for either the Hamza itself or the

letter preceding it, the Hamza takes a Kasra/Fatha/Dama diacritic,
respectively. The rules for determining the diacritic of Hamza are of
notorious complexity.
 Challenges in Arabic Natural Language Processing 65

In transcribing to Arabic, it is difficult to determine the Hamza seat as

well as the short vowel that it follows. These types of the Hamza are of a
complex nature and need special handling by the computational system.
Al-Hamza orthographic variants are non-standard ways to spell a
specific variant of a name, like "‫ "ﺍﻻﻣﺎﺭﺍﺕ‬instead of "‫( "ﺍﻹﻣﺎﺭﺍﺕ‬Al-Emarat,
Emirates), in which the Hamza is omitted and bare Alef is used instead.
Though the difference between these variants cannot be strictly defined,
based on “statistical and linguistic analysis” of Modern Standard Arabic
orthography37, they are both occur frequently. For example, the capital of
The United Arab Emirates, "‫( "ﺃﺑﻮﻅﺒﻲ‬Abu Dhabi) can be written in
different ways. According to statistics from Google, the most frequent
ones are: "‫"ﺃﺑﻮﻅﺒﻲ‬, ,"‫ "ﺍﺑﻮﻅﺒﻲ‬and "‫ "ﺑﻮﻅﺒﻲ‬with 13,800,000, 9,400,000, and
1,400,000 occurrences, respectively.

Defective Verb Ambiguity

Defective (weak) verb (“‫ )”ﺍﻟﻔﻌﻞ ﺍﻟﻤﻌﺘﻞ‬is any verb that its root has a long
vowel as one of its three radicals. These long vowels will go through a
change when the verb is conjugated. For example, consider the case of a
negated present tense verb that is preceded by the apocopative particle
Lam—“‫”ﺣﺮﻑ ﺍﻟﺠﺰﻡ ﻟﻢ‬. In Arabic, this particle is used for negating a present
tense verb form which is understood as a negated past form18. It is one of
the defining features of Modern Standard Arabic, and is not used in any
dialects. Being able to use this word properly and effectively will bring
Arabic language to a higher level.
Table 2 presents examples of this verb forms. The negative past tense
verb causes ambiguity by having misspelling in writing skills31. When the
apocopative particle Lam precedes a past tense verb, the verb changes to
the present tense form by: 1) attaching a suitable present tense letter, 2) by
omitting the long vowel in the verb, and 3) adding a short vowel to the last
letter. Although apocopative particle Lam is used for the past tense, it can
never be used with the perfective verb itself; rather it is only used before
imperfective verbs.
66 Khaled Shaalan et al.

Table 2. Examples of negated past tense verb form.

Verb Transliteration Sentence Change applied to the present form of the verb
‫ﺩﻋﺎ‬ Da-aa ‫ﻉ‬
ُ ‫ﻟﻢ ﻳﺪ‬ Omit the last long vowel “ ‫ ”ﻭ‬and add the present
tense letter “ ‫“ﻱ‬
‫ﺳﻌﻰ‬ Sa-aa ‫ﻟﻢ ﻳﺴ َﻊ‬ Omit the last long vowel “ ‫ “ﻯ‬and add the present
tense letter “ ‫“ﻱ‬
‫ﺻﻠﻰ‬ Sala ‫ﻟﻢ ﻳﺼ ِﻞ‬ Omit the last long vowel “ ‫ “ﻱ‬and add the present
tense letter “ ‫“ﻱ‬
‫ﺯﺍﺭ‬ Zara ‫ﻳﺰﺭ‬
ْ ‫ﻟﻢ‬ Omit the middle long vowel "‫ "ﺍ‬and add the present
tense letter “ ‫“ﻱ‬

2.1.2. Nonappearance of capital letters

Arabic has no uncommon sign rendering the recognition of a Named
Entity (NE) all the more difficult32. On the other hand, English, in line with
numerous other Latin script-based dialects, has a particular marker in
orthography, in particular upper casing of the underlying letter, and
showing that a word or succession of words is a named substance. Arabic
does not have capital letters; this trademark speaks to an extensive
hindrance for the basic task of Named Entity Recognition in light of the
fact that in different languages, capital letters speak to a vital highlight in
distinguishing formal people, places or things19. Along these lines, the
issue of distinguishing appropriate names is especially troublesome for
Arabic. For instance, in English, capital letters are used, e.g. “Adam”, but
no capital letter in the same name in Arabic, e.g. “‫”ﺁﺩﻡ‬.
Another reality about Arabic to consider is that the vernacular has no
capital letters (e.g. for proper names: the names of people, countries,
months, days of the week); therefore, cannot make usage of acronyms.
This can lead to confusion, especially during Information Extraction in
general and Named Entity Recognition in particular. It makes it difficult
to see names of substances. For example, the NE “‫ ”ﺍﻻﻣﺎﺭﺍﺕ ﺍﻟﻌﺮﺑﻴﺔ ﺍﻟﻤﺘﺤﺪﺓ‬has
the acronyms UAE in English but not in Arabic. Therefore, it is common
to resolve the nonappearance of capital letters by analyzing the context
surrounding the Named Entity.
 Challenges in Arabic Natural Language Processing 67

2.1.3. Inherent ambiguity in named entities

Most Arabic proper nouns (NEs) are indistinguishable from forms that are
common nouns and adjectives (non-NEs) which might cause ambiguity.
For example, the noun “‫( ”ﺍﻟﺠﺰﻳﺮﺓ‬Aljazeera) can be recognized as an
organization name or a noun corresponding to island. Nevertheless, Arabic
names that are derived from adjectives are usually ambiguous, which
presents a crucial challenge for some Arabic NLP applications such as
Arabic Named Entity Recognition. As an example, consider the word “‫”ﺃﻣﻞ‬
(Amal), which means “hope”, and can be confused with the name of a
person. In the following two sentences, the word “Amal” means two
different senses:

1. “‫ ”ﺍﻟﺸﺒﺎﺏ ﻫﻢ ﺃﻣﻞ ﺍﻟﺒﻠﺪ‬which means: the youth is the hope of the

country.
2. “‫ ”ﺃﻣﻞ ﺑﻨﺖ ﺟﻤﻴﻠﺔ‬which means: Amal is a beautiful girl.

Remedies to resolve this type of ambiguity might not necessarily fix all
problems33,34. For example, consider the sentence “‫( ”ﺭﺃﻳﺖ ﺃﻣﻞ‬I saw
hope/Amal) which have either meaning.

2.1.4. Vowels
In written Arabic, there are two types of vowels: diacritical symbols and
long vowels. Arabic text is dominantly written without diacritics which
leads to major linguistic ambiguities in most cases as an Arabic word has
different meaning depending on how it is diactritized. A diacritic sign
(Tashkeel Or Harakat) is not an orthographic letter. It is formed as
diacritical marks above or below a consonant to give it a sound. Ref. 35
presented a good survey of recent works in the area of automatic
diacritization. There are three groups of diacritics32,36. The first group
consists of the short vowel diacritics such as Fatha ( َ◌), Dhamma ( ُ◌), and
Kasra (◌).
ِ The second group represents the doubled case ending diacritics
(Nunation or tanween) such as Tanween Fatha ( ً◌),Tanween Kasra (◌), ٍ
and Tanween Damma ( ٌ◌). These are vowels occurring at the end of
nominal words (nouns, adjectives and adverbs) indicating nominal
indefiniteness. The third group is composed of Shadda ( ّ◌) and Sukuun (◌) ْ
68 Khaled Shaalan et al.

diacritics. Shadda reflects the doubling of a consonant whereas Sukuun

indicates the absence of a vowel and reflects a glottal stop.
Diacritics could also be classified into two main groups based on their
functions. The first group includes the lexeme diacritics that determine the
Part of Speech (POS) of a word as in ‫َﺐ‬ َ ‫( َﻛﺘ‬wrote, kataba) and, ْ‫( ُﻛﺘُﺐ‬books,
kutub), and also the meaning of the word such as “‫ﺳﺔ‬ َ ‫( ” َﻣﺪَ َﺭ‬school,
madarasa) and “‫ﺳﺔ‬ ‫ﺭ‬
َ ِ َ ‫ﺪ‬ ‫ﻣ‬
ُ ” (teacher/female, almudarisa). The second category
represents the syntactic diacritics that reflect the syntactic function of the
word in the sentence. For example, in the sentence "َ‫( "ﺯَ ﺍ َ َﺭ ﺍ َ ْﻟ َﻮﻟَﺪُ ﺍﻟ َﺤ ِﺪﻳﻘَﺔ‬The
boy visited the garden, zar aalwalad alhadiqa), the syntactic diacritic
“Fatha” of the word " َ‫( "ﺍﻟ َﺤ ِﺪﻳﻘَﺔ‬the garden, alhadiqa) reflects its “object”
role in the sentence. While in sentence “ُ‫َﺖ ﺍ َ ْﻟ َﺤﺪَﻳﻘَﺔ‬ ْ ‫( ”ﺗ َﺰَ ﻳَﻨ‬Spruced up the
garden, tazayanat aalhadayqa) the same word occurs as a “subject” hence
its syntactic diacritic is a “Damma”. A text without diacritics adds layers
of confusion for novice readers and for automatic computation. For
example, the absence of diacritics is a serious obstacle to many of the
applications such as text to speech (TTS), intent detection, and automatic
understanding in general. Therefore, automatic diacritization is an
essential component for many Arabic NLP applications.
The long vowels in English, that is “a”,”e”, “i", “o” and “u”, are the
ones which are clearly spelled out in a text whereas in Arabic they are not.
There are no exact matches between English and Arabic vowels; they may
differ in quality, and they may behave differently under certain
circumstances. All letters of the Arabic alphabet are consonants except
three letters: “ ‫( ﺍ‬Alef), “‫( ”ﻭ‬Waw), and the letter “‫( ”ﻱ‬Ya'a) which are used
as long vowels or diphthongs, and they also play a role as weak
consonants5. The long vowel can appear at the beginning, in the middle,
or at the end of a word, and it has many forms of pronunciation. Table 3
presents a homographic issue with the aid of an example:
“‫ ﻭﻟﻜﻦ ﺃﻣﻪ ﻟﻢ ﺗﺴﺘﺴﻢ‬، ‫( ”ﻗﺎﻟﻮﺍ ﺇﻧﻪ ﻟﻢ ﻳﻌﺶ‬They said that he did not live, but his mother
did not give up, Qalo Anaho lam ya’esh lakin Amah lam tastaslim). The
acoustic and language models of the speech processing systems should
deal with long vowel issues.
 Challenges in Arabic Natural Language Processing 69

Table 3. Homographic issue for long vowels.

Word Transliteration Meaning Marks

‫ﻗﺎﻟﻮﺍ‬ Qalo Said-they No pronunciation for the letter Alef at the
end
‫ﻟﻜﻦ‬ Lakin But No appearance of the letter Alef in the
middle but pronounced

2.1.5. Lack of uniformity in writing styles

The high level of ambiguity of the Arabic script poses special challenges
to developers of NLP areas such as Morphological Analysis, Named Entity
Extraction and Machine Translation. These difficulties are exacerbated by
the lack of comprehensive lexical resources, such as proper noun
databases, and the multiplicity of ambiguous transcription schemes. The
process of automatically transcribing a non-Arabic script into Arabic, is
called Arabization. For example, transcribing an NE such as the city of
Washington into Arabic NE produces variants such as “ ، ”‫ﻭﺍﺷﻨﺠﻄﻦ‬
‫ “ﻭﺷﻨﻄﻦ‬، ”‫ “ﻭﺍﺷﻨﻐﻄﻦ‬، ”‫”“ﻭﺍﺷﻨﻄﻦ‬. Arabizing is very difficult for many
reasons; one is that Arabic has more speech sounds than Western European
languages, which can ambiguously or erroneously lead to an NE having
more variants. One solution is to retain all versions of the name variants
with a possibility of linking them together. Another solution is to
normalize each occurrence of the variant to a canonical form; this requires
a mechanism (such as string distance calculation) for name variant
matching between a name variant and its normalized form19.

2.2. Arabic morphology

An additional property of Arabic that should be noted is that Arabic is an
exceptionally morphological rich lingo. Its vocabulary can be easily
amplified using a framework that allows for a creative use of roots and
morphological samples4,17,28,37,38,39,40. According to Ref. 41, referred to in
Ref. 20, there are 85% of words from tri-demanding roots and there are
around 10,000 free roots. Hence, Arabic is highly derivational and
inflection results in high inflections in morphology28,31,37,42. Arabic is
known for its templatic morphology where words involve roots and
illustrations in the form of patterns, and fastened with affixes.
70 Khaled Shaalan et al.

2.2.1. Morphology is intricate

Arabic is a Semitic language that has a powerful morphology and a
flexible word order. It is difficult to put a border between a word and
sentence; yielding morpho-syntactic structure combinations for a word
along the dimensions of parts of speech, inflection, declension, clitics,
among other features13,43. Arabic morphology and sentence structure give
the ability to incorporate a broad number of adds to each word which
makes the combinatorial expansion of possible words.
Arabic is highly derivational. All the Arabic verbs are derived from a
base of three or four characters’ root verb. Essentially, every one of the
descriptors gets from a verb and every one of them are inferences too50.
Deductions in Arabic are quite often templatic; hence, we can say simply
that: Lemma = Root + Pattern. Additionally, in case of a general deduction
we can realize the significance of a lemma on the off chance that we know
the root and Lemma, which have been utilized to determine it44. Table 4
depicts examples of the composite relation “Lemma = Root + Pattern”,
demonstrating a case of producing two Arabic verbs from the same
classification and their inference/derivation from the same pattern. Notice
that the Arabic root is consonantal whereas the pattern is the vowel(s)
attached to a root.

Table 4. Illustration of Arabic Language in the derivational stage.

Lemma Transliteration = Root Transliteration + Pattern

‫ﻣﻔﺘﻮﺡ‬ Maftooh ‫ﻓﺘﺢ‬ Fath ‫ﻡ_؟ _ ؟ ﻭ_؟‬
‫ﻣﺪﺭﻭﺱ‬ Madroos ‫ﺩﺭﺱ‬ Daras

2.2.2. Morphology declension

Arabic is highly inflectional. The prefixes can be articles, relational words
or conjunctions, though the suffixes are by and large protests or
individual/possessive anaphora. As stated by Ref. 45, both prefixes and
suffixes are permitted to be mixes, and along these lines a word can have
zero or more affixes, i.e. Word = Prefix(es) + Lemma + Suffix(es). Arabic
verb morphology is central to the construction of an Arabic sentence
because of its richness of form and meaning. A more complicated example
would be words that could represent an entire sentence in English such as
 Challenges in Arabic Natural Language Processing 71

“‫( ”ﻭﺳﻴﺤﻀﺮﻭﻧﻬﺎ‬and they will bring it, wasayahdurunaha). This word can be
written in this form:

‫ ﻫﺎ‬+ ‫ ﻭﻧـ‬+ ‫ ﺣﻀﺮ‬+ ‫ ﻱ‬+ ‫ ﺱ‬+‫ﻭﺳﻴﺤﻀﺮﻭﻧﻬﺎ = ﻭ‬

(wa+sa+ya+hdr+runa+ha, and+will+bring+they +it )

In this example, the Lemma “‫( ”ﺣﻀﺮ‬hadr) accepts three prefixes: “‫( ”ﻭ‬wa),
“‫( ”ﺱ‬sa), and “‫( ”ﻱ‬ya) and two suffixes: “‫( ”ﻭﻥ‬wa noun), and “‫( ”ﻫﺎ‬ha).
Thereby, because of the complexity of the Arabic morphology, building
an Arabic NLP system is a challenging task.
The early step in analyzing an Arabic text is to identify the words in
the input sentence based on its type and properties, and outputs them as
tokens. There might be a problem in segmentation where some word
fragments that should be parts of the lemma of a word and were mistaken
to be part of the prefix or suffix of the word; thus, were separated from the
rest of the word as a result of tokenization. This problem arises with
Named Entities Recognition where the ending character n-grams of the
Named Entity were mistaken for objects or personal/possessive anaphora,
and were separated by tokenization19. Moreover, the POS tagger used for
the training and test data may have produced some incorrect tags,
incrementing the noise factor even further.
Another morphological challenge highlighted by Ref. 46, with regard
to relationships between words. The syntactic relationship that a word has
with alternate words in the sentence shows itself in its inflectional endings
and not in the spot in connection to alternate words in that sentence. For
example, “‫( ”ﺍﻟﻤﻌﻠﻢ ﺍﻟﻤﺨﻠﺺ ﻳﺤﺘﺮﻣﻪ ﻁﻼﺑﻪ‬Al Mu’alim al-mukhlis yahtarimaho
Tulabaho, the faithful teacher is respected by his students), the suffix
pronoun “‫( ”ـﻪ‬Heh) in the two words “‫( ”ﻳﺤﺘﺮﻣﻪ‬yahtarima-ho, respected-
him), and “‫( ”ﻁﻼﺑﻪ‬Tulaba-ho, students-his) refers to the word “‫( ”ﺍﻟﻤﻌﻠﻢ‬Al
Mu’alim, teacher-the).
Generally, Arabic computational morphology is challenging because
the morphological structure of Arabic also comprises a predominant
system of clitics. These are morphemes that are grammatically
independent, but morphologically dependent on another word or phrase47.
Subsequently, one can naturally conclude that this proportion is higher for
Arabic information than for different languages with less perplexing
72 Khaled Shaalan et al.

morphology that the same word can be joined to various appends and
clitics and thus, the vocabulary is much greater. The following Arabic
words: “‫”ﻣﻜﺘﻮﺏ‬, (Maktoob, Written) “‫”ﻛﺘﺎﺑﺎﺕ‬, (Kitabat, Writings), “‫”ﻛﺎﺗﺐ‬
(Katib, Writer) “‫( ”ﻛﺘﺎﺏ‬Kitab, Book), “‫( ”ﻛﺘﺐ‬Kutob, Books) , “‫”ﻣﻜﺘﺐ‬
(Maktab, Office) , “‫( ”ﻣﻜﺘﺒﺔ‬Maktabah, Library), “‫( ”ﻛﺘﺎﺑﻪ‬Kitabah, Writing)
are derived from the same Arabic three consonants trilateral with the origin
verb “‫( ”ﻛﺘﺐ‬Ktb, Wrote). They also refer to the same concept. To extract
the stem from the words, there are two types of stemming. The first type
is light stemming which is used to remove affixes (prefixes, infixes, and
suffixes) that belong to the letters of the word “‫( ”ﺳﺄﻟﺘﻤﻮﻧﻴﻬﺎ‬sa'altamuniha);
where they are formed by combinations of these letters. The second type
is called heavy stemming (i.e. root stemming) which is used to extract the
root of the words and includes implicitly light stemming48,49.

2.2.3. Annexation
Another morphologic challenge in Arabic language is that we can
compose a word to another by a conjunction of two words. This
conjunction can be with nouns, verbs, or particles. Although it is not
common in traditional Arabic language, it is used in Modern Standard
Arabic. Usually, the compound word is semantically transparent such that
the meaning of the compound word is compositional in the sense that the
meaning of the whole is equal to the meaning of parts put together50. For
example, the word “‫( ”ﺭﺃﺳﻤﺎﻟﻴﺔ‬capitalism, rasimalia) comes from compound
of two nouns “‫( ”ﺭﺃﺱ ﺍﻟﻤﺎﻝ‬capital, ras almal); the word “‫( ”ﻣﺎﺩﺍﻡ‬as long as,
madam) comes from the compound of a particle “‫( ”ﻣﺎ‬ma) and a verb “‫”ﺩﺍﻡ‬
(dam), and the word “‫( ”ﻛﻴﻔﻤﺎ‬however) comes from the compound of two
particles “‫( ”ﻛﻴﻒ‬kayf) and “‫( ”ﻣﺎ‬ma). The meaning of a compound word is
important for understanding the Arabic text, which is a challenge to POS
tagging and applications that require semantic processing51.

2.3. Syntax is intricate

Historically, as Islam spread, the Arab grammarians wanted to lay down
the basis of grammar rules that prevents the incorrect reading of the Holy
Qur’an. Arabic syntax is intricate. Automating the process that makes the
 Challenges in Arabic Natural Language Processing 73

computer analyze the Arabic sentences is truly a challenging problem from

the computer perspective34.
Arabic grammar distinguishes between two types of sentences: verbal
and nominal. Verbal sentences usually begin with a verb, and they have at
least a verb (“‫”ﻓﻌﻞ‬, faeal) and a subject (“‫”ﻓﺎﻋﻞ‬, faeil). The subject as well
as the object can be indicated by the conjugation of the verb, and not
written separately. For example, the conjugated verb “‫( ”ﺷﺎﻫﺪﺗﻚ‬I saw you,
saw-I-you, shahidtuk) has a subject and an object suffix pronouns attached
to it. Another example of a verbal sentence is “‫( ”ﻳﺪﺭﺱ ﺍﻟﻮﻟﺪ‬studying the
boy, “yadrus alwald”). This type of sentence is not applied in English
sentences. All the English sentences begin with a subject, and followed by
a verb, for example, “the boy is studying”.
In Arabic, a nominal sentence begins with a noun or a pronoun. The
nominal sentence has two parts: a subject or topic “‫”ﻣﺒﺘﺪﺃ‬, (mubtada) and a
predicate “‫”ﺧﺒﺮ‬, (khabar). The nominal sentences have two types: with or
without a verb. The nominal verbless sentence is a typical noun phrase.
When the nominal sentence is about being, which in some languages such
as English requires the presence of the linking verb ‘to be’ (i.e. copula) in
the sentence. This verb is not given in Arabic. Instead, it is implied and
understood from the context. For example, “‫( ”ﺍﻟﻄﻘﺲ ﺟﻤﻴﻞ‬alttaqs jamil) has
two nouns without a verb; its English translation is “The Weather [is]
wonderful”. This can be confusing to second language learners who speak
European languages and are used to have a verb in each sentence52,53.
Arabic grammar allows complex sentence structure formation which is
discussed in the following subsections.

2.3.1. Multi word expressions

Multi word expressions are very important constructs because their total
semantics usually cannot be determined by adding up the semantics of the
parts. For example, the multi words expression “‫( ”ﺑﺎﻟﺤﺪﻳﺪ ﻭﺍﻟﻨﺎﺭ‬by force,
bialhadid walnnar) consists of two words that have the literal meaning
“‫( ”ﺣﺪﻳﺪ‬iron, hadid) and “‫( ”ﻧﺎﺭ‬fire, nar). Another example is the medical
terminology “‫( ”ﻓﻘﺮ ﺍﻟﺪﻡ‬Anemia, faqar alddam) that consists of two words
that has the literal meaning “‫( ”ﻓﻘﺮ‬poor, faqar) and “‫( ”ﺩﻡ‬blood, dam). These
non-decomposable lexicalized phrases are syntactically-unalterable units
74 Khaled Shaalan et al.

that are unable to capture the effects of inflectional variation. Thus, they
can cause problems in Machine Translation, Information Retrieval, Text
Summarization, among other NLP applications. Such expression is termed
as idiomatic multi word expressions. Other multi words expressions are
words that co-occur together more often than not, but with transparent
compositional semantics such as “‫( ”ﺭﺋﻴﺲ ﺍﻟﺪﻭﻟﺔ‬The president of the
country, rayiys alddawla). As such, they do not pose a challenge in NLP
applications. Such expressions could be of interest if we categorize them
to types as in Named Entity Recognition, i.e. contextual cues.

2.3.2. Anaphora resolution

Anaphora Resolution is specifically concerned with matching up
particular entities or pronouns with the nouns or names that they refer to.
This is very important since without it a text would not be fully and
correctly understood, and without finding the proper antecedent, the
meaning and the role of the anaphor cannot be realized. Anaphora occurs
very frequently in written texts and spoken dialogues. Almost all NLP
applications such as Machine Translation, Information Extraction,
Automatic Summarization, Question Answering, etc., require successful
identification and resolution of anaphora54.
Anaphora Resolution is classically recognized as a very difficult
problem in NLP. It is one of the challenging tasks that is very time
consuming and requires a significant effort from the human annotator and
the NLP system in order to understand and resolve references to earlier or
later items in the discourse.

Ambiguous Anaphora
The pronominal anaphora is a very widely used type in Arabic language
as it has empty semantic structure and does not have an independent
meaning from their antecedent; the main subject. This pronoun could be a
third personal pronoun, called “‫( ”ﺿﻤﻴﺮ ﺍﻟﻐﺎﺋﺐ‬damir alghayib) in Arabic,
such as “‫ ”ﻫﺎ‬/hA/ (her/hers/it/its), “‫ ”ﻩ‬/h/ (him/his/it/its), “‫ ”ﻫﻢ‬/hm/
(masculine: them/their), and “‫ ”ﻫﻦ‬/hn/ (feminine: them/their).
 Challenges in Arabic Natural Language Processing 75

As an example that shows the challenges of pronominal anaphora to

NLP tasks, consider the result of using Google Translate© to translate two
Arabic sentences into English55:

‫ ﻓﺄﻋﻄﻴﺘﻬﺎ ﺍﻟﻄﻌﺎﻡ‬، ‫ﺭﺃﻳﺖ ﺍﻟﻘﻄﺔ‬

Transliteration: ra'ayt alquttah, fa'aetiatuha alttaeam

 Google translation: I saw the cat, so I gave her food
 Correct translation: I saw the cat, so I gave it food

‫ ﻓﺄﻋﻄﻴﺘﻬﺎ ﺍﻟﻄﻌﺎﻡ‬، ‫ﺭﺃﻳﺖ ﺍﻟﻄﻔﻠﺔ‬

Transliteration: ra'ayt alttaflah, fa'aetiatha alttaeam

 Google translation: I saw the little girl, so I gave her food
 correct translation: I saw the little girl, so I gave her food

The machine translation system fails to identify the correct antecedent

indicated by the third personal pronoun “‫ ”ﻫﺎ‬/hA/ (her/hers/it/its) and thus
external knowledge is needed in order to correctly identify this antecedent.
There are differences between Arabic and English pronominal systems and
Arabic is rich in morphology. The Arabic third person pronouns are
commonly encliticized which make them ambiguous. Arabic pronominal
does not differentiate linguistically between the value of the humanity
feature, i.e. ±human. As a result, both the -HUMAN FEMININE noun
“‫( ”ﺍﻟﻘﻄﺔ‬the cat) and the +HUMAN FEMININE noun “‫( ”ﺍﻟﻄﻔﻠﺔ‬the little
girl), causes ambiguity in the translated English sentence.

Hidden Anaphora
Another major kind of anaphora is hidden anaphora. It is restricted to the
subject position when there is no present noun or pronoun acting as the
subject. This is evident in the following sentence: “‫ ﻣﻌﻘﺪﺓ‬،‫”ﺍﻟﻤﻼﺣﻈﺔ ﻋﻠﻰ ﺍﻟﻠﻮﺡ‬
(The note on the board, complex) where the pronoun “‫ ”ﻫﻲ‬is not presented
in the sentence, i.e. “‫ ﻫﻲ ﻣﻌﻘﺪﺓ‬، ‫”ﺍﻟﻤﻼﺣﻈﺔ ﻋﻠﻲ ﺍﻟﻠﻮﺡ‬, which is called “zero
anaphora”. The human mind can determine the hidden Anaphora
(antecedent) but it causes grammatical mistakes in automated NLP
systems.
76 Khaled Shaalan et al.

2.3.3. Syntactically flexible text sequence

Syntactically-flexible expressions exhibit a much wider range of syntactic
variability and types of variations possible are in the form of Verb-Subject-
Object constructions13.
Arabic is generally a free word request language. While the essential
word order in Classical Arabic and Modern Standard Arabic is verb-
subject-object (VSO), they likewise permit subject-verb-object (SVO),
object-subject-verb(OSV) and object-verb-subject (OVS). It is basic to
utilize the SVO in daily papers features. Arabic vernaculars display the
SVO request. Word order disparity is depicted in Table 5. This makes the
sentence generation of Arabic NLP applications a challenge. For example,
in a question-answering system, the answer to the question “‫”ﺃﻳﻦ ﻛﺘﺎﺏ ﻫﺪﻯ؟‬
(where is Hoda’s book? 'ayn kitab hudaa?) could be any sentence that is
shown in Table 5 which indicates that Huda sold the book.

Table 5. Word order disparity.

Examples in Arabic Transliteration English Translation in English Order

‫ﺑﺎﻋﺖ ﻫﺪﻯ ﺍﻟﻜﺘﺎﺏ‬ sold Huda-NOM book-ACC Huda sold the book VSO
‫ﻫﺪﻯ ﺑﺎﻋﺖ ﺍﻟﻜﺘﺎﺏ‬ Huda-NOM sold book-ACC Huda, she sold the book. SVO
‫ﺍﻟﻜﺘﺎﺏ ﻫﺪﻯ ﺑﺎﻋﺘﻪ‬ DEF-book-NOM Huda-NOM sold-it The book, Huda sold it. OSV
‫ﺍﻟﻜﺘﺎﺏ ﺑﺎﻋﺘﻪ ﻫﺪﻯ‬ DEF-book-NOM sold-it Huda-NOM The book, Huda sold it. OVS

It is interesting to make a note of the placement of the word “‫”ﻛﺘﺎﺏ‬

(Book, kitab) in Table 5. VSO does not topicalize any constituent as old
data, which as a starting sentence in a talk cannot contain new components.
Confirms that VSO does not concentrate a specific constituent, as opposed
to different requests, which cannot be utilized in light of the fact that they
just center a specific constituent53.
Arabic case framework neglects to unmistakably stamp linguistic
contentions. This particularly happens when the case marker, which is
constantly included toward the end of the noun, cannot be incorporated on
the grounds that the noun closes with a long vowel as opposed to a
consonant. When this happens, elucidation of word request turns out to be
entirely VSO, contrary to VOS.
Additional proof originates from a study of syntactic structures in the
dialect, in which we find that VSO has the best dissemination. Bland
 Challenges in Arabic Natural Language Processing 77

inserted provisions, notwithstanding, may display both SVO and VSO

orders56.

2.3.4. Agreement
Agreement is a major syntactic principle that affects the analysis and
generation of an Arabic sentence which is very significant to difficult NLP
applications such as Machine Translation and Question Answering13,47.
Agreement in Arabic is full or partial and is sensitive to word order
effects1. An adjective in Arabic usually follows the noun it modifies
“‫( ”ﺍﻟﻤﻮﺻﻮﻑ‬almawsuf) and fully agrees with respect to number, gender,
case, and definiteness, e.g. “‫( ”ﺍﻟﻮﻟﺪ ﺍﻟﻤﺠﺘﻬﺪ‬The diligent boy, alwald
almujtahad) and “‫( ”ﺍﻷﻭﻻﺩ ﺍﻟﻤﺠﺘﻬﺪﻭﻥ‬The diligent boys, al'awlad
almujtahidin). The verb is marked for agreement depending on the word
order of the subject relative to the verb, see Figure 1.

Fig. 1. Agreement patterns in verb-subject vs. subject-verb word order.

The verb in Verb-Subject-Object order agrees with the subject in

gender, e.g. “‫ ﺍﻷﻭﻻﺩ‬/ ‫( ”ﺟﺎء ﺍﻟﻮﻟﺪ‬came the-boy/the-boys, ja' alwalad/ ja'
78 Khaled Shaalan et al.

al'awlad) versus “‫ ﺍﻟﺒﻨﺎﺕ‬/ ‫( ”ﺟﺎءﺕ ﺍﻟﺒﻨﺖ‬came the-girl/the-girls, ja'at albint/

ja'at albanat). In Subject-Verb-Object (SVO) order, the verb agrees with
the subject with respect to number and gender, e.g. “‫ ﺍﻷﻭﻻﺩ ﺟﺎءﻭﺍ‬/ ‫”ﺍﻟﻮﻟﺪ ﺟﺎء‬
(came the-boy/the-boys) versus "‫ ﺍﻟﺒﻨﺎﺕ ﺟﺌﻦ‬/ ‫( "ﺍﻟﺒﻨﺖ ﺟﺎءﺕ‬came the-girl/the-
girls). In Aux-subject-verb word order, the auxiliary agrees only in gender
while the main verb agrees in both gender and number, e.g. “ ‫ﻛﺎﻧﺖ ﺍﻟﺒﻨﺖ ﺗﺄﻛﻞ‬
‫ ﻛﺎﻧﺖ ﺍﻟﺒﻨﺎﺕ ﺗﺄﻛﻠﻦ ﺍﻟﻄﻌﺎﻡ‬/ ‫( ”ﺍﻟﻄﻌﺎﻡ‬the-girl was/the-girls were eating the food,
kanat albint takul alttaeam/ kanat albanat takulun alttaeam). If the subject
precedes the auxiliary, then both verbs agree with it in both gender and
number “‫ ﺍﻟﺒﻨﺎﺕ ﻛﻦ ﻳﺄﻛﻠﻦ ﺍﻟﻄﻌﺎﻡ‬/ ‫( ”ﺍﻟﺒﻨﺖ ﻛﺎﻧﺖ ﺗﺄﻛﻞ ﺍﻟﻄﻌﺎﻡ‬albint kanat takul
alttaeam / albanat kunn yakuln alttaeam).
Some other agreements also exist between the numbers and the
countable nouns57. Number–counted noun agreement is governed by a set
of complex rules for determining the literal number that agree with the
counted noun with respect to gender and definiteness. In Arabic, the literal
generation of numbers is classified into the following categories: digits,
compounds, decades, and conjunctions. The case markings depend on the
number–counted name expression within the sentence. In the following
example, the number, “‫( ”ﺧﻤﺲ‬five [masc.sg]) and the (broken plural)
counted noun “‫( ”ﻣﺘﺎﺣﻒ‬museums [fem.pl]) need to agree in gender and
definiteness:

‫ﺍﻷﻭﻻﺩ ﺯﺍﺭﻭﺍ ﺧﻤﺴﺔ ﻣﺘﺎﺣﻒ‬

al'awlad zaruu khmst matahif

the-boys visited-they five.fem.sg museum.fem.pl
The boys visited five museums

3. Conclusion

Arabic as a language is both challenging and interesting. In this paper, we

delved into the basics of word and sentence structure, and relationships
among sentence elements. This should help readers appreciate the
complexity associated with Arabic NLP. The challenges of Arabic
language were depicted by giving examples under MSA. It was found that
although Arabic is a phonetic language in the sense that there is one-to-
 Challenges in Arabic Natural Language Processing 79

one mapping between the letters in the language and the sounds with
which they are associated. An Arabic word does not dedicate letters to
represent short vowels. It requires changes in the letter form depending on
its place in the word, and there is no notion of capitalization. As for MSA
texts, short vowels are optional which makes it even more difficult for non-
native speakers of Arabic to learn the language and present challenges to
analyze Arabic words. Morphologically, the word structure is both rich
and compact such that it can represent a phrase or a complete sentence.
Syntactically, the Arabic sentence is long with complex syntax. Arabic
Anaphora has increased the ambiguity of the language, as in some cases
the Machine Translation system fails to identify the correct antecedent
because of the ambiguity of the antecedent. External knowledge is needed
to correct the antecedent. Moreover, Arabic sentence constituents (free
word order) can be swapped without affecting structure or meaning, which
adds more syntactic and semantic ambiguity, and requires analysis that is
more complex. Nevertheless, agreement in Arabic is full or partial and is
sensitive to word order effects.
Arabic language differs from other languages because of its complex
and ambiguous structure that the computational system has to deal with at
each linguistic level.

References

1. Farghaly and K. Shaalan, Arabic Natural Language Processing: Challenges and

Solutions, ACM Transactions on Asian Language Information Processing (TALIP),
the Association for Computing Machinery, ACM, 8(4):1/22 (2009).
2. R. Al-Shalabi and R. Obeidat, Improving KNN Arabic Text Classification with N-
grams based Document Indexing, In Proceedings of the Sixth International Conference
on Informatics and Systems, Cairo, Egypt, pp. 108/112 (2008).
3. L., Abd El Salam, M. Hajjar and K. Zreik, Classification of Arabic Information
Extraction methods, MEDAR 2009, 2nd International Conference on Arabic Language
Resources and Tools, pp. 71/77, Cairo, Egypt (2009).
4. N. Farra, E. Challita, R. A. Assi and H. Hajj, Sentence-level and document-level
sentiment mining for Arabic texts, In Data Mining Workshops (ICDMW), 2010 IEEE
International Conference on, IEEE, pp. 1114/1119 (2010).
5. K., Dave, S., Lawrence and D. M. Pennock, Mining the peanut gallery: Opinion
extraction and semantic classification of product reviews, In Proceedings of the 12th
international conference on World Wide Web (pp. 519-528), ACM (2003).
80 Khaled Shaalan et al.

6. S. Ghosh, S. Roy and S. Bandyopadhyay, A tutorial review on Text Mining

Algorithms, International Journal of Advanced Research in Computer and
Communication Engineering, 1(4), (2012).
7. F. Harrag, E. El-Qawasmeh and P. Pichappan, Improving Arabic text categorization
using decision trees, In First International Conference on Networked Digital
Technologies, NDT'09, IEEE, pp. 110/115 (2009).
8. J. Wiebe and E. Riloff, Creating subjective and objective sentence classifiers from
unannotated texts, In Computational Linguistics and Intelligent Text Processing, pp.
486-497, Springer, Berlin Heidelberg (2005).
9. N. Y. Habash, Introduction to Arabic natural language processing, Synthesis Lectures
on Human Language Technologies, 3(1), 1/187 (2010).
10. N. Habash, Syntactic preprocessing for statistical machine translation, In Proceedings
of the 11th MT Summit XI, pp. 215/222 (2007).
11. K. Shaalan, M. Magdy and A. Fahmy, Analysis and Feedback of Erroneous Arabic
Verbs. Journal of Natural Language Engineering (JNLE), Cambridge University Press,
UK, 21(2):271/323 (2015).
12. S. Izwaini, Problems of Arabic machine translation: evaluation of three systems. The
British Computer Society (BSC), London, pp. 118/148 (2006).
13. S. Ray and K. Shaalan, A Review and Future Perspectives of Arabic Question
Answering Systems, IEEE Transactions on Knowledge and Data Engineering,
28(12):3169-3190, IEEE, (2016). DOI: 10.1109/TKDE.2016.2607201.
14. M. Korayem, D. Crandall, and M. Abdul-Mageed, Subjectivity and sentiment analysis
of Arabic: A survey, In Advanced Machine Learning Technologies and Applications,
Springer Berlin Heidelberg, pp. 128/139 (2012).
15. N. Habash and O. Rambow, MAGEAD: a morphological analyzer and generator for
the Arabic dialects, In Proceedings of the 21st International Conference on
Computational Linguistics and the 44th annual meeting of the Association for
Computational Linguistics, Association for Computational Linguistics (ACL), pp.
681/688 (2006).
16. Elgibali, Investigating Arabic: current parameters in analysis and learning, Studies in
Semitic Languages and Linguistics series, Vol. 42, Brill. (2005).
17. Abdel Monem, K. Shaalan, A. Rafea, and H. Baraka, Generating Arabic Text in
Multilingual Speech-to-Speech Machine Translation Framework, Machine
Translation, Springer, Netherlands, 22(4): 205/258 (2008).
18. K. Ryding, A Reference Grammar of Modern Standard Arabic. Cambridge University
Press, New York (2005).
19. K. Shaalan, A Survey of Arabic Named Entity Recognition and Classification,
Computational Linguistics, MIT Press, USA, 40(2):469/510 (2014).
20. P. Daniels, The Oxford Handbook of Arabic Linguistics, The Arabic writing system,
Ed. J. Owens (2013).DOI:10.1093/oxfordhb/9780199764136.013.0018.
21. M. N. Al-Kabi, I. M. Alsmadi, A. H. Gigieh, H. A. Wahsheh and M. M. Haidar,
Opinion Mining and Analysis for Arabic Language, IJACSA) International Journal of
Advanced Computer Science and Applications, 5(5), 181/195 (2014).
22. H. Abo Bakr, K. Shaalan and I. Ziedan, and I., A Hybrid Approach for Converting
Written Egyptian Colloquial Dialect into Diacritized Arabic, In the Proceedings of The
 Challenges in Arabic Natural Language Processing 81

6th International Conference on Informatics and Systems, INFOS2008, the special

track on Natural Language Processing, 27-29 March, Cairo, Egypt (2008).
23. E. Refaee and V. Rieser, An Arabic twitter corpus for subjectivity and sentiment
analysis, In Proceedings of the Ninth International Conference on Language Resources
and Evaluation (LREC’14), Reykjavik, Iceland, European Language Resources
Association (ELRA) (2014).
24. S. Siddiqui, A. Abdel Monem and K. Shaalan, Sentiment Analysis in rabic, Natural
Language to Information Systems: 21st International Conference on Applications of
Natural Language to Information Systems (NLDB 2016), Eds. E. Métais, F. Meziane,
F. Saraee, V. Sugumaran, and S. Vadera, Lecture Notes in Computer Science (LNCS
9612), Chapter 41, pp. 409/414, Springer, Berlin, Heidelberg (2016).
25. S. Siddiqui, A. Abdel Monem and K. Shaalan, Towards Improving Sentiment Analysis
in Arabic, In Proceedings of the International Conference on Advanced Intelligent
Systems and Informatics 2016, Volume 533 of the Series Advances in Intelligent
Systems and Computing, Eds. A E. Hassanine, K. Shaalan, T. Gaber, A. Ahmad, F.
Tolba, pp. 114/123, Springer (2017).
26. Al-Sughaiyer and I. Al-Kharashi, Arabic Morphological Analysis Techniques: A
Comprehensive Survey, Journal of the American Society for Information Science and
Technology, 55(3):189–213 (2004).
27. Darwish, Building a Shallow Arabic Morphological Analyzer in One Day. In
Proceedings of the ACL Workshop on Computational Approaches to Semitic
Languages, Philadelphia, PA, pp. 1–8 (2002).
28. R. Beesley, Finite-state morphological analysis and generation of Arabic at Xerox
Research: Status and plans in 2001, In ACL Workshop on Arabic Language
Processing: Status and Perspective, Vol. 1, pp. 1/8 (2001).
29. R. Ibrahim, A. Khateb and H. Taha, How Does Type of Orthography Affect Reading
in Arabic and Hebrew as First and Second Languages? Open Journal of Modern
Linguistics, 3(1):40/46 (2013).
30. Said, M. El-Sharqwi, A. Chalabi, and E. Kamal, A Hybrid Approach for Arabic
Diacritization, Natural Language Processing and Information Systems: 18th
International Conference on Applications of Natural Language to Information
Systems, NLDB 2013, pp. 53-64, Salford, UK, 2013, Springer, Berlin, Heidelberg,
June 19-21 (2013).
31. Attia, P. Pecina, Y. Samih, K. Shaalan and J. Van Genabith, Arabic Spelling Error
Detection and Correction. Journal of Natural Language Engineering (JNLE),
Cambridge University Press, UK, 22(5):751/773 (2016).
32. Oudah and K. Shaalan, NERA 2.0: Improving coverage and performance of rule-based
named entity recognition for Arabic, Journal of Natural Language Engineering (JNLE),
23(3):441/472, Cambridge University Press, UK (2017). DOI:
10.1017/S1351324916000097.
33. Oudah and K., Shaalan, Studying the impact of language-independent and language-
specific features on hybrid Arabic Person name recognition, Language Resources &
Evaluation, 51(2):351/378, Springer (2017). DOI:10.1007/s10579-016-9376-1.
34. E. Othman, K. Shaalan and A. Rafea, Towards Resolving Ambiguity in Understanding
Arabic Sentence, In the Proceedings of the International Conference on Arabic
82 Khaled Shaalan et al.

Language Resources and Tools, NEMLAR, 22nd–23rd Sept., Egypt, pp. 118/122
(2004).
35. Azmi and R. Almajed, A survey of automatic Arabic diacritization techniques, Natural
Language Engineering, Cambridge University Press, UK, 21(3):477/495 (2015).
36. S. Abu-Rabia, The Role of Vowels in Reading Semitic Scripts: Data from Arabic and
Hebrew, Reading and Writing: An Interdisciplinary Journal, 14, 39/59 (2001). DOI:
10.1023/A:1008147606320.
37. Farghaly, Three Level Morphology for Arabic, presented at the Arabic Morphology
Workshop, Linguistics Summer Institute, Stanford, CA, (1987).
38. T. McCarthy, The critical theory of Jurgen Habermas, Studies in Soviet Thought,
Springer, Berlin Heidelberg, 23(1):77/79 (1982).
39. Soudi, G. Neumann and A. Bosch, Arabic computational morphology: knowledge-
based and empirical methods, vol. 38, Springer, Dordrecht (2007).
40. Shoukry and A. Rafea, Sentence-level Arabic sentiment analysis, 2012 International
Conference on Collaboration Technologies and Systems (CTS), Denver, CO, USA,
2012, pp. 546/550 (2012). DOI: 10.1109/CTS.2012.6261103.
41. S. S. Al-Fedaghi and F. Al-Anzi., A New Algorithm to Generate Arabic Root-Pattern
forms, In Proceedings of the 11th national Computer Conference and Exhibition, pp.
391/400 (1989).
42. N. De Roeck and W. Al-Fares, A morphologically sensitive clustering algorithm for
identifying Arabic roots, In Proceedings of the 38th Annual Meeting on Association
for Computational Linguistics, Association for Computational Linguistics, pp. 199/206
(2000).
43. S. Mesfar, Towards a cascade of morpho-syntactic tools for Arabic natural language
processing, In Computational Linguistics and Intelligent Text Processing, Springer
Berlin Heidelberg, pp. 150/162 (2010).
44. Y., Benajiba, M. Diab and P. Rosso, Arabic named entity recognition using optimized
feature sets, In Proceedings of the Conference on Empirical Methods in Natural
Language Processing, Association for Computational Linguistics, pp. 284/293 (2008).
45. Y. Benajiba, P. Rosso and M. J. Bened, ANERsys: An Arabic Named Entity
Recognition system based on Maximum Entropy, In Proc. of CICLing-2007, Springer-
Verlag, LNCS series (4394), pp. 143/153 (2007).
46. K. Thakur, Genitive Construction in Hindi. M. Phil Thesis, University of Delhi, India
(1997).
47. K. Shaalan, Arabic GramCheck: A Grammar Checker for Arabic, Software Practice
and Experience, John Wiley & sons Ltd., UK, 35(7):643-665 (2005).
48. M. N. Al-Kabi, S. Kazakzeh, B. Abu Atab, S. Al-Rababah and S. Alsmadi, A Novel
Root based Arabic Stemmer, Journal of King Saud University, Computer and
Information Sciences, 27(2):94–103 (2015). DOI: 10.1016/j.jksuci.2014.04.001
49. H. K. AlAmeed, S. O. AlKitbi, A. A. AlKaabi, K. S. AlShebli, N. F. AlShamsi, N. H.
AlNuaimi, and S. S. AlMuhairi, Arabic Light Stemmer: A new enhanced approach, In
Proceedings of the Second International Conference on Innovations in Information
Technology (IIT'05), Dubai, UAE (2005).
50. W. M. Amer. (2010). Compounding in English and Arabic: A contrastive study,
Technical Report, available online at:
 Challenges in Arabic Natural Language Processing 83

http://site.iugaza.edu.ps/wamer/files/2010/02/Compounding-in-English-and-
Arabic.pdf
51. S. Elkateb, W. Black, P. Vossen, D. Farwell, H. Rodríguez, A. Pease and M. Alkhalifa,
Arabic WordNet and the challenges of Arabic, In Proceedings of Arabic NLP/MT
Conference, London, UK (2006).
52. K. Shaalan, An Intelligent Computer Assisted Language Learning System for Arabic
Learners. Computer Assisted Language Learning: An International Journal, Taylor &
Francis Group Ltd., 18(1 & 2):81/108 (2005).
53. Hammo, A. Moubaiddin, N. Obeid, and A. Tuffaha, Formal Description of Arabic
Syntactic Structure in the Framework of the Government and Binding Theory,
Computacion y Sistemas, 18(3):611/625 (2014).
54. S. Hammami, L. Belguith and A. Hamadou, Arabic Anaphora Resolution: Corpora
Annotation with Co-referential Links, The International Arab Journal of Information
Technology, 6(5):481/489 (2009).
55. R. Al-Sabbagh and K. Elghamry, Arabic Anaphora Resolution: A Distributional,
Monolingual and Bilingual Approach, Faculty of Al-Alsun, Ain Shams University,
Cairo, Egypt (2002).
56. S. Usama, On issues of Arabic syntax: An essay in syntactic argumentation, Brill’s
Annual of Afroasiatic Languages and Linguistics, pp. 236/280 (2011).
57. M. Shquier and T. Sembok, Word agreement and ordering in English-Arabic machine
translation, 2008 International Symposium on Information Technology, IEEE Explore,
Kuala Lumpur, pp. 1/10 (2008). DOI: 10.1109/ITSIM.2008.4631625.

View publication stats