Arabic POS Tagset: Morphology and MSA Annotation

Arabic presents a particularly rich environment for part-of-speech tagging (POS tagging). Its morphological structure, productive use of prefixes and suffixes, grammatical gender, number, definiteness, case and verbal inflection mean that identifying a word simply as a noun, verb or particle is often only the beginning of linguistic analysis.

At the broadest grammatical level, Arabic words can be organized into three fundamental categories:

• ism (اسم) — nominal

• fiʿl (فعل) — verb

• ḥarf (حرف) — particle

These categories provide an important foundation for describing Arabic grammatical structure. In contemporary Natural Language Processing (NLP), however, they can be expanded into more detailed computational tagsets capable of representing the linguistic distinctions required for corpus annotation, morphological analysis and automatic language processing.

For Modern Standard Arabic (MSA), effective POS annotation requires particular attention to morphology. A written Arabic word may contain several meaningful components, making the relationship between an orthographic word and a computational token considerably more complex than simple whitespace segmentation suggests.

Why Arabic POS Tagging Is Different

Arabic words can combine lexical stems with conjunctions, prepositions, definite articles and pronominal clitics.

Consider:

وبالكتاب — wa-bi-l-kitāb

This written form can contain several components:

و + ب + ال + كتاب

wa + bi + al + kitāb

roughly:

and + with/by + the + book

Although these elements appear together orthographically, they perform different grammatical functions.

An Arabic NLP system therefore needs to distinguish between the orthographic word and its morphological segments. Depending on the annotation framework and NLP task, segmentation may occur before or as part of POS tagging.

This is why Arabic tokenization cannot always be reduced to splitting text at spaces.

The Arabic Nominal Category: Ism — اسم

The category ism (اسم) is broader than the English grammatical category traditionally called “noun.” It encompasses several nominal and noun-related classes that computational tagsets may distinguish separately.

A practical MSA-oriented tagset may represent categories such as:

Category	Tag	Arabic	Description
Noun	N / NOUN	اسم	Common noun
Proper noun	PN / PROPN	اسم علم	Proper name
Adjective	ADJ	صفة	Adjective
Pronoun	PRON	ضمير	Personal or other pronoun
Demonstrative	DEM	اسم إشارة	Demonstrative
Relative pronoun	REL	اسم موصول	Relative expression
Adverb	ADV	ظرف	Adverbial expression
Time expression	T	ظرف زمان	Temporal adverbial
Location expression	LOC	ظرف مكان	Locative adverbial

The precise labels depend on the annotation scheme being used. A project following Universal Dependencies, for example, uses universal categories such as NOUN, PROPN, ADJ, PRON and ADV, while more Arabic-specific annotation systems may preserve finer distinctions.

The objective is not necessarily to force Arabic grammar into an English-oriented classification system. A useful Arabic tagset should capture linguistically meaningful distinctions while remaining suitable for computational processing.

Proper Nouns and Arabic NLP

The distinction between common and proper nouns presents an interesting computational problem.

English orthography provides capitalization:

London versus city

Arabic has no equivalent capitalization system:

لندن
مدينة

Consequently, an NLP system cannot use an initial capital letter as evidence that an Arabic token represents a proper noun.

Recognition must instead depend on combinations of lexical knowledge, morphological information, syntactic context and contextual probability.

This becomes particularly important in Named Entity Recognition (NER), where systems attempt to identify people, organizations, geographical locations and other named entities.

POS tagging and morphological analysis can therefore provide useful features for more advanced semantic tasks.

Pronouns and Attached Clitics

Arabic pronouns demonstrate particularly clearly why morphological segmentation matters.

Personal pronouns can occur independently:

أنا — anā — I

هو — huwa — he

هي — hiya — she

But Arabic also makes extensive use of attached pronouns.

Consider:

كتابه — kitābuhu

This can be segmented as:

كتاب + ه

kitāb + hu

book + his

The pronominal suffix carries grammatical information that an NLP system may need to identify independently from the noun.

Similarly:

رأيته — raʾaytuhu

contains information corresponding approximately to:

رأيت + ه

I saw + him

A single orthographic form therefore contains lexical, verbal and pronominal information.

Accurate Arabic annotation must be capable of representing this internal structure.

Adjectives — صفة

Arabic adjectives participate in a rich agreement system.

An adjective associated with a noun can reflect grammatical features including gender, number, definiteness and case, depending on the construction and annotation level.

Consider:

كتاب جديد

a new book

and:

الكتاب الجديد

the new book

The relationship between the noun and adjective is expressed through more than simple word order.

For computational processing, identifying ADJ is therefore only one layer of analysis. Morphological features can provide additional information about the grammatical relationship between the adjective and the noun it modifies.

This illustrates an important principle of Arabic corpus annotation:

POS + morphological features provide substantially more information than POS alone.

Adverbs of Time and Place

Arabic grammatical description gives particular importance to expressions associated with ẓarf (ظرف).

Two useful distinctions are:

ظرف زمان — adverbial of time

and

ظرف مكان — adverbial of place

Examples of temporal expressions include:

اليوم — today

غداً — tomorrow

while spatial expressions may include words such as:

فوق — above

تحت — below

These distinctions have practical computational applications.

Temporal expressions are important for event extraction, information retrieval, question answering and timeline construction, while locative expressions can contribute to spatial analysis and information extraction.

Depending on the computational annotation scheme, these expressions may be represented through categories such as ADV, NOUN or ADP, with their syntactic function captured separately.

Verbs: Fiʿl — فعل

The second fundamental Arabic category is:

fiʿl (فعل) — verb

A computational tag such as V or VERB identifies the general category, but Arabic verbs contain considerably more information.

Arabic verbal morphology can encode:

• person

• gender

• number

• aspect

• voice

• mood

Consider:

كتب — kataba — he wrote

كتبت — katabat / katabtu — she wrote / I wrote

يكتب — yaktubu — he writes/is writing

نكتب — naktubu — we write

These forms demonstrate how changes in morphology encode grammatical information that English frequently expresses through separate pronouns or auxiliary constructions.

For Arabic NLP, identifying a token simply as VERB therefore leaves much of its linguistic information unexplained.

Verbal Mood

The Arabic imperfect verb can also interact with grammatical mood.

Depending on the syntactic environment, computational annotation may distinguish categories corresponding to the:

indicative,
subjunctive, and
jussive.

Particles and surrounding grammatical structures can affect these forms.

This demonstrates why POS tagging, morphological analysis and syntax are closely connected in Arabic. Identifying a preceding particle can contribute to interpreting the grammatical properties of the verb that follows it.

Particles: Ḥarf — حرف

The third fundamental category is:

ḥarf (حرف) — particle

Particles form an especially important area for Arabic computational analysis because relatively short forms can encode significant grammatical relationships.

Computational systems may distinguish categories such as:

Category	Possible Tag
Preposition	P / ADP
Coordinating conjunction	CONJ / CCONJ
Subordinating conjunction	SCONJ
Particle	PART
Negative particle	NEG
Interrogative particle	INTG
Vocative particle	VOC

Modern universal annotation frameworks may group some traditional Arabic distinctions differently. For example, Universal Dependencies distinguishes categories such as ADP, CCONJ, SCONJ and PART.

An Arabic-specific system may choose a finer level of granularity when the distinction is important for the intended application.

Beyond POS: Morphological Features

Part-of-speech information alone cannot represent the richness of Arabic morphology.

An MSA morphological annotation framework may therefore associate tokens or segments with features such as:

Person: first, second, third
Gender: masculine, feminine
Number: singular, dual, plural
Case: nominative, accusative, genitive
Definiteness: definite, indefinite
Aspect: perfective, imperfective
Mood: indicative, subjunctive, jussive
Voice: active, passive

The dual is particularly noteworthy.

Arabic grammatically distinguishes:

كتاب — book

كتابان — two books

كتب — books

Number is therefore not restricted to the singular/plural distinction familiar from English.

This morphological richness has significant implications for computational analysis.

Roots, Lemmas and Surface Forms

Arabic NLP must also distinguish between the surface form, stem, lemma and root.

Consider the consonantal root:

ك ت ب — k-t-b

It participates in the formation of numerous lexical items associated with writing, including:

كتب — kataba — wrote

كتاب — kitāb — book

كاتب — kātib — writer

مكتبة — maktaba — library

These words are related, but they are not interchangeable.

A root captures an abstract morphological relationship, while a lemma provides a canonical lexical form associated with an inflected word.

This distinction is extremely important in applications such as information retrieval, lexicography, corpus analysis and search.

Over-aggressive root extraction can collapse semantically distinct words, while purely surface-based matching may fail to recognize meaningful morphological relationships.

POS Tagging and Syntactic Analysis

Morphological annotation becomes more powerful when connected with syntactic parsing.

A useful computational representation can be viewed as several interconnected levels:

orthographic word → morphological segmentation → lemma → POS → morphological features → syntactic dependencies

Consider:

قرأ الباحث المقالة

qaraʾa al-bāḥithu al-maqālata

The researcher read the article.

A POS analysis identifies the verb and nominals. Morphological analysis contributes gender, number, case and other information. Syntactic analysis then establishes relationships such as the predicate, subject and object.

Each level contributes different information about the same sentence.

Modern Arabic Annotation Resources

Contemporary Arabic NLP uses several important annotation and morphological-analysis frameworks.

The Arabic Treebank has provided annotated Modern Standard Arabic material for research into morphological analysis, POS tagging and syntactic parsing.

Universal Dependencies Arabic provides Arabic treebanks using a cross-linguistically consistent system of POS categories, morphological features and dependency relations.

Resources and tools associated with CAMeL Tools also provide computational support for MSA morphological analysis and disambiguation.

These resources illustrate different approaches to the same fundamental challenge: representing Arabic linguistic structure in forms that both preserve meaningful grammatical distinctions and support computational processing.

Arabic Varieties and POS Annotation

Arabic NLP becomes still more interesting when contemporary varieties are considered.

Modern Standard Arabic coexists with regional and spoken varieties that differ in vocabulary, morphology, pronunciation and syntactic patterns.

A POS tagset designed for formal MSA text cannot automatically be assumed to capture every phenomenon found in Moroccan, Egyptian, Levantine, Gulf or other Arabic varieties.

Code-switching introduces another challenge. Digital Arabic content may combine MSA, regional varieties, foreign words, transliteration and Arabizi within the same conversation.

Future Arabic annotation frameworks therefore need to balance standardization with linguistic variation.

Why Arabic POS Tagsets Matter for NLP

Reliable Arabic POS and morphological annotation supports numerous language-technology applications.

These include:

Machine Translation — morphological and syntactic features help systems interpret relationships that must be transferred between languages.

Information Retrieval — morphological normalization can improve matching between queries and documents.

Named Entity Recognition — proper-noun information can contribute to identifying people, organizations and locations.

Sentiment Analysis — negation, pronouns and modifiers can substantially change the interpretation of sentiment.

Information Extraction — morphological and syntactic relationships help systems identify entities, events and relations.

Dependency Parsing — POS and morphological features provide important information for identifying grammatical relationships.

Language Models — Arabic morphology presents important questions about tokenization, segmentation and representation.

Corpus Linguistics — POS tags allow researchers to investigate grammatical patterns across large collections of MSA and other contemporary Arabic texts.

Conclusion

Arabic part-of-speech tagging demonstrates the close relationship between grammar, morphology and computational linguistics.

The fundamental categories ism (اسم), fiʿl (فعل) and ḥarf (حرف) provide a useful high-level foundation, while computational language processing expands these categories into detailed machine-readable representations.

Nouns, proper nouns, adjectives, pronouns, adverbs, verbs, conjunctions, prepositions and particles can be identified alongside morphological properties such as gender, number, person, case, definiteness, aspect, mood and voice.

Most importantly, Arabic reminds us that a written word is not necessarily a single computational unit. Prefixes, articles, stems and pronominal suffixes may need to be represented separately in order to capture the linguistic structure encoded within an orthographic word.

For researchers working in Modern Standard Arabic, Arabic varieties, Natural Language Processing, corpus linguistics, morphological analysis, syntactic parsing, information retrieval and language modelling, POS annotation provides an essential bridge between Arabic linguistic analysis and modern language technology.

	Tag	Arabic Name	Description
Nouns	N	اسم	Noun
	PN	اسم علم	Proper noun
Derived nominals	ADJ	صفة	Adjective
	IMPN	اسم فعل أمر	Imperative verbal noun
Pronouns	PRON	ضمير	Personal pronoun
	DEM	اسم اشارة	Demonstrative pronoun
	REL	اسم موصول	Relative pronoun
Adverbs	T	ظرف زمان	Time adverb
	LOC	ظرف مكان	Location adverb