Investigating the Syntactic Structure of Arabic Sentences Asma Moubaiddin
Abeer Tuffaha, Bassam Hammo and Nadim Obeid
Department of Linguistics, Faculty of Foreign Languages The University of Jordan, Amman, Jordan
King Abdullah II School for Information Technology The University of Jordan, Amman, Jordan
[email protected]
Abstract—In this paper, we employ the Government and Binding theory (GB) to present a system that analyzes the syntactic structure of some simple Arabic sentences structures. We consider different word orders in Arabic and show how they are derived. We include an analysis of Subject-Verb-Object (SVO), Verb-Object-Subject (VOS), Verb-Subject-Object (VSO), nominal sentences, nominal sentences with inna (or sisters), and question sentences. We use the analysis to develop syntactic rules for a fragment of Arabic, such that we include two sets of rules (1) rules on sentences structures that do not account for case and (2) rules on sentences structures that account for Noun Phrases (NPs) case. We present an implementation of the grammar rules in prolog. The results of testing the system t were reasonable with a high accuracy especially when the input sentences are tagged with identification of end cases. Keywords—component; formatting; style; styling; insert (key words)
I.
INTRODUCTION
Syntactic analysis is essential in Natural Language Processing (NLP). It is based on parsing sentences to identify their components. It plays an essential role in many applications, such as grammar checking, text generation and machine translation. Parsing Arabic texts is challenging [2, 15] because Arabic language has rich morphology due to its highly inflectional nature, highly flexible word order and frequent use of clitics that are attached to words. The emphasis in this paper is on the use of Government and Binding theory (GB) [4, 9] in analyzing the syntactic structure of some simple Arabic sentences' structures. We consider different word orders in Arabic and show how they are derived. We shall include an analysis of SVO, VOS, VSO, nominal sentences, nominal sentences with inna (or sisters), and question sentences. We use the analysis to develop syntactic rules for a fragment of Arabic, such that we include two sets of rules (1) rules on sentences structures that do not account for case and (2) rules on sentences structures that account for NPs case. We present an implementation of the grammar rules in prolog. Due to space limitation, we shall not present some important computational steps and we shall not present the structure of the lexicon needed to build the implemented system. 978-1-4673-2821-0/13/$31.00 ©2013 IEEE
In section 2 we present a brief introduction to Arabic. In section 3 we discuss the notion of syntactic analysis. In section 4 we give a brief presentation of GB. In section 5 we present an analysis of some aspects of Arabic syntax. Sections 6 and 7 are dedicated to the parser and its implementation . II.
BRIEF INTRODUCTION TO ARABIC
Arabic is a Semitic language that has a rich morphology and flexible word order. In this paper we are concerned with Modern Standard Arabic (MSA) version used in modern writing and understood by Arabic language speakers. Arabic grammar distinguishes between two types of sentences, verbal and nominal. Nominal sentences have two parts: a subject ( )ﻣﺒﺘﺪأand a predicate ()ﺧﺒﺮ. When the nominal sentence is about being, i.e. if the verb of the sentence is ‘to be’ in English, this verb is not given in Arabic. Arabic morphology is based on roots and patterns through which words are derived. An Arabic word may be composed of a stem consisting of a base root and a pattern which defines its semantic and syntactical role. Moreover, affixes and clitics are often attached to words. Affixes include inflectional markers for tense, gender, and number. Clitics include prepositions, conjunctions, determiners, and possessive pronouns. Some of the characteristics and/or challenges of Arabic language: (1) It has a relatively free word order, it is not unusual to find each of VSO, SVO and VOS word orders within an Arabic text as in: (1a) ﻗﺮأ اﻟﻤﻌﻠّﻢ اﻟﺪرس qara?-a al-mu؟alim-u al-dars-a: VSO (1b) اﻟﻤﻌﻠ ُﻢ ﻗﺮأ اﻟﺪرس al-mu؟alim-u qara?-a al-dars-a: SVO (1c) ﻗﺮأ اﻟﺪرس اﻟﻤﻌﻠﻢ qara?-a al-dars-a al-mu؟alim-u: VOS All of these are grammatical and imply the same meaning (The teacher read the lesson). (2) Arabic is a clitic or clitic-directed language. Clitics are morphemes that have the syntactic characteristics of a word but are morphologically bound to other words (e.g., coordinating conjunctions, the definite article, many prepositions and particles, and a class of pronouns that attach themselves either to the start or end of words) as in: آﺘﺒﻨﺎ:katabnaa (we wrote) which is made up of the verb
(آﺘﺐ: katab) and the clitic (ﻧﺎ: na) that acts as the subject for the verb ()آﺘﺐ. (3) The omission of diacritics (syntactic marks) in most written Arabic Texts. (4) Arabic is a pro-drop language. The subject can be omitted leaving any syntactic parser with the challenge to decide whether or not there is an omitted pronoun in the subject position. (5)
Homographs of words with/without the same pronunciation are often produced. They have different meanings and usually different POS. For example, the word ( )ذهﺐcan be interpreted as thahab-a (verb) meaning went or thahab-un (noun) meaning gold. III.
GVERNMENT AND BINDING THEORY
GB is an approach to Universal Grammar which includes rules and principles that apply to all languages. However, while certain grammatical principles and rules are universal, there is a lot of variation between different languages such as different ordering for subject (S), verb (V) and object (O). It is agreed that every language has a basic word order, and all other word orders result from the movement of sentence constituents and this movement is restricted by some rules and principles. Words are organized hierarchically into bigger units called phrases. Phrase constituents include: (1) IP - Inflectional Phrase: a phrase headed by I. I/INFL stands for Inflection, and it consists of tense, and number and gender agreement (AGR) elements. (2) CP - Complementizer Phrase: a phrase headed by a complementizer (C). C takes an IP (INFL Phrase) as its complement and heads the maximal projection CP (3) NP - Noun Phrase: a phrase headed by a noun (N). (4) VP - Verb Phrase: a phrase headed by a verb (V). (5) AP - Adjectival Phrase: a phrase headed by an adjective (A). (6) PP - Prepositional preposition (P).
Phrase:
a
phrase
headed
by a
The main principles of GB include: (1) Government which is concerned with the syntactic relations in a sentence and has a main application in case assignment. (2) Theta Theory which is concerned with describing the thematic relations between arguments and predicates. (3) Predicates and arguments: The arguments are the participants minimally involved in the activity or state expressed by the predicate. (4) Case Theory which is concerned with the assignment of abstract case: nominative, accusative, and genitive to words, based on their positions in the sentence.
(5) X-Bar Theory which is concerned with phrase formation. It states that all phrases are headed by a lexical head (noun, verb, adjective, or preposition) . - Complements combine with X to form X' projections, adjuncts combine with X' to form X'. The specifier combines with the topmost X' to form the maximal projection X''/XP.
- D-structure and S-structure: All sentences are represented in terms of both forms, the D-structure and S-structure. Dstructure encodes the predicate-argument relations and the thematic properties of the sentence and it is built upon the basic word order. S-structures accounts for the surface ordering of the sentence constituents. - NP-Movement: GB assumes that the different word orders arise from the movement of sentence constituents. Hence, a basic word order is assumed, and all other word orders are derived. IV.
ANALYSIS OF ARABIC SYNTAX
A. Constituents Structure in Arabic We first analyze the lexical formation of the smaller phrase constituents that make up a sentence. - Noun Phrase (NP): an NP head is a noun and it can be represented as: NP → N Example: آﺘﺎب- kitab: a book NP → NP NP Example: آﺘﺎب اﻟﺒﻨﺖ- Kitab-u al-bint-i: The girl's book NP → NP Conj NP Example: اﻟﻠﻴﻞ واﻟﻨﻬﺎر- Al-layl-u wa al-nahar-u: the night and the day NP → NP AP Example: آﺘﺎب ﻣﻔﻴﺪ-Kitab-un mufid-un: A useful book - Adjectival Phrase (AP): an AP head is an adjective and it can be represented as: AP → A Example: ﻣﻔﻴﺪ-mufid-un: useful AP → A AP Example: أﺳﻮد ﺻﻐﻴﺮ- aswad-un sagheer-un: black small AP → AP Conj AP Example: أﺳﻮد و ﺻﻐﻴﺮ- aswad-un wa sagheer-un: black and small - Verb Phrase (VP): a VP head is a verb and it can be represented as: VP → V Example: ﻗﺮأ- qara?-a: he read VP → V NP Example: أآﻞ اﻟﺘﻔﺎﺣﺔ- akal-a altufahat-a: he ate the apple V → V PP
Example: ذهﺐ إﻟﻰ اﻟﻤﺪرﺳﺔ- thahab-a ila almadrasat-i: he went to school VP → VP PP Example: ب ﻋﻠﻰ اﻟﻄﺎوﻟﺔ َ وﺟﺪت اﻟﻜﺘﺎ- wajadt-u al-kitab-a ؟la altawilat-i: I found the book on the table - Prepositional Phrase (PP): A PP head is a preposition and it can be represented as: PP → P NP Example: ﻓﻲ اﻟﻤﻜﺘﺒﺔ- fi almaktabat-i: In the library B. Arabic Word Order We assume that the basic word order is SVO. In SVO order, I(NFL) assigns a NOM case to the subject at [Spec,IP] position (through the percolation of I(NFL) to IP), and the verb which heads the VP assigns an ACC case to its object. For example: (1a) اﻟﻤﻌﻠّﻢ ﻗﺮأ اﻟﺪرس (1b) al-mu؟lim-u qar?-a al-dars-a (1c) the teacher-nom+masc. read-past+masc.+sing. the lesson-acc (1d) the teacher read the lesson In this sentence 'al-mu؟lim-u' receives a NOM case from I(NFL), and the noun 'al-dars-a' is assigned an ACC case from its governing verb qara?.(cf Figure 1). S- structure: [IP al-mu؟lim-u [I' [VP qara?-a aldars-a]]]
(2a) ﻗﺮأ اﻟﺪرس اﻟﻤﻌﻠ ُﻢ (2b) qara?-a al-dars-a al-mualim-u (2c) read-past the lesson-acc the teacher-nom (2d) the teacher read the lesson At D-structure, the subject at [Spec,IP] receives a NOM case from I(NFL) , the verb's object receives an ACC from the verb. VSO order is obtained by moving the verb to empty [C,CP] (Head to Head movement) leaving its co-indexed trace (the accusative case of VP's internal NP is assigned through the verb's trace) [1] (cf. Figure 3 and Figure 4). C. Nominal Verbless Sentences A typical NP contains two nouns. For example, (3a) ٌب ﻣﻔﻴﺪ ُ اﻟﻜﺘﺎ (3b) al-kitab-u mufid-un (3c) the book-nom useful-nom (3d) the book is useful To explain the grammaticality of this type of sentences, we can assume that there is a hidden verb, such that this verb carries the meaning of is and it occupies V. Accordingly, the phrase will be tensed, which allows I(NFL) to assign a nominative case to NP at [Spec,IP]. However, the hidden verb will fail to govern its internal argument. To solve this issue we will adopt the default case approach mentioned in [10] which states that NPs with no case assigner are possible in Arabic and they are assigned a nominative case.
The other word orders are the result of movements applied on the basic word order. VOS results from the subject adjunction to the end of VP. Hence, it will receive a NOM case from I(NFL). And to satisfy the EPP (Extended Projection Principle), we can assume [Spec,IP] to be occupied by PRO (cf. Figure 2).
Figure 3: S-structure for a VSO sentence Figure 5 shows the structure of a nominal sentence: 'alkitab-u' receives a NOM case from I(NFL), and 'mufid-un' is assigned the default NOM case. Figure 1. ' al-mu؟lim-u qar?-a al-dars-a'
Figure 2. 'qara?-a al-dars-a al-mu؟alim-u'
Figure 4. D-structure for a VSO sentence
D. Nominal sentences with Inna (or sisters) Inna and sisters particles (ن وأﺧﻮاﺗﻬﺎ ّ )إcan occur in nominal sentences. They include ( ﻞ ّ ﻟﻌ، ﻟﻴﺖ،ّ أن،ّ إن- inna, anna, layta, la؟lla..). Inna and sisters are complementizers that assign an Accusative case to their noun governees. Below are some examples: (4a) ٌب ﻣﻔﻴﺪ َ ن اﻟﻜﺘﺎ ّإ (4b) Inna al-kitab-a mufid-un (4c) the book-acc useful-nom (4d) the book is useful In Arabic, in the existence of inna and sisters particles at [C,CP], the NP at [Spec,IP] is assigned an ACC case from the complementizer particle. We can assume that inna and sisters complementizers are stronger than tensed I(NFL) and prevents it from assigning the nominative case to NP at [Spec,IP]. In this sentence, 'inna' is a complementizer that assigns an ACC case to its governee 'al-kitab-u' and the hidden verb (is) fails to assign an accusative case to 'mufid-un'. Therefore, it is assigned the default NOM case.
At D-structure, the question word 'maatha' is base generated at the object position. Note that this question begins with a question word followed by a verb. In order to explain sentence grammaticality, we assume that the question word is moved to [Spec,CP], and the verb is moved to [C,CP] (cf. Figure 8 and Figure 9).
Figure 6. S-structure for a question with 'man'
Figure 5. Nominal Sentence E. Questions Questions in Arabic usually start with a question word like maatha ( ﻣﻦ: what), mata ( ﻣﺘﻰ:when), and ayna ( أﻳﻦ:where). Both man (who) and maatha (what ) can be used to ask about the subject or the object.
Figure 7. D-structure for a question with 'man'
If the question is about the subject, at D-structure the question word is placed at [Spec,IP], at S-structure it is moved to [Spec,CP]. If the question is a about the object, at D-structure the question word is placed at the object's node under VP, and to produce the S-structure, the question word is moved to [Spec,CP]. Consider the following examples: Question on subject: (5a) ﻣﻦ ﻗﺮأ اﻟﺪرس؟ (5b) man qara'a aldars-a? (5c) Who read the lesson? (5d) Who read the lesson? At D-structure, the question word 'man' is base generated at [Spec,IP]. At S-structure it moves to [Spec, Cp] (cf. Figure 6 and Figure 7). Question on object: (6a) ﻣﺎذا ﻗﺮأ اﻟﻄﺎﻟﺐ؟ (6b) Maatha qara'a altalib-u? (6c) What read the student (6d) What did the student read?
Figure 8. S-structure for a question with 'maatha' Questions on VP adjuncts: ayn-a (where), mata (when), and kayf-a (how), are usually used to ask about a VP's adjunct. At D-structure the question word occupies the VP's adjunct position. Consider the following example: (7a) أﻳﻦ ﺳﺎﻓﺮ ﻣﺤﻤﺪ؟ (7b) Ayna safar-a mohammad? (7c) Where traveled mohammad (7d) Where did mohammad travel? At D-structure, the question word 'ayna' is base generated at VP's adjunct position. At S-structure, we assume that the question word is moved to occupy [Spec,CP], and the verb is moved to [C,CP].
correctness and (T) (resp (F)) indicates that there is (resp. not) cased analysis. ﻣﻮﺳﻰ ﻳﺤﺐ ﻟﻌﺐ آﺮة اﻟﻘﺪم (1) [ip: np[vp: verb np]]
(F)
POST1: [noun,verb,noun,noun,noun] [ip: np[vp: verb np]] (F) POST2: [noun_prop,verb,noun_acc,noun_gen,noun_gen]
(2) [ip:noun_prop[vp: verb np_acc]] (T) Both analyzes are correct
ﺖ إﻟﻰ اﻟﻤﺪرﺳ ِﺔ ُ ذهﺒﺖ اﻟﺒﻨ
POST1: [verb,noun,prep,noun]
(3) [cp: verb[ip: np[vp:t [pp: prep np]] (F) (4) [ip:pro [vp:[vp: verb np][pp: prep np]]] (F) Figure 9. D-structure for a question with 'maatha' Yes/No questions: Yes|No questions use the question word 'hal'. (8a) هﻞ ﺳﺎﻓﺮ ﻣﺤﻤﺪ؟ (8b) Hal safar-a mohammad? (8c) travel mohammad? (8d) Did mohammad travel? In this sentence, at D-structure, we assume that the question word ‘hal’ is base generated as an IP adjunct. To produce the S-structure, the verb is moved to [C,CP], and the question word is moved to [Spec,CP]. The figures for questions on VP adjunct and yes/no questions are not provided due to lack of space. V.
THE PARSER
The input is a sequence of tags for a sentence (with or without diacritics) and the output is the structure for syntactically correct sentences. We adopted a top-down approach for parsing sentences, with support to recursive rules. The implementation is restricted to (1) Basic sentence structures for SVO, VOS, and VSO sentences where the subject is an NP and the Object is an NP. (2) Sentences followed by a PP adjunct. (3) Nominal sentences made up from NP(s), or NP followed by PP, and nominal sentences preceded by Inna (or sisters particles). (4) Question Sentences staring with a question word followed by a VSO order sentence. VI.
SYSTEM IMPLEMENTATION AND RESULTS
We have employed SICStus Prolog 3.12.2 [16] to implement the parser. The implemented system was tested on 500 sentences (using caseless and cased analysis). The input are tagged [3] sequences of sentences in prolog queries. Sentences were tested against both grammar rules bases, the cased and the caseless grammar rules. The results were as expected. Below we present some sample sentences (whose tags were given as input to the application) in Arabic, along with their resulting syntax structures. The bold text signifies result
POST2: [verb, noun_nom,prep, noun_gen]
(5) [cp: verb[ip: np_nom[vp:t [pp: prep np_gen]]]] (T)
Cased analysis (5) is correct. The caseless analysis (4) is incorrect because the analyzer cannot differentiate between a subject an object np. ٌﺗﻐﻴﺐ ﻋﻦ اﻟﺼﻒ ﻣﺤﻤﺪ
POST1: [verb,prep,noun,noun]
(6) [ip:pro [vp: verb[pp: prep np]]] (F) (7) [ip:[vp:[vp: verb[pp: prep np]] np]] (F) POST2: [verb,prep,noun_gen,noun_nom]
(8) [ip:[vp:[vp: verb[pp: prep np_gen]] np_nom]] (T)
Analysis (6) is incorrect. ٌاﻟﺒﻴﺖ ﺟﻤﻴﻞ
POST1: [noun,noun]
(7) [ip: np[vp:e np]] (F) POST2: [noun_nom,noun_nom]
(8) [ip: np_nom[vp:e np_nom]] (T) Both analyzes are correct.
هﻞ ﺳﺎﻓﺮ ﻣﺤﻤﺪٌ ؟
POST1: [q_word,verb,noun]
(9) [cp: q_word[cp: verb[ip: np]]] (F) POST: [q_word,verb,noun_nom]
(10) [cp: q_word[cp: verb[ip: np_nom]]] (T) Both analyzes are correct. ٌﺗﻐﻴﺐ ﻋﻦ اﻟﺼﻒ ﻣﺤﻤﺪ POST1: [verb,prep,noun,noun] (11)[ip:pro [vp: verb[pp: prep np]]] (F) (12) [ip:[vp:[vp: verb[pp: prep np]] np]] (F) POST2: [verb,prep,noun_gen,noun_nom] (13) [ip:[vp:[vp: verb[pp: prep np_gen]] np_nom]] (T) Cased analysis (13) is correct whereas caseless (12) is incorrect.
ٌإن اﻟﺠ َﻮ ﺟﻤﻴﻞ POST1: [func_word,noun,noun] (14) [cp: func_word[ip: np[vp:e np]]] (F) POST: [func_word,noun_acc,noun_prop] (15) [cp: func_word[ip: np_acc[vp:e noun_prop]] (T) Both analyzes are correct. هﻞ ﺳﺎﻓﺮ ﻣﺤﻤﺪٌ ؟ POST1: [q_word,verb,noun] (16) [cp: q_word[cp: verb[ip: np]]] (F)
accuracy especially when the input sentences are tagged with identification of end cases. It is important to note that the system is far from complete. We intend to test it on a standard corpus and compare it with similar systems. However, the proposed system is flexible and can be extended such that further modifications can be applied. We hope to enhance the system by using a morphological analyzer that would provide important features about the words such as clitics identification, number and gender features identification and by adding more rules to deal with more sentences structures and to cover other syntactic features such as subject-verb agreement on number and gender, words clitics and cases that are represented as suffixes to nouns.
POST2: [q_word,verb,noun_nom] (17) [cp: q_word[cp: verb[ip: np_nom]]] (T)
REFERENCES
Both analyzes are correct. When using caseless analysis, grammatical sentences were parsed correctly and usually the parser produces many possible analyses, especially when the sentence has a verb. This multiplicity in analyses can be mainly referred to the following reasons: (C1) VSO | VOS | Nominal sentence: NP ambiguity: the parser has no clue on where the subject and object start and end. (C2)VSO | VOS The parser is not able to determine whether an NP is a subject or an object. (C3) SVO| VS The parser is not able to determine if there is hidden subject (PRO) and sentence has an SVO order, or if the sentence is VS with an intransitive verb. When using cased analysis, grammatical sentences were parsed correctly with the precision in determining the exact sentence structure highly enhanced and usually producing one matching syntactic structure for the input sentence. VII. CONCLUSION We have analyzed the syntactic structure of some simple Arabic sentences' structures based on GB. We have considered different word orders in Arabic and shown how they are derived. We have included an analysis of SVO, VOS, VSO, nominal sentences, nominal sentences with inna (or sisters), and question sentences. We have used the analysis to develop syntactic rules for a fragment of Arabic, such that we include two sets of rules on sentences structures that do not account for case and rules on sentences structures that account for NPs case. We have presented an implementation of the grammar rules in prolog. The results were reasonable with a high
[1] [2]
[3]
[4] [5] [6]
[7]
[8]
[9] [10]
[11]
[12] [13] [14]
[15] [16]
J. Al-Bayaty, Adjunction in Arabic, Case and Chain theory, Ph.D thesis, 1990, Simon Fraser Univ. B. Bataineh and E. Bataineh, An Efficient recursive Transition Network Parser for Arabic Language, Proc. World Congress on Engineering Vol II, 2009. F. Ben Fraj, C. Ben Othmane Zribi and M. Ben Ahmed, ArabTAG: a Tree Adjoining Grammar for Arabic Syntactic Structures, ACIT, 2009, Tunisia. C. A. Black, A step-by-step introduction to the Government and Binding theory of syntax, 1999, University of North Dakota. K. Daimi, Identifying Syntactic Ambiguities in Single-Parse Arabic Sentence, Computers and the Humanities 35, 2001, 333-349. M. Diab, K. Hacioglu and D. Jurafsky, Automatic Processing of Modern Standard Arabic Text, Arabic Computational Morphology, 2007, 159 – 179. Y. ElHadj, I. Al-Sughayeir and A. Al-Ansari, Arabic part-of-speech tagging using the sentence structure, Proc. Second International Conference on Arabic Language Resources and Tools, 2009. N. Habash and O. Rambow, Extracting a Tree Adjoining Grammar from the Penn Arabic Treeban, Proc. Traitement Automatique du Langage Naturel (TALN-04), 2004. L. Haegeman, Introduction to Government and Bonding Theory, 1991, Blackwell. M. Homeidi, The Notion of Governor in Modern Standard Arabic (MSA) and English: a Contrastive Perspective, J. King Saud Univ., 15, 2003, 49 - 62. D. Jurafsky and J. Martin, Speech and Language processing: An Introduction to Natural Language Processing Computational Linguistics, and Speech Recognition, 2000, Prentice Hall. A Lobeck, Discovering Grammar: An Introduction to English Sentence Structure, 2000, Oxford University Press. E. Othman, K. Shaalan and A. Rafea, Towards Resolving Ambiguity in Understanding Arabic Sentences, Proc. NEMLAR, 2004. E. Othman, K. Shaalan and A. Rafea, A Chart Parser for Analyzing Modern Standard Arabic Sentence, MT IX Workshop on Machine Translation for Semitic Languages: Issues and Approaches, 2003. K. Shaalan, K., Arabic GramCheck: a grammar checker for Arabic, Software—Practice And Experience, 35, 2005, pp. 643–665. SICStus Prolog User’s Manual, Intelligent Systems Laboratory, 2005, Swedish Institute of Computer Science.
