<?xml version="1.0"?><!DOCTYPE article SYSTEM "/project/take/software/searchbench_offline_processing/paperxml_generator/aclextractor/src/python/../resource/dtd/paperxml.dtd"><article><header><firstpageheader><page local="1" global="686"/><title>Analyses Tools for Non-head Structures</title><pubinfo>Proceedings of Recent Advances in Natural Language Processing,pages 686-691, Hissar, Bulgaria, 12-14 September 2011.</pubinfo><author surname="Boukedi" givenname="Sirine"><org  name="Faculty of Sciences Economy and Sciences Faculty of Sfax Management of Sfax"/></author><author surname="Haddar" givenname="Kais"><org  name="Faculty of Sciences Economy and Sciences Faculty of Sfax Management of Sfax"/></author></firstpageheader><frontmatter><p><b>Analyses Tools for Non-head Structures</b></p><p><b>Sirine Boukédi Kais Haddar</b></p><p>Faculty of Sciences Economy and Sciences Faculty of Sfax</p><p>Management of Sfax Kais.haddar@fss.rnu.tn Sirine.boukedi@gmail.com</p></frontmatter><abstract>Syntactic analysis is a fundamental phase in NLP (Natural Language Processing) do­main. This phase occurs in several applica­tions and at different levels. Moreover, it wasn't spilled in domain research, espe­cially for Arabic language. In fact, most of researchers working on Arabic language treated simple structures and neglected complicated ones such as relatives, coordi­nation, ellipse and juxtaposition. In this con­text, the present work lies within the con­struction of a HPSG (Head-driven Phrase Structure Grammar) grammar treating Ara­bic coordination. The established grammar is specified on TDL (Type Description Lan­guage) and experimented with a parser gen­erated by LKB (Linguistic Knowledge Building) system. </abstract></header><body><section number="1" title="Introduction"><p>The study on Arabic language showed that coor­dination is one of particular structures. It is fre­quent in different corpus and occurs with many other phenomena. The interaction with the other structures makes the study very delicate. For this reason, it wasn't spilled in research domain.</p><p>Based on a large literature, most of existing researchers treated coordination structure for Roman languages except some works such as (Haddar, 2000) and (Maaloul <i>et al., </i>2004). In fact, Arabic coordination is very complicate. It covers many forms and different structures. Therefore, there is a big problem in the categori­zation of Arabic coordination.</p><p>Moreover the last researchers found a problem in the choice of the adequate formalism repre­senting the different forms of coordination struc­tures. But most of related works used HPSG. The choice of this formalism is justified. In fact, HPSG offers a complete representation for lin­guistic entries.</p><p>Therefore, our work aims to find an adequate typology classifying Arabic coordination struc­tures and to construct a HPSG grammar repre­senting the different forms of coordination. This grammar will be validated on LKB system.</p><p>In the present paper, we start by describing some related works treating coordination struc­ture. Then, we adapted HPSG grammar to repre­sent the different forms of our phenomenon, based on a proposed typology. It should be noted that the established grammar treated simple sen­tences and complex ones representing the differ­ent forms of coordination except cases of interac­tion with ellipse phenomenon. Finally, we vali­dated our grammar on LKB system after specifi­cation in TDL. According to the obtained results, we evaluate our grammar and we enclose our work by a conclusion and some perspectives.</p></section><section number="2" title="Previous works"><p>The study on previous works showed that re­searchers on coordination structure started since 1970, such as (Hudsan, 1976), (Postal, 1974) and (Rau, 1985), for many languages. The different researches used various grammars. Some works used the GCCA (Applicative Categorical Com-binatory Grammar), other works used GI (Inter­active Grammar) and other ones were based on HPSG Grammar. But most of them, used this last one (HPSG formalism).</p><p>For French language, we can mention (Biskri and Desclés, 2006) who studied coordination structure of similar constituents, based on GCCA grammar. Moreover, (Le Roux and Perrier, 2006) studied constituent and non constituent structures based on XMG tools, a compilation tool, and used the GI formalism.</p><p>For Portuguese language, (Villavicencio <i>et al., </i>2005) studied the coordination of nominal phras<page local="2" global="687"/></p><p>es. They identified different strategies of analyses based on the HPSG formalism.</p><p>For Bulgarian language, we can mention the work of (Osenova and Simov, 2005) who studied the coordination phenomenon and its interaction with ellipse forms based on HPSG grammar. It should be noted that the formalization was en­coded in XML.</p><p>According to our research, the study showed that most of the related works treated the coordi­nation of Roman language. But, there is some works treating Arabic coordination such as proposed typology is similar in most of the re­lated works. The difference between them ap­pears in the choice of the grammar and the anal­ysis tools.</p><doubt alpha="52.3" length="44" tooSmall="False" monospace="0.0">(Haddar, 2000) and (Maaloulet al.,2004). The</doubt><p>For Arabic works, for example, Haddar (2000) studied syntactic analyses of Arabic co­ordination based on ATN (Augmented Transition Network) and (Maaloul <i>et al., </i>2004) studied the coordination of Arabic constituent based on HPSG grammar. This grammar was tested and validated on a constructed system, AICOO.</p><p>Based on the proposed typology, these related researches working on Arabic coordination didn't treat all forms of this structure. Therefore, the originality of our work is to construct a HPSG grammar covering all the possible forms of coordination and its interaction with the other phenomena such as the ellipse one. In the follow­ing paragraph, we present the proposed typology of Arabic coordination that we adopted from the related works.</p></section><section number="3" title="Proposed typology for Arabic coordi­nation"><p>According to some linguists such as (Abdelwaphenomenon joins two or several elements with a particle of coordination (conjunction). In Arabic, there exist nine conjunctions ( <b><i>'Cß^ '&lt;J&gt;=&gt;- </i></b>'j 6 'V 'jl</p><doubt alpha="57.4" length="47" tooSmall="False" monospace="0.0">hed, 2004) and (Dahdeh, 1992), the coordination</doubt><p>Based on some related works (Haddar, 2000) and (Maaloul <i>et al. </i>, 2004), coordination structure in Arabic can be subdivided, like Roman lan­guages, in two categories: coordination of con­stituent and coordination of non constituent. The first category covers cases when the conjunction joins two or several well formed constituents. These constituents can have similar or different categories. The figure 1 represents the coordina­tion of similar constituents. The figure 2 repre­sents the coordination of different constituents.</p><doubt alpha="100.0" length="2" tooSmall="False" monospace="0.0">PN</doubt><doubt alpha="50.0" length="12" tooSmall="False" monospace="0.0">N     CONJ N</doubt><p><i>'adl</i><b><i>-Aistaadha </i></b><i>via 'aaltalaamydha The professor </i><b><i>and </i></b><i>the pupils</i></p><figure caption="Figure 1. Coordination of constituents having simi­lar categories."></figure><p>As shown in Figure 1, the conjunction" j, and" joins two compounds having the same category, two defined nouns "^Vl, the professor" and "^^l, the pupils".</p><doubt alpha="100.0" length="3" tooSmall="True" monospace="0.0">piI</doubt><doubt alpha="58.3" length="12" tooSmall="False" monospace="0.0">PN     CONJN</doubt><doubt alpha="0.0" length="3" tooSmall="False" monospace="0.0">/ \</doubt><doubt alpha="66.7" length="3" tooSmall="False" monospace="0.0">N P</doubt><doubt alpha="50.0" length="16" tooSmall="False" monospace="0.0">4a iySk-- 5 ^JMA</doubt><p><b><i>mitsriaaN\\ a hi charaahatiN Quic&amp;ly and with goimmmdizc</i></b></p><figure caption="Figure 2. Coordination of constituents having dif­ferent categories."></figure><p>However in Figure 2, the same conjunction joins two different compounds. The first one "l^-j^, quickly" is an adverb, the second one ", with gourmandize" is a reduction phrase " — j^?, reduction phrase".</p><p>For the second category, coordination of non constituent, the conjunction joins two or several constituents where one of them is incomplete. In fact, it represents the case where there is interac­tion with the ellipse phenomenon.</p><p>According to some references, there exist four forms of ellipse: Right Node Raising, Left Node Raising, Gapping and VP-ellipse. The first form: Right Node Raising, designed the case when the first element that should be at the right of the second compound, is missed.</p><p>The second form: the Left Node Raising de­signed the case when the second element is missed in the left of the second compound of coordination phrase.</p><p>For the third form: Gapping, it represented when there exist discontinuities in the second compound of the coordination phrase.</p><page local="3" global="688"/><p>Finally, for the last form, VP-ellipse, it repre­sented the case when the verbal phrase is missed and replaced by a proverb like also".</p><p>Based on the proposed typology for the Arabic coordination, we adapted the HPSG grammar. In fact, based on some references such as (Godard, 2006), the coordination phenomenon is a non head structure. Its representation differs from other phenomena. So it necessitates a particular structure. In the following paragraph, we present the HPSG grammar and the different modifica­tions brought to this formalism to represent Ara­bic language. Then, we present the HPSG struc­ture of Arabic coordination.</p></section><section number="4" title="HPSG for the Arabic language"><p>HPSG is a unification grammar (Pollard and Sag, 1994). It is characterized by a reliable modeling of the universal grammatical principles and a complete representation of linguistic knowledge.</p><p>HPSG grammar is based on two essential components: AVMs (Attribute Value Matrix) and a set of immediate domination schemata (DI schemata). An AVM is based on a set of features characterizing a lexical entry. The DI schemata, describe a syntactic phenomenon. It should be noted that to compose the various phrases, a set of principles should be verified (i.e., HFP Head Feature Principle).</p><p>HPSG grammar was conceived for Roman languages. To use it for Arabic language, we pre­sent in the following paragraph the modifications made to HPSG. These modifications were made on the features and schemata level.</p><subsection number="4.1" title="Arabic features"><p>Referring to previous projects (Elleuch, 2004), (Bahou <i>et al, </i>2005) and (Abdelkader <i>et al., </i>2006), we have kept some features and have added some others according to the proposed type's hierarchy. As example, we present, in ta­ble 1 below, the features characterizing the Ara­bic particle.</p><p>fact, it didn't specify any constraint to the con­joint compounds.</p><p>The modifications brought to this formalism cover not only the features but also the different schemata of the HPSG grammar. In the follow­ing paragraph, we present as example the con­ceived schema for Arabic coordination.</p></subsection><subsection number="4.2" title="Arabic schemata"><p>HPSG grammar is based on six schemata. In this work, we adapted each schema to represent an Arabic syntactic phenomenon (the simple one). In the context of our work; we present the con­ceived schema for Arabic coordination.</p><p>To represent coordination structure, a compli­cate phenomenon, we have represented, at first, the simple one. In fact, coordination interacts with different others phenomenon. All other rep­resentations were headed structure. However, the coordination has a particular structure. In fact, according to some references, the coordination is a non headed structure. Godard (2003) showed that the conjunction can't be the head of the phrase. In fact, a coordination particle is non op­erative. Thus, it can't specify conjunct elements.</p><p>Therefore, we developed a ternary non headed rule for the coordination phenomenon to obtain the representation below:</p><table caption="Table 1. Arabic particle features"></table><p>Indeed, an Arabic particle can be operative particles or non operative. The coordination par­ticles are classified as non operative particles. In</p><doubt alpha="100.0" length="2" tooSmall="True" monospace="0.0">ST</doubt><doubt alpha="33.3" length="9" tooSmall="True" monospace="0.0">VAL---.-,</doubt></subsection></section><section number="0" title="0 Figure 3. Coordination schema"><p>As represented in figure 3, this structure doesn't contains three non headed daughters: two Fils-conj representing the two compounds of the coordination phrase <i>('aalwaladu, the boy) </i>and '•^1, <i>('aalbintu, the girl) </i>and the Fils-conjonction representing the coordination parti-clej, (wa, <i>and).</i></p><p>To validate our constructed grammar with LKB system, we specified it in TDL (Type De­scription Language). The choice of LKB plat­form is justified. In fact, it generates automati­cally a reliable parser. Some related works such as (Garcia, 2005) used this system and they ob­tained good results.</p><p>In the following paragraph, we give an idea about the specification of the constructed HPSG.</p><table class="main" frame="box" rules="all" border="1" regular="False"><tr class="row"><td class="cell"></td><td class="cell"></td><td class="cell"></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p><b>Features</b></p></td><td class="cell"><p><b>Possible values</b></p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p>PFORM</p></td><td class="cell"><p>- Non operative lW«</p><p>- Operative lU*</p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p>NATP</p></td><td class="cell"><p>- elision particle j?­- Subjunctive</p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"></td><td class="cell"></td><td class="cell"></td></tr></table><page local="4" global="689"/></section><section number="5" title="TDL specification"><p>According to (Krieger and Schäfer, 1994), the TDL syntax presents an important similitude with the HPSG representation. Therefore, the TDL specification was simple. At the present time, our grammar covers the first category of coordination mentioned in section 3: coordina­tion of constituents.</p><p>To specify this grammar, we specified the lex­ical entries AVMs, the type hierarchy and the syntactic rules representing the different forms of coordination and all possible simple sentences (verbal and nominal).</p><p>In the following paragraph, we present an ex­ample of TDL specification of an AVM and some schemata.</p><subsection number="5.1" title="TDL specification of an AVM"><p>From a HPSG representation, the TDL specifica­tion of an AVM is very simple. We present in the following figure the specification TDL of "ii*, that" (hadha).</p><figure caption='Figure 4. TDL Specification of "that" (hadha, ü*)'></figure><p>In fact, the symbol ":=" designates that "ii*", <i>(hadha, this) </i>is an instance of indeclinable nouns. The different constraints are added by the symbol &amp;. The feature structures are delimited by brackets [ ]. Besides, the various attributes values are separated by commas "," and the full stop ". " designates the end of the AVM.</p></subsection><subsection number="5.2" title="TDL specification of a schema"><p>To specify the syntactic rule of Arabic coordina­tion, we started by rules representing simple phenomenon. For the coordination structure, we specified two different rules. The first one repre­sents verbal phrases and sentences. The second one represents nominal phrases and sentences. In the following figure we present the TDL specifi­cation of the coordination of nominal phrases.</p><p><b>regle_coordination_nom := regie-tern-sans-t &amp; [SS.LOC[CAT[TETE[DEFINI oui, DEC tdec],</b> <b>[LOC[CAT[TETE nom s[DEC fdec],</b> <b>[LOC[CAT[TETE particule_non_operative,</b> <b>COMPS &lt;#comps2&gt;]]]]]&gt;].</b><b></b></p><doubt alpha="64.0" length="114" tooSmall="False" monospace="0.0">VAL [TOPIC &lt; &gt;, SPR &lt;#nontetel&gt;, COMPS &lt;#compsl, #comps2&gt;]], CONT.IND [NOMB duel]], BRS.BRS-NTETE &lt;[SS fnontetel S</doubt><doubt alpha="53.5" length="43" tooSmall="False" monospace="0.0">VAL  [COMES &lt;#compsl&gt;]]]]], [SS #nontete2 &amp;</doubt><doubt alpha="58.3" length="84" tooSmall="False" monospace="0.0">VAL [SPR &lt;#nontetel&gt;]]]]], [SS[LOC[CAT[TETE nom &amp;  [DEC Idee], VAL [SPR &lt;#nontete2&gt;,</doubt><figure caption="Figure 5. TDL Specification of the coordination rule"></figure><p>As represented in this figure, this rule joins nominal phrases. It extends from the type regle-tern-sans-t. This type of rules represents non headed structures. In fact, before implementing this syntactic rule, we specified this type of rules. (Figure 6):</p><p><b>regle-tern-sans-t := cegle-ternaire £ [BRS struc-sans-tete £</b></p><p><b>[BRS-NTETE &lt;  SI,   #2,   S3  &gt;] , ARGS </b><i>&lt; </i><b>SI,   #2,   #3  &gt;].</b></p><figure caption="Figure 6. TDL Specification of the type rule regle-tern-sans-t"></figure><p>In fact, regle-tern-sans-t is a ternary rule hav­ing two non-headed daughters joined with a par­ticle of coordination.</p><p>Following the phase of specification TDL, we tested the adapted HPSG grammar with the LKB system. In the next paragraph, we give an idea about this system. Then we describe the experi­mentation and the evaluation of this grammar.</p></subsection></section><section number="6" title="Experimentation and evaluation"><p>LKB system is a parser generation tool, proposed by (Copestake, 2002). This system can run on Windows or on UNIX. But the version on Win­dows can't support Unicode. In fact, LKB is written in LISP using Motif. Therefore, we have added Trollet (TROndheim LingLab Engineering Tool), a tool for multilingual grammar develop­ment. It is easy to extend and can be used only on UNIX system. Therefore we installed Ubunto system. Then we install LKB and Trollet. Thus LKB is embedded in Trollet and invisible for the user. This tool replaced the LKB window.</p><p>It should be noted that the LKB is based on two types of files: TDL files and LISP files. The first type represents the grammar's files (i.e., types.tdl, rsynt.tdl, lexique.tdl). The second type represents files to parameterize the LKB system. Among these files, we can especially mention the file: "script.lsp'. It is a very important file. It<page local="5" global="690"/></p><p><b>103 </b>1 <b><i>ihi*l\ </i></b><b>iLjïl </b><b><i>Ji </i>t^ùJI aiJI <i>jfai\ </i></b><b>jbJI       </b><b>U,Li jCijJI </b>IJjJI Aul <b>488</b></p><doubt alpha="52.2" length="356" tooSmall="False" monospace="0.0">16 1 i ijL. IjjJI jill 489 25 1 oh^ LaiJI ^ 490 16 1 üjjj äi bujI iiaij I,s^jj491 231^kJI jüjljijUJI lj,JI JkLil 492 11 1 ijUI iJjJI jUl 493 18 0 iijlj buiJIol494 301fiU\vtii , iLJI ji IUI tfiii 495 291 u&gt;JI&amp;JI £&gt;•iUli 1)&gt;)I irf « 30 1jJjJIuUl ojjjjfiiJjJI Jj jjj 497 |25 1 £*JI itj« -, ijUJI&lt;iß498 25 1 udjjl iiuji    iJ,JI Jul 499 16 1 It iljj lUJIIii500</doubt><doubt alpha="53.1" length="49" tooSmall="True" monospace="0.0">;;; Total CPU tine:1290isecs ;;; Mean edges:29.35</doubt><doubt alpha="45.5" length="22" tooSmall="True" monospace="0.0">::: Mean parses:1.11_|</doubt><p>charges the grammar on LKB. In fact, it indicates the name and the root of each grammar file. In the following paragraph, we describe the stages of syntactic analysis. Then, we present the ex­perimentation of the constructed grammar.</p><subsection number="6.1" title="Stages of syntactic analyses"><p>After charging the constructed grammar on LKB, this system offers a generated parser to analyze a simple sentence or a corpus of sentences. But, it should be noted that this corpus must be seg­mented in sentences. In fact, LKB system didn't have a module segmenting the tested corpuses.</p><p>To analyze a simple phrase, we have to unroll the "parse" menu and choose the "parse input" order. Thus, the LKB system generates a zone of text to type the sentence and as result, a deriva­tion tree appears. The following figure represents the result of a sentence's analysis.</p><doubt alpha="57.1" length="7" tooSmall="True" monospace="0.0">VÎHB -5</doubt><figure caption="Figure 7. Derivation's tree of a verbal sentence"></figure><p>The derivation's tree in figure 7 is a represen­tation of the following verbal sentence:</p><doubt alpha="60.0" length="60" tooSmall="False" monospace="0.0">&lt;^lc öjjil^ll C±n]l jj?-^l d±L]4*\-i\àj ^bjjJi Aijli Ju^JjXl</doubt><p><i>ahmadu the sick boy and fatimatu the girl of the old neighbour, the little girl, are slept on the new couch</i></p><p>As shown in Figure 7, with TROLLET, we re­solved the problem of transliteration. Contrary to the LKB on Windows, this system on UNIX can now support Arabic language except that the reading direction of the tree is false. Indeed an Arabic sentence must be read from the right to the left. At the present time, we are looking for a solution for this problem.</p><p>The choice of this sentence as example is jus­tified. It represents the interaction of the coordi­nation structure with various syntactic phenom­ena (i.e. annexing, description). In fact, the sub­ject (circled in red in the figure) is a coordination of constituents joining two nominal phrases.</p><p>To analyse a corpus of sentences with LKB, we should specify to it two paths: the path of the file containing the sentences (corpus.txt) com­posing the corpus, and the path of the file that will contain the results (results.txt). In</p><figure caption="Figure 8. Extract of results.txt"></figure><p>Figure 8 represents our test corpus containing 500 sentences which are extracted from different linguistic resources and containing coordination structures using different particles (-à It should be noted that in "results.txt", LKB repre­sents the total time of analyses and for each ex­ample; it represents the number of tree's deriva­tion and the number of nodes in the creation of the derivation's tree.</p><p>According to the obtained results, we evalu­ated our work. This evaluation is described in the following paragraph.</p></subsection><subsection number="6.2" title="Evaluation"><p>To test our HPSG grammar, we used a corpus of 500 sentences. This corpus was created from a lexicon of 2000 words. It covers different sen­tences containing coordination structures using different particles. As we have mentioned before, LKB system or TROLLET didn't have a seg­mentation module to cut sentences composing the corpus. At the present time, it is done manu­ally. Therefore, we missed much time in this phase. But, we are looking for a segmentation tool toad it to our system.</p><p>At the present time, we treated only constitu­ent's cases and we are working on ellipse phe­nomenon and its interaction with the coordina­tion one. According to the file "results.txt", we obtained the following results.</p><table caption="Table 2. Obtained results" class="main" frame="box" rules="all" border="1" regular="False"><tr class="row"><td class="cell"></td><td class="cell"></td><td class="cell"></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p><b>Number of deri-</b></p></td><td class="cell"><p><b>Number of sentences</b></p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p><b>vation's trees (n)</b></p></td><td class="cell"><p><b>having n analysis</b></p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p>0</p></td><td class="cell"><p>50</p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p>1</p></td><td class="cell"><p>420</p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p>2</p></td><td class="cell"><p>30</p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"><p></p></td><td class="cell"><p>500</p></td><td class="cell"></td></tr><tr class="row"><td class="cell"></td><td class="cell"></td><td class="cell"></td><td class="cell"></td></tr></table><page local="6" global="691"/><p>In fact, 84% of sentences were analyzed cor­rectly. The failure cases (0 analyzes) are due to the absence of rules treating some particular syn­tactic phenomena (i.e., relative phenomenon, coordination phenomenon). In fact, at the present time we treated constituent structures. We have not yet treating ellipse phenomenon and its inter­action of the coordination phenomenon. The am­biguous cases (2 analyzes) are due to a no pre­cise specification of the constraints specification of some syntactic rules.</p></subsection></section><section number="7" title="Conclusion and perspectives"><p>In this article, we proposed a typology for coor­dination structure in Arabic language. Based on this hierarchy, we adapted the HPSG grammar. In fact, we defined a particular structure for this phenomenon. Then we specified syntactic rules treating simple sentences and coordination struc­tures. The specified grammar was validated with the LKB system. The experimentation was done on a corpus of 500 sentences. According to the obtained results, we evaluated our grammar.</p><p>As perspectives, we are going to treat ellipse phenomenon and its interaction with coordina­tion structures. Then, we will treat other particu­lar phenomena and specify more constraints to eliminate the ambiguous cases. Moreover we consider developing lexical rules to make our lexicon extensional. Furthermore, we aim to con­struct a converter permitting to convert the lexi­cal entries of XML in TDL in order to facilitate the development of the lexicon.</p></section><references><p>Abdelkader A., Haddar K., and Ben Hamadou A. 2006, <i>study and analyses of nominal sentences in HPSG, </i>TALN, Louvain, 379-388.</p><p>Abdelwahed A. 2004, <i>'alkalima fy 'attourath 'alli-saany 'alaraby, <b>tjHjMl t^l^M djljUl ^LKJl</b></i>, Aladin library, Sfax - Tunisie, 1-100.</p><p>Bahou Y., Hadrich Belguith L., Aloulou C. and Ben Hamedou A. 2005, <i>Adaptation and implementation of HPSG grammars to parse non-voweled Arabic texts, </i>Faculty of Economics and Management of</p><p>Sfax.</p><p>Blache P. 2001, <i>Propriety grammar : constraints for NL, </i>Hermès Sciences, Paris.</p><p>Biskri I., Desclés J., 2006, <i>coordination of different categories in French, </i>Quebec university, Canada.</p><p>Bourigault D.and Fabre C. 2000, <i>Linguistic approach for syntactic analyse of corpuses, </i>Sémantisue et corpus, 131-151.</p><p>Copestake A. 2002, <i>Implementing Typed Feature Structure Grammars, </i>CSLI Publications, Stanford University.</p><p>Dahdah A. 1992, <b><i>j Jj'A&gt;. ^ ^njMl Âilll jcljif^, </i></b>Librairie de Nachirun ebanon, 5ème edition.</p><p>Elleuch S. 2004, <i>syntactic analyse of arabic language based on HPSG formalism, </i>DEA memory on In­formation system and new technology, 55-88.</p><p>Garcia O. 2005, <i>Une introduction à l'implémentation des relatives de l'espagnol en HPSG-LKB, </i>Re­search memory.</p><p>Godard D. 2003, <i>Syntacyic problems of coordination and recent propositions in syntagmatic grammars, </i>study journey in the university of Paris 7.</p><p>Haddar K. 2000, <i>Formel caracterization of Arabic ellipse and recouvrement processus in Arabic lan­guage, </i>University of Tunis II - Sciences Faculty of Tunis.</p><p>Hudson R. 1976, <i>Conjunction Reduction, gapping and right-node raising, </i>535-562.</p><p>Krieger H. and Schäfer U. 1994, <i>TDL: A Type De­scription Language for HPSG, </i>Part 1 and Part 2, Research Report, RR, 94-37.</p><p>Le Roux J., Perrier G. 2006, <i>coordination modelisa­tion in Interactive grammars, </i>TAL, Volume 47, n° 3, 89-113.</p><doubt alpha="61.7" length="47" tooSmall="False" monospace="0.0">Maaloul H., Haddar K., and Ben Hamadou A. 2004,</doubt><p><i>Arabic coordination: HPS analyse, </i>MCSEAI 2004, 8th maghrébine conference on GL and IA, Sousse,</p><p>Tunisie : 487- 498.</p><p>Osenova P. and Simov K., <i>Special Linguistic phe­nomena in the Bulgarian HPSGbased Treebank, </i>BulTreeBank Project, Linguistic modelling labora­tory, IPP, Bulgarian Academy of Sciences.</p><p>Pollard C. and Sag I. 1994, <i>Head-drive phrase struc­ture grammars, </i>CSLI series, Chicago University Press.</p><p>Postal P. 1974, <i>On raising, </i>MIT Press, Cam­bridge/MA.</p><p>Rau F. L. 1985, <i>the understanding and generation of ellipses in a natural language system, </i>Berkeley Ar­tificial Intelligence Research project.</p><p>Villavicencio A., Sadler L. and Arnold D. 2005, <i>An HPSG account of closest conjunct Agreement in NP Coordination in Portuguese, </i>Proceedings of the HPSG05 conference, Department of Informa­tions, University of Lisbon.</p></references></body></article>