The Treatment of Gerund Forms for Arabic Nouns with LKB System

Samia Ben Ismail

, Sirine Boukédi

and Kais Haddar

ISITCom Hammam Sousse, Sousse University, Miracl laboratory, Tunisia

National Engineering School, Gabes University, Miracl laboratory, Tunisia

Faculty of Science of Sfax, Sfax University, Miracl laboratory, Tunisia

Keywords: Arabic Gerund Forms, Head-Driven Phrase Structure Grammar (HPSG), Linguistic Knowledge Building

(LKB) System, Type Description Language (TDL).

Abstract: The treatment of morphological phenomena is important in Natural Language Processing (NLP), especially

using a unification grammar. In Arabic grammar, the gerund is considered one of the most delicate

morphological structures since it changes the grammatical category. Thus, we present in this paper, a Head-

driven Phrase Structure Grammar (HPSG), treating Arabic gerund forms. The elaborated grammar is

specified with Type Description Language (TDL) and validated on Linguistic Knowledge Building (LKB)

system. The obtained results were encouraging, which proves the effectiveness of our system.

1 INTRODUCTION

Arabic language is a Semitic language that is very

rich by morphological phenomena (i.e. inflectional

and derivational). Thus, the automatic treatment of

Arabic morphological forms is primordial for

syntactic analyzer. It contributes to the construction

of extensional lexicon with a wide coverage and

guarantees the reusability of the resources, mostly

using a unification grammar. In fact, this kind of

formalism offers complete representation with a

minimum number of rules. Among the most

complicated derivational forms, we find the gerund

for Arabic nouns.

However, works treating Arabic gerunds

especially with HPSG are very limited or almost

non-existent. Indeed, gerund has several forms and

its treatment isn’t evident. It is based on several

criteria (i.e., the type the scheme of verb and

semantic aspect), that’s make difficult to find the

hierarchy type representing the proposed patterns.

In this context, we are interested in generating

the gerund within LKB. To do this, we begin by

studying the Arabic gerund to generate its different

characteristics. Based on this study, we identified the

different constraints characterizing each type of

Arabic gerund. These constraints were described

using AVMs including a set of features. To validate

the constructed HPSG, the elaborated patterns were

specified in TDL. The originality of our work

appears in the use of highly theoretical formalisms

like HPSG to model TALN phenomena and

applications. In addition, relying on the object-

oriented paradigm derived from TDL to specify the

hierarchy of types of words shows an interaction

between computer science and linguistics.

Moreover, the lack of researchers treating the Arabic

morphology especially gerund forms with LKB

platform represents another novelty.

In the present paper, we begin by describing and

discuss some previous works treating Arabic

morphological aspect. After that, we present a

detailed linguistic study about the Arabic gerund.

According to this study, we present, in the next

section, the elaborated HPSG grammar for Arabic

gerund and its TDL specification. Then, we

experiment and evaluate the different gerund forms

with LKB system. Finally, we conclude our work

and we give our perspectives.

2 PREVIOUS WORK

The literature showed that there exist two main

approaches used in NLP domain: statistical and

linguistic ones. Moreover, despite works are

evolved, Arabic language is still among languages

that have less linguistic resources. The lack appears

during the grammar generation with such formalism

especially for morphological analyzer.

Ismail, S., Boukédi, S. and Haddar, K.

The Treatment of Gerund Forms for Arabic Nouns with LKB System.

DOI: 10.5220/0006932302150222

In Proceedings of the 10th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2018) - Volume 2: KEOD, pages 215-222

ISBN: 978-989-758-330-8

215

The research work (Shahrour, 2016) develops an

Arabic syntactic parser called CamelParser. This

system is based on the art state of MADAMIRA

morphological disambiguated (Pasha, 2014). It

integrates also the notion of optimization using an

adapted version of MaltOptimizer (Ballesteros,

2012). In addition, this system produces several

output formats such as plain text and tree file that it

is in ‘.fs’ format. The CamelParser system was

evaluated with 35,750 words. To further enrich the

evaluation, the authors evaluate its system on two

steps: Parsing Accuracy and Morphological

Disambiguation Accuracy. First, this system is

compared to the Baseline Parser (Shahrour, 2015).

Moreover, the authors have calculated some

accuracy metrics that are labeled attachment,

unlabeled attachment, and label accuracy. In fact,

CamelParser achieves 83.8%, 86.4%, and 93.2%,

respectively. However Baseline Parser achieves

81.6%, 84.6%, and 92.0% respectively. For the

second step, the authors compare the performance of

the morphological disambiguation between its

system and MADAMIRA system. In fact, they

calculate two types of metric: full word

diacritization accuracy and all morphological feature

selection. Indeed, CamelParser attains 90.8% and

88.7% while MADAMIRA attains 88.1% and 86.1%

for these two metrics respectively.

For (Khalifa, 2016), YAMAMA (Yet Another

Multi-Dialect Arabic Morphological Analyzer) is a

multi-dialect Arabic morphological analyzer

combines the rich output of MADAMIRA with fast

and simple out-of-context analysis. This system is

motivated by FARASA approach (Abdelali, 2016).

Moreover, it uses the same database of

MADAMIRA. Then, it creates the maximum

likelihood model selecting for each word the most

frequent analysis. Next, these selected analyses are

saved in a dictionary that is loaded once when the

system runs. For out-of-vocabulary words,

YAMAMA ranks for each words all of the analyses

using the language models of the lemma and the

Buckwalter POS tag. In fact, the analyses include all

the morphological and lexical features as in

MADAMIRA. Moreover, YAMAMA’s output is in

the same format as MADAMIRA’s. To evaluate this

system, the authors make two types of experiments.

The first is the targets accuracy and speed while the

second is the targets machine translation quality.

Indeed, they compare its system with tow systems

MADAMIRA and FARASA. As result, YAMAMA

is five times faster than MADAMIRA but FARASA

is four times faster than YAMAMA.

These works, describing above, construct their

system to analyze the Arabic morphology and

syntactic aspect. However, the output representation

is not with a standard structure, although, the

standardization is considered as a major factor in the

reusability of NLP applications.

Thus, we find other works using HPSG to

analyze Arabic language such as (Shadiqul Islam,

2010), (Ben Ismail, 2017a) and (Ben Ismail, 2017b).

(Shadiqul Islam, 2010) proposes an HPSG

representation for Arabic derived nouns. In fact, this

works treat eight types of noun derived from a verb

such as Gerund and Active participle. The authors

describe the morphology aspect of these Arabic

nouns by extending the features of MORPH, SYN

and SEM. This work treats derived noun just from

trilateral sound Form I (i.e the schema is











 / fa’ala)

within TRALE platform. Moreover, for the Gerund

type, he treats just one type of gerund that it is

Gerund-Mojared.

Next, (Ben Ismail, 2017a) construct an HPSG to

generate the extensional forms of Arabic language.

In fact, this work treats the verb conjugation and

noun regular plural. It generates all the forms of verb

and noun. In fact, it adds 10000 verbs (canonical

forms) and 500 nouns (singular forms). Moreover,

the description of the extensional forms is based on

morphological rules added to the elaborated

grammar. As evaluation, the authors calculate

performance percent of the system to verify the full

description of generated forms which attains 87%.

Furthermore, the authors in (Ben Ismail, 2017b)

treat the derivational forms for Arabic (such as

Active participle, Passive participle) with HPSG

grammar. This work used a linguistic approach to

develop their morphological rules. Although, the

elaborated grammar constructs all the derived forms

for all types of verb but didn’t construct gerund

forms. These two works described above use LKB to

generate their HPSG grammars.

Thus, for Arabic language, most of research are

focused either on the construction of its system or

the generation of regular forms for Arabic words

represented with HPSG formalism. So, we observe

that gerund forms are treated in some works but not

with a complete manner. In fact, all the treated forms

are manipulated only for simple cases.

For this in the following section, we start by

presenting a linguistic study on Arabic gerund to

detail the manner of construction.

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

216

3 LINGUISTIC STUDY ON

ARABIC GERUND

Referring to (Ammar, 1999) and (Dahdah, 1992),

the gerund is a derivational noun obtained from an

Arabic verb. It expresses the principal idea of an

action, or an action that has no reference time. As

represented in Figure 1, there exist seven sub-types.

Figure 1: Gerund type hierarchy.

Each sub-type of gerund, illustrated in Figure 1,

has specific criteria representing the base for its

construction. First, the Gerund-Mojared “ ”

can be obtained from an un-augmented verb ( 

 / fi’l mojared). In addition, this type of Gerund

has several types of forms (i.e. regular and

irregular). To construct regular forms, for example,

we are based on the schema of verb by adding it’s

semantic or transitivity (i.e. transitive or intransitive)

descriptions. Table 1 illustrates some regular forms

of Gerund-Mojared.

Table 1: Example of Gerund-Mojared forms.

Scheme of

verb

Semantic / Transitive Examples











 (fa’ala)

Semantic

A trade

meaning

 –>













to plant-> the

planting

A voice

meaning

–> 

to scream-> the

screaming

Transitive

intransitive

 –> 

to sit-> the

seating

As shown in Table 1, for example, if an Arabic

verb has the scheme fa’ala and its semantic

description is a voice meaning, the gerund-Mojarad

is obtained according the pattern fu’oul which adds

the letter “w/” before the last letter of the verb.

While, the irregular forms for the gerund-

Mojarad can have several patterns either by adding

one or more letters such as “(,w )/ (,y) / (,t) /

(,A)”, either by modification of vocalization, or by

the combination of two cases of change. For

example, the two verbs ( <to open> /  <to

plant>) have the same schema (fa’ala_yaf’alou







_

) and the same type “Intact” but we can’t

apply the same rule to obtain the Gerund-Mojarad

( 







 <the opening> /











 <the planting>). In fact,

the first forms of gerund-Mojarad (i.e. 







 <the

opening>) is an irregular form obtained by

modifying the vocalization.

However, most of gerund-Almazid “”

forms can be obtained with regular method applied

to an augmented Arabic verb (  / fi’l mazid).

In fact, this type has 15 patterns according to the

schema of verb.

By the way, for the gerund-Mimi “

 

”,

we add generally, the prefix “” to an Arabic verb.

If the verb has three letters, the vocalization of the

prefix will be either “



/ma” or either “



/mou”.

For the gerund-Alsinaa’i“

 

”, it is a

noun that can be obtained from a participle noun or

active noun or superlative noun or proper noun or

another type of gerund such as the gerund-Mimi. In

fact, for each type of this noun, we add the sign of

the feminine, the letter “ /t”. For example the active

participle “ /Scientist/’aalimoun” will be “



 /

International/’aalamiyatun” for the gerund-Alsinaa’i.

The Gerund-Almarra “



 ” means that the

verb is applied only once. In fact, if the verb is

composed by three letters, this type of gerund

became transformed to the schema “ / fi’latun”.

But otherwise, this type became transformed

according to schema of its verb with adding the

letter “ / t” at the end. Moreover, with the same

way, we can obtain the gerund-Alnaw’i “ ”

but we can distinguish between them if they appear

in a sentence as illustrated in examples (1) and (2).

 

Wathabtu wathbatan, I leaped a leap

(1)

  

Wathabtu wathbata alasadi, I leaped the

lion’s leap

(2)

The gerund-Almarra should be a simple word as

shown in sentence 1. However, the gerund-Alnaw’i

should have after it a word describing the way of

this type of gerund as illustrated with the sentence 2.

According to this linguistic study, we conclude

that each sub-type of gerund has specific patterns.

Indeed, we need for some patterns the semantic level

of the verb mainly for the gerund-Mojarad

construction. In the next section, we describe the

elaborated Arabic HPSG representing the gerund

patterns.

The Treatment of Gerund Forms for Arabic Nouns with LKB System

217

4 ELABORATED HPSG

GARAMMAR FOR ARABIC

GERUND

HPSG (Pollard, 1994) is a unification grammar

based on typed feature structure called AVM

(Attribute Value Matrix) and a set of schemata.

According to some previous works such as (Ben

Ismail, 2017b), (Shadiqul Islam, 2010), (Boukedi,

2014) and based on our linguistic study (Dahdah,

1992), we adapted the HPSG representation for

Arabic gerund.

As shown in Figure 2, the used lexical rule

transforms the verb “ / zar’a/ to plant” to the

gerund-Mojarad “/ ziraa’atun/ the planting”.

Indeed, this rule adds the feature “ARGS” to

describe the original verb. In fact, this verb must

have the meaning of “hirfa” in its semantic

description presented in RELN feature. Moreover, it

must be a triliteral verb with the scheme “fa’ala”

represented respectively in RADICAL and

SCHEME features. Besides, all types of gerund are a

derived noun obtained by a specific pattern

described in the feature SingSCHEME.

Figure 2: AVM of the gerund-Mojarad “ziraa’atun” after

application of a morphological rule.

As we mentioned in our linguistic study, the

gerund-Alsinaa’i can be obtained for example from an

active participle that it is also a derived form. This

type of form is obtained from an Arabic verb. In fact,

the HPSG representation of this transformation is

illustrated in the following figure (Figure 3).

Figure 3: AVM of the gerund-Alsinaa’i < ‘aalamiyyatun>

after application of a morphological rule.

Figure 3 shows the HPSG representation of the

derived form by detailing the process of

transformation (verb -> active participle -> gerund-

Alsinaa’ii). In fact, each step of this process is

represented in the feature “ARGS”.

While for the HPSG representation of the

gerund-Alnaw’i and the gerund-Almarra is shown in

the following figure.

Figure 4: AVM description of the gerund-Alnaw’i

“wathbatun”.

As illustrated in Figure 4, these two sub-types

need in the HPSG representation the feature ROOT,

a constraint feature used for syntactic analyzer.

Moreover, to distinguish between them, we must add

the specification schemata in the HPSG

representation. In fact, if the feature SPR is not an

empty element, the type of gerund is either “gerund-

alnaw’i” or “gerund-Almarra” in contrary (i.e. SPR

is an empty element).

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

218

After the HPSG representation of gerund, we can

validate the elaborated grammar on LKB system. In

addition, we should specify it with a description

language. In the next section, we present the TDL

specification of HPSG Arabic gerund.

5 TDL SPECIFICATION

To implement the proposed HPSG grammar for

Arabic gerund within LKB system, it is necessary to

specify it in TDL. Indeed, TDL syntax (Krieger,

1994) is very similar to HPSG and based on typed

features connected by a set of principles, especially

inheritance.

Figure 5 below, illustrates the type specification

of the Arabic gerund in TDL.

nom := tete&[MAJ "nom",

NTYPE ntype,

NFORM nform].

nom_variable:= nom &

[NFORM ,

NGENRE ngenre,

NRADICAL nradical,

NAT nat,

SingSCHEME nscheme,

ADJ boolean,

ROOT string,

DEFINI boolean].

nom_variable_derive:= nom_variable & [NFORM

_].

nom_origine:= nom_variable_derive & [NTYPE

].

 := ntype.

 =:



_.

 =:_.

 =:_.

 =:_.

 =:



_.

 =:_.

Figure 5: Type specification of gerund.

In Figure 5, Arabic gerund is inherited from the

variable derived noun that is inherited from the

variable noun type and the noun that is a general

type. As well as the Arabic noun is inherited from

the base sign “tete”. Since the gerund is a derived

form from an Arabic verb, it is necessary also to

specify the type hierarchy of Arabic verbs (i.e.

type.tdl and lex-type.tdl). Figure 6 shows the type

specification of Arabic verbs.

lex-verb := lexeme & [SS [LOC [CAT [TETE verbe,

VAL [ SUJ < [LOC [CAT [TETE nom,

VAL[COMPS< >]],

CONT.IND #indice ],

NONLOC.REL <! !>] > ],

MARQUE non-marque],

CONT [IND referentiel,

RELS <! [ARG1 #indice,

RELN rel-sem-verb] !>]],

NONLOC.REL <! !>]].

verbe := tete &

[MAJ "verbe", TYPE type,

RADICAL radical,

SCHEME scheme,

VFORM vform,

ROOT string].

Figure 6: Type specification of Arabic verb.

As we mentioned in our linguistic study, the

patterns construction of Arabic gerund need a

complete description of the verb (illustrated in

Figure 5). In fact, the semantic description is

represented in the feature “RELN”. Moreover, the

morphological description is represented in the

features: “TYPE”,”RADICAL”, and “SCHEME”,

“VFORM” and “ROOT.

After the type specification of each type of word

(i.e. noun and verb), we specify all entries with its

features and values in the file “lexicon.tdl”. Figure 7

shows the entry specification of an Arabic verb.

 := lex-verb-complet-sain-intact &

[PHON <! "" !>,

SS.LOC [CAT[TETE[ RACINE "",

SCHEME















_











,

RADICAL 



_,

VFORM ]],

CONT.RELS<![RELN ]!>]].

Figure 7: Verb “” with TDL syntax.

As shown in Figure 7, the verb “/ zaraa/ to

plant

” is an instance from “lex-verb-complet-sain-

intact”. This lexical rule contains all the verb of type

intact “”. Each verb is specified by phonetic

feature “PHON”, morphological and semantic

features. In fact, this verb is a triliteral “



_”

and transitive “” verb. Besides, its semantic

description is a craft “

/ hirfa”. This constraint is

necessary in the specification of the morphological

rules to generate the gerund-Mojarad.

We give in

Figure 8 an example of these morphological rules.

The Treatment of Gerund Forms for Arabic Nouns with LKB System

219

%(letter-set (!f                 ))

%(wild-card (?t ))

%(wild-card (?a ))

%(wild-card (?u  ))

gernud-trileteral-nu1 :=

%suffix(!f ?a!f?t)

l2m-flex &

[SS [LOC.CAT [TETE [NFORM _, NTYPE





_, SingSCHEME ger2, RACINE #string,

NAT _ ,NGENRE _, DEFINI , DEC no-

dec]]],

ARGS < [SS[LOC[CAT[TETE [TYPE TYPE1,

SCHEME scheme_mas, RACINE #string]],

CONT.RELS <![ RELN ]!>]]]>].

gernud-trileteral-nu12 :=

%prefix(!p !p?u)

m2m-flex &

[SS [LOC.CAT [TETE [NFORM _, NTYPE





_, SingSCHEME 



, ADJ  ,RACINE

#string,DEFINI , NAT 

_ ,NGENRE

_,NRADICAL __ ,DEFINI ]]],

ARGS<[SS.LOC.CAT.TETE[NFORM_,

NTYPE 



_, SingSCHEME ger2,RACINE

#string, NAT _ ,NGENRE _, DEFI ]]>].

Figure 8: Example of morphological rule applied to

generate a gerund-Mojarad.

As shown in Figure 8, we added two letters to

the last letter of the verb. They belongs to the set of

letter called “!f”: one before “/a” and another after

“/t”. This rule is applied to a set type of verbs

combined in the type “TYPE1” such as intact verb.

Moreover, these types of verb must belong to a set

of scheme regrouped also in “scheme_mas”. This

rule combines all forms of gerund-Mojarad

generated in the same manner.

As the gerund can be a derived form from a verb,

it can be also a derived form from an active

participle (a derived form from a verb). This type of

gerund is called the gerund-Alsinaa’i (Figure 9).

active-participle-trileteral :=

%prefix (!p !p?a) l2m-flex &

[SS[LOC.CAT[TETE [NFORM _, NTYPE

_,RACINE #string, DEFINI , DEC

_]]],

ARGS < [SS.LOC.CAT.TETE[TYPE type, RADICAL





_,RACINE #string]] >].

gerung-sina3ii :=

%suffix(* 



) m2m-flex &

[SS[LOC.CAT [TETE [NFORM _, NTYPE

_, SingSCHEME 



,DEFINI ,ADJ 

,RACINE #string, DEFINI , NAT _

,NGENRE _,NRADICAL __ ,DEFINI

]]], ARGS<[SS.LOC.CAT.TETE[NTYPE _,

RACINE #string]] >].

Figure 9: Example of morphological rule applied to

generate a gerund-Alsinaa’i.

In Figure 9, we give two morphological rules

generating the active participle from a verb then a

gerund-Alsinaa’i from an active particle. These two

rules are related and specified with a chronological

order. This order is treated through morphological

operations “l2m-flex and m2m-flex” that allow the

concatenation respectively between a lexeme/word

and word/word with adding specific and proper

constraints to generate each step.

gerung-mara-naw :=

%suffix(* )

l2m-flex &

[SS[LOC.CAT [TETE [NFORM _,

NTYPE ntype_masd, SingSCHEME , ADJ 

,RACINE #string, DEFINI , NAT _

,NGENRE _,NRADICAL __ ]]],

ARGS < [SS.LOC.CAT.TETE[TYPE type,

RADICAL 



_, RACINE #string]] >].

Figure 10: Example of morphological rule applied to

generate a gerund-Alnaw and a gerund-almarra.

Figure 10 illustrates the morphological rule that

generate the two type of gerund: gerund-Alnaw’i and

gerund-Almarra. These types of gerund have as

SingSheme “ / fi’latun”. In fact, we add the letter

“ /t” as a suffix to an Arabic verb.

During the steps of specification, we created five

TDL files; three for the type specification, one for

the lexicon and one for the morphological rules

specification containing 33 rules to generate the

Arabic gerund. Therefore, in the following, we

present our obtained results with LKB.

6 EXPERIMENTATION AND

EVALUATION

To experiment and evaluate the established

grammar, we used LKB (Linguistic Knowledge

Building). This system applies its own algorithms

and generates a reliable analyzer (Copestake, 2002).

It is used to validate unification grammars based on

constraints and feature structures. In fact, this

platform is composed from two types of files: lisp

files (i.e. files system configuration) and TDL files

representing the elaborated grammar. Moreover,

LKB is compatible with several operating systems

such as Windows, Linux, and even Solaris.

In our work, we developed 5 TDL files

describing the gerund Arabic grammar such as the

lexicon file “lexicon.tdl”. This file contains 10000

verbs as lexemes. Besides, as we already mentioned,

our morphological rules are specified in the file

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

220

“rlex.tdl”. In fact, since to inheritance principle that

is based in HPSG formalization, we specified our

rules with the optimization aspect. Therefore, we

grouped different types of verbs constructing the

gerund with the same manner. In fact, in Table 2, we

show the number of rules for each type of gerund.

Table 2: Rule numbers of gerund types.

Gerund type Number of rules

Gerund-Mojarad 16

Gerund-Almazid 34

Gerund-Alsinaii 1

Gerund-Mimi 5

Gerund-Almarra/Alnawa 1

As shown in Table 2, the number of rule for

gerund-Mojarad specified is 16. After application of

the added rules, LKB platform adds automatically,

nine morphological features describing this gerund

form such as NTYPE and SingSCHEME. Moreover,

this platform generates an adequate derivation tree

that proves the effectiveness of our system. Thus,

Figure 10 illustrates an example of our result

obtained with LKB. It shows the generation of the

gerund form “ / ziraa’atun/ the planting” from

the canonical form of verb “/zara’a/to plant”. We

can note also in this figure all the adding

morphological features. Moreover, the description of

the gerund’s origin verb is added in the feature

“ARGS”. In fact, for this example, it is an intact

verb defining in the lexical rule called “lex-verb-

complet-sain-intact”.

Figure 11: Example of derivation tree of gerund-Mojarad.

At the same way, the generation of gerund-

Alsinaa’i is illustrated in Figure 12. As we

mentioned above, this type of gerund can have two

generation steps. In fact, the full process of

generation is represented in the following figure.

Figure 12: Example of derivation tree of gerund-ALsinaii.

Figure 12 illustrates the generation of the

gerund-Alsinaa’i “



 /’aalamiyatun”. In addition,

all the process of generation is represented in the

feature “ARGS”. In fact, this feature describes the

active participle “ /’aalim” and the verb “

/’alima”. Moreover, all the morphological features

of this type of gerund are represented.

As we indicate in figures above, HPSG

morphological representation of all types of gerund

is generally complete excepting some cases. The

obtained average, average of performance (P), of the

correct features automatically added is given in

Table 3.

P= (total number of correct features

automatically added) / (total number of features

automatically added)

(3)

Table 3: Average of performance.

Gerund type Average of performance

gerund-Mojarad 100%

Gerund-Almazid

100%

gerund-Alsinaii 100%

Gerund-Mimi

100%

Gerund-Almarra/Alnawa

88%

Total

96%

Table 3 presents the performance of our

generated system. In fact, the obtained values (96%)

prove the effectiveness of our proposed

transformation system. However, the percent of

failure is because of the ambiguous information for

the morphological feature “NTYPE” for the gerund

type: gerund-almarra/alnawii. In fact, for example,

The Treatment of Gerund Forms for Arabic Nouns with LKB System

221

for the gerund “the leap/ ”, our system can

generate this form but its HPSG representation is not

completed. Indeed, the proper value of the feature

NTYPE is ambiguous; it can be either gerund-

Alnaw’i or gerund-Almarra at the same time. In fact,

as we showed above, these two derived type have

the same process of generation. So, these

ambiguities can be eliminated just with syntactic

rule as we mentioned in our linguistic study and we

illustrated in our HPSG representation.

7 CONCLUSION

In this paper, we have developed a system to

generate all type of gerund within LKB. Based on

linguistic approach, this system elaborates Arabic

HPSG grammar specified in TDL. For the

experimentation and the evaluation phases, we tested

several types of gerund. Therefore, as shown in the

evaluation phase, our system can represent all the

morphological features of each type of gerund that

prove its effectiveness.

As perspectives, we aim to treat other irregular

morphological phenomena such as Arabic

agglutination. In addition, we plan to extend our

Arabic HPSG grammar to treat all types of

morphological phenomena. Moreover, we aim to

integrate, in our system, syntactic rules to test our

established grammar on Arabic corpora.

REFERENCES

Abdelali, A., Darwish, K., Durrani, N., Mubarak, H.,

2016. Farasa: A fast and furious segmenter for Arabic.

In Proceedings of the 2016 Conference of the North

American Chapter of the Association for

Computational Linguistics: Demonstrations, pp 11–

16, San Diego, California.

Ammar, S., Dichi Y., 1999.

   



 

Bescherelle collection. Hatier, Paris, ISSN 0990 3771.

Ballesteros, M., Nivre, J., 2012. Maltoptimizer: an

optimization tool for maltparser. In Demonstrations,

13th Conference of the European Chapter of the

Association for Computational Linguistics, pp 58–62.

Ben Ismail, S., Boukédi, S., Haddar, K., 2017a. LKB

generation of HPSG extensional lexicon. In

AICCSA’17, 14th ACS/IEEE International Conference

on Computer Systems and Applications, pp. 944-950,

Hammamet-Tunisia.

Ben Ismail, S., Boukédi, S., Haddar, K., 2017b.

Transformation system to generate derivational forms

of an Arabic verb with HPSG. In ICEMIS’17: 3th

International Conference on Engineering & MIS/

IEEE, pp 1-5, Monastir-Tunisia.

Boukedi, S., Haddar, K., 2014. HPSG Grammar Treating

of Different Forms of Arabic Coordination. Research

in Computing Science 86, pp 25-41.

Copestake, A., 2002. Implementing Typed Feature

Structure Grammars. CSLI Publications, Stanford,

CA.

Dahdah, A., 1992. mʿǧm qwāʿd āllġt ālʿrbyt fy ǧdāwl w

lwḥāt, Librairie de Nachirun Lebanon, 5ème édition,.

Khalifa, S., Zalmout, N., Habash, N., 2016.YAMAMA:

Yet Another Multi-Dialect ArabicMorphological

Analyzer. COLING’16, 26th International Conference

on Computational Linguistics: System Demonstra-

tions, pp 223–227, Osaka, Japan, December 11-17.

Krieger, H., Schäfer, U., 1994. TDL : A Type Description

Language for HPSG. Part 2: User guide”, Technical

reports, Deutsches Forschungszentrum für Künstliche

Intelligenz, Saarbrücken, Germany.

Pasha, A., Al-badrashiny, M., Diab, M., R. El Eskander,

A. Kholy., Habash, N., Pooleery, M., Rambow, O.,

Roth, R. M., 2014. MADAMIRA: a fast,

comprehensive tool for morphological analysis and

disambiguation of Arabic”, In: Proc. 9th Lang.

Resour. Eval. Conf, pp 1094–1101.

Pollard, C., Sag, I., 1994. HeadDriven Phrase Structure

Grammars, CSLI Lecture Notes,Chicago University

Press.

Shadiqul Islam, Md., Shariful Islam Bhuyan, Md., Ahmed,

R., 2010. Arabic Nominals in HPSG: A Verbal Noun

Perspective. In HPSG’10, 17

International

Conference on Head-driven Phrase Structure

Grammar, pp 158-178.

Shahrour , A., Khalifa, S., Habash, N., 2015. Improving

Arabic diacritization through syntactic analysis. In

Proceedings of the 2015 Conference on Empirical

Methods in Natural Language Processing,

EMNLP’15, pp1309–1315, Lisbon, Portugal.

Shahrour, A., Khalifa, S., Taji, D., Habash, Nizar, 2016.

CamelParser: A System for Arabic Syntactic Analysis

and Morphological Disambiguation. COLING’16,

26th International Conference on Computational

Linguistics: System Demonstrations, pp 228–232,

Osaka, Japan, December 11-17.

KEOD 2018 - 10th International Conference on Knowledge Engineering and Ontology Development

222