An Open Source Urdu Resource Grammar

Shafqat M Virk Muhammad Humayoun Aarne Ranta

Department of Applied IT Laboratory of Mathmatics Department of CS & Eng
University of Gothenburg University of Savoie University of Gothenburg
virk@chalmers.se mhuma@univ-savoie.fr aarne@chalmers.se

Abstract from the dialect of Delhi region called khari boli

(Masica, 1991).
We develop a grammar for Urdu in We develop a grammar for Urdu that addresses
Grammatical Framework (GF). GF is a problems related to automated text translation
programming language for defining using an Interlingua approach and provide a way
multilingual grammar applications. GF to precisely translate text. This is described in
resource grammar library currently Section 2. Then we describe different levels of
supports 16 languages. These grammars grammar development including morphology
follow an Interlingua approach and (Section 3) and syntax (Section 4). In Section 6,
consist of morphology and syntax we discuss an application in which a semantics-
modules that cover a wide range of driven translation system is built upon these
features of a language. In this paper we components.
explore different syntactic features of the
Urdu language, and show how to fit them 2. GF (Grammatical Framework)
in the multilingual framework of GF. We
also discuss how we cover some of the GF (Grammatical Framework, Ranta 2004) is a
distinguishing features of Urdu such as, tool for working with grammars, implementing a
ergativity in verb agreement (see Sec programming language for writing grammars
4.2). The main purpose of GF resource which in term is based on a mathematical theory
grammar library is to provide an easy about languages and grammars1. Many
way to write natural language multilingual dialog and text generation
applications without knowing the details applications have been built using GF. GF
of syntax, morphology and lexicon. To grammars have two levels the abstract and the
demonstrate it, we use Urdu resource concrete syntax2. The abstract syntax is
grammar to add support for Urdu in the language independent and is common to all
work reported in (Angelov and Ranta, languages in GF grammar library. It is based on
2010) which is an implementation of common syntactic or semantic constructions,
Attempto (Attempto 2008) in GF. which work for all the involved languages on an
appropriate level of abstraction. The concrete
1. Introduction syntax is language dependent and defines a
mapping from abstract to actual textual
Urdu is an Indo-European language of the Indo- representation in a specific language2. GF uses
Aryan family, widely spoken in south Asia. It is the term ‘category’ to model different parts of
a national language of Pakistan and one of the speech (e.g verbs, nouns adjectives etc.). An
official languages of India. It is written in a abstract syntax defines a set of categories, as
modified Perso-Arabic script from right to left. well as a set of tree building functions. Concrete
As regards vocabulary, it has a strong influence syntax contains rules telling how these trees are
of Arabic and Persian along with some linearized. Separating the tree building rules
borrowing from Turkish and English. Urdu is an (abstract syntax) from linearization rules
SOV language having fairly free word order. It (concrete syntax) makes it possible to have
is closely related to Hindi as both originated
________________________________ multiple concrete syntaxes for one abstract. This
1
http://www.grammaticalframework.org
2
In given example code ’fun’ and ’cat’ belongs to abstract syntax, ’lin’ and ’lincat’ belongs to concrete syntax

153
Proceedings of the 8th Workshop on Asian Language Resources, pages 153–160,
Beijing, China, 21-22 August 2010. 2010
c Asian Federation for Natural Language Processing
makes it possible to parse text in one language syntax shows how these words (parts of speech)
and translate it to multiple languages. are grouped together to build well formed
Grammars in GF can be roughly classified into phrases. In this section we show how this works
two kinds: resource grammars and application and is implemented for Urdu.
grammars. Resource grammars are general
purpose grammars (Ranta, 2009a) that try to 4.1 Noun Phrases (NP)
cover the general aspects of a language
linguistically and whose abstract syntax encodes When nouns are to be used in sentences as part
syntactic structures. Application grammars, on of speech, then there are several linguistic
the other hand, encode semantic structures, but details which need to be considered. For
in order to be accurate they are typically limited example other words can modify a noun, and
to specific domains. However, they are not nouns have characteristics such as gender,
written from scratch for each domain, but they number etc. When all such required details are
use resource grammars as libraries (Ranta grouped together with the noun, the resulting
2009b). structure is known as noun phrase (NP). The
Previously GF had resource grammars for 16 basic structure of Urdu noun phrase is, “(M) H
languages: English, Italian, Spanish, French, (M)” according to (Butt M., 1995), where (M) is
Catalan, Swedish, Norwegian, Danish, Finish, a modifier and (H) is the head of a NP. Head is
Russian, Bulgarian, German, Interlingua (an the word which is compulsory and modifiers can
artificial language), Polish, Romanian and or cannot be there. In Urdu modifiers are of two
Dutch. Most of these languages are European types pre-modifiers i.e modifiers that come
languages. We developed resource grammar for before the head for instance (

 kali: bli:
Urdu making it the 17th in total and the first “black cat”), and post-modifiers which come
south Asian language. Resource grammars for after the head for instance (
 tm sb “you
several other languages (e.g. Arabic, Turkish, all”). In GF resource library we represent NP as
Persian, Maltese and Swahili) are under a record
construction.
lincat NP : Type = {s : NPCase => Str ; a :
3. Morphology Agr} ;

In GF resource grammars a test lexicon of 350 where

words is provided for each language. These
words are built through lexical functions. The NPCase = NPC Case | NPErg | NPAbl
rules for defining Urdu morphology are |NPIns|NPLoc1NPLoc2
borrowed from (Humayoun et el., 2006), in |NPDat;|NPAcc
which Urdu morphology was developed in the Case = Dir | Obl | Voc ;
Functional Morphology toolkit (Forsberg and Agr = Ag Gender Number UPerson ;
Ranta, 2004). Although it is possible to Gender = Masc | Fem ;
automatically generate equivalent GF code from UPerson = Pers1| Pers2_Casual
it, we wrote the rules of morphology from | Pers2_Familiar | Pers2_Respect
scratch in GF, to receive better abstractions than | Pers3_Near | Pers3_Distant;
are possible in generated code. Furthermore, we Number = Sg | Pl ;
extend this work by including compound words.
However, the details of morphology are beyond Thus NP is a record with two fields, ’s’ and ’a’.
the scope of this paper, and its focus is on ‘s’ is an inflection table and stores different
syntax. forms of a noun.
The Urdu NP has a system of syntactic cases
4. Syntax which is partly different from the morphological
cases of the category noun (N). The case
While morphological analysis deals with the markers that follow nouns in the form of post-
formation and inflection of individual words, positions cannot be handled at lexical level

154
through morphological suffixes and are thus the corresponding forms of adjective and
handled at syntactic level (Butt et el., 2002). common noun from their inflection tables using
Here we create different forms of a noun phrase selection operator (‘!’). Since CN does not
to handle case markers for Urdu nouns. Here is a inflect in degree but the adjective does, we fix
short description of the different cases of NP : the degree to be positive (Posit) in this
construction. Other modifiers include possibly
• NPC Case: this is used to retain the adverbs, relative clauses, and appositional
original case of Noun attributes.
• NPErg: Ergative case with case marker A CN can be converted to a NP using different
‘ne: ‫’ ﮯ‬ functions: common nouns with determiners;
• NPAbl: Ablative with case marker ‘se: proper names; pronouns; and bare nouns as mass
‫’ ﮯ‬ terms:
• NPIns: Instrumental case with case
marker ‘se: ‫’ ﮯ‬ fun DetCN : Det -> CN -> NP (e.g the boy)
• NPLoc1: Locative case with case fun UsePN : PN -> NP (e.g John)
fun UsePron : Pron -> NP (e.g he)
marker ‘mi: ɳ ‫’ﮟ‬
fun MassNP : CN -> NP (e.g milk)
• NPLoc2: Locative case with case
marker ‘pr ’ These different ways of building NP’s, which
• NPDat: Dative case with case marker are common in different languages, are defined
‘kʋ ’ in the abstract syntax of the resource grammar,
• NPAcc: Accusative case with case but the linearization of these functions is
marker ‘kʋ ’ language dependent and is therefore defined in
the concrete syntaxes.
And ‘a’ (Agr in the code sample given in
previous column) is the agreement feature of the 4.2 Verb Phrases (VP)
the noun that is used for selecting the
appropriate form of other categories that agree A verb phrase is a single or a group of words
with nouns. that act as a predicate. In our construction Urdu
A noun is converted to an intermediate category verb phrase has following structure
common noun (CN; also known as N-Bar)
which is then converted to NP category. CN lincat VP = {
deals with nouns and their modifiers. As an s : VPHForm => {fin, inf: Str} ;
example consider adjectival modification: obj : {s : Str ; a : Agr} ;
vType : VType ;
fun AdjCN : AP -> CN -> CN ; comp : Agr => Str;
embComp : Str ;
lin AdjCN ap cn = { ad : Str } ;
s = \\n,c =>
ap.s ! n ! cn.g ! c ! Posit ++ cn.s ! n ! c ; where
g = cn.g
}; VPHForm =
VPTense VPPTense Agr
The linearization of AdjCN gives us common | VPReq HLevel | VPStem
nouns such as (
 ‫ ﭨﮉا‬ʈȹn ɖa pani: “cold
water”) where a CN (
 pani: “water”) is and
modified by an AP ( ‫ ﭨﮉا‬, ʈȹn ɖa “cold”). VPPTense = VPPres |VPPast |VPFutr;
Since Urdu adjectives also inflect in number, HLevel = Tu |Tum |Ap |Neutr
gender, case and degree, we need to concatenate
the appropriate form of adjective that agrees
with common noun. This is ensured by selecting

155
In GF representation a VP is a record with question is the complement of the verb. In that
different fields. The most important field is ‘s’ case complement of the verb comes at the very
which is an inflectional table and stores different end of clause e.g (ʋo khta he: kh ʋo dʋɽti: he: ‫وﮦ‬
forms of Verb. ‫“ ﮩ ﮨﮯ ہ وﮦ دوڑ
ﮨﮯ‬he says that she runs”).
At VP level we define Urdu tenses by using a We have two different fields named ‘compl’ and
simplified tense system, which has only three ‘embCompl’ in the VP to deal with these
tenses, named VPPres, VPPast, VPFutr. In case different situations.
of VPTense for every possible combination of ‘vType’ field is used to store information about
VPPTense and agreement (gender, number, type of a verb. In Urdu a verb can be transitive,
person) a tuple of two string values {fin, inf : intransitive or double-transitive (Schmidt R. L.,
Str} is created. ‘fin’ stores the coupla (auxiliary 1999). This information is important when
verb) , and ‘inf’ stores corresponding form of dealing with ergativity in verb agreement. The
verb. VPStem is a special tense which stores the information about the object of the verb is stored
root form of verb. This form is used to create the in ‘obj’ field. All this information that a VP
full set of Urdu tenses at clause level (tenses in carries is used when a VP is used in the
which the root form of verb is used, i.e. construction of a clause.
perfective and progressive tenses). Handling A distinguishing feature of Urdu verb agreement
tenses at clause level rather than at verb phrase is ‘ergativity’. Urdu is one of those languages
level simplifies the VP and results in a more that shows split ergativity at verb level. Final
efficient grammar. verb agreement is with direct subjective except
The resource grammar has a common API in the transitive perfective tense. In transitive
which has a much simplified tense system, perfective tense verb agreement is with direct
which is close to Germanic languages. It is object. In this case the subject takes the ergative
divided into tense and anteriority. There are only construction (subject with addition of ergative
four tenses named as present, past, future and case marker (ne: ‫) ﮯ‬.
conditional, and two possibilities of anteriority However, in the case of the simple past tense,
(Simul , Anter). This means it creates 8 verb shows ergative behavior, but in case of
combinations. This abstract tense system does other perfective tenses (e.g immediate past,
not cover all the tenses in Urdu. We have remote past etc) there are two different
covered the rest of tenses at clause level, even approaches, in first one auxiliary verb (tʃka $)
though these tenses are not accessible by the is used to make clauses. If (tʃka $) is used,
common API, but still can be used in language verb does not show ergative behavior and final
specific modules. verb agreement is with direct subjective.
Other forms for verb phrases include request Consider the following example
form (VPReq), imperative form (VPImp). There
are four levels of requests in Urdu. Three of ‫ ﮨﮯ‬$ %&' ‫ﮍ ب‬
them correspond to (tʋ , tm
 , a:p ‫ ) پ‬honor lɽka Direct ktab Direct xri:d Root tʃka aux_verb he:
levels and the fourth is neutral with respect to The boy has bought a book
honorific levels. .
The Urdu VP is a complex structure that has The second way to make the same clause is
different parts: the main part is a verb and then
there are other auxiliaries attached to verb. For ‫* ﮨﮯ‬%&' ‫ﮍﮯ ﮯ ب‬
example an adverb can be attached to a verb as a
lɽke: ne: Erg ktab Direct_Fem xri:di: Direct_Fem he:
modifier. We have a special field ‘ad’ in our VP
The boy has bought a book
representation. It is a simple string that can be
attached with the verb to build a modified verb.
In Urdu the complement of a verb precedes the In the first case the subject (lɽka, ‫“ ﮍ‬boy”) is
actual verb e.g (‫ وﮦ دوڑ  ﮨ
ﮨﮯ‬ʋo dʋɽna tʃahti: in direct case and auxiliary verb agrees to
subject, but in second case verb is in agreement
he: “she want to run”), here (‫ ﮨ‬tʃahna “want”) with object and ergative case of subject is used.
is complement of verb ( ‫ دوڑ‬dʋɽna “run”), However, in the current implementation we
except in the case where, a sentence or a follow the first approach.

156
In the concrete syntax we ensure this ergative When a comparative AP is created from an
behavior through the following code segment in adjective and a NP, constant “se: ‫ ” ﮯ‬is used
GF. However the code given here is just a between oblique form of noun and adjective. For
segment of the code that is relevant. example linearization of above function is

case vt of { lin ComparA a np = {

VPPast => case vp.vType of { s = \\n,g,c,d => np.s ! NPC Obl ++ "se:"
(Vtrans| VTransPost) => <NPErg, vp.obj.a> ++ a.s ! n ! g ! c ! d ;
_ => <NPC Dir, np.a> };
};
_ => <NPC Dir, np.a> 4.4 Clauses
};
e.g in case of simple past tense if verb is A clause is a syntactic category that has variable
transitive then ergative case of noun is used and tense, polarity and order. Predication of a NP
agreement is with object of verb. In all other and VP gives simplest clause
cases direct case of noun is used and agreement
is with subject of verb. fun PredVP : NP -> VP -> Cl ;
A VP is constructed in different ways; the
simplest is The subject-verb agreement is insured through
agreement feature of NP which is passed to verb
fun UseV : V -> VP ; as inherent feature. A clause is of following type

where V is the morphological category and VP lincat Clause : Type = {s : VPHTense =>
is the syntactic category. There are other ways to Polarity => Order => Str} ;
make a VP from other categories, or
combinations of categories. For example Here VPHTense represents different tenses in
Urdu. Even though current abstract level of
fun AdvVP : VP -> Adv -> VP ; common API does not cover all tenses of Urdu,
we cover them at clause level and can be
An adverb can be attached to a VP to make an accessed through language specific module. So,
adverbial modified VP. For example (i:haɳ ‫&ﮩں‬ VPHTense is of following type
  )
VPHTense = VPGenPres | VPPastSimple
4.3 Adjective Phrases (AP) | VPFut | VPContPres
| VPContPast | VPContFut
Adjectives (A) are converted into the much | VPPerfPres | VPPerfPast
richer category adjectival phrases (AP) at syntax | VPPerfFut | VPPerfPresCont
level. The simplest function to convert is | VPPerfPastCont
| VPPerfFutCont | VPSubj
fun PositA : A -> AP ;
Polarity is used to make positive and negative
Its linearization is very simple, since in our case sentences; Order is used to make simple and
AP is similar to A e.g. interrogative sentences. These parameters are of
following forms
fun PositA a = a ;
Polarity = Pos | Neg
There are other ways of making AP for example Order = ODir | OQuest

fun ComparA : A -> NP -> AP ; PredVP function will create clauses with
variable tense, polarity and order which are

157
fixed at sentence level by different functions, fun IdetQuant : IQuant -> Num -> IDet ;
one is. fun PrepIP : Prep -> IP -> IAdv ;

fun UseCl : Temp -> Pol -> Cl -> S 5. Example

Here Temp is syntactic category which is in the As an example consider the translation of
form of a record having field for Tense and following sentence from English to Urdu, to see
Anteriority. Tense in the Temp category refers how our proposed system works at different
to abstract level Tense and we just map it to levels.
Urdu tenses by selecting the appropriate clause.
This will create simple declarative sentence, He drinks hot milk.
other forms of sentences (e.g Question
sentences) are handled in Questions categories Figure 1 shows the parse tree for this sentence.
of GF which follows next. As a resource grammar developer our goal is to
provide correct concrete level linearization of
4.5 Question Clauses and Question this tree for Urdu.
Sentences

Common API provides different ways to create

question clauses. The simplest way is to create
from simple clause

fun QuestCl : Cl -> QCl ;

In Urdu simple interrogative sentences are

created by just adding “ki:a ” at the start of a
direct clause that already have been created at
clause level. Hence, the linearization of above
function simply selects appropriate form of
clause and adds “ki:a ” at the start. However
this clause still has variable tense and polarity
which is fixed at sentence level e.g Figure 1. Parse tree of an example sentence
fun UseQCl : Temp -> Pol -> QCl -> QS The nodes in this tree represent different
categories and its branching shows how a
Other forms of question clauses include clauses particular category is built from other categories
made with interrogative pronouns (IP), and/or leaves (words from lexicon). In GF
interrogative adverbs (IAdv), and interrogative notation these are the syntactic rules which are
determiners (IDet), categories. Some of the declared at abstract level. For example category
functions for creating question clauses are CN can be built from an AP (adjectival phrase)
and a CN. So in GF representation it has
fun QuestVP : IP -> VP -> QCl (e.g who following type signature.
walks)
fun QuestIAdv : IAdv -> Cl -> QCl (e.g why fun AdjCN : AP -> CN -> CN ;
does he walk)
A correct implementation of this rule in Urdu
IP, IAdv, IDet etc are built at morphological concrete syntax ensures correct formation of a
level and can also be created with following
functions. common noun (‫م دوده‬. grm dʋdȺ “hot milk”)
from a CN (‫ دوده‬dʋdȺ “milk”) modified by an
fun AdvIP : IP -> Adv -> IP Adjective ( ‫م‬. , grm “hot”).

158
A NP is constructed from this CN by one of the implemented in a different but related
NP construction rules (see section 4.1 for formalism.
details). A VPSlash (object missing VP) is build Like the GF resource library, Pargram project
from a two place verb ( pi:ta “drinks”). This (Butt et el., 2007) aims at building a set of
VPSlash is then converted to VP through parallel grammars including Urdu. The
function grammars in Pargram are connected with each
other by transfer functions, rather than a
fun ComplSlash : VPSlash -> NP -> VP ; common representation. Further, the Urdu
grammar is still one of the least implemented
Resulting VP and NP are grouped together to grammars in Pargram at the moment. This
make a VP (‫م دوده  ﮨﮯ‬. ʈgrm dʋdȺ pi:ta he: project is based on the theoretical framework of
“drinks hot milk”). Finally clause (‫م دوده  ﮨﮯ‬. lexical functional grammar (LFG).
Other than Pargram, most work is based on LFG
‫ وﮦ‬ʋh grm dʋdȺ pi:ta he: “he drinks hot milk”) is and translation is unidirectional i.e. from
build from NP (‫ وﮦ‬ʋh “he”) which is build from English to Urdu only. For instance, English to
pronoun (‫ وﮦ‬ʋh “he”) and VP (‫م دوده  ﮨﮯ‬. Urdu MT System is developed under the Urdu
grm dʋdȺ pi:ta he: “drinks hot milk”). Language Localization Project (Hussain, 2004), (Sarfraz
dependent concrete syntax assures that correct and Naseem, 2007) and (Khalid et el., 2009).
forms of words are selected from lexicon and Similarly, (Zafer and Masood, 2009) reports
word order is according to rules of that specific another English-Urdu MT system developed
language. While, morphology makes sure that with example based approach. On the other
correct forms of words are built during lexicon hand, (Sinha and Mahesh, 2009) presents a
development. strategy for deriving Urdu sentences from
English-Hindi MT system. However, it seems to
6. An application: Attempto be a partial solution to the problem.

An experiment of implementing Controlled 8. Future Work

languages in GF is reported in (Angelov and
Ranta, 2010). In this experiment, a grammar for The common resource grammar API does not
Attempto Controlled English (Attempto, 2008) cover all the aspects of Urdu language, and non-
is implemented and then ported to six languages generalizable language-specific features are
(English, Finnish, French, German, Italian, and supposed to be handled in language-specific
Swedish) using the GF resource library. To modules. In our current implementation of Urdu
demonstrate the usefulness of our grammar and resource grammar we have not covered those
to check its correctness, we have added Urdu to features. For example in Urdu it is possible to
this set. Now, we can translate Attempto build a VP from only VPSlash (VPSlash
documents between all of these seven languages. category represents object missing VP) e.g (‫ﮨﮯ‬
The implementation followed the general recipe  kȹata he:) without adding the object. This
for how new languages can be added (Angelov rule is not present in the common API. One
and Ranta, 2009) and created no surprises. direction for future work is to cover such
However the details of this implementation are language specific features.
beyond the scope of this paper. Another direction for future work could be to
include the causative forms of verb which are
7. Related Work not included in the current implementation due
to efficiency issues.
A suite of Urdu resources were reported in
(Humayoun et el., 2006) including a fairly 9. Conclusion
complete open-source Urdu morphology and a
small fragment of syntax in GF. In this sense, it The resource grammar we develop consists of
is a predecessor of Urdu resource grammar, 44 categories and 190 functions3 which cover a
fair enough part of language and is enough for

______________________________
3
http://www.grammaticalframework.org/lib/doc/synopsis.html
159
building domain specific application grammars Grammar Development. Proceedings of the
including multilingual dialogue systems, Conference on Language & Technology 2009.
controlled language translation, software Masica C., 1991. The Indo-Aryan Languages,
localization etc. Since a common API for Cambridge, Cambridge University Press, ISBN
multiple languages is provided, this grammar is 9780521299442.
useful in applications where we need to parse
and translate the text from one to many other Ranta A., Grammatical Framework: A Type-
languages. Theoretical Grammar Formalism. The Journal of
However our approach of common abstract Functional Programming 14(2) (2004) 145–189.
syntax has its limitations and does not cover all Ranta A. The GF Resource Grammar Library
aspects of Urdu language. This is why it is not A systematic presentation of the library from the
possible to use our grammar for arbitrary text linguistic point of view. to appear in the on-line
parsing and generation. journal Linguistics in Language Technology,
2009a.
10. References Ranta A. Grammars as Software Libraries. From
Semantics to Computer Science, Cambridge
Angelov K. and Ranta A. 2010. Implementing University Press, Cambridge, pp. 281-308, 2009b.
controlled Languages in GF. Controlled Natural Rizvi, S. M. J. 2007. Development of Algorithms and
Language (CNL) 2009, LNCS/LNAI Vol. 5972 Computational Grammar of Urdu. Department of
(To appear) Computer & Information Sciences/ Pakistan
Attempto 2008. Project Homepage. Institute of Engineering and Applied Sciences
attempto.ifi.uzh.ch/site/ Nilore Islamabad. Pakistan.
Butt M., 1995. The Structures of Complex Predicate Sarfraz H. and Naseem T., 2007. Sentence
in Hindi Stanford: CSLI Publications Segmentation and Segment Re-Ordering for
English to Urdu Machine Translation. In
Butt M., Dyvik H., King T. H., Masuichi H., and Proceedings of the Conference on Language and
Rohrer C. 2002. The Parallel Grammar Project. Technology, August 07-11, 2007, University of
In Proceedings of COLING-2002 Workshop on Peshawar, Pakistan.
Grammar Engineering and Evaluation. pp. 1-7.
Schmidt R. L., 1999. Urdu an Essential
Butt, M. and King, T. H. 2007. Urdu in a Parallel Grammar,Routledge Grammars.
Grammar Development Environment'. In T.
Takenobu and C.-R. Huang (eds.) Language Sinha R., and Mahesh K., 2009. Developing English-
Resources and Evaluation: Special Issue on Asian Urdu Machine Translation Via Hind., Third
Language Processing: State of the Art Resources Workshop on Computational Approaches to
and Processing 41:191-207. Arabic Script-based Languages (CAASL3) in
conjunction with The twelfth Machine Translation
Forsberg M., and Ranta A., 2004. Functional Summit. Ottawa, Ontario, Canada.
Morphology. Proceedings of the Ninth ACM
SIGPLAN International Conference of Functional Zafar M. and Masood A., 2009. Interactive English
Programming, Snowbird, Utah. to Urdu Machine Translation using Example-
Based Approach. International Journal on
Humayoun M., Hammarström H., and Ranta A. Computer Science and Engineering Vol.1(3),
Urdu Morphology, Orthography and Lexicon 2009, pp 275-282.
Extraction. CAASL-2: The Second Workshop on
Computational Approaches to Arabic Script-based
Languages, July 21-22, 2007, LSA 2007
Linguistic Institute, Stanford University. 2007
Hussain, S. 2004. Urdu Localization Project.
COLING:WORKSHOP ON Computational
Approaches to Arabic Script-based Languages,
Geneva. pp. 80-81
Khalid, U., Karamat, N., Iqbal, S. and Hussain, S.
2009. Semi-Automatic Lexical Functional

160