Developing links of compound sentences for parsing through marathi link grammar parser

International Journal on Natural Language Computing (IJNLC) Vol. 3, No.5/6, December 2014
DOI : 10.5121/ijnlc.2014.3601 1
DEVELOPING LINKS OF COMPOUND SENTENCES
FOR PARSING THROUGH MARATHI LINK GRAMMAR
PARSER
Vaishali B. Patil1
and B. V. Pawar2
1
Institute of Management Research and Development,Shirpur, Maharashtra 425405, India
2
School of Computer Sciences, North Maharashtra University,Jalgaon, Maharashtra
425001, India
ABSTRACT
Marathi is a verb-final language with a relatively free word order. Complex Sentences is one of the major
types of sentences which are used commonly in any language. This paper explores the study of complex
sentence structure of Marathi language. The paper proposes various links of complex sentence clauses and
modelling of the complex sentences using proposed links in the Link Grammar Framework for parsing
purpose.
KEYWORDS
Marathi Complex Sentences, Link Grammar, Marathi Link Grammar Parser
1. INTRODUCTION
Link Grammar is a formal grammatical system defined on the basis of natural language property
which states that if arcs are drawn connecting each pair of words that relate to each other, then the
arcs will not cross [16]. This property is called as planarity. A parsing system has been developed
to capture many phenomenon of English grammar by providing roughly seven hundred
definitions that includes the word of the language and their linking requirements and an algorithm
[6] for parsing sentences according to the given grammar.
A given sentence is accepted by a system if the linking requirements of all the words in a
sentence are satisfied (connectivity property), none of the links between the words cross each
other (planarity property) and there exists at most one link between any pair of words (exclusion
property).
Parsed output is very fundamental requirement for natural language processing (NLP)
applications like Information retrieval, Information extraction, Question Answering, etc.
especially in Machine translation [17]. Indian languages are resource deficient languages as it
does have very limited electronically managed tools like morphological analyzer, part of Speech
tagger, parser etc. Marathi language is also not an exception however since last decade there are
numerous efforts has been witnessed among this we have gone through [3, 4, 5, 12, 13, 14, 15].
Our proposed Marathi link Grammar parser is one attempt to develop such tools which will be
helpful in various applications wherever it suits better. Following figure will give quick glimpse
of our proposed system.

2
Figure 1 Block Diagram of Proposed Marathi Link Grammar Parser
Our proposed Marathi link grammar parser is rule based parsing system which contains link
database, the handcrafted rules and an algorithm to get parsed output if one exists. So far by
studying Marathi noun phrases, verb phrases and subject/object to verb agreement we have
proposed 31 links [8, 9, 10]. Based on computational Panini grammar [1] we proposed Karaka
links [11] which defines the relation between nominal words with verb of a sentence summarized
in table1. Karka relations are the relations of nominal that participate in the action specified by
the particular verb mentioned in the sentence. Links between any pair of words gives the
functional association between that pair of words. For eg consider the sentence “Ram aamba
khato (राम आंबा खातो : Ram eats mango)” by our proposed system links between words will be
established between verb karta and verb Karma as sentence consists it. Hence Ka_karta link
will be established on khato (खातो: eats) and Raam (राम : proper Noun) and Ka_karma link will
be established on khato (खातो: eats) and aamba (आंबा: Mango) word pairs.
Table 1: Karaka and its links
Karaka Link Functionality
Karta Ka_Karta Verb to Subject
Karma Ka_Karma Verb to Object
Karan Ka_karna Verb to Instrument of the Activity
Adhikarna Ka_Adhikarna Verb to time and place of the activity
Aapadan Ka_Aapadan Verb to word which gives separation meaning
Sampradan Ka_Sampradan Verb to word which gives donation meaning
The task of our system is building links by judging each individual word’s role in the sentence. A
system gets complete linkage if it satisfies all the rules laid as per link grammar framework i.e.
Planarity, Connectivity and Exclusion.
Input
Sent.
Parsed
Output
Pre
Process
Apply
Parsing
Algo.
Post
Process
Link
Dictionary
Lexicon /
wordNet

3
2. COMPOUND SENTENCES IN MARATHI
In Marathi language, coordination is of two type sentence coordination and constituent
coordination [2][7]. There are three major coordinators namely Conjunctives, Disjunctive and
Adversative.
2.1. Sentence Coordination
Any number of sentences can be coordinated with “aani” (आǔण : and) which is always placed
before the last conjunct. In a sequence of more than two sentences, all preceding sentences before
the last are simply juxtaposed as given in following example:
Ex 1: babu aala aani lili ghari geli
(बाबू आला आǔण िलली घरȣ गेली : Babu left and Lili came home)
Ex 2:babu aala, lili ghari geli aani lagech minila phone kela.
(बाबू आला, िलली घरȣ गेली आǔण लगेच िमनीला फोन के ला : babu left, Lili came home and
immediately phoned Mini)
Sentence coordination is used to express various semantic distinctions such as contrast,
contingence, sequential events and even casual connections.
2.2 Constituent/word level Coordination
Various parts of speech can be coordinated at constituent level. Nouns of all categories may be
coordinated. Pronouns, adjectives, adverbs and active and passive verbs can also be coordinated.
While coordinating within a sentence part of speech follows certain agreement rules on the
conjoining category. Following are few examples on constituent level coordination,
Ex 3: Noun (Subject) Coordination
lili sudha aani mini gharat hotya.
(िलली सुधा आǔण िमनी घरात हो×या : Lili, Sudha and Mini were in the house)
Ex 4: Noun (object) Coordination
liliNe aambe keli aani peru khalle
(लीलीने आंबे के ळȣ आǔण पेǾ खाãले : Lili ate mangoes, bananas and guavas )
Ex 5: pronoun coordination
mi aani tu udya baget jau
(मी आǔण तू उƭा बागेत जाऊ : I and you will go in the garden tomorrow)
Ex 6: Adjective Coordination
lili jara bavali aani vedi aahe
(िलली जरा बावळȣ आǔण वेडȣ आहे : Lili is a little bit disorderly and crazy)
Ex 7: Adverb Coordination
lili halu halu aani mand swarat bolate
(िलली हळू हळू आǔण मंद ःवरात बोलते : Lili speaks slowly and in a low voice)
Ex 8: verb coordination
chor kholit shirala aani lagech pakadala gela (चोर खोलीत िशरला आǔण लगेच पकडला गेला :
Thief entered the room and was immediately caught)

4
2.3 Conjunctive Coordination
The basic conjunctive coordinator is “aani” (आǔण : and) with alternates such as wa (व : and ,
ankhi (आणखी : and), aankhin (आणखीन : and), aanik (आǔणक : and), an (अन : and). The first
alternate i.e. wa (व)is a perso-Arabic borrowing. It is used mostly in literary styles however; its
use is increasing in Modern Marathi. The rest are used in conversational speech. All examples
mentioned in section 2.1 and 2.2 are confined to conjunctive coordinator “aani”(आǔण).
2.4 Disjunctive structures
There are three disjunctives, kinva (Ǒकं वा :or), ka/ki (का/Ǒक : gives meaning of or) and athava
(अथवा : or) all expressing the sense of ‘or’. The first, kinva (Ǒकं वा : or) is prevalent. The second,
ka/ki (का/Ǒक : gives meaning of or) is used in interrogatives and in subordinate clauses
expressing the sense of ‘whether’. The last is confined to the formal language. In both sentence
and constituent coordination kinva (Ǒकं वा : or) is placed immediately before the last sentence or
constituent as the case may be. It may also appear before each sentence or sentential constituent.
It is never placed in the beginning of the first sentence or first sentence constituent. Although
kinva (Ǒकं वा : or) allows a juxtaposed sequence like aani (आǔण: and), unlike aani (आǔण: and) it
may however not be totally absent from the sequence. The last placement of kinva (Ǒकं वा: or) is
obligatory. Following is one example,
Ex 9: lili ghari geli asel kinva baget basali asel.
(िलली घरȣ गेली असेल Ǒकं वा बागेत बसली असेल : Lili may have gone home or she may be
sitting in the garden)
2.5 Adversative Structures
The three adversative coordinators pan (पण : but), parantu (परंतु : but) and tathapi (तथाǒप :
but) expressing the sense of ‘but’ are semantically identical except in their usage. The last one is
used mostly in formal contexts. The first two are nearly exchangeable. Adversative conjunctions
encode a contrast with various semantic implications, for example
Ex 10: lili hushar aahe pan abhyas karat nahi
(िलली हुशार आहे पण अßयास करत नाहȣ : Lili is intelligent but does not study )
3. DEVELOPING LINKS FOR MARATHI COMPOUND SENTENCES
We have adopted two level linking schemes specifically considering complex sentences and
compound sentences. The challenge in dealing such sentences is crossing of the links. Crossing of
the links occurs due to violating planarity rule which states that links drawn between two words
shall not cross any other link connecting any pair of words. Planarity cannot always be preserved
in free word order languages. Considering Marathi compound sentences, we observed that
coordination either sentential coordination or constituent coordination is used majorly.

5
In two level linking the compound sentences are split in two levels. The upper level deals with
coordinators encountered in a sentence to be parsed and lower level deals with the inner clause
placed before any coordinators. We have proposed various link types for compound sentences
based on their functionality. These functional link names helps in identifying the roles it is
playing linking between two distinct sentences or constituents. Following diagram illustrates two
level linking scheme,
Figure 2 Two level linking in compound sentence
The links proposed to connect sentential and constituent structures are summarized in a table
below followed by brief description of each identified structure with an illustration,
Table 2: Proposed Links for Complex Sentence structures
Sr No Link Name Functionality of link
1 SeCC Sentence to Conjunctive Coordinator
2 CCSe Conjunctive Coordinator to Sentence
3 SeDC Sentence to Disjunctive Coordinator
4 DCSe Disjunctive Coordinator to Sentence
5 SeAC Sentence to Adversative Coordinator
6 ACSe Adversative Coordinator to Sentence
7 SCC Subject to Conjunctive Coordinator
8 CCS Conjunctive Coordinator to Subject
9 OCC Object to Conjunctive Coordinator
10 CCO Conjunctive Coordinator to Object
11 AjCC Adjective to Conjunctive Coordinator
12 CCAj Conjunctive Coordinator to Adjective
13 AvCC Adverb to Conjunctive Coordinator
14 CCAv Conjunctive Coordinator to Adverb
15 SDC Subject to Disjunctive Coordinator
16 DCS Disjunctive Coordinator to Subject
17 ODC Object to Disjunctive Coordinator
18 DCO Disjunctive Coordinator to Object
19 AjDC Adjective to Disjunctive Coordinator
20 DCAj Disjunctive Coordinator to Adjective
21 AvDC Adverb to Disjunctive Coordinator
22 DCAv Disjunctive Coordinator to Adverb
SeCC CCSe
babu aala
Ka_karta
lili ghari
babu aala aani
geli
Ka_karta
lili ghari geli
Ka_adhikaran

6
3.1. CO1
The structure identified in following figure is two sentences coordinated with conjunctive
coordinator “aani” (आǔण). In this structure links proposed to connect sentence 1 to conjunctive
coordinator is SeCC and from conjunctive coordinator to sentence 2 it is CCSe.
Figure 3 Compound sentence structures 1
3.2. CO2
In this structure of compound sentence two separate sentences are connected with Disjunctive
Coordinator “kinva” (Ǒकं वा). The links proposed to connect sentence 1 to disjunctive coordinator
is SeDC and DCSe link is proposed to connect disjunctive coordinator to sentence 2.
Figure 4 Compound Sentence structure 2
3.3 CO3
The following figure shows the compound sentence structure 3 in which sentence 1 is connected
with sentence 2 with an Adversative coordinator “pan”(पण). Links proposed to connect it are
sentence 1 to adversative connector is SeAC and adversative connector to sentence 2 is ACSe.
Figure 5 Compound sentence structure 3
3.4 CO4, CO5, CO6 and CO7
In Constituent coordination subject, object, adjective and adverb can be connected together in a
sentence. In this category of coordination such constituent or word level coordination is
connected with conjunctive coordinator. Following figure shows these structures,
Figure 6 Compound sentence structure 4: subject coordination with Conjunctive
SeCC CCSe
Sent 1 aani Sent 2
SeDC DCSe
Sent 1 kinva Sent 2
SeAC ACSe
Sent 1 pan Sent 2
SCC CCS
Subject 1 aani Subject 2

7
Figure 7 Compound sentence structure 5; Object coordination with Conjunctive
Figure 8 Compound sentence structure 6; Adjective coordination with Conjunctive
Figure 9 Compound sentence structure 7; Adverb coordination with Conjunctive
3.5 CO8, CO9, CO10 and CO11
As discussed in above section constituent coordination subject, object, adjective and adverb
builds links with disjunctive coordinator too, which is illustrated in following figure
Figure 10: Compound sentence structure 8; Subject coordination with Disjunctive
Figure 11: Compound sentence structure 9; Object coordination with Disjunctive
Figure 12: Compound sentence structure 10; Adjective coordination with Disjunctive
Figure 13: Compound sentence structure 11; Adverb coordination with Disjunctive
OCC CCO
Object 1 aani Object 2
AjCC CCAj
Adjective 1 aani Adjective 2
SDC DCS
Subject 1 kinva Subject 2
ODC DCO
Object 1 kinva Object 2
AjDC DCAj
Adjective 1 kinva Adjective 2
AvDC DCAv
Adverb 1 kinva Adverb 2
AvCC CCAv
Adverb 1 aani Adverb 2

8
We have modelled compound sentences in the form of possible valid linkage and proposed
various links to connect the sentential and constituent structures in appropriate way. Our system
identifies 11 such compound sentence structures.
4. CONCLUSIONS
In this paper we have proposed links for compound sentence structure of Marathi language in
Link Grammar framework. By studying compound sentence structure of Marathi language, links
were developed to build connection on the sentential and constituent level. Total 22 new links are
proposed. More such structures will be studied and links will be developed by using this
framework.
REFERENCES
[1] A. Bharati, V. Chaitnya & R. Sangal (1995) Natural Language Processing: A Paninian Perspective,
New Delhi: Prentice-Hall of India.
[2] R. V. Dhongade & K. Wali (2009) Marathi, Amsterdam/Philadelphia: John Benjamins Publishing
Company.
[3] H. Gune, M. Bapat, M. Khapara & P. Bhattacharya, (2010), “Verbs are where all the action lies:
experiences of shallow parsing of a morphologically rich language”, In Proceedings of the 23rd
International Conference on Computational Linguistics: Posters (COLING '10). Association for
Computational Linguistics, Stroudsburg, PA, USA, pp 347-355.
[4] S. R. Kolhe & B. V. Pawar (2007) “A Connectionist Approach for Learning Regular Grammars”,
Journal of Computer Society of IndiaVol.37 Issue No. 3, PP. 79-86
[5] S. R. Kolhe & B. V. Pawar (2010) “Learning Subset of Natural Language (Marathi) Grammar Using
Neural Networks”, International Journal of Computer Engineering and Computer Applications, ISSN
0964-4983, Vol. 02, No.3, pp 24-31.
[6] J. Lafferty, D. Grinberg & D. Sleator (1995) “A Robust Parsing Algorithm for Link Grammars”,
Technical Report CMU-CS-95-125.
[7] R. Pandharipande (1997) MARATHI, London: Rutledge Publication.
[8] V. B. Patil & B. V. Pawar (2011) “Developing Verb Phrase links for Marathi link grammar parser”,
ICGST International journal on Artificial Intelligence and Machine learning, ISSN 1687-4846, Vol.
11, No.2, 2011, pp 33-38.
[9] V. B. Patil & B. V. Pawar (2011) “Developing Subject/Object links with Verb for Marathi Link
Grammar Parser”, In the proceedings of the National Conference on Advances in Computing (NCAC-
2011), ISBN 978-81-910591-2-0, 2011, pp 285-288.
[10] V. B. Patil & B. V. Pawar (2012) “Developing links for Marathi noun phrase morphology for
proposed Link Grammar parser”, Karpagam Journal of Computer Science, Vol. 6, No. 6, ISSN 0973-
292, 2012, pp 291-299.
[11] V. B. Patil & B. V. Pawar (2013) “Influence of Karaka Relation in the framework of Marathi Link
Grammar Parser”, in National Conference on Advances in Computing (NCAC’13), ISBN 978-81-
910591-7-5, 2013, pp 255-258.
[12] H. B. Patil, A. S. Patil & B. V. Pawar (2014) “Part-of-Speech Tagger for Marathi Language using
Limited Training Corpora”, International Journal of Computer Applications, 0975 – 8887, pp 33-37.
[13] B. V. Pawar (2001) “LA Grammar Formalism and Parsing of Simple Marathi Sentences using LA
algorithm”, Indian Linguistics, Journal of Linguistic society of India, Vol. 62, No.1-4, pp 141-154.
[14] B. V. Pawar (2004) “Comparison of results of Tomita’s algorithm and LA algorithm for parsing
Marathi sentences”, Indian Linguistics, Journal of Linguistic society of India, Vol. 65, No1-4, pp 133-
140.
[15] B. V. Pawar & N. S. Chaudhari (2000) “Marathi language Grammar Parsing using Tomita’s
approach”, Indian Linguistics, Journal of Linguistic society of India, Vol. 61, No.1-4, pp 69-96.

9
[16] D. D. K. Sleator & D. Temperley (1991) “Parsing English with a Link Grammar”, Technical Report
CMU-CS-91-196.
[17] G.V.Garje & G.K.Kharate (2013) “Survey of Machine Translation Systems in India”, International
Journal on Natural Language Computing, Vol. 2, No. 4, pp 47-67.
Authors
Vaishali B. Patil has completed Bachelor of Computer Science in 2000 and Master of
Computer Application (MCA) in 2003 from North Maharashtra University, India. She is
currently pursuing her Ph.D in Computer Science at School of Computer Sciences, North
Maharashtra University, India under the supervision of Dr B. V. Pawar and has total 11
years of teaching experience in PG course at Institute of Management Research and
Development, Shirpur (MS), India. She is a member of Computer Society of India. Her research interests
include Natural Language Processing, Information Retrieval and Intelligent Tutoring Systems.
Dr B. V. Pawar Received his B. E. (Production) degree in 1986 from VJTI, Mumbai
University, Mumbai, (India) and M.Sc.(Computer Science) degree in 1988 from Department
of Computer Science, Mumbai University, Mumbai, (India). He received his Ph.D. degree
in Computer Science in 2000 at North Maharashtra University, Jalgaon (India). He is having
26 years teaching experience. Presently he is working as Professor and Director, School of
Computer Science at North Maharashtra University, Jalgaon (India). He is member of various professional
bodies like CSI & LSI. He has been recognized as a Ph.D. guide for the subjects Computer Science,
Information Technology & Computer Engineering by various Universities in the state of Maharashtra
(India). Till date he has guided 05 students towards their Ph.D. degree. His research areas include pattern
recognition, neural networks, Natural Language Processing, Web Technologies & Information Retrieval.
His work has been published in various international and national journals and conferences.

Developing links of compound sentences for parsing through marathi link grammar parser

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Developing links of compound sentences for parsing through marathi link grammar parser (20)

Recently uploaded (20)

Developing links of compound sentences for parsing through marathi link grammar parser