A Novel Approach for Rule Based Translation of English to Marathi

Advanced Computational Intelligence: An International Journal (ACII), Vol.2, No.4, October 2015
DOI:10.5121/acii.2015.2401 1
A NOVEL APPROACH FOR RULE BASED
TRANSLATION OF ENGLISH TO MARATHI
Amruta Godase1
and Sharvari Govilkar2
1
Department of Information Technology (AI & Robotics), PIIT, Mumbai University,
India
2
Department of Computer Engineering, PIIT, Mumbai University, India
ABSTRACT
This paper presents a design for rule-based machine translation system for English to Marathi language
pair. The machine translation system will take input script as English sentence and parse with the help of
Stanford parser. The Stanford parser will be used for main purposes on the source side processing, in the
machine translation system. English to Marathi Bilingual dictionary is going to be created. The system will
take the parsed output and separate the source text word by word and searches for their corresponding
target words in the bilingual dictionary. The hand coded rules are written for Marathi inflections and also
reordering rules are there. After applying the reordering rules, English sentence will be syntactically
reordered to suit Marathi language.
KEYWORDS
Syntax analysis, Bilingual, Multilingual, Named Entity Recognition, Word Sense Disambiguation,
Morphological Synthesizer, Transliteration
1. INTRODUCTION
This paper presents a novel approach for rule based translator English to Marathi Machine aided
translation system. Machine Translation (MT) is the central areas of focus of Natural Language
Processing. Machine translation is important for breaking the language barrier among the
multilingual country and for facilitating the inter-lingual communication. If we succeed to this,
then we can say that exact translation is done by system.
India which is the largest democratic country where more than 30 languages and 2000 dialects
used for the communication by the Indians. Because of this different culture and multilingual
environment there is a big requirement for translation for the transfer of information and sharing
of the ideas, thoughts and facts.
Various MT approaches are exists for developing MT system: 1) Direct based MT 2) Rule based
MT 3) Interlingua based MT 4) Statistical based MT 5) Example based MT 6) Knowledge based
MT 7) Principle based MT 8) Online Interactive MT 9) Hybrid based MT.
Direct Machine Translation is simplest approach in which a direct word to word translation is
done (1). A Rule-Based Machine Translation (RBMT) system includes collection of various
rules, a bilingual lexicon or dictionary, and software programs to process the rules (2). Interlingua
based approach, this translation consists of two stages, the source Language (SL) which is first
converted in to the Interlingua (IL) form a then finally translate into target language. The main
advantage of this approach is that the analyzer and parser of SL script is independent of the
generator for the Target Language (TL) script and which requires complete resolution of
ambiguity in source language text(3). Statistical machine translation (SMT) is a statistical

2
framework which is based on the knowledge and statistical models which are extracted from
bilingual corpora and this is a data oriented structure (4). Basic idea of example based MT is to
reuse the examples of already existing translations (5). Knowledge-Based Machine Translation
(KBMT) is closely related to Interlingua approach and which requires complete understanding of
the source text prior to the translation into the target text. KBMT is implemented on the
Interlingua architecture (6). Principle-Based Machine Translation (PBMT) Systems are totally
based on the Principles & Parameters Theory of Chomsky‘s Generative Grammar and which
formally applies parsing method. In this, the parser generates a tree which shows detailed
syntactic structure along with lexical, phrasal, grammatical information (7). In online interactive
translation system, the user has full rights to give suggestion for the correct translation which is
very advantageous for improving the performance of MT system. This approach is very useful,
where the context of a word is not that much clear or unambiguous and where multiple possible
meanings for a particular word (8). By combining the advantages of statistical framework and
rule-based MT methodologies, a new approach was emerged, which is namely called as “hybrid-
based approach”. The hybrid approach used in a number of different ways (9).
This paper is organized into 4 sections. Section 1 discuss an introduction of MT, Section 2 gives
brief idea of major MT systems related work in India in tabular format; section 3 introduces the
proposed approach to build a MT systems and finally we conclude the paper in the next section.
2. RELATED WORK & LITERATURE SURVEY
In this section we look at some major Machine translation systems of India. Most of the
researchers concentrate on Rule based approach because Rule based approach is an easy to build
and which is always extensible and maintainable. English to Devnagari Translation is done by
M.L.Dhore [1]. The author proposes a hybrid approach and system is specifically developed only
for Banking Domain. System translates User Interface labels of commercial web based interactive
applications. Devika P, Sayli W. presents a MT system which translates an English sentence to
Marathi sentences of equivalent meaning [2]. Abhay A, Anuja G. dealing with rule based
translation of assertive sentences [3]. In this system author going through various processes. A
novel approach for Interlingual example based translation is developed by K.Balerao, V.Wadne
[4]. Transmuter MT is developed for Tourism domain by G. Gajre [5]. ANUVAADAK MT [6]
has been hosted online for public access by IIIT Bombay. The System enables translation
between different Indian Languages and also provides transliteration support for input of system.
SAAKAVA [7] is the websites which carries out translation of an English sentence into Marathi.
They are now developing a computer programme with the help of certain dictionary and will try
to understand English sentence and then translates the same sentences into Marathi by applying
all the rules of Marathi grammar. Google translate is a multilingual service which translate
written text from one language to another [8]. It supports 90 languages and many more. The
Google translation algorithm is based on statistical analysis and largely depends on a solid
corpus.
3. SYSTEM OVERVIEW
Like translation done by human, MT does not simply substituting words in one language for
another, but the complex linguistic knowledge; morphology (how words are built from smaller
units of meaning), syntax(grammar), semantics(meanings) and understanding of concepts such as
ambiguity. The translation process stated as:
1. Decoding the meaning of the source text and
2. Re-encoding this meaning in the target language.

3
The idea is to translate an input document by going through various phases such as pre-
processing, syntax, semantic and lexical phases and finally translating the documents into target
language using various mapping rules. The input to the system is a single text document in
English Natural Language (NL) and output will be a translated in Marathi NL. The proposed
approach consists of 3 phases:
Pre-processing phase, Transfer & generation phase and Post-processing phase. Following
Diagram shows the proposed approach.
Algorithm:
Input: Accept a digital document as input in English NL.
Output: Translate document in Marathi NL.
1. Accept a text file as Input.
2. For each sentence in input document do,
3. Apply POS tagging & generate the parse tree for each sentence then,
4. Apply NER rules on each sentence.
5. Perform WSD on lemmas to understand the exact meaning of the lemmas.
6. Use a bilingual dictionary to obtain appropriate translation and transliteration of lemmas.
7. Obtain the proper form of words using Inflections.
8. Represents the sentence based on target language grammar rules.
3.1 Pre-processing Module:
This is the first phase of any machine translation process. This phase is about to make MT
process easier and qualitative. The source text may contains figures, diagrams, formulas,
flowchart etc. that do not require any translation. So only translation portion should be identified
here. It consists of 3 main processes: Syntax analysis, Named Entity Recognition and Word sense
Disambiguation.
Figure 3.1 Proposed Approach

4
3.1.1 Syntax Analysis
Syntax analysis exploits the result of morphological analysis to build a structural representation
of a sentence. Parser is an algorithm which developed a syntactic structure like tree for a given
input. Parser is used for 4 main purposes: To give the parse tree structure of sentences, for Part-
of-speech (POS) tagging of English sentences, for stemming the words of English sentences and
for chunking of words.
S S
NP VP NP VP
CN AV CV CN CV
DEF-ART N is MV NP DEF-ART N NP
MV
The boy drinking tea एक मुलगा चहा पीतआहे
Figure 3.2 English to Marathi Translation of The boy is drinking tea.
3.1.2 Named Entity Recognition
Named Entity Recognition (NER) gives sequences of words in a text which are the names of
things. It comes with well-engineered feature extractors for Named Entity and for defining
feature extractors. Stanford NER tool and Open NLP tool are available for doing the tasks.
Various rules are exist for Named entity Recognition:
I) Rules for creating Person’s Name
a) Look for Proper Nouns.
b) Contextual words like {men, books, author of, co-author, read, worked, state, city, country,
university, college, school, island of, hero, hospital, born, establish, started, saints, founded,
chairman of , director} if came then it will consider as proper noun.
c) If set of capitalized word include a set of letters followed by (.), followed by mostly one
(rarely two) capitalized words, then the whole set is considered as name.
d) If one of the capitalized words appears subsequently, the probability for it belongs to name.
e) If the set of words or one of capitalized words appear at the beginning of a sentence, it will
considered as name.
f) If preposition belongs to {by, of, friend, colleagues, to, co-author, with, men, persons,
emperor, men like, sage, as}, the probability for it to be name increases.
g) If the word immediately after the capitalized word(s) (i.e. the post-position) is belongs to set
{said, told} the probability for it to be name increases.
h) An apostrophe’s (‘s) to a capitalized word, then the probability it consider as name.
II) Rules for creating Place /Institute /Organization Name Index
b) If a Preposition comes immediately after a Name, it is likely to be a Place or Organization or
Institute.
c) Possible set of preposition for potential Place or Organization {from, in, at, to, for, of}
III) Rules for creating Date Index

5
b) Words like Century, decade, AD., BC., during, before, after, until if came in sentences then
probably it’s a year.
3.1.3 Word Sense Disambiguation
Word Sense Disambiguation (WSD) is defined as assigning the correct sense to a word according
to its context. The disambiguation is done by applying construct which is called as restriction. For
example,
back bencher => student who is not serious in studies and always sitting at back of the class.
This additional sense will be included in the dictionary with the help of some rule patterns.
The Verb get off
Main
Word
Multiple
Meaning
Meaning Examples
get off leave थान
करणे
We got off after breakfast
get off be saved वाचणे lucky to get off with a scar only
get off send पाठवणे Get these parcels off by the first post
get off stop बंद करणे get off the subject of alcoholism
get off stop with
objwork
काम
थांबवणे
get off (work) early tomorrow
The Noun Shadow
Main Word Multiple
Meaning
Meaning Examples
shadow darkness अंधार the place was now in shadow
shadow patch काळे डाग shadows under the eyes
shadow atmosphere छायेत country in the shadow of war
shadow hint संक
े त the shadow of things to come
shadow close company सावल the child was a shadow of her mother
shadow deterrent छाया a shadow over his happiness
shadow ghost भूत seeing shadows at night
3.2 Transfer Generation Module:
This is the second phase of machine translation system. This module composes the meaning
representation and assigns them the linguistic inputs. Here exact word mapping and transliteration
is done with the help of dictionary.

6
3.2.1 Language Representation
It is necessary to have dictionary. The bilingual dictionary is use for the purpose of a lexicon by
storing root words together along with their morphological properties and meaning of English
words in Marathi, equivalent synonyms. Dictionary database is endless. Therefore we can extend
it according to our need.
The pre-processing of dictionary undergoes various stages depending on the data: 1) Font-
converting 2) Aligning 3) POS tagging 4) Adding Synonyms.
Each database has five fields: source, target, category and Synonym.
i. The field ‘source’ stores the English words.
ii. The field ‘target’ stores the Marathi words.
iii. The field ‘POS category’ stores the Part-of-Speech category of the source and target
words.
iv. The field ‘synonym’ stores the synonym of Marathi words for that particular English
word.
Table 3.1: Typical Entries for English words in a Sample Bi-lingual Dictionary
Source POS category Target Equivalent Synonyms
after adjective नंतर काह वेळाने
abed adverb %बछायावर अंथ'णावर
book noun पुतक (यवथा करणे
across preposition आरपार पलकडे
3.2.2 Transliteration
Transliteration involves separating tokens form the input query and mapping the characters of
each token to its equivalent Marathi characters.
Algorithm:
1) Once the English sentence is entered , the tokens/words from the query are separated as
below
a) From the start of string, identify all the vowels positions one by one till the end of string.
b) If vowel position is found then the word before the vowel is identified as a token
2) For each token identified in the above step:
a) For each character in the token, its equivalent English character(s) are mapped.
b) We have used below mapping between English alphabets with its equivalent Marathi
alphabets.
c) Once for a complete English string, the tokens are identified and transliterated then the
transliterated text will be provided as input to next module.
For eg. Shivaji (Sh +i) + (v+a )+( j+i) )श + वा + जी )शवाजी
3.3 Post-processing Module:
This is the last and main phase of any machine translation process. Once the text is represented in
target language, the target text is to be reformatted after post-editing. Post-editing is done to make

7
sure that the quality of translation is upto the mark. Post-editing should continue till the MT
system reach human like. It consists of two main sub-modules.
3.3.1 Marathi Morphological Synthesizer
The language which follows SOV structure use postpositions. Hence while translating an English
sentence (SVO structure) to a Marathi sentence (SOV structure), we need to change the position
that is we have to convert preposition to postposition.
Words can be classified in two types depending on inflections [6]:
Inflectional Words: Noun, Pronoun, Adjective, Verb
Non-Inflectional Words: Adverb, Interjection Conjunction
The words are inflected base on the changes in Gender i.e. Masculine, Feminine, Neuter or
Multiplicity i.e. Singular, Plural or Tense i.e. Present, past, Future and case.
For Example: Shivaji came to house to meet Jijabai but she was not there.
English Word Marathi Lemma Inflected Words Rules to get inflected
Shivaji )शवाजी )शवाजी ------
Came येणे आला Verb, 3rd
person,
Masculine, Singular,
Past tense
to ला
house घर घर Case table, Locative,
Singular
to ला भेटायला Case table, Dative
Singular
meet भेटणे
Jijabai िजजाबाई िजजाबाईला Case table, Accusative
Singular
but पण पण -----
she ती ती -----
was होती न(हती -----
not नाह
there 1तथे 1तथे -----
So for implementation of the inflection we need to store the following information to the
database.
English words, POS Tag, Gender, Tense, Multiplicity, Degree and Case.

8
3.3.2 Target Language Sentence Generation
To obtain a Marathi output from input English sentences, the differences in the syntactic
structure must be determined using rules i.e. mapping rules. A set of mapping rules in terms of
syntactic structure can be used for English-Marathi MT. Following table shows the reordering
rules for English-Marathi translation where, there are very few rules presented to give the idea of
how the production rules work.
Table 3.2 Reordering Rules
English Patterns Marathi Patterns
R1 S NNP+VBD+DT+NN+IN+
NNP’+NNP”+IN’+NNP”’
Shivaji was the founder of Maratha empire
in India.
R1’ S NNP+(IN’+NNP’”)+NNP’+
(NNP”+IN)+(DT+NN)+VBD
)शवाजी भारतात मराठा सा2ा3याचे
संथापक होते.
R2 S NNP+VBD+VBN+IN+
CD+IN’+NNP’
Shivaji was born in 1627 at Poona.
R2’ S NNP+ VBN+(NNP’+IN’)+
(CD+IN)+VBD
)शवाजींचा जम पुणे येथे 1627 म4ये झाला.
R3 S PRP+NN+NNP+NNP’+
VBD+DT+NN’
His father Shahaji Bhonsle was a Jagirdar
R3’ S PRP+NN+NNP+NNP’+DT+
NN’+VBD
6यांचे वडील शहाजी भोसले एक जहागीरदार
होते
R4 S NNP+VBD+IN+NNP’
Shivaji lived at Bijapur
R4’ S NNP+NNP’+IN+VBD
)शवाजी 7वजापूर येथे राहत होते
R5 S PRP+NN+NNP+NNP’+VBD+
DT+RB+JJ+NN’
His mother Jija Bai was a very pious lady
R5’ S PRP+NN+NNP+NNPJJ+NN’+
VBD
6यांची आई िजजा बाई एक अ1तशय धा)म8क
वृ6तीची म:हला होती.
R6 S NN+VBD+VBG+NNS
Seema was peeling potatoes
R6’ S NN+NNS+VBG+VBD
सीमा बटाटे सोलत होती
R7 S NNP+VBZ+DT+NN+TO+
VB+TO’+NN’
Knowledge is the way to go to heaven
R7’ S NNP+(TO’+NN’)+(TO+VB)+
(DT+NN)+VBZ
;ान वगा8कडे जायाचा राता आहे
R8 S PRP+VBZ+DT+JJ+NNP
It is a costly pen
R8’ S PRP+DT+JJ+NNP+VBZ
तो एक महाग पेन आहे
4. CONCLUSION
This paper presents a novel approach for English to Marathi language translation. This proposed
system follows transfer approach with rule based translation. The main focus revolves around
morphological synthesizer. We have presented a complete architecture for this MT system and
several algorithms for this system. Proposed system uses sentence construction rules both for
English and Marathi language. It uses a Stanford parser for tagging and chunking which
eliminates the need of unnecessary computations. English to Marathi dictionary will be used to
support efficient translation. The goal of developing such translation system is to make the
resources available to everyone.

9
ACKNOWLEDGEMENT
I wish to express my deep sense of gratitude to our guide Prof. Sharvari Govilkar, for providing
me with the idea of this topic and for her invaluable support, guidance and supervision. I would
like to thank my Principle and Head of Department for their invaluable support.
REFERENCES
[1] Devika Pishartoy, Priya, Sayli Wandkar (2012) “Exteneding capabilities of English to Marathi
machine Translator”
[2] Abhay A, Anita G, Paurnima T, Prajakta G (2013), “Rule based English to Marathi translation of
Assertive sentence”
[3] Krushnadeo B, Vinod W, S.V.Phulari, B.S.Kankate (2014), “A novel approach for Interlingual
example-based translation of English to Marathi”
[4] G.V.Gajre, G.Kharate, H. Kulkarni (2014), “Transmuter: An approach to Rule-based English to
Marathi Machine Translation” (2014)
[5] Pushpak B. Jignashu P. “Interlingua Based English-Hindi Machine Translation and Langauge
Divergence”
[6] Charugatra Tidke, Shital B, Shivani P (2013) “Inflection Rules for English to Marathi Machine
Translation”
[7] www.cflit.iitb.ac.in/indic-translator/
[8] www.saakava.com/index.aspx
[9] https://translate.google.co.in
Authors
Ms. Amruta Godase is Pursuing M.E. in Information Technology with Specialization
in Artificial Intelligence Robotics from Mumbai University. She received her
polytechnic from MSBTE B.E from Mumbai University.
Sharvari Govilkar is Associate professor in Computer Engineering Department,at PIIT,
New Panvel, University of Mumbai, India. She has received her M.E in Computer
Engineering from University of Mumbai. Currently She is pursuing her PhD in
Information Technology from University of Mumbai. She is having Sixteen years of
experience in teaching. Her areas of interest are Text Mining, Natural language
processing, System programming Compiler Design etc.

A Novel Approach for Rule Based Translation of English to Marathi

More Related Content

Similar to A Novel Approach for Rule Based Translation of English to Marathi (20)

Recently uploaded (20)

A Novel Approach for Rule Based Translation of English to Marathi