ISOLATING WORD LEVEL RULES IN TAMIL LANGUAGE FOR EFFICIENT DEVELOPMENT OF LANGUAGE TOOLS

International Journal on Natural Language Computing (IJNLC) Vol.8, No.1, February 2019
DOI: 10.5121/ijnlc.2019.8103 29
ISOLATING WORD LEVEL RULES IN TAMIL
LANGUAGE FOR EFFICIENT DEVELOPMENT OF
LANGUAGE TOOLS
Suriyah M, Aarthy Anandan, Anitha Narasimhan and Madhan Karky
Karky Research Foundation, India
ABSTRACT
With the advent of social media, the amount of text available for processing across different natural
languages has become enormous. In the past few decades, there has been tremendous increase in the
number of language processing applications. The tools for natural language computing of various
languages are very different because each language has its own set of grammatical rules. This paper
focuses on identifying the basic inflectional principles of Tamil language at word level. Three levels of
word inflection concepts are considered – Patterns, Rules and Exceptions. How grammatical principles for
word inflections in Tamil can be grouped in these three levels and applied for obtaining different word
forms is the focus of this paper. These can be made use of in a wide variety of natural language
applications like morphological analysis, morphological generation, word level translation, spelling and
grammar check, information extraction etc. The tools using these rules will account for faster operation
and better implementation of Tamil grammatical rules referred from [த ொல்த ொப்பியம் |
tholgaappiyam] and [ நன்னூல் | nannool] in NLP applications.
KEYWORDS
Natural language processing, Rule based approach, word level rules, Tamil tool, language tools
1. INTRODUCTION
The number of social media platforms for various purposes like blogging, micro-blogging, photo
sharing, social networking etc., has risen steadily. The number of people who are members of the
same has also increased. For instance, the micro-blogging website Twitter has about 330 million
users (on 1st January 2018) and about 500 million tweets (on 24th January 2017) are sent on an
average per day [1]. This in turn, has paved way for collection and processing of various forms of
data – text, image, video etc., generated with high velocity and veracity. Focusing on the
processing of text, the posts of users in different natural languages have to be analysed; this
avenue of NLP has many challenges. One of them is to analyse text and interpret the meaning
conveyed. This is not possible without a tool which has knowledge of all the grammatical rules
followed in the natural language considered. Different tools need to be used for processing
different natural languages because of the differences that exist in terms of the syntax and
grammatical rules. In this work, a three level grouping of grammatical principles for word
inflections in Tamil language is proposed for better grammatical analysis.

30
Inflection in grammar refers to the modification of a root word to convey various characteristics
like tense, gender, number, person, case etc. Inflection expresses one or more grammatical
categories with a prefix, suffix or infix, or another internal modification such as a vowel change
[2].
For example, consider the Tamil word [கிளிஞ்சல் ள் | kiLinjalgaL | sea shell <plural>].
Here, [கிளிஞ்சல் | kiLinjal | sea shell] is the root word which is a noun in singular form. The
suffix [ ள் | kaL |] has been added to obtain the inflected version of the word in the plural form.
[கிளிஞ்சல் தளொல் | kiLinjalgaLaal | sea shell <plural><instrumental case>] is another
inflected form of the same noun where the root word கிளிஞ்சல் is added with two suffixes –
[ ள் | kaL |] and [ஆல் | aal |] to denote plural and instrumental case respectively.
[கிளிஞ்சல் ளிடமிருந்து | kiLinjalgaLidamirundhu | sea shell <plural><locative
case><ablative case>] is yet another form of the same root word [கிளிஞ்சல் | kiLinjal | sea
shell] with three suffixes. The suffixes are [ ள் | kaL |] denoting plural, [இடம் | idam |]
denoting locative case and [இருந்து | irundhu |] denoting ablative case.
EXAMPLE 1: SAMPLE NOUN INFLECTIONS
1.1 கிளிஞ்சல் ள் ← கிளிஞ்சல் <root word> <noun> + ள் <plural>
kiLinjalgaL ← kiLinjal + kaL
sea shell<plural> ← sea shell + plural suffix
1.2 கிளிஞ்சல் தளொல்← கிளிஞ்சல்<root word><noun> + ள்<plural> +
ஆல்<instrumental case> kiLinjalgaLaal ← kiLinjal + kaL + aal
sea shell <plural><instrumental case> ← sea shell + plural suffix + instrumental case
suffix
1.3 கிளிஞ்சல் ளிடமிருந்து ← கிளிஞ்சல் <root word> <noun> + ள் <plural> +
இடம் <locative case> + இருந்து <ablative case> kiLinjalgaLidamirundhu ← kiLinjal
+ kaL + idam + irundhu
sea shell<plural><locative case> <ablative case> ← sea shell + plural suffix + locative case
suffix +ablative case suffix
Example 2: Sample verb inflections
2.1 பொடுவொய் ← பொடு<verb> + வ் <future tense> + ஆய் <second person> <singular>
paaduvaay ← paadu + v + aay
sing <imperative> ← to sing + future tense + second person + singular
2.2 பொடி ்த ொண் டிருந் ொள் ← பொடு<verb> + இ<past tense marker> + ்<sandhi> +
த ொண் டிரு<auxiliary verb> + ந் ்<past tense marker> + ஆள் <third person femine
singular suffix > paadikkoNdirundhaaL ← paadu + i + k + koNdiru + nth + aaL she was
singing ← to sing + past continuous + third person + singular + feminine

31
2.3 பொடுவொர் ள் ← பொடு<verb> + வ் < future tense marker > + ஆர் < pronominal
third person singular > + ள் < plural suffix >
paaduvaargaL ← paadu + v + aar + kaL they will sing ← to sing + future tense + third
person + plural
In this work, three levels of word inflection rules are considered – Rules, Patterns and Exceptions.
How grammatical rules for word inflections in Tamil can be grouped in these three levels and
applied for obtaining different word forms is the focus of this paper. For the sake of simplicity, in
this paper we call a morphological ending, a pattern. A Pattern can be a number or case ending
associated with a noun or a person-number-gender marker associated with a verb. We have a set
of Rules and Exceptions associated with each pattern. A Rule will define how the pattern can add
to a root word or how it can separate from a morphological variation to obtain the root. An
Exception to a rule will handle special cases where the rule cannot be applied directly. These
three levels of grammatical forms discussed in this paper facilitate simple and efficient
implementation of grammatical rules in basic NLP tools like morphological analyser,
morphological generator, word level translator etc.
2. LITERATURE REVIEW
The three main types of morphological relations are Inflection, Derivation, and Compounding.
Inflectional morphology modifies the properties of lexemes, while maintaining the basic meaning
of the lexeme [3].
Singh and Sarma aim at reducing the number of affixes and word generation and re-adjustment
rules describing nominal inflection by using inflectional class information to analyse various
forms of Hindi nouns. They propose that the inflectional behaviour of nominal forms should be
used entirely in the formation of nominal classes in Hindi [4].
To Ramscar, it appears that inflection is carried out through analogical reminding based on
semantic and phonological similarity and that a rule-based route is not necessary to account for
past tense inflection [5].
Agglutination is a process of morphology in which complex words are formed by joining
morphemes together without changing the spelling or phonetics [6]. Prominent south Indian
languages like Tamil, Telugu, Malayalam, Kannada and Tulu are agglutinative [7].
Reddiah suggests that in addition to having similar sounds, all Dravidian languages are related
closely in grammatical aspects too. There are similarities in prefix, suffix, numbers, gender and
syntax and that all the Dravidian Languages are formed from one main Dravidian language,
which may be Tamil by most scholars’ opinions [8].
Anand kumar et al propose a machine learning based approach for morphological analysis using
SVM. This is a supervised model which needs voluminous training data to get good results [9].
But rule based approaches do not have this overload of having enormous tagged training data.
TamilComputingLab morphological analyser is a conventional rule based one [10]; Piripori is the
Tamil morphological analyser built on the principles discussed in this paper [11]. It was observed
from the working of the above two morphological analysers that when morphological rules are

32
grouped and streamlined following the approach mentioned in this work, the speed of processing
and the accuracy with respect to precision and recall are increased compared to the conventional
rule based approach. The grouping of inflectional rules in Tamil accounts for faster and more
accurate operation of language tools; hence becoming necessary for better tools and applications.
3. DESCRIPTION OF INFLECTIONAL RULES IN TAMIL
Tamil script has 12 vowels [உயிதரழு ்து ் ள் | uyirezhuthukkaL], 18 consonants
[தமய்தயழு ்து ் ள் | meyyezhuthukkaL] which combine to form 216 combinational
letters [உயிர்தமய்தயழு ்து ் ள் | uyirmeyyezhuthukkaL] summing up to 247 letters.
Tamil has more than 4 lakh words. These words are classified as [தபயர்ச்தசொல் | peyarchol |
noun], [வினனச்தசொல் | vinaichol | verb], [இனடச்தசொல் | idaichol | preposition and post
position], [உரிச்தசொல் | urichol | adjective], [இல ்கியச்தசொல் | ilakkiyachol | literary
words]. Apart from the classes mentioned, Tamil has the ability to adapt words from other
languages; these words are called [வடதசொல் | vadasol | words adapted from Sanskrit],
[தினசச்தசொல் | thisaichol | word adapted from other languages like Urdu, Portuguese,
Telugu etc] [12]. These words combine with various prefixes and suffixes to yield their different
inflectional forms. When these combinations happen, certain alphabet transformations may (or
may not) happen. This work focuses on the collection of different prefixes and suffixes added to
Tamil root words (which are nouns or verbs) and the different grammatical transformations which
may happen during inflection. [த ொல்த ொப்பியம் | tholgaappiyam] and [நன்னூல் |
nannool] are the sources of the inflectional rules discussed in this paper.
4. PROPOSED WORK
4.1 PATTERNS
These are suffixes added to root words to get the different morphological variations of the words
considered. These patterns contain morphological information of number and case for nouns;
tense, number, case, person and gender for verbs.
Example 3 (Inflectional forms of a noun): The Tamil root word, [மலர் | malar | flower], has
been taken as a sample to demonstrate some of the inflectional forms of nouns in Table 1. [மலர்
| malar | flower] is a noun whose inflected forms are obtained by mere suffixing of appropriate
patterns. Some other nouns may require some transformation, addition or removal of certain
letters. This is discussed in the Rules section.

34
transformation of one or more alphabet to obtain the inflected form when combining with a
pattern. Such examples are discussed in the “Rules” part.
Table 2. Different inflectional forms of the verb [பொர் | paar | see]
Pattern Implication
of the
pattern
Inflectional split of the
pattern
Inflected form of the
verb
Description
[ப்பொய் | ppaay
]
<second
person>
<singular>
<future tense>
ப்(சந்தி) +
ப்(எதிர்த ொல
இனடநினல ) + ஆய்
(முன்னினல ஒருதன)
[பொர்ப்பொய் | paarppaay |
you will see ]
[பொர் + ப் +
ப் + ஆய் |
paar + p + p +
aay]
[ப்பொர் ள்|
ppaargaL]
<third person>
<plural> <future
tense>
ப்(சந்தி) + ப்
(எதிர்த ொல
இனடநினல) + ஆர் ள்
(படர் ்ன
பலர்பொல்விகுதி)
[பொர்ப்பொர் ள் |
paarppaargaL | they will see
]
[பொர் +ப் +
ப் + ஆர் ள்
| paar +p + p
+ aargaL]
[ப்பொள்| ppaaL] <third person>
<singular>
<future tense>
<female>
இனடநினல ) +ஆள்
(தபண் தபொல் விகுதி)
[பொர்ப்பொள் | paarppaaL |
she will see]
[பொர் +ப் +
ப் +ஆள் |
paar +p + p
+aaL ]
[ப்பீர்| ppeer] <second
person>
<plural> <future
tense>
இனடநினல ) + ஈர்
(முன்னினல பன் னம
விகுதி)
[பொர்ப்பீர் | paarppeer |
you will see ]
[பொர் + ப் +
ப் + ஈர் |
paar + p + p +
eer ]
[ப்பார்| ppaar] <third person>
<plural> <future
tense>
இனடநினல ) + ஆர்(
படர் ்ன பலர்பொல்
விகுதி)
[பொர்ப்பொர் | paarppaar |
they will see]
[பொர் +ப் +
ப் + ஆர் |
paar +p + p +
aar]
[ப்பபன் | ppaen] <first person>
<singular>
<future tense>
இனடநினல ) +
ஏன் ( ன் னம ஒருனம)
[பொர்பபன் | paarppaen| i
will see]
[பொர் +ப் +
ப் + ஏன் |
paar +p + p +
aen]
[ப்பபொம்|
ppoam]
<first person>
<plural> <future
tense>
இனடநினல ) + ஓம்
( ன் னம பன் னம
விகுதி)
[பொர்ப்பபொம்| paarppoam|
let us see]
[பொர் +ப் +
ப் + ஓம் |
paar +p + p +
oam ]
[ப்பொன் | ppaan] <third person>
<singular>
<future tense>
<masculine>
இனடநினல ) + ஆன்
(ஆண் பொல் விகுதி)
[பொர்ப்பொன் | paarppaan |
he will see ]
[பொர் +ப் +
ப் + ஆன் |
paar +p + p +
aan]
[ப்பீர் ள்|
ppeergaL]
<second
person>
<plural> <future
tense>
ப்(சந்தி) + ப் (எதிர் ொல
இனடநினல) + ஈர் ள்
(முன்னினல பன் னம
விகுதி)
[பொர்ப்பீர் ள் |
paarppeergaL | you will see]
[பொர் +ப் +
ப் + ஈர் ள் |
paar +p + p +
eergaL]

35
4.2. RULES
Each pattern has a set of Rules associated with it which records the list of all possible
transformations which may occur when that pattern combines with different words to form the
inflected forms of the same.
When a root word combines with a specific pattern (suffix with some morphological
information), the former may have a change in one or more letters, or may have one or more
letters added to it to form the final inflected form. These transformations are governed by
grammatical principles described in [13], [14] and [15]. These are classified as “Rules” under
each pattern i.e., each pattern has an associated set of Rules governing the transformations which
may occur when the pattern combines with a word to form the inflected form of the latter.
The relationship between patterns and rules is illustrated in the Figure 1.
Let
n - number of patterns
Rules(x) - the set of rules for the pattern ‘x’
Rule x,y - Rule number ‘y’ of the pattern ‘x’
npx - number of rules for pattern ‘x’
Each of the ‘n’ patterns have a set of rules associated with it. A pattern is a suffix which can be
affixed with a word to form its morphological variants. Each of the rules under a pattern are
numbered with a unique number also denoting the pattern it is associated with.
Fig 1. Relationship between patterns and rules

38
[த ொள்|
koL| have ]
[ஆன் |
aan]
<third person>
<singular>
<past tense>
<masculine>
[த ொண் டொன் |
koNdaan| he had]
[த ொண் ட +
ஆன் | koNda
+ aan]
[த ொள் +
ஆன் | koL
+ aan]
[சொப்பிடு |
chaappidu | eat
]
[ஆன் |
aan]
<third person>
<singular>
<past tense>
<masculine>
[சொப்பிட்டொன்
| chaappittaan he
ate]
[சொப்பிட்ட
+ ஆன் |
chaappitta +
aan]
[சொப்பிடு
+ ஆன் |
chaappidu +
aan]
The verb [வொ | vaa | come] combines with the pattern [ந் ொன் | ndhaan] to form [வந் ொன் |
vandhaan| he came] i.e., the alphabet [வொ | vaa] changes to [வ | va]. The same verb combines
with the pattern [கி றொ ன் | kiRaan] to form [வருகி றொ ன் | varugiRaan| he comes]. In this
inflection, the alphabet [வொ | vaa] changes to [வ | va] and the alphabet [ரு | ru] gets added as
well.
5. RESULTS AND DISCUSSION
280 noun patterns and 880 verb patterns were collected. The number of rules for nouns, verbs and
compound words are 50, 124 and 54 respectively. Each pattern is given a specific ID. All the
rules applicable for a specific pattern are grouped with the ID allotted for the pattern. These
inflection rules form a repository which can be used to form a variety of NLP tools like
morphological analyser, morphological generator, word level translator, spelling and grammar
checker etc.
The grouping and classification of inflectional rules in Tamil discussed in this paper have been
applied to build Piripori [11], a Tamil morphological analyser. A list of 232292 Tamil words
were taken from different sources and were input to Piripori and Tamil Computing Lab analyser.
In Table 6, the analysis times for 137144 Tamil words for Tamil Computing Lab analyser [10]
and Piripori [11] are tabulated. The time taken by Piripori to analyse the 232292 words given as
input is lesser than that taken by Tamil Computing Lab analyzer. The graphical representation of
the data is given in Figure 2.
Table 6. Analysis times of TaCoLa and Piripori analysers

39
Fig 2. Comparison of analysis times of Tamil Computing Lab analyser and Piripori
To compare precision and recall of the two analysers, a standard POS tagged document [16] was
taken from LDCIL website as test data. The file consisted 164 words out of which 121 were
nouns and 43 were verbs. Precision and recall of analysis of noun and verbs in the
abovementioned file were calculated separately. The results of the same are tabulated in Table 7.
Table 7. Data for precision and recall calculation
Tacola Piripori
No. of nouns correctly classified 85 110
No. of verbs correctly classified 20 25
No. of nouns classified as verbs 9 0
No. of verbs classified as nouns 12 5
Precision and Recall are calculated using the following formulae.
Precisionnoun = (No. of nouns correctly analysed) / (No. of words analysed as nouns)
Precisionverb = (No. of verbs correctly analysed) / (No. of words analysed as verbs)
Recallnoun = (No. of nouns correctly analysed) / (No. of nouns)
Recallverb = (No. of verbs correctly analysed) / (No. of verbs)
The results of the calculation are tabulated in Table 8 and represented graphically in Fig 3.
Table 8. Precision and Recall of nouns and verbs
TaCoLa Piripori
Precision of Nouns 0.87628866 0.9565217
Recall of Nouns 0.702479339 0.9090909
Precision of Verbs 0.689655172 1
Recall of Verbs 0.465116279 0.5813953

40
The precision and recall values of nouns and verbs of Piripori, which is based on the grouping of
grammatical rules discussed in this paper are higher than those of Tamil Computing Lab analyser.
Thus with the application of the grouping of grammatical rules into Patterns, Rules and
Exceptions as discussed in this paper has led to a better performing Tamil morphological analyser
in terms of analysis time, precision and recall compared to another analyser following the
traditional rule based approach.
Fig 3. Comparison of Precision and Recall
.
6. CONCLUSION
Three levels of Tamil word inflection rules are considered in this work – Rules, Patterns and
Exceptions. How grammatical rules for word inflections in Tamil can be grouped in these three
levels and applied for obtaining different word forms is the focus of this paper.
Efficient application of syntax and grammatical rules of a natural language in a natural language
application or tool is very important for minimal error. This work focuses on the collection of
inflectional rules in Tamil grammar referring [13], [14] and [15] thus forming a repository which
can be used in various NLP tools and applications. There has been great improvement in retrieval
speed and accuracy with respect to precision and recall when these rules were used to build a
morphological analyser compared to the traditional rule based approach.
ACKNOWLEDGEMENTS
The authors would like to thank Subalalitha C.N, Kanniammal.R, Manju Bashini P,
Elanchezhiyan K, Rajapandian C, Swathi P.G and Sindhuja G for their guidance and contribution
to this work

41
REFERENCES
[1] Omnicore.[Online]. Available: Https://Www.Omnicoreagency.Com/Twitter-Statistics/
[2] L.J.Brinton, The Structure Of Modern English: A Linguistic Introduction. Amsterdam, Philadelphia,
PA: John Benjamins, 2000.
[3] UC Sandiego Linguistics Department.[Online]. Available:
Http://Grammar.Ucsd.Edu/Courses/Lign120/08-Intro_Rev.Pdf
[4] S. Singh And V. M Sarma, “Hindi Noun Inflection And Distributed Morphology” In Proceedings Of
The 17th International Conference On Head-Driven Phrase Structure Grammar, 2010, Pp. 307321
[5] M. Ramscar , “The Role Of Meaning In Inflection: Why The Past-Tense Does Not Require A Rule,”
Cognitive Psychology, Vol. 45, No. 1, Pp. 45–94, 2002.
[6] Wikipedia.[Online]. Available: Https://En.Wikipedia.Org/Wiki/Agglutination
[7] Wikipedia.[Online]. Available: Https://En.Wikipedia.Org/Wiki/Agglutinative_Language
[8] S. C. Reddaiah. “Dravidian Languages And Its Fundamental Grammar,” Indian Journal Of Research,
Vol. 3, No. 2, Pp. 164-166, 2014.
[9] Anand Kumar M, Dhanalakshmi V, Soman K.P And Rajendran S, “A Sequence Labeling Approach
To Morphological Analyzer For Tamil Language”, International Journal On Computer Science And
Engineering, Vol. 2, No. 6, Pp. 1944 – 1951, 2010
[10] P. Anandan, K. Saravanan, R.Parthasarathi And T. V. Geetha, “Morphological Analyzer For Tamil”
In Proceedings Of International Conference On Natural Language Processing, 2002
[11] Suriyah M, Aarthy Anandan, Anitha Narasimhan And Madhan Karky, “Piripori - Morphological
Analyser For Tamil” In International Conference On Artificial Intelligence, Smart Grid And Smart
City Applications, 2019.
[12] [ ளஞ்சியம் | Kalanjiyam].[Online]. Available: Http://Store.Tamillexicon.Com
[13] Maanikkavaasakan, Tholkaappiyam, Chennai, TN : Uma Padhippagam, 2010
[14] A. Manikkam, Nannool Kaandigaiyurai,Chennai, TN : Poompuhar Padhippagam, 1988
[15] Seeni Naina Muhammad, Nalla Tamizh Ilakkanam, CITY, TN : Adayalam Padhippagam, 2013
[16] Linguistic Data Consortium For Indian Languages. [Online]. Available:
Http://Www.Ldcil.Org/Standardstextpos.Aspx

42
AUTHORS
Suriyah M is a post graduate in Computer Science and Engineering with 4 years’ experience in teaching
graduate engineering students. Now she is a research writer with Karky Research Foundation.
Aarthy Anandan is a post graduate in Computer Applications. Now she is working as Software Developer
with 4 years’ in Karky Research Foundation.
Anitha Narasimhan is a post graduate in Computer Applications. Now she is working as Software
Developer with 4 years’ in Karky Research Foundation.
Madhan Karky is a Ph.D in Computer Science. He is the Founder and Research Head of Karky Research
Foundation, a nonprofit educational research organization which focuses on language computing and
language literacy.

ISOLATING WORD LEVEL RULES IN TAMIL LANGUAGE FOR EFFICIENT DEVELOPMENT OF LANGUAGE TOOLS

More Related Content

What's hot (16)

Similar to ISOLATING WORD LEVEL RULES IN TAMIL LANGUAGE FOR EFFICIENT DEVELOPMENT OF LANGUAGE TOOLS (20)

Recently uploaded (20)

ISOLATING WORD LEVEL RULES IN TAMIL LANGUAGE FOR EFFICIENT DEVELOPMENT OF LANGUAGE TOOLS