Chapter-1

Computational Morphology and Sanskrit

1.0 Introduction
Morphology is a branch of linguistics that deals with word formation, analysis and
generation. It is concerned with how words are formed or created in a language from
smaller units in a systematic way. It has to find as well as to describe the mechanism
behind this process. Each language has morphemes which are its smallest meaningful
units. Realization of morphemes as part of a word is called morph. In formation of words
from these smaller units, certain processes such as inflection, derivation and
compounding etc. are involved. Also involved is morphotactics that determines how
morphs should be put together to form words.

Computational Morphology (CM) is application of morphological rules in the field of

computational linguistics. CL is an emerging area in Artificial Intelligence that is related
to accomplish linguistic tasks through computational methods. CM deals with the
processing of words in both their graphemic and phonemic forms. Its most basic task can
be defined as taking a string of characters or phonemes as input and delivering an
analysis as output. It has various practical applications from low-level applications to
speech and language processing systems.1

Low-level applications

Low-level applications of CM are simpler tasks such as hyphenization, spelling

correction, stemming etc.

1
Harald Trost, Entry under “Morphology” in Oxford Handbook of Computational Linguistics

4
Hyphenation is almost exclusively done automatically. Segmenting words correctly into
their morphs helps to solve the task. The major problem is spurious segmentations.

Spelling correction, in its simplest form, can be a mere comparison of the input against a
list of word forms. But such a list will never contain all the words occurring in a text and
enlarging the list will include more and more obscure words that will match with typos
thus preventing their detection. Therefore, most systems use a root lexicon, plus a
relatively small set of affixes and simple rules to cover morphotactics.

Stemmers are used in information retrieval to reduce as many related words and word
forms as possible to a common canonical form which can then be used in the retrieval
process. However, this canonical form is not necessarily the base form. The main
requirement is – like in all the above tasks – robustness.

Another application is text-segmentation in Chinese, Japanese or Korean. In these

languages words in a sentence are not separated by blanks or punctuation marks.
Morphological analysis can be used to perform the task of word separation.

Natural language applications

Natural language processing (NLP) is a subfield of artificial intelligence and linguistics.

It studies the problems of automated generation and understanding of natural human
languages. Natural language generation systems convert information from computer
databases into normal-sounding human language, and natural language understanding
systems convert samples of human language into more formal representations that are
easier for computer programs to manipulate.2 CM has a wider scope of being applied in
natural language processing systems involving parsing and/or generating natural
language utterances in written or spoken form. Such applications can range from message

2
http://en.wikipedia.org/wiki/Natural_language_processing

5
and information extraction to dialogue systems and machine translation. For many
current applications, only inflectional morphology is considered.

In a parser, morphological analysis of words is an important prerequisite for syntactic

analysis. Properties of a word the parser needs to know are its part-of-speech category
and the morphosyntactic information encoded in the particular word form. Another
important task is lemmatization, i.e., finding the corresponding dictionary form for a
given input word, because for many applications a lemma lexicon is used to provide more
detailed syntactic (e.g, valency) and semantic information for a deep analysis.

In generation, on the other hand, the task is to produce the correct word form from the
base form plus the relevant set of morphosyntactic features.

Speech applications

A text-to-speech system takes (electronically stored) text as input and produces speech
from it. Morphological analysis helps to solve two different tasks in such systems. One is
to guide the grapheme-to-phoneme conversion. Characters are often ambiguous with
respect to their translation into phonemes. Finding out the underlying morphological
structure is necessary for solving the task correctly.

A less obvious application is the use of morphological analysis to help in determining the
part-of-speech category of words. This is an important prerequisite of syntactic analysis
which is the basis for coming up with a correct prosody.

Speech recognition is a field where morphological analysis can become ever more
important. At the moment most available systems make use of full form lexicons and
perform their analysis on a word basis. Increasing demands on the lexicon size on the one
hand and the need to limit the necessary training time on the other hand will make
morph-based recognition systems more attractive.

6
1.1 Morphological Analysis and Morphological Analyzers

Morphological analysis is the process in which a word is analyzed into its root word and
associated morphemes. Morphological analyzers are programs used to provide
morphological analysis of a language. They include recognition engine, lexicon/thesaurus,
and algorithms to find out stem within an input word and identifying its suffixes.3 They
take as input a sentence, i.e. a sequence of words and retrieve related grammatical
information such as lexical category, gender, number, person, tense etc. of every word.
For example, for nouns they provide its gender, number and case information along with
the root word. For verbs, information related to tense, aspect, mood etc. is retrieved. If a
word has more than one meaning, the analyzer is supposed to return grammatical
information for every meaning separately. 4

It is the basic tool used in spell checker, grammar checker, parser and machine translation
systems. To build a syntactic representation of the input sentence, a parser must map each
word in the text to some canonical representation and recognize its morphological
properties.

For the purpose of analysis of morphologically rich languages, the root and the
morphemes of each word has to be identified. The combination of a surface form and its
analysis as a canonical form and inflection is called a lemma. The main problems
involved in the process are:

1. Morphological alternations: In a language the same morpheme may be realized in

different ways depending on the context.

2. Morphotactics: Stems, affixes, and parts of compounds do not combine freely, a

morphological analyzer needs to know what arrangements are valid.

3
http://www.acroterion.ca/Morphological_Analysis.html
4
Akshar Bharati, Vineet Chaitanya and Rajeev Sangal. "Natural Language Processing: A Paninian
Perspective", Prentice-Hall of India, New Delhi, 1995.

7
There are many approaches for morphological analysis. A popular approach to
morphological alternation is the cut-and-paste method in which the canonical form is
derived by removing and adding letters to the end of a string. Another one is the finite
state technology, which was introduced in early 1980s and is applied for automatic
recognition and generation of word forms. It maintains that rules for morphological
alternations can be implemented by finite-state transducers. It was also widely recognized
that possible combinations of stems and affixes can be encoded as a finite-state network.
An automaton containing inflected word forms can be upgraded to a morphological
analyzer, for example, by adding a code to the end of the inflected form that triggers
some predefined cut-and-paste operation to produce the lemma. Instead of cutting and
pasting it at runtime, the entire lemma can be computed in advance and stored as a finite-
state transducer whose arcs are labeled by a pair of forms. The number of nodes in this
type of network is small, but the number of arc-label pairs is very large as there is one
symbol for each morpheme-allomorph pair. A more optimal lexical transducer can be
developed by constructing a finite-state network of lexical forms, augmented with
inflectional tags, and composing it with a set of rule transducers. Lexical transducers can
be constructed from descriptions containing any number of levels. This facilitates the
description of phenomena that are difficult to describe within the constraints of the two-
level model. Because lexical transducers are bidirectional, they are generally non-
deterministic in both directions. If a system is only to be used for analysis, a simple
finite-state network derived just for that purpose may be faster to operate. 5

A morphological generator performs the reverse function of an analyzer. For every root
and its features provided as input, it generates the word-forms in the given paradigms.
Morphological analyzer and morphological generator are two essential and basic tools for
building any language processing application for a natural language.

5
http://www.languageinindia.com/aug2006/parsingrajendran.pdf

8
1.1.1 India’s linguistic scenario

The Indian subcontinent is remarkably rich in languages. Linguistic diversity is a unique

component of the cultural and political structure of this area. The 200 languages
enumerated in the Census are a linguistic subtraction of over 1,600 mother tongues
reported by the people indicating their perception of their linguistic identity and linguistic
6
difference. The linguistic diversity in India is marked by the fluidity of linguistic
boundaries between dialect and language, between languages around state borders and
between speech forms differentiated on cultural and political grounds. In spite of
diversity, linguistic identity is thin because of the large size of population of the country.
Some of the minor languages have more speakers than many European languages. The
linguistic differences between the Indian languages, particularly at the grammatical and
semantic levels, are less than expected, given their different historical origins. The mutual
convergence of these languages due to their intensive and extensive contact have made
India one linguistic area. Inter-translatability between those languages is therefore very
high.
The languages of India can be historically categorized into four major language families:
Indo–European : This is the most prominent language family. Most of the major Indic
languages leaving Dravidian languages belong to it. The Indo – European has the sub–
families, Indo–Aryan and Dardic/Kashmiri. Major languages are Hindi, Sanskrit, Punjabi,
Sindhi, Gujarati, Marathi, Bengali, and many other languages.
Dravidian: There are four prominent Dravidian languages – Telugu, Kannada, Tamil and
Malayalam. All of them are spoken in the southern most part of the country.
Austro-Asiatic: It has Munda and Mon-Khmer/Khasi as sub-families.
Sino – Tibetan: Tibeto-Burman and Thai/Kempti are sub-families of Sino-Tibetan.
The Indo-European which is commonly called 'Indo-Aryan’ has the largest number of
speakers followed by Dravidian, Austro-Asiatic, also called 'Munda' and Sino-Tibetan
which is commonly called the 'Tibeto-Burman'.

6
http://www.languageinindia.com/july2005/morphologynortheast1.html

9
Andamanese: Andamanese is the fifth language family of India.7 It comprises of the four
language groups represented by the four speech communities of the Great Andamanese,
the Jarawa, the Onge and the Sentinelese. Great Andamanese arguably is language family
in itself.8 All these languages are primitive languages and the speaking community of
these languages are the primitive tribes who have been living in these islands for
thousands of years, untouched from civilized society. They are people of the Negrito
stock and it is not clearly known where they came here from.

1.1.2 Development of Language Technology in India

Technology Development in Indian Languages (TDIL)9 was initiated by the Department

of Information Technology, Ministry of ICT, Government of India with the objective of
developing Information Processing Tools and Techniques to facilitate human-machine
interaction without language barrier; creating and accessing multilingual knowledge
resources; and integrating them to develop innovative user products and services. Since
then, TDIL has been facilitating and supporting the development of language technology
resources in all Indian languages, and promoting their dissemination. Towards this goal it
has established 13 Resource Centers for Indian Language Technologies (RCILTS) in
10
March 2000 at different Universities and Institutes. These Resource centers are aimed
at:

Improving the quality of life of people of India by enabling to use Information
Technology through Indian Languages..

Developing new products and services for processing information in Indian
Languages.

Facilitating research in technology areas such as Machine Translation (MT),
Optical Character Recognition (OCR), Text-to-Speech (TTS), and Speech
Recognition in Indian Languages that leads to product development.

7
Abbi, Anvita. 2006. Endangered Languages of Andaman Islands. Lincom Europa.
8
Ibid.
9
http://tdil.mit.gov.in/
10
http://tdil.mit.gov.in/languagetechnologyresourcesapril03.pdf

10
List of Resource Centers

Resource Centre Languages Associated With

Indian Institute of Technology, Kanpur Hindi, Nepali

Indian Institute of Technology, Mumbai Marathi, Konkani
Indian Institute of Technology, Guwahati Assamese, Manipuri
Indian Institute of Science, Bangalore Kannada, Sanskrit (Cognitive Models)
Indian Statistical Institute, Kolkata Bengali
Jawaharlal Nehru University, New Delhi Foreign Languages (Japanese,
Chinese) & Sanskrit (Language
Learning Systems)
University of Hyderabad, Hyderabad Telugu
Anna University, Chennai Tamil
MS University, Baroda Gujarati
Utkal University, Oriya
Department of Computer Science and Application Punjabi
Thapar Institute of Engg. & Tech., Patiala
ERDCI, Trivendrum Malayalam
CDAC, Pune Urdu, Sindhi, Kashmiri

A map showing main resource centers of TDIL across the country11

11
Source: TDIL website, http://tdil.mit.gov.in/

11
A few other funding agencies are also supporting the language technology efforts.
Several individuals in India and abroad and some industries have also been developing
different language products/resources. An example of a morphological verb analyzer can
be cited here that has been developed at JNU by a research scholar12 for an extremely
endangered language of Great Andamanese. As there are so many distributed efforts and
with so many languages it is very difficult to keep track of the activities in this area. So,
TDIL has tried to evolve an indexing system for the language technology resources, so
that a suitable portal can be created to offer a variety of services to persons concerned
with Indian language technologies. If all the resource centres follow this indexing and
share their documents with TDIL, it would be easy to keep track of the progress and
would be possible to encourage more people to participate in the development activities.

1.1.3 Survey of Morphological Tools for Indian languages

Morphological analyzer is an integral part of any Natural Language Processing system,

especially in the context of Indian languages where morphology plays significant role due
to high inflectional and derivational nature of these languages. For fixed word-order
languages, the semantics of a word are primarily governed by its absolute and relative
position within a sentence. But for free word-order languages, the position of words in
the sentence cannot provide much clue about its semantics. As in the case of Indian
languages, which are mostly free order, the semantics (part of speech and other
subtleties) are heavily dependent on the surface realization of the word. Therefore,
morphological analysis is inevitable to develop any NLP tool for Indian languages.
Unlike English and various foreign languages, most of the Indian languages can be
characterized by a rich system of inflections (vibhakti), derivation and compound
formation. The numbers of word are derived from the root word by some specific
orthographic rule in the Indian languages. A competent morphological analyzer is needed
for a machine to deal with the lexicons of these languages.

12
Choudhary, Narayan Kumar, 2006, “Developing a Computational Framework for the Verb Morphology
of Great Andamanese”, Unpublished dissertation at Centre for Linguistics, JNU.

12
The task of the morphological analyzer is to identify the structural components of a word,
and glean information about it. While Indian languages in general are morphologically
richer than languages like English, Dravidian languages are a lot more complex. Clearly,
there is no way to list all forms of all words in a dictionary. Therefore, morphology is
essential.

The development of Natural Language Processing in India has done significant progress
since its beginning. A variety of tools have been developed for different purposes. These
include machine translation systems, morphological analyzers and generators, POS
taggers, spell checkers, text processors, word nets and many other tools. Given the
morphological richness of these languages, numerous NLP tools related to morphology
have been developed. The range covered is from major languages like Hindi, Telugu and
Bengali to languages like Manipuri and Assamese. The survey of morphological
analyzers, generators and other tools for morphological understanding of various Indian
languages is given below. It mainly focuses on morphological analyzers and generators,
taggers, spell-checkers and text-processors. However, efforts and tools of government
institutions only have been covered leaving behind the initiatives of private sector which
also is putting a lot of effort in this field.

1.1.3.1Morphological Analyzers for Indian languages:

Morphological Tagger for Hindi

Vasu Renganathan from University of Pennsylvania has developed a Hindi

Morphological Tagger13. One has to enter a Hindi word or sentence and press 'Tag' to get
tagged words as output. It is available in Unicode and ITRANS compatibility.
The source code is written in PERL. The site also provides some of the data files

Morphological Analyzer for Telugu

13
http://ccat.sas.upenn.edu/plc/tamilweb/hindi.html

13
Telugu is highly inflectional and agglutinative language. It exhibits features like vowel
harmony, sandhi etc. Clitics, particles, vocatives are all part of the word. Auxiliary verbs
are used in various combinations to indicate complex aspects.

RCILTS for Telugu, Univ. of Hyderabad claims to develop the only large scale
morphological analyzer system built for Telugu.14 It has been built, tested against corpora
and refined over the past 10 years. The system uses a root word dictionary of 64,000
entries and a suffix list categorized into a number of paradigm classes. The basic
methodology is to look for suffixes, remove them taking care of sandhi changes and then
cross checking with the dictionary. Inflection, derivation, external sandhi are all handled.
There is also a separate morphological generator that can put together the roots and
affixes to construct complete word forms.

For POS Tagging to Telugu, a morphological analyzer developed by G. Uma Maheswara

Rao from University of Hyderabad is being used. It gives the way to analyze each and
every word in the Telugu corpus with the different possibilities.

Kannada Morphological Analyzer

RCILTS, Hyderabad has also developed a Kannada morphological analyzer and

generator using their Network and Process Model.15 A finite state network captures in a
declarative and bidirectional fashion all the affixes, their ordering and the various
permitted combinations. The process component takes care of sandhi changes when
affixes are added or removed. This model makes it possible to develop a morphological
analyzer, test it against a corpus and then get a generator of comparable performance with
no extra effort since the same network is used both for analysis and generation. In this
model, as it is claimed, a complete and detailed analysis is made at the level of each affix.

14
TDIL magazine ‘vishvabharata’ July 2003, accessible@ tdil.mit.gov.in/Telugu-UOHJuly03.pdf
15
Ibid.

14
Morphological Analyzer and Synthesizer for Hindi 16
17
Media Lab Asia Research Laboratory at IIT Kharagpur, WB has developed a
morphological analyzer for Hindi. It can identify the tense, aspect, modality, person,
gender and number along with the root (‘dhaatu’) of an inflected verb form. For this, the
morphological analyzer uses a Directed Acyclic Morphological Structure. For nouns, the
analyzer has to determine its vibhakti (inflection), suffixes and prefixes. It analyzes the
lexical word groups corresponding to noun and determines the karaka (semantic role).

The group also aims to perform the decompositions for sandhi and samaasa (conjugating
and compounding words) so as to have a powerful vocabulary for the system, and a
generalized prefix and suffix handler. The DAM structure developed is based on the fact
that possible inflections of a word can be identified by scanning the word backward from
the end, one step at a time.

A morphological synthesizer has also been developed to generate the structural word
given the root of the word, and the tense, aspect, mode, number, gender, and person.
Hindi is a GNP language - i.e. the inflected forms of the verbs are mainly inflected using
the gender, number and person information.

Morphological Analyzer and Synthesizer for Bangla 18

19
Media Lab Asia Research Laboratory at IIT Kharagpur has also designed a
morphological analyzer and synthesizer for Bengali. The morphological analyzer can
identify the tense, aspect, modality and person of an inflected verb form.

Bengali is a classifier language - where the inflection takes place mainly due to the tense,
aspect and mode. For second and third persons, there is the concept of familiarity. The
morphological analyzer has been modeled as finite state transducers which are given a

16
http://www.mla.iitkgp.ernet.in/technology.html
17
http://www.mla.iitkgp.ernet.in/Resource/index.html
18
http://www.mla.iitkgp.ernet.in/technology.html
19
http://www.mla.iitkgp.ernet.in/Resource/index.html

15
root and TAM and GNP information as input. The transducer’s job is to generate the
correct inflected form of the word by adding the appropriate suffix for Bengali.

Verbs are classified according to the vowels present in them. The modus operandi of the
synthesizer is to determine the category of the root, given an input. It then accordingly
synthesizes the ultimate form of the word depending on the input TAM GNP parameters.
The number of exceptions is limited, and therefore easily traceable.

Oriya Morphological Analyzer (OMA)

PG Department of Computer Science and Application, Utkal University, Bhubaneswar

has developed Oriya Morphological Analyzer (OMA), a computational model for the
analysis and generation of Oriya language.20 The major contents of OMA are:

i) Pronoun Morphology (PM),

ii) Inflectional Morphology (IM) and
iii) Derivational Morphology (DM).

The OMA has many constituent parts 21 such as OriNet Database (OD), which stores the
Oriya lexicon (Only root words) whereas OMA Engine (OE) processes the system and
Morphological Parsing (MP) parses the word according to orthographic rule. Decision
Tree (DT) decides to classify the morphemes and different functional programmes by use
of OMA.

The OMA system has been designed on the basis of Object-Oriented Approach (OOA) so
that different functions can be easily added to or deleted from it. Pronoun Morphology
and Inflection Morphology have been implemented in the OMA in such a manner that it
successfully runs with the OriNet system, Oriya Spell Checker (OSC) and Oriya
Grammar Checker (OGC). The OSC handles any type of word (derived, inflectional or
root) using the OMA. It also provides sufficient interface for applications involved in
Oriya Machine Translation (OMT), Word-Net for Oriya (OriNet), Oriya Spell Checker

20
http://www.ilts-utkal.org/oriyamorphological.htm

16
(OSC) and Oriya Grammar Checker (OGC). All these developments have been worked
out on the basis of the syntactic approach of Sanskrit language.

Developers have developed and implemented the Decision Tree (DT) and its respective
algorithm of each type of morphology, through which OMA runs successfully. While
performing morphological analysis, OMA not only deals with the study of words but also
its morphemes. The OMA system is useful for various applications in developing NLP
tools. For example, in OMT, there is need of root words, which is obtained through the
OMA. Similarly, it has typical use in OriNet, the Word-Net for Oriya language, for
searching any type of lexicon. Its Morphological Parser (MP) successfully handles
ambiguous morphemes and provides different alternatives for them.

Morph Analyzer for Assamese and Manipuri22

RCILTS at Department of Computer Science & Engineering, Indian Institute of

Technology Guwahati Assam23 has developed morphological analyzers for Assamese and
Manipuri. They are knowledge tools and work as an aid to MT system and Spell Checker.

Assamese Morphological Analyzer is based on the Stemming technique in which

affixes are added/ deleted according to the linguistic rules. The derived words are verified
with the existing Corpus/Dictionary to treat as valid words. This analyzer currently works
with twenty linguistic rules, though more rules can be added without alteration in code.
Modules are available in the form of API’s for customization.

Manipuri Morphological Analyzer also uses stemming technique and does

morphological processing with the help of the grammatical rules and the dictionaries of
roots and affixes. A root dictionary containing 3000 root words and an affix dictionary
containing 55 affixes are being used. Identification of linguistic rules for nouns and
pronouns has also commenced. A model tagger is added to tag and analyze word. The

21
http://www.ilts-utkal.org/publication/NLP/Abstract/abs-12.pdf
22
tdil.mit.gov.in/TDIL-OCT-2003/morph%20analyzer.pdf

17
tagger tags the lexical category of the root and the grammatical category of the affixes.
The code is written using Perl script and the user interface has been developed using
Perl/Tk. The dictionaries are stored in MS-Access database.

Morphological Analysis in Tamil

Tamil is a verb final, relatively free-word order, morphologically rich and agglutinative
language. Computationally, each root word in Tamil can take a few thousand inflected
word-forms. Subject-verb argument is required for the grammaticality of a Tamil
sentence. Tamil allows subject and object drop as well as verb less sentences. In addition,
the subject of a sentence or a clause can be a possessive Noun Phrase (NP) or an NP in
nominative or dative case. As Tamil is an agglutinative language, each root word can
combine with multiple morphemes to generate word forms.

The following is the list of computational morphological analysis attempted and/or

implemented for Tamil:24

1. Rajendran’s Morphological Analyzer for Tamil: This was one of the very first
efforts towards building a morphological analyzer for Tamil. It was initiated by
anusaraka group of researchers under whose guidance Rajendran, Tamil University
prepared this morphological analyzer for Tamil for Translating Tamil into Hindi at the
word level.

2. AUKBC Morphological Parser for Tamil: 25 AUKBC NLP team under the
supervision of Rajendran prepared a Morphological parser for Tamil. The API Processor
of AUKBC makes use of the finite state machinery like PC Kimmo. It parses, but does
notgenerate.

23
http://www.iitg.ernet.in/rcilts
24
“Parsing in Tamil: Present State of Art”, S. Rajendran, Ph.D.
http://www.languageinindia.com/aug2006/parsingrajendran.pdf
25
http://www.au-kbc.org/frameresearch.html

18
3. Vaishnavi’s Morphological Generator and Analyzer for Tamil: Vaishanvi has built
generators and analyzers for Tamil morphology. The generator implements the item and
process model of linguistic description. It works by the synthesis method of PC Kimmo.
The analyzer uses a hybrid model for Tamil. It is theoretically rooted in a blend of IA and
IP models of morphology. It constitutes an in-built lexicon and involves a decomposition
of words in terms of morphemes within the model to realize surface well-formed words-
forms. Thus it tries to define a transformation depending on the morphemic nature of the
word stem. The analysis involves a scanning of the string from the right to left periphery
scanning each suffix at a time stripping it, and reconstructing the rest of the word with the
aid of phonological and mopho-phonemic rules exemplified in each instance. This goes
on till the string is exhausted. For the sake of comparison AMPLE and KIMMO models
are implemented.

5. RCILTS-T’s Morphological analyzer for Tamil: 26 Resource Centre for Indian

Language Technological Solutions-Tamil, Anna University, Chennai has prepared a
morphological analyzer for Tamil named as Atcharam. It takes a derived word as input
and separates it into root word and associated morphemes. It uses a dictionary of 20,000
root words based on fifteen categories. It has two modules - noun and verb analyzer
based on 125 rules. It uses heuristic rules to deal with ambiguities. It can handle verb and
noun inflections. It is also used for extracting the root words from inflections in Tamil
Search Engine and Online Tamil Dictionary. The system was developed using Java on
Windows 2000 platform. It is expandable as more morphological rules can be added and
Dictionary size can be increased. It has been tested with a file from CIIL Corpus
producing approximately 75% result.

6. RCILTS-T’s Morphological generator for Tamil:

Resource Centre for Indian Language Technological Solutions-Tamil has also prepared a
morphological generator for Tamil. It is named as Atchayam. It generates words when
Tamil morphs are given as input. It has two major modules – noun and verb generators.
The noun section handles suffixes like plural markers, oblique form, case markers and

26
http://ns.annauniv.edu

19
postpositions. The verb section takes tense and PNG makers, relative and verbal
participle suffixes, and auxiliary verbs. It uses sandhi rules and125 morphological rules.
It handles adjectives and adverbs. It has word and sentence generator interfaces.

Others

Ganesan’s Morphological Analyzer for Tamil to analyze CIIL corpus. It uses
phonological and morphophonemic rules and takes into account morphotactic
constraints of Tamil. An efficient morphological parser has also been built.

Kapilan’s Morphological Analyzer is especially built for analyzing Tamil
Verbal Forms.

Deivasundaram’s Morphological parser built for a Tamil Word Processor. He
too makes use of phonological and morphophonemic rules and morphotactics
constraints of the language for developing his parser.

Ramasamy’s Morphological Generator for Tamil: Ramasamy has prepared a
morphological generator for Tamil.

Winston Cruz’s Parsing and Generation of Tamil Verbs makes use of GS
morph method for parsing Tamil verbs which does morphotactics by indexing.
The algorithm simply looks up two files to see if the indices match or not. The
processor generates as many forms as it parses and uses only two files.

Dhurai Pandi’s Morphological Generator and Parsing Engine for Tamil
Verb Forms: It is a full-fledged morphological generator and a parsing engine on
verb patterns in modern Tamil.

Punjabi Morphological Analyzer and Generator

The Advanced Centre for Technical Development of Punjabi Language, Literature and
Culture, Punjabi University, Patiala has developed a Morphological Analyzer and
Generator for Punjabi.27 As generator, the software displays the list of all the possible
word forms of any Punjabi root word, along with their respective grammatical
information. As analyzer, it identifies the grammatical attributes of any Punjabi word and

27
http://www.advancedcentrepunjabi.org/punjabi_mor_ana.asp

20
can also be used to search for any Punjabi word in it to know its root and other
grammatical information.

The application areas of the software include, automatic spelling and grammar checking,
natural language understanding, machine translation, speech recognition, speech
synthesis etc part of speech tagging, parsing. The common man can also get in-depth
information about the Punjabi words from the software. Knowing the grammatical
information of a word helps in its proper and error free use in sentences. It can also help
the beginners to learn new words and the specialists to create new terminology.

The database used in the software consists of more than 1.72 lakh Punjabi words,
grouped into 22 word classes such as noun, personal pronoun, reflexive pronoun, verb,
inflected and uninflected adverb, inflected and uninflected adjective, conjunction,
interjection etc.

Morphological Analysers for multiple languages

Akshara Bharathi Group 28 at Indian Institute of Technology, Kanpur, India and

University of Hyderabad, Hyderabad, India has developed morphological analyzers for
different languages. These are available for online download for Hindi, Marathi, Telugu,
Kannada and Punjabi. Downloaded system needs Linux Operating System, Perl, Perl
enabled vim(only for Telugu), GDBM, Flex. The site claims that Telugu has 95%
coverage (for arbitrary text in modern standard Telugu) and Hindi has 88% coverage.

IIT Bombay

RCILTS (CFILT - Marathi), IIT-Bombay, led by Dr. Pushpak Bhattacharya is a

happening place for NLP in India. With an objective to enable Indian language
information processing on internet, this project funded by MIT, Govt. of India has been
working on creating Lexical Knowledge Bases like Wordnet , tools like MorphA and
POS taggers for Hindi and Marathi and MTS with Universal Networking Language

28
http://ltrc.iiit.net/showfile.php?filename=onlineServices/morph/index.htm

21
(UNL) as interlingua. Wordnet for Hindi and Marathi with lot of literature on it can be
downloaded from the CFILT portal 29 . It also offers an overview of the UNL based
Interlingua approach to MT with its online versions of English to Hindi MTS, Hindi to
UNL conversion system, UNL to Hindi generation system. This centre is developing a
POS tagger for Hindi using a data driven approach, making use of graphical algorithm
Conditional Random Fields. This algorithm based tagger makes use of the spelling of the
words, their lexical attributes and the suffixes to achieve an accuracy of about 90% on the
BBC Hindi news corpora. It has a tagset of 30 tags with two pseudo tags S-START and
S-END marking the sentence boundaries.30

1.1.3.2 Spell-Checkers for Indian languages

Tel-Spell: Spell Checker for Telugu

RCILTS-Telugu, Univ. of Hyderabad has developed a spell checker for Telugu named
Tel-Spell.31 The system has been tested on large scale corpora and enhancements and
refinements have been going on for years. The new version is far simpler, more
transparent, portable, well documented, and conforms to standards. Research work has
also been taken up on developing stemming algorithms for Telugu. A pure corpus based
statistical stemming algorithm has been developed. The performance of this stemmer for
the spell checking application has been studied in various combinations with dictionary
and morphology based approaches. A 10 Million word corpus developed has been used to
build and test the performance. The performance of the system has been found to be
satisfactory both in terms of detection and correction of spelling errors. The Telugu spell
checker has also been integrated into AKSHARA advanced multilingual text processing
system.

Bangla Spell-checker

29
Resource Centre for Indian Language Technology Solutions (CFILT) - http://www.cfilt.iitb.ac.in/
(accessed: 14.10.2006)
30
Srivastava, Manish et al., 2006, ‘Conditional Random Field Based POS Tagger for Hindi’, in the
Proceedings of MSPIL, IIT-B, Mumbai
31
tdil.mit.gov.in/Telugu-UOHJuly03.pdf

22
RCILTS-Bengali Indian Statistical Institute, Kolkata has developed a Bangla Spell-
Checker. 32 In this spell-checker, only non-word errors are considered and not word
errors like substitution, deletion, insertion, and transposition error. In Bangla, errors may
occur due to phonetic similarity of characters or typographical mistakes. In this spell-
checker, the main technique of error detection is based on matching the candidate string
in the normal as well as in the reversed dictionary. Moreover, this approach is combined
with a phonetic similarity key based approach where phonetically similar characters are
mapped into a single symbol and a nearly-phonetic dictionary of words is formed. Using
this dictionary, phonetic errors can be easily detected/ corrected. Here a candidate string
first passes through the phonetic dictionary. If the word is not found in the dictionary and
also failed to give suggestion then it tries to divide the word in root part and suffix part by
separately verifying both. If an error is found, the spell-checker tries to provide
suggestions. If it fails, it checks whether the string is a conjunct word generated by
appending two noun words and suffix. Option for adding new words permanently or
temporarily is provided in the spell checker.
The spell-checker uses several files containing root words and suffix words. The main
dictionary contains about 60,000 root-words and 100,000 inflected words. Noun and verb
suffix files are also used. The spell-checker works fast and almost correctly detects the
non-word errors. However, it makes about 5% false alarm due to conjunct words formed
by euphony and assimilation as well as proper nouns in the corpus.

Gujarati Spell Checker

RCILTS-Gujarati, Maharaja Sayajirao University, Baroda is developing a Gujarati Spell

Checker. 33 It is based on a morphological analysis of Gujarati language. Morphological
analyzer covers the analysis of Nouns and Verbs of the language and generates effective
suggestions not greater then ten. All the non-Gujarati words along with different symbols
are ignored and are not checked. The Gujarati Spell Checker is integrated along with
Multi-Lingual Text Editor for the convenience to user. Currently the software is under

32
tdil.mit.gov.in/Bengali-ISIKolkataJuly03.pdf
33
tdil.mit.gov.in/Gujarati-MSUniversityBarodaJuly03.pdf

23
rigorous testing to find all the faults in the spell checker and morphological analyzer.
Developers have plans to increase the size of the root dictionary covering the maximum
possible words of the language and improving the algorithms for the correctness and
efficiency of the spell checker. They also aim at building an independent spell checker
which can be used on Unicode or ISCII compatible text.

NERPADAM - Malayalam Spell Checker

RCILTS-Malayalam, C-DAC, Thiruvananthapuram has designed a software subsystem
named Nerpadam 34 that can be integrated with Microsoft word as a macro or the
Malayalam editor style-pad developed by them, to check the spelling of words in a
Malayalam text file. While running as a macro in word, it functions as an offline spell
checker and can be use with a previously typed text file only. Both off line and online
checking are possible when it is integrated with the text editor. It generates suggestions
for wrongly spelt words. The system adapts a rule cum dictionary-based approach for
spell checking.

It incorporates a fully developed morphological analyser for Malayalam. This module

splits the input word into root word, suffixes, post positions etc. and checks the validity
of each using the rule database. Finally it checks the dictionary to find the root word. If
anything goes wrong in this checking it is detected as an error and the error word is
reprocessed to get 3 to 4 valid words, which are displayed as suggestion. The user can
add new words into a personalized data base file, which can be added to the dictionary.

Tamil Spell Checker

RCILTS for Tamil at Anna University, Chennai has developed a spell checker for Tamil
that takes care of the rich morphological structure of Tamil. After tokenizing the
document into a list of words, each word is passed to the morphological analyzer which
analyzes only the correct words. The morphological analyzer tries to split the suffix. In

34
tdil.mit.gov.in/Malayalam-CDACThiruvananthapuramJuly03.pdf

24
case of error, it passes the word to spelling verification and correction phase to correct the
mistake. When the correction of errors is completed, root word and all components are
sent to morphological generator (for word forming), which then generate the possible
corrected words as suggestions.

The Spell checker for Tamil helps the user to identify most of errors, which may occur
while typing. The tasks implemented in Tamil Spell checker are Case marker,
postposition checking and adjective checking for nouns, PNG marker checking for verbs,
Adverb checking, and adjacent key error checking.

Spell-checker for Punjabi

RCILTS-Punjabi at Thapar Institute of Engineering & Technology, Patiala 35 has

developed a spell-checker for Punjabi. Developers have retained multiple spellings for
the commonly used words and have used Harkirat Singh’s Shabad Jor Kosh as the base.
Additional words were taken from the dictionaries published by Punjabi University and
the State Language Department and the corpus developed by CIIL, Mysore. For
generating the suggestion list, a study was conducted to discover the most common errors
made by Punjabi typists. A list of similar sounding words and consonants was also
complied and a suggestion list using this knowledge was generated based on reverse edit
distance. In most cases, the right suggestion is presented as a default suggestion and the
user just needs to confirm the default suggestion and proceed with the next error.
Otherwise, the user needs to scroll through a list of suggestions and pick one as the right
one. The spell checker supports text typed in any of the popular Punjabi fonts as well as
ISCII encoded files. It is nearly complete now.

25
1.1.3.3 Text-processors for Indian languages

AKSHARA: An Advanced Multi-Lingual Text Processor

AKSHARA 36 is an advanced multi-lingual text processing tool. Various dictionaries,
morphological analyzers, spell checkers, OCR systems, TTS systems; text processing
tools have been included in it. Several text processing tools, Telugu spell checker and
Telugu TTS have also been integrated. AKSHARA encodes texts in a standard character
encoding scheme such as ISCII or UNICODE. Mapping to fonts is done only for the
purposes of display and printing - all other operations are performed on character
encodings. Attributes are included in an open XML style markup language called
Extensible Document Definition Language (XDL) developed by us. This makes it easy to
convert to and from various other encoding schemes thereby ensuring highest levels of
portability and platform independence.
A unique feature of AKSHARA is that it understands the script grammar and warns if
one tries to build ungrammatical syllables. AKSHARA has been successfully used to
clean up all the corpora at CIIL, Mysore.
AKSHARA is platform independent and also claimed to be also robust and reliable.
AKSHARA has been successfully used to develop a 10 Million word corpus of Telugu.
Full support for Regular Expressions and Finite State Machines is being integrated.
AKSHARA is unique in providing multilingual email sending as well as receiving
facilities. AKSHARA also enables one to develop interactive web pages in Indian
languages and English. These web pages work across platforms and browsers. All this is
made possible through WILIO technology. AKSHARA is freely available.

Bilingual Punjabi/English Word Processor:

A bilingual Punjabi/English word processor named Likhari 37 has been developed by

RCILTS-Punjabi at Thapar Institute of Engineering & Technology, Patiala. Likhari
supports word processing under the windows environment and allows typing and

35
http://tdil.mit.gov.in/Punjabi-TIETPatialaJuly03.pdf
36
http://ildc.gov.in/telugu/htm/Akshara.htm

26
processing in Punjabi Language through the common typewriter keyboard layout. It has
MS-Word compatible features and commands. It provides a number of features that make
the use of Punjabi Language on a computer easy and provides a number of tools to
increase the efficiency of the user. These tools include Bilingual Spell Checker with
suggestion list, on-screen keyboard layouts with composition reference for Punjabi
language typing, bilingual search and replace, sorting as per the language, alphabetical
order, technical glossaries and onscreen Bilingual dictionaries.
The main features of Likhari are:

Very simple user interface

Online active Keyboard for users who do not know how to type in Punjabi.

Choice of Phonetic, Remington and Alphabetic Keyboard layouts with
composition reference.

Bilingual Spell Checker for Punjabi and English.

Bilingual Search and Replace.

Support for sorting the text in English or Punjabi as per alphabetical order.

Extensive help at various levels to make it easy for the user to learn.

Nashir – Word processor for Urdu and other languages:

RCILTS-Urdu, Sindhi & Kashmiri, Centre for Development of Advanced Computing,
Pune has developed a word-processor named Nashir. 38 It is said to be capable of creating
documents in Perso-Arabic languages, and at the same time to layout complete
newspapers and magazines in Urdu, Sindhi, Kashmiri, Arabic and Persian. Each
document of the Nashir consists of a number of pages on which text blocks, graphics, etc.
can be placed. Nashir is also well suited for publishing segment. Spellchecker support
which uses a base-dictionary is added in the full version.

Nashir supports Nastaliq True Type fonts (presently 2 fonts) and Naskh fonts (presently
12 fonts) along with fonts for Sindhi and Kashmiri. It supports C-DAC & Phonetic
Keyboards as well as user-defined keyboards. It provides various drawing objects and

37
http://tdil.mit.gov.in/Punjabi-TIETPatialaJuly03.pdf
38
tdil.mit.gov.in/UrduSindhiKashmiri-CDACPuneJuly03.pdf

27
also supports the OLE Automation. Nashir supports Urdu, Sindhi & Kashmiri along with
English. A transliteration engine (uTRANS) has also been implemented which converts
an “.aci” (ISCII file) into transliterated version in Urdu script (Naskh/ Nastaliq). Rule
based transliteration has been developed for Hindi & Punjabi. The user can save the
document as HTML page, and thus Naskh as well Nastaliq scripts can be viewed on the
Internet.
Phonetic keyboard is supported. Both Horizontal and Vertical kerning is provided to
manually adjust a text. It has Horizontal and Vertical rulers in the GUI, dynamic font
settings for the Urdu and English fonts.

1.2 Sanskrit Morphology

1. 2.0 Sanskrit morphology

Sanskrit has two-fold morphology- nominal and verbal. In Sanskrit, a syntactic unit is
called pada. A Sanskrit sentence, according to Cordona39 (1988), can be represented in
following formula
(N-En)p…(V-Ev)p.
Pada can be nominal (subanta) or verbal (ti
anta). These forms are produced by
inflecting the stems and hence they are part of Sanskrit inflectional morphology. The
derivational morphology in Sanskrit includes formation of derived nominal and verbal
stems. Both of them can be formed by either nominal stems or verb roots.

1. 2.1 Inflection
Sanskrit is very rich in inflections. The inflection morphology involves formation of two
kinds of words or padas: subanta padas (nominal words) and ti
anta padas (verb forms).
The Pā
inian analysis for Sanskrit has categorized each and every usable word under
these two categories. A word cannot be used in the language unless it is one of them.
Even the indeclinable words (avyayas) are first assigned to the first category, and later
the characteristics of it are deleted. The two categories of padas are named so according

28
to the set of affixes which is affixed to them. These are called sup suffixes and ti
suffixes respectively, thus creating sup+anta = subanta and ti
+anta = ti
anta padas.

Nominal inflection
Nominal inflection in Sanskrit includes all non-verb categories, i.e. subanta-padas.
Sanskrit subanta forms can be potentially very complex. They can include primary
(kdanta) and secondary (taddhitānta), feminine forms (strīpratyayānta) and compound
nouns (samāsa). They can also include upasargas and avyayas etc. According to Pā
ini,
there are 21 morphological suffixes called sup suffixes to be attached to the nominal
bases (prātipadika).

Pā
ini has listed 21 sup suffixes:40

su, au, jas, am, au
, śas,
ā, bhyām, bhis, e, bhyām, bhyas, asi, bhyām, bhyas, as, os,
ām, i, os, sup.
These suffixes are grouped in seven sets of three affix each- (su, au, jas) (am, au
, śas)
(
ā, bhyām, bhis) (e, bhyām, bhyas) (asi, bhyām, bhyas) (as, os, ām) (i, os, sup).41
The 7 sets basically indicate seven cases and the vibhakti-s used for their denotation. The
three in each set are meant to indicate three numbers singular, dual and plural 42
respectively. These suffixes are added to the prātipadikas43 (any meaningful form of a
word, which is neither a root nor a suffix) to obtain inflected forms (subanta padas). The
further modifications of the sup affixes appended to the prātipadika depends mainly on
the gender and end character of the base.
prātipadikas are of two types: primitive and derived. The primitive bases are stored in
gaapāha (collection of bases with similar forms) while the latter are formed by adding
the derivational suffixes. They denote unity, duality and plurality respectively. Some
words are only in the singular always, like eka(one), some are always dual like dvi (two),
aki (eyes) etc. and some are always plural like apa (water), dārā (wife) etc.

39
George Cardona, 1988 Pā
ini, His Work and its Traditions, vol. i (Delhi: MLBD, 1988)
40
svaujasamauchasābhyāmbhisebhyāmbhyasasibhyāmbhyasasosāyossup
41
supa
42
dvyekayordvivacanaikavacane
43
arthavadadhāturapratyaya prātipadikam 1/2/45; kttaddhitasamāsāśca 1/2/46

29
Verbal inflection
The ti
terminations are 18 in number. They are added to the verb root which can be
primitive or derived. The ti
terminations are divided in 2 equal sets in two padas:
parasmai and ātmane. The terminations in each set are again viewed as 3x3, i.e
combination of three persons (first, second and third) with three numbers.

1.2.2 Derivation

Sanskrit is rich in derivation. Both subanta padas and ti
anta padas, before they are
inflected, generally undergo certain derivational operations. Derivation includes suffixes
which are added to both nominal stems and verbal roots to form new words, again
nominal or verbal. In other words, newly derived nominal words or verb roots can be
formed by nominal stems as well as from verbal stems. The suffixes are well categorized
with their meanings and environments.

Nominal Derivatives:
Nominal derivatives are the words formed by adding derivational suffixes to nominal
words. These new words can be nominal stems as well as verbal stems. The nominal
stems which are derived from the nominal stems can be mainly categorized under
taddhitāntas and strīpratyayāntas. The nouns derived from verb roots are called kdantas.
These are derived by affixing kt suffixes to the verb roots.

The nominal forms formed from other nominal stems are called taddhitāntas- derived by
adding taddhita suffixes to nominal stems. The secondary derivative affixes are called
taddhita, which derive secondary nouns from prātipadikas. For example - dāśarathī,
gaua etc. Pā
ini has listed many taddhita suffixes some of which are- a, aka–c, ac, a–,
a, at, i han, īyasun, kan, hak, ha–, tamap, tarap, tayap, tal, tyap, tral, dvayasac, fak,
matup, mātrac, yak, yat, ya–, āc, kha, gha, cha, uraca, hak, ha–, han, na, ha, va, vatup
etc. For example, dākī, kva, aśvaka, viśvajanīnam, katriya, mālīya, raivatika,
dā ika, laghutama, gurutara, gārgyāyaa, iha, balavān etc.

30
Sanskrit also has eight feminine suffixes āp, cāp āp,
ī,
īn,
īp, u
and ti etc. and the
words ending in these suffixes are called strīpratyayānta. For example - ajā, gaurī,
mūikā, indrāī, gopī, a ādhyāyī, kurucarī, yuvatī, karabhorū etc.

The nominal morphology of Sanskrit distinguishes nouns and adjectives on the one hand
and pronouns on the other. There is no absolute distinction between nouns and adjectives
with respect to inflection, but modifiers take the number and gender of the terms they
qualify.

Pronouns are personal (first and second person), demonstrative (deictic), interrogative
and relative. Pronouns other than personal pronouns observe gender distinctions. There
are also endings particular to pronouns.

Sanskrit also has indeclinable terms, including particles such as the connective ca ‘and’
(sentence and nominal connective) and the negative particle na, as well as terms such as
yad, yadi ‘if’, tad, tarhi ‘then’, karhi ‘when?’, tata thence, yata whence, kuta
whence?, yadā when, tadā then, kadaa when?, yatra where, tatra there, kutra where?,
derived from pronominals.

There are also preverbs, which regularly occur immediately preceding a verb or another
preverb – although in vedic they can be separated from a verb or follow it- as well as pre-
and post-positional terms like adhi, anu, which co-occur with particular case-forms.

Indeclinable NPs
44
avyaya subanta-padas remain unchanged under all morphological conditions .
45
According to Pā
ini [2.2.82] , affixes cāp, āp, āp, (feminine suffixes) and sup are
deleted by luk when they occur after an avyaya. Pā
ini gives definitions for identifying
certain word forms as avyayas which also include compound forms.

44
sad tri u lige u sarvāsu ca vibhakti u |
vacane u ca sarve u yanna vyeti tadavyayam || - gopatha brāhma
a
45
avyayādāpsupa 2/4/82

31
Verbal Derivatives:

Verbal derivatives may be nominal stems or verbal stems derived from verb roots or
stems. The nouns that are derived immediately from verbs comprehend a great variety of
terms. There are two principal classes of their classification:
• Adjectives (Attributives)
• Names (Substantives)
These different nouns are formed by affixing certain terminations to the verb root, which
is modified in a greater or lesser degree, and then forms the inflective base. These
terminations are numerous. The greater number has a very limited application while
others comprehend a large class of words. The primary derivatives are called kdanta.
The primary affixes are to be added to verbs to derive substantives, adjectives or
indeclinable kt. For example pa hitavyam, pātavya, pa hanīya, pacelima, jeyam, deyam,
karttā, kumbhakāra, janamejaya, pā haka, pa hantī, gantum, khāditum, svapnam
gati, gatvā,vihāya, ādāya etc.

Verbal derivatives also include participles, infinitives, gerund etc.

1.2.3 Compounding

Simple words (padas), whether substantives, adjectives, verbs or indeclinables, when

added with another subanta pada form samāsa (compound). Sanskrit samāsas are
divided into four categories, some of which are divided into sub-categories. The four
main categories of compounds are as follows:
(1) Adverbial or avyayībhāva,
(2) Determinative or tatpurua,
(3) Attributive or bahuvrīhi and
(4) Copulative or dvandva.
dvandva and tatpurua compounds may be divided into sub-categories also. tatpuru a has
dvigu and karmadhāraya as its sub-categories.

32
1.2.4 Sanskrit Verb-Morphology:

The verb forms are derived from verb-roots or dhātus. These dhātus are encoded with the
core meaning of the verb. These can be primitive46 or derived47. Primitive verb-roots,
which are around 2000 in number, have been listed in a lexicon named ‘dhātupā ha’.
They are divided in 10 groups/classes called gaas. All the verb-roots of a group undergo
somewhat similar inflectional process. Derived verb-roots may be derived from primitive
verb-roots or from nominal forms. Prefixes also play an important role as they can
change the meaning of a verb root. These roots then have to undergo various inflectional
suffixes that represent different paradigms. In this process, the base or root also gets
changed. The chart given on the next page gives an overview of Sanskrit verb roots.

Verb-roots

Primitive Derived
(2000 listed)
Derived from verbs Denominatives
Divided in
10 gaas Causal

Desiderative

Frequentative

Derived Verb-roots:
Derived from verb-roots:
1. Causals (ijanta)- Causals are formed by adding ‘ic’ suffix to a primitive verb
root. They convey the sense that a person or thing causes another person or thing
to perform the action or to undergo the state denoted by root.

46
bhūvādayo dhātava (Pāņini 1/3/1)

33
 2. Desideratives (sannanta)- Desiderative of a primitive verb root is formed by
adding ‘san’ affix to it. It conveys the sense that a person or thing wishes to
perform the action or is about to undergo the state indicated by the desiderative
form. Any basic verb-root or its causal base may have a desiderative form.

3. Frequentatives (yaanta)- Frequentatives verbs import repetition or intensity of

the action or state expressed by the root from which it is derived. Can be of two
types:

a. Ātmanepada Frequentative (ya
anta) – ‘ya
’ affix added

b. Parasmaipada Frequentative (ya
luganta) – ‘ya
’ is added but deleted

An illustration is given below of formation of derived verb-roots from a primitive verb

root bhū.
 --- (+
ic) ---- bhāvay (to cause someone or something to be)
bhū ( to be )----------- --- (+ san) ---- bubhū (to desire to be)
 --- (+ ya) ---- bobhūy (to be repeatedly)
(ya deleted) ---- bobho/bobhav
These derived verb-roots, however, undergo similar operations, with some specifications,
to form verb forms.

Roots derived from nominal words:

A large number of Sanskrit verb-forms can be derived from nominal words. These are
known as ‘nāmadhātus’ (denominatives). Taking a nominal word as head, various
derivational suffixes are added to these to form nominal verb-roots. The sense conveyed
by the nominal verb root depends upon the suffix added to it. Yet, denominatives
commonly import that a person or thing behaves or looks upon or wishes for or resembles
a person or thing denoted by the noun. These denominatives, however, can be
innumerable as there is no end to nominal words in Sanskrit.

47
sanādyantā dhātava (Pāņini 3/1/32)

34
Process of formation of Sanskrit verb forms:
A Sanskrit verb root may take various forms. There are ten lakāras that represent Tense,
Aspect and Mood. Inflectional terminations are 18 in number. These are divided in two
groups – Parasmaipada and Ātmanepada, each having 9 affixes which is a combination of
3 persons x 3 numbers. A verb is conjugated in either pada, though some of the roots are
conjugated in both. For each different lakāra, a root is affixed with these 9 terminations.
Again, there are three voices- Active, Passive and Impersonal. Transitive verbs are used
in the Active and Passive voices while intransitive verbs are conjugated in the Active and
Impersonal voices. Addition of one or more of 22 prefixes (upasargas) to verb roots can
result in more variety of forms. Derivative verb roots, both derived from verb roots as
well as nominal words, also follow same process to form verb forms. There can be some
specific rules and exceptions in some cases. A chart given here gives a rough estimate of
all the possible verb-forms of Sanskrit. 

VR [2000]
├ Causal ((ijanta)
├ Desiderative (sannanta)
└ Frequentative (ya(anta/ya(luganta)
+ One normal form
Denominatives
| ↓
TAM [10 lakāras]
↓
┌────────────┐
Active Voice Passive/Impersonal Voice
↓ ↓
10x9(3x3) forms 10x9(3x3) forms
↓ ↓
22(+) upasarga 22(+) upasarga

48
Mishra Sudhir K., Jha, Girish N., 2004, Identifying Verb Inflections in Sanskrit morphology, In proc.
of SIMPLE 04, IIT Kharagpur, pp. 79-81.

35
The verb roots of gaa 1, 4, 6 and 10 adopt certain terminations when ti
affixes are
added to them. Consequently, the verb roots of these class forms base ending in ‘a’. The
ti
affixation also influences the verb root and it undergoes several morpho-phonemic
changes such as its end vowel is gunated. The verb root can adopt certain more
operations resulting in the final verb-form.

1.2.5 Survey of Morphological tools for Sanskrit

C-DAC

Deśika 49 (Natural Understanding System), a Software Package, developed by Indian

Heritage Group, C-DAC, Bangalore led by P. Ramanujan, claims to generate and analyze
plain and accented written Sanskrit texts using grammar rules of Pā
ini’s A ādhyāyī,
with an exhaustive database of Amarakośa and heuristics of semantics and contextual
processing from Nyāya and Mī āmsā Śāstra-s. It also claims to analyze Vedic texts. The
analysis module of the Software is called as a general purpose Sanskrit parser which
claims to dissolve and identify the compound and combined word forms, though the
present version of the Software downloadable from the TDIL (Technology Development
for Indian Languages) website has Subanta Generation Module only. This module has
two modes: choose mode declines for the prātipadika-s already present in the list and
edit mode declines for the entries which are not present in the list for which suitable
gender and paradigm should be selected. Deśika is also supposed to include Vedic
processing and śābdabodha50.

Sanskrit Authoring System (VYASA)51 claims to be a robust multilingual document

editor with transliteration facility among Indian Languages and Roman, sorting and

49
Deśika (Natural Language Understanding System), http://tdil.mit.gov.in/download/Desika.htm
50
C-DAC R&D Activies: Development of Desika A Natural language Understanding (NLU) system for
Sanskrit. [1990 – 1994], http://www.cdac.in/html/ihg/activity.asp, accessed on July 16, 2006
51
C-DAC, R&D Activities: Developing Sanskrit Authoring System (VYASA),
http://www.cdac.in/html/ihg/activity.asp and The House Magazine of C-DAC, Pen to Paper Developing
Sanskrit Authoring System – VYASA, http://www.cdac.in/html/connect/3q2000/art10a.htm

36
searching facilities, indexing, and concordance. It also claims to provide tools for
analyses at morphological, syntactic and semantic levels. Tools for
searching/indexing/sorting, lexical updation, lexical tagging, extraction/indexing of
quotations in commentaries/explanations, transliteration facility, word split programs for
sandhi and samāsa, poetry analysis (textual/metric/statistical), statistical tools like
concordance, thesauri, and electronic dictionaries are also included. However, the system
is not available anywhere to evaluate or check.

Academy of Sanskrit Research, Melkote

Academy of Sanskrit Research (ASR), Melkote under the directorship of Prof. M. A.

Lakshmitatachar has been working on NLP in Sanskrit and other Indian languages for
more than 10 years. The software ‘śābdabodha’ developed at ASR, Melkote is available
on TDIL website. It is claimed to be an interactive application built to analyze the
semantic and syntactic structure of Sanskrit sentences. The software works on DOS 6.0
or higher with GIST (Graphic based Intelligence Script Technology) shell on Windows
95 platform. The software is said to include a user interface. For an input Sanskrit
sentence or a file the program will check for the syntactic compatibility and otherwise
gives all the morphological details with POS information having nouns, pronouns,
adjectives, participles, indeclinables, indeclinable participle and verbs tags of the
components of the sentence. It claims to process all types of sentences of Sanskrit, and
can handle generation and analysis of subanta of more than 26,000 stems, ti
anta
conjugations of roots, in two voices, ten lakāras and three modes viz. kevala ti
anta,
ijanta and sananta. It also handles the generation and analysis and identification of case
inflected forms of 11 types of kdanta of 150 roots. Apart from this it is said to have a
database of 690 avyayas, 26,000 nominal stems, 600 verbal roots, kdanata forms of 600
verbal roots, 5 taddhita suffixes. For handling the semantic analysis, a matrix of 52 sets
of nouns with their synonyms amounting to 300 nouns, 27 actions denoted by nearly 200
verbs are said to have been prepared. 52 This institution is also working with 20 odd
software tools like Bodha (Sentence disambiguation system according to śābdabodha of

52
Language Processing Tools: TDIL website, http://tdil.mit.gov.in/nlptools/ach-nlptools.htm

37
navīna nyāya system), Śemuī (subanta generator/analyser), Prajñā (ti
anta
generator/analyser), Cetanā (kdanta generator/analyser), Pāini (sandhi joiner
according to Pā
inian rules) etc. which are to be released.53

Rashtriya Sanskrit Vidyapeetha (RSV), Tirupati

RSV Tirupati has put much effort in developing linguistic resources for NLP in Sanskrit.
Prof. K.V. Ramakrishnamacharyulu, (presently V.C. of Rashtriya Sanskrit Sansthan,
Jaipur) and Dr. Srinivasa Varkhedi along with Prof. Vineet Chaitanya and Amba P.
Kulkarni have initiated many projects and have developed many tools like pada-ccheda,
which isolates Sanskrit compound words into its components, which works on Sanskrit
ISCII text in Linux environment. It is also working on developing Sanskrit morphological
analyzer. An initial analyzer developed in collaboration with IIIT-H is already online.
Apart from this it is also concentrating on kdanta and ti
anta analyzers and also
generators for subanta, ti
anta and samāsa. RSV Tirupati along with C-DAC Bangalore,
Ahobila Mutt Sanskrit College Madhurantakam Tamil Nadu, Poorna Prajna
Samshodhana Mandiram Bangalore, Chinmaya International Foundation Veliyanad
Kerala, ASR Melkote Karnataka, IIIT-H, Dept. of Sanskrit H.S.Gour University Sagar
Madhya Pradesh have combined initiative to develop a large Sanskrit Corpus (presently
this is not online).54

RCILTS – Utkal University

Resource Centre for Indian Language Technology Solutions (RCILTS) – Oriya Centre at
the Department of Computer Science and Application, Utkal University, led by Prof.
Sangamitra Mohanty and funded by Ministry of Information Technology (MIT), has been
working on Oriya-English-Hindi Trilingual Dictionary, Oriya and Sanskrit WordNet,
Trilingual Word Processor, Grammar and Spell checkers and MorphAs for Oriya and

53
Academy of Sanskrit Research, Melkote, http://www.sanskritacademy.org/About.htm
54
RSV Tirupati, http://rsvidyapeetha.ac.in and http://www.sansknet.org

38
also English to Oriya Machine Translation System. It has received IPR for four of their
products from Ministry of Human Resource and Development (MHRD), Govt. of India.

San-Net (Sanskrit WordNet)

RCILTS-Utkal 55 claims to have developed a proto-type of Sanskrit WordNet using

Navya-Nyāya and Pā
inian Grammar. It has 300 Sanskrit words (250 Nominal words and
50 Verbal words) having synonymy, antonym, hyponymy, hypernymy, holonymy and
meronymy relations with their analogy, etymology, and definitions. It also claims that a
standard Knowledge Base (KB) that has been developed for analyzing
syntactic, semantic and pragmatic aspects of any lexicon.

The Sanskrit Heritage Site

Dr. Gerard Huet, Director, INRIA has developed various computational tools for
Sanskrit which are available online56. The Declension Engine declines all the nominal
inflectional forms with the ‘compound base’ for a query word given with its gender
information. It also gives the ‘compound base’ of the query word. The Conjugation
Engine gives all the possible forms of the verb root in its ātmane and/or parasmai
terminations, in kartari and karmai/bhāve voices with its Desiderative forms in eight
lakāra-s. It also gives few participle and indeclinable kt forms57. He claims that from
535 roots, his engine generates 14554 noun forms, 113935 root verbal forms, 203281 root
participial forms, 14311 iic and periphrastic forms, and 737 indeclinable forms, totaling
roughly half a million forms58. Lemmatizer and Sanskrit Readers are the analyzers.
While the lemmatizer tries to tag a given simple inflected noun or a verb (without
upasarga-s), the Sanskrit Reader Companion analyses a given phrase or a simple
sentence, segments it into individual words, tags each word and parses the input. Modular
transducers are applied to constraint the lexical analyzer to recognize the stream of forms
as a regular expression over 14 phases, specifying Sanskrit's morphology geometry to get

55
RC-ILTS – Computer Science & Application, Utkal University, http://www.ilts-utkal.org/sanskrit.htm
56
The Sanskrit Heritage Site, Huet, Gerard. http://sanskrit.inria.fr/
57
Ibid.

39
right composition of compound chunks. Over generation of segmentation and ambiguity
of tags are checked by semantic role analysis similar to Pā
ini’s kāraka theory and also
by governance patterns of verbs.59

The Sanskrit Library

The Sanskrit Library Project, directed by Dr. Peter M. Scharf, Classics Dept., Brown
University, has developed a web-based reading room holding Kramapā ha, a Sanskrit
independent-study reader on texts like Pa–catantra, Rāmopākhyāna and Pā
ini’s
A ādhyāyi. The reader gives the text in Devanagari with Roman transliteration, sandhi
analysis, a detailed grammatical analysis and lexical analysis of the text, with notes and
translation. As part of independent-study texts, a detailed classification of grammatical
categories (tag set) is made.
A nominal inflection generator tool is also available which generates nominal inflections
for a given word with gender and class information.60

IIIT Hyderabad

Language Technology and Research Centre (LTRC), IIIT-H, is one of the leading
NLP centres in India. IIT Kanpur and later LTRC, IIIT-H, have been the pioneers in the
field of language technology and to initiate Pā
inian approach to NLP in India. LTRC
has several ongoing activities with Govt. of India, Carnegie Mellon University's
Language Technology Institute, UPENN, HP Labs, Google, Nokia and TCS besides
several academic institutions in India. Prof. Rajeev Sangal and Prof. Vineet Chaitanya
with the Akshar Bharati61 group have developed many NLP tools for Indian languages
like MorphA-s (morphological analyzers), Anusaraka-s62 (language access tool), Shakti

58
Huet, Gerard, 2006, Parsing Sanskrit by Computer, (abstract) in the Proceeding of the 13th World
Sanskrit Conference, Edinburgh, UK
59
Ibid.
60
The Sanskrit Library Reading Room - http://cgi-user.brown.edu/cgi-user/sanskrit/login
61
Akshar Bharati, is a personification of groups (@ IITK, IIIT-H, University of Hyderabad, etc.) working
on NLP with special emphasis to Indian Languages giving due attention to Indian theories of grammar and
language.
62
Anusaaraka or a Language Accessor is a computer software which renders text from one Indian language
into another (Kannada-Hindi, Marathi-Hindi, Punjabi-Hindi and Telugu-Hindi). It produces output not in

40
Machine translation System. MorphA-s for Hindi, Telugu, Marathi, Kannada and Punjabi
were developed as part of Anusaraka-s and now presently they are made available online
and also can be downloaded from their website. Telugu MorphA is said to have coverage
of 95% for any arbitrary modern standard Telugu text and Hindi MorphA has 88%
coverage. 63 MorphA-s for these languages were developed based upon the paradigm
model.64 For a given word the MorphA gives the root word with its feature information
like gender, number, person, tense etc.65

Sanskrit MorphA was developed by Vinish Jain, an M.Tech student of IIIT-H under the
guidance of Amba P. Kularni, which was a further development of an earlier working
Sanskrit MorphA developed at ASR Melkote. This MorphA was developed using
paradigm approach in the model of other MorphA-s at IIIT-H which used a lexicon
derived from Monier William’s dictionary. This could only handle subanta-s. Later with
the collaboration of RSVP Tirupati, the present Sanskrit MorphA66 was developed by
Amba P. Kulkarni and V.Sheeba. This MorphA has separate modules to handle subanta,
ti
anta, kdanta and samāsa. Each word is filtered through all the four modules and all
possible answers are produced. subanta module contains 222 paradigms for nouns and
pronouns with a root dictionary of around 1.5 M words extracted from Monier Williams
dictionary. ti
anta module has a database of ~10 M verb forms (no prefixed verbs).
kdanta module can handle ~42 K of 20 types of kdanta. This approach has the
advantage of developing further without disturbing the main program, as the program is
independent of data and also more modules can be added.67 This MorphA can handle
only sandhi-free text and cannot filter the multiple tags and disambiguate it for a single
word in a given context.

the target language but close to it, which is in comprehensible to the reader who is trained to read the
output.
63
Morphological Analysers – IIIT Hderabad – http://www.iiit.net/ltrc/morph/morph_analyser.html
64
Bharati, Akshar et al, 1999, Natural Language Processing: A Paninian Perspective, Prentice Hall Pvt.
Ltd. New Delhi, pp.39-43.
65
Ibid..
66
RSVP, http://rsvidyapeetha.ac.in/~anusaraka (accessed: 13.10.2006)

41
Jawaharlal Nehru University (JNU)

The RCILTS – Sanskrit, Japanese, Chinese unit of JNU, under the leadership of Prof.
G.V.Singh claims to have developed web based Sanskrit Language Learning System for
the use of scholars for designing Knowledge based systems based on the Indian traditions.
The unit has developed a computational module of A ādhyāyī of Pā
ini, Sanskrit-
English lexicon, English-Sanskrit lexicon and a lexicon of Nyāya terms. It also says that
it has made some efforts on the sandhi analysis system.68

Girish Nath Jha69 developed a Nominal Inflection Generator for Sanskrit for his M. Phil.
dissertation. The program, written in prolog, generates all the inflections of subanta given
a Sanskrit word with gender and ending letter information.

Special Centre for Sanskrit Studies, JNU:

Special Centre for Sanskrit Studies, JNU, established in 2002, has undertaken the task of
developing NLP tools for Sanskrit. Under the guidance of Dr. Girish Nath Jha, the
students have developed certain useful programs for computational processing of Sanskrit
texts. A list of works is given below:

The NLP team at the centre has developed an Online Multilingual Amarakosha (OMA)
under a project funded by UGC under UPOE program. It is Unicode based software
which supports seven languages- Sanskrit, Hindi, Kannada, Punjabi, Bangla, Oriya and
English and allows the user to search the synonym from one language to another. The
output displays the grammatical and semantic category of the word, its base word,
reference and ontological information. The software also provides the facility to enter and

67
Bharati Akshar, Amba Kulkarni, V Sheeba, “Building a Wide Coverage Sanskrit Morphological
Analyzer: A Practical Approach”, in the Proceedings of First National Symposium on Modelling and
Shallow Parsing of Indian Languages, 2-4 April 2006, IIT Bombay, Mumbai
68
RCILTS, JNU – Achievements: http://tdil.mit.gov.in/SanskritJapaneseChinese-JNUJuly03.pdf (accessed
15.10.2006)
69
Jha, Girish Nath, 1993, ‘Morphology of Sanskrit Case Affixes: A Computational Analysis’, M.Phil.,
submitted to JNU, New Delhi.

42
edit the data by language experts. The software is expected to be extended as a
multilingual interface, search engine and text processing tool in near future.70

R. Chandrashekhara71, developed tag-sets for POS tagging of Sanskrit text as part of his
Ph.D. thesis. This tagger (POST) which is an online system run on Apache Tomcat
platform using Java Servlet takes sandhi-free classical Sanskrit prose text as input and
provides the tagged text as output. The system is very important for the further R&D on
the Sanskrit-Indian Languages Machine Translation Systems (MTS).

Sudhir Kumar Mishra72, has recently completed his Ph.D. research on Kāraka Analyzer
for Laukika Sanskrit prose text based on Pā
ini’s Kāraka formulations. Kāraka rules are
of central importance in Sanskrit syntactic structure. As part of this research work he also
worked on identification of verb inflections in Sanskrit morphology73.

Subash Chandra74 developed a Sanskrit Subanta Recognizer and Analyser System (SRAS)
for his M.Phil. dissertation. SRAS is also an online system run on Apache Tomcat
platform using Java Servlet and MSSQL server 2005 as back end. This system has been
developed according to Pā
inian formulation which accepts only non-joint (sandhi-rahita)
Sanskrit text in Devanāgarī script and fully depends on both the rule base, example base
and a database of other linguistic resources. The system has been tested with some
selected corpora and claims to give an average accuracy of 91.65%.

Research work is also being done on learning Sanskrit language using e-learning
approach 75 , and sandhi analyzer applying Pā
inian and some heuristic rules 76 , online

70
System is accessible @ http://sanskrit.jnu.ac.in
71
Chandrashekhar, R, 2006, ‘POS Tagging for Sanskrit’, submitted for Ph.D degree at SCSS, JNU.
72
Mishra, Sudhir Kumar, 2007, ‘Sanskrit Karaka Analyzer for Machine Translation’, submitted for Ph.D.
degree at SCSS, JNU
73
Mishra, Sudhir Kumar & Girish Nath Jha, 2005, ‘Identifying Verb Inflections in Sanskrit Morphology’,
in the proceedings of SIMPLE-05, IIT-Kharagpur, pp.79-81
74
Chandra, Subash, 2006, ‘Machine Recognition and Morphological Analysis of Subanta-padas’, submitted
for M.Phil degree at SCSS, JNU
75
Bhowmik, Preeti & Jha, Girish Nath, 2006, ‘Sanskrit Language Pedagogy: an e-learning approach’, In
the Souvenir Abstracts of 28th AICL, BHU, Varanasi, p.150.
76
Kumar, Sachin & Jha, Girish Nath, 2006, ‘Issues in sandhi processing of Sanskrit’, In the Souvenir
Abstracts of 28th AICL, BHU, Varanasi, p.129.

43
indexing of Ādiparva of Mahābhārata77, computational analysis of Sanskrit gender and
analysis of derived nouns in Sanskrit.78

Sanskrit to English Translator:

Aparna Subramanian of School of Computer Science, Devi Ahilya Vishwavidyalaya,

Indore worked on a project titled “Sanskrit to English Translator”79 for her M.Sc. degree.
It performs sandhi viccheda of Sanskrit words and then translates them to English. It has
two components: morphological parser and translation generator. The morphological
parser module consists of a set of transducers that transform the provided input text to a
set of acceptable output. It also gives the parse of these words. The second part of the
parser is a Viccheda module, which applies reverse sandhi rules to split the combined
Sanskrit words into separate basic words. The translator is also a two part module, which
first structures the English sentence according to the grammar using the parse information.
The second generates equivalent English words according to the morphological details.
The two combined together give the needed translated sentence.

77
Mani, Diwakar, & Jha, Girish Nath, 2006, ‘Online indexing of Ādiparva of Mahābhārata’, In the
Souvenir Abstracts of 28th AICL, BHU, Varanasi, p. 125.
78
Singh, Surjit Kumar & Jha, Girish Nath, 2006, ‘Strategies for Identifying and Processing Derived Nouns
in Sanskrit’, In the Souvenir Abstracts of 28th AICL, BHU, Varanasi, p. 131.
79
Subramanian, Aparna, Jan 2005, “Sanskrit to English Translator”, Language in India, Vol. 5:1.

44