Chapter 5

Karaka Theory of Interpretation

येना>र समा(नाय म?धग(य महे वरात ् ।

कृ !नं 5याकरणं ो तं त!मै पAणनये नमः।।

The above verse means, ‘I bow my salutations to Mahamuni Panini, who with the
blessings of God Maheshwara, developed the Vyakarana (Grammatical Framwwork)’
[Jha, 2014].

Grammar is the important tool to analyze any language. The Dependency Grammar,
studied in the earlier chapter, is able to give the relation among the words of the sentence
in terms of the dependency label, which is mostly syntactic in nature. In case of Text
Summarization, syntax of a language is important alongside the semantics are also
important role to decide the saliency of a certain topic in the text. In the earlier chapters,
we have discussed the statistical features of Purva and Uttar Mimansa along with the
syntactic features like Vakya for selection of important sentences. This sentence selection
strategy if further gets combined with the Syntacto-semantic frame work; it may enrich
the task of Text Summarization. In this chapter, a detailed discussion on the Syntacto-
semantic framework of Sanskrit popularly known as Karakas is presented which will be
used later as sentence selection strategy for Text Summarization.

99
5.1 Sanskrit Karaka Theory

Knowledge representation is one of the crucial tasks in Artificial Intelligence (AI).

In this view the studies of knowledge representation using Karaka theory is getting
evolved as one of the most useful tools in association with the semantic nets. A rigorous
thought process is given to the fact that being a Natural Language (NL) can Sanskrit be
used for the tasks of AI, like Machine Translation (MT) [Briggs, 1985; Kadvany, 2007;
Kelly, 2015]. The well known scientist Briggs opines that the early machine translation
methods were not successful because they lack in the semantic approach hence the
computational models for MT should be constructed by considering the cognition behind
the actual thinking process [Briggs, 1985].

In order to achieve the human like cognition, the fluent methods that are capable of
extracting the semantics from the syntax should be researched. The classical Paninian
Grammar (PG) is one of such tools that facilitate the task of obtaining the semantics
through syntactical framework. In case of Text Summarization, the human cognition
focuses upon the identification and interpretation of important sentences, which broadly
is known as Shabdabodha. The Shabdabodha theory tries to acquire the verbal
knowledge through the application of Karaka theory [Tatacharya Ramanuja, 2006]. The
subsequent sections elaborate how the Karakas participate in the action of this
Shabdabodha.

Sanskrit is one of the oldest and the richest language of India with its origin in the
Indo-European languages family through the Indo-Aryan branch. Panini’s Astadhyayi
(P), which is a complete description of Sanskrit Grammar, has fixed the form of Sanskrit
language decisively. It consists of near about four thousand aphorisms (Sutras) of the
greatest compactness i.e. economy of language. Panini's condensed Sutra style is its
characteristic which is not found outside the Paninian grammatical schools [Bhate and
Kak, 1993; Shastri, 2008].

In Paninian grammar, an extensive and perfect interpretation of Phonology,

Morphology, Syntax and Semantics is available. It ideally explores all the concepts of

100
theories of scientific analysis and description of a language. With these strong basics, the
link between the modern Western Schools of Linguistics and the ancient Indian
grammarian school is considerably very close [Kiparsky and Staal, 1969; Bhate and Kak,
1993]. Paninian grammar (PG) focuses on information while analyzing the language
structures. It observes the language from the speaker and hearer’s or reader’s and writer’s
point of view. During communication between them, the information is coded in the
language strings. While analyzing these language strings one is expected to decode this
information. The computational framework of Panini can be considered as a tool, which
concerned about the coding and extraction of these languages constructs [Bharati et. al.,
1996]. It works on various levels of language analysis to achieve the meaning of the
sentence from hearer’s perspective. To achieve the desired meaning, the grammar
analysis may divide itself internally into the various levels as per the need.

From learning cognition view, the hearer or the reader understands the entire
sentence first and then they concentrate on the meaning of the individual word. Hence,
the analysis works on two stages i.e. syntax and morphology. Paninian grammatical
framework considers that the words and sentences are isolated. When they come in a
group through a formal process then only they contribute to the overall meaning of the
sentences. Thus one can say that, sentence carries the real syntactic meaning and not the
individual words used in it. This is the essence, which makes it more meaningful to use
Paninian Grammar for Text Summarization [Sarkar, 2003].

5.1.1 Linguistic Analysis in Paninian Grammar

As Panini has accepted the existence of words only within the sentences, it is clear
that the Paninian grammar is sentence based and not merely word based. Hence, it
automatically proves the importance of the sentences.

During the sentence analysis, the Paninian grammar theory does not propose any
separate syntactic level. Instead, it assumes various levels of sentence analysis for the
actual written/spoken sentence for its meaning interpretation. Each level in this analysis
is more refined than the previous one. Broadly, two levels of analysis can be considered,
as surface level and the semantic level [Bharati et. al., 1996]. The first one is the actual

101
string or the sentence and second one represents the speaker’s actual intention i.e. his real
thought for the sentence. The early level known as Vyakaran levels makes minimal use of
world knowledge and more use of morphological knowledge. The later levels make
greater use of world knowledge or semantics.

Moreover the Paninian framework also considers two more crucial levels known as
Karaka and Vibhakti level (figure 5.1) in between the surface level and the semantic level
[Bharati et. al., 1996] etc. Never the less, it is notable that even the surface level also
considers the notion of Vivaksha i.e. speaker’s view point to a certain extent, which has a
relation to pragmatics. As shown in the figure (5.1), the surface level captures the written
or the spoken sentences as it is. In between this surface level and the upper level, i.e. the
Karaka level there exist the Vibhakti level, which is purely syntactic. At this level, the
case endings are used to form the local groups of the words.

Figure 5.1 Paninian levels of language processing [Bharati et. al., 1996]

At Vibhakti level, a noun group is formed containing a noun, which contains the
instances of noun or pronoun etc. These instances are included with their respective
Vibhakti. According to Sanskrit grammar, Vibhakti is the word suffix, which helps to
find out the participants, gender as well as the form of the word. Vibhakti for verbs
includes the verb form and the auxiliary verbs. Vibhakti gives Tense Aspect modality
details of the word which is popularly known as TAM. These labels are syntactic in
nature and are determined from the verbs or the auxiliary verbs. Hence, the Vibhakti

102
plays an important role in Sanskrit. We intend to use this grammatical framework for
English but English is not inflectionally rich language; therefore, Vibhakti has a very little
role in English.

After Vibhakti, at the next Karaka level, the relation of the participant noun, in the
action, to the verb is determined. On contrary to Vibhakti, the Karaka relations are
Syntacto-semantic [Cardona, 1976; Bharati et. al., 1996]. These relations are established
in between the verb and other constituent nouns that are present in the sentences.
Through these relations, the Karakas try to capture the information from the semantics of
the texts. Thus, Karaka level processes the semantics of the language but represents it at
the syntactic level. Hence, it acts as a bridge between semantic and syntactic analysis of a
language. The next section elaborates the concept of Karaka in terms of its roles in
achieving the action described by the verbs.

5.2 The Concept of ‘Kriya’ and ‘Karaka’

In Paninian Grammar, the sentence analysis is carried out by considering centrality

of the verb. Here the verb is considered as the binding element of the sentences. In
Karaka theory a sentence is defined with reference to Kriya i.e. action and Karaka i.e.
participatory factor) and Anvaya i.e. their relations [Bhatta, VSLM 1963].

Briggs view about the Karaka was not complete, in his presentation it was
elaborated that in a sentence, semantic nodes are directly connected among themselves
with the connectors (i.e. especially by using the Karakas) [Briggs, 1985]. This creates an
impression that Karakas are directly involved in the main action. On the contrary,
Vacaspati Mishra (C. 9th A. D.) in his Nyayavartika-Tatparya-Tika comments that [Dash,
1992]:

धानB+या साधने हे तु अवांतरB+या वशेषयु ते कारकशEद: वतते ।

This means that, to achieve the primary action indicated by the verb the Karakas
mostly participate through their auxiliary actions with certain causal factors [Dash, 1992].
Consider the following sentence:

103
 ‘Shambhavi Lekhanya patram likhati’
Shambhavi writes the letter with Pen.
Shambhavi (Nom) lekhanya (pen) (Inst) patram (letter) (Acc) likhati (writes) (verb.)

The Vakyarthabodha i.e. meaning of a sentence can be obtained by analyzing the

exact wordily situations in a sentence. In the above sentence, the pen becomes the
effective instrument in carrying out the task of writing. A general observation shows that
while performing the action with an instrument, the subject i.e. Shambhvai is completing
the task of writing. Thus, the main activity of writing is carried out by the auxiliary
actions like movement of a pen. In this sentence, the Karta Karaka i.e. Shambhavi is
directly involved in the task through the auxiliary actions given by the Karana Karaka
i.e. Pen. Paninian Grammar has used this idea to obtain the Syntacto-semantic
information of the sentences.

In the scheme of Karaka, the analysis considers the verb as the central element and
its requirements for the argument are analyzed. Both, arguments i.e. verb and adjuncts i.e.
the participants are annotated and analyzed considering the verb meaning. These
relationships are termed as Karaka. The Karakas describe how the adjuncts i.e. the
constituents participate in the central action described by verbal. Panini has prescribed
various Karakas in various meanings which cannot be conveyed without using a
sentence. This shows that Karaka is a syntactic and semantic category as well. Therefore,
[Bharati et. al., 1996] claims that the notion of Karakas is the bridging gaps between the
surface and semantic level. They take the help of surface level syntactic as well as
morphological information and include the local semantic information of the verb in the
sentence. Thus, they are able to capture a specific level of semantics.

In Ashtadhyayi, Panini has mentioned Karaka in the sutra P.1.4.23 [Shastri, 2008].
These Karakas are applied to thirty-two sutras (aphorisms) as ‘Adhikara’ (Authority)
[P.1.4.24-55, Shastri, 2008]. One of the eminent grammarians who have a commentary
on ‘Paninian Ashtadhyayi’ is Patanjali. He also agrees with the concept of Karaka put
forth by Panini. Patanjali presented his view on Vakya (i.e. a sentence) while explaining
the view of Paninian concepts. He opines that Kriya i.e. action is the most important in

104
Vakya (i.e. in a complete sentence). Patanjali’s view about Kriya and Karaka is explained
in the subsequent section [Joshi, 1968].

5.2.1 Kriya

According to Patanjali, without Kriya, there cannot be Karaka. Patanjali comments

that Kriya is a matter of notion. In a sentence, an object is clear-cut visible candidate but
like object, the Kriya cannot be made visible. According to Patanjali’s logic Karakas are
many not one. Therefore, if Panini has to indicate many Karakas here by using the term
Karaka, he would have used plural number but he used singular number. In a Vakya, the
Karakas are many but Kriya is one. This shows that Panini wants to point out Kriya by
the term Karaka because in Vakya, Karakas are many but Kriya is one. So, the use of
singular number (Karaka) by Panini can be proved only if Kriya has to be conveyed by
the use of the term Karaka. Panini says Karma is the most desired to be achieved from
the agent’s point therefore Patanjali suggests to remove the superlatives degree and wants
Karma to be defined as ‘desired to be achieved by an agent’ [Joshi,1968].

5.2.2 Definition of Karaka

The above concept of Karaka can be elaborated in detail. As discussed earlier while
dealing with the syntax and semantics of the language, the Sanskrit grammarians have
given much importance to Karaka. It is clear by the use of the terms like:

• B+या<व य वं कारक वम ् । [Joshi, 1968; P. 1.4.22, Shastri, 2008;

Ramanuja Tatacharya, 2006] (State of direct relation with a verb)

• B+या नGपादक वं कारक वम ् । [Ramanuja Tatacharya, 2006] (Sense of

performing an action)

• B+याजनक वं कारक वम ् । [Ramanuja Tatacharya, 2006] (Sense of

producing an action)

105
 The etymological meaning of the word Karaka is ‘one who does something’, i.e.
one who performs an action. The Karaka and the Kriya, i.e. the cases and verb are
bounded with the sense of mutual requirement i.e. Aakanksha [Jha, 1969]. The one who
performs an action, accepts an action, or otherwise helps to perform an action is known
as a Karaka. There is a mutual expectancy in between the action i.e. Kriya and the
adjuncts i.e. Karaka. The presence of one calls for the existence of the other. In other
words Kirya and Karaka are mutually exclusive (Poorak) [Jha, 1969].

In Sanskrit, there are around six types of Karaka relations. With these six types of
Karakas it tries to capture the numerous semantic relations ions and the participants.
However, these six relations are sufficient to broadly classify such type of semantic
relations. The six Karakas accepted by the Sanskrit Grammarians are such as: Karta or
Karttr, Karmma or Karma, Karana, Sampradana, Apadana and Adhara or Adhikarana
[Bhatt, 1963]. These Karakas are derived by analyzing the direct and continuous or
mutual relation between the verb and the nominal. Panini enlists all these Karakas in the
following sutras:

कता कम च करणम ् सं दानं तथैवच ।

अपादाना?धकरण म याहु: कारकाAण षH ।। [Bhatta, 1963]

Some of the grammarians other than Panini consider the Sambandha, i.e. Possessive
or Genitive and Sambodhana, i.e. Vocative as the Karakas. However, Paninian theory
does not consider it as a valid Karaka because they do not have direct and continuous
relation with the verb.

While naming the Karakas we have followed the naming scheme proposed by
Akshar Bharati [Bharati et. al., 1996]. During English translation of the Sanskrit or Hindi,
the words are spelled based on their previous usage in literature. Therefore, ‘Karaka’ is
not spelt as ‘Kaarak’ or ‘Karta’ is not spelt as ‘Kartaa’.

106
5.3 Meaning of various Karakas

The Karta, Karma and Karana are considered as the foremost Karakas while
Sampradana, Apadana and Adhikarana Karakas are known as the influenced Karakas
[Dash, 1992].

Figure 5.2 Karaka representation chart

5.3.1 Karta Karaka

The Sanskrit grammarians propose that The Karta Karaka is the premier one
according to action and it is used to perform an action independently of its own. Karta
may also get the action performed by other nominatives with control and authority. Thus,
an action indicated in a sentence is entirely dependent upon the Karta-Karaka. In the
Mahabhashyam with the following verse, Patanjali states the importance of the Karta
Karaka as:

Bकं पुनः धानं? कता । कथं पुनJायते कता इ त?

यत ् सवLषु साधनेषु सि<नMहतेषु कता वत यता भव त ।। [Abhyankar, 1962]

This means that, the action is directly dependent on the participant hence the
participant that greatly affects the action is known as Swatantra (i.e. independenct)
participant. Therefore, the participant on which the action is mostly dependent as
compared to the other participant i.e. Karkas is known as the Karta Karaka. Thus Karta,
the Swatantra or the most independent participant, is the Ashraya (or the locus) of action.

107
In this regard the activity either resides in or rises from the Karta only. Panini explains
this in the form of sutras as:

!वतंN: कता । [P.1.4.54, Shastri, 2008]

Karta is Swatantra i.e. is not dependent on any karaka or it is not induced by any
other Karaka. For the efficient achievement of the primary action Apadana, Adhikarana
and Sampradana Karakas look upon the Karta (Agent) [Jha, 1969].

5.3.2 Karma Karaka

The way Karta is the Aashraya (locus) of action, in the same manner Karma is the
Aashraya (locus) of the result, if it is different from Karta. Panini [P.1.4.49, Shastri,
2008] opines that the most desirable nominative by the Karta Karaka is the Karma
Karaka. When the Karta carries out any activity, the result of that activity rests in the
Karma. The verbal roots can specifically indicate it. Panini characterizes Karma in the
following sutra:

• कतरOिPसततमं
ु कम । [P.1.4.49, Shastri, 2008]

• तत ् धा वथतावQछे दकफलशा ल वं तत ् धा वथकम वं । [Shastri, 1942]

This means, the result or effect of an action performed by the Karta primarily rests
on it (i.e. Karma), which is clearly determined by the meaning of the verbal root.

As the Karma (object) is the basis of outcome of the primary action, it is one of the
important prominent Karaka. Many times during the identification of the main action, the
Karma Karaka is identified. The auxiliary activities of the instrument i.e. ‘Karana
Karaka’ become effective with reference to the Karma only. To achieve the action the
way Karana Karaka is involved, in the same manner some other Karakas such as
‘Apadana, Adhikarana and Sampradana’ are also involved with some auxiliary actions.
These are explained in the following subsections.

108
5.3.3 Karana Karaka

The Sanskrit grammarians comment on the importance of Karana Karaka distinctly

by pointing out that Karana is the most important tool by means of which the action is
achieved [P.1.4.45, Shastri, 2008]. Karana is the only direct participant in the action.
Karta and Karma also are directly dependent on the Karana for performing the action.
When the Karana Karaka executes its auxiliary actions then only the main action is
executed by the Karta Karaka. This is why the Karana is considered as the efficient
mean in action accomplishment and its supremacy is unquestioned. This is well explained
in the following Sanskrit sutra:

साधकततमं करणम ् l [P 1.4.42, Shastri, 2008]

The ancient book archive Karakollasam [Bharata, 1924] also states the importance
of Karaka in the following verse as:

+Tयते साUयते कता यदा? य वदं त तत ् l

करणं त) वधा बाWयमा/यंतरम प !मत
ृ म ् ll [Bharata, 1924]

The above verse means the Karaka which helps in carrying out the Kriya is known
as Karana Karaka. Which in turn means Karana Karaka helps in attaining the desired
result ascertained by the Kriya i.e. action [Jha, 1969].

5.3.4 Sampradana Karaka

From its etymological meaning, the word Sampradana can be interpreted as ‘He to
whom something is given properly’ is Sampradana.

कमणायम भ ै त स सं दानम ् l [P 1.4.32, Shastri, 2008]

This means Sampradana is that Karaka which the Karta wants to connect with the
object of the action. This reveals that Sampradana is the Karaka which the Karta has in
view and achieves through the Karman and Karana [Jha, 1969].

109
ति +या कारणीभूतेन तत ् B+या कमणा यं संबंधी कतुम भ ै त कता स सं दानम ्। [Shastri,

1942]

This verse further means that, ‘A Sampradana Karaka denotes the motive or
purpose of an action towards which the subject use to proceed through the object’.

5.3.5 Apadana Karaka

Panini defines Apadana by the following Sutra:

धव
ृ मपाये अपादानम ् । [P.1.4.24, Shastri, 2008]

About Apadana Karaka Panini stated that, as when ablation or separation is

affected by a verbal action, the point of separation is called Apadana. Karaka Darsanam
also states that,

अपादOयते अ!मात ् तदपादानम ् । [Jha, 1969]

This means that during the execution of the action whenever the task of separation
from a certain entity is executed then whatever remains unmoved or constant is known as
Apadana. Thus, an Apadana denotes the starting point of an action of separation. The
entity from which something gets separated or is separated out is known as Apadana
[Jha, 1969].

5.3.6 Adhikarana Karaka

An Adhikarana is used to extend help in performing an action by being the base or

location of the verbal activities carried out by the subject. In brief, as stated in Paninian
Ashtadhyayi it may be said that the base or location of the verbal activities is called

Adhikarana: आधारो?धकरणम ् । [P.1.4.45, Shastri, 2008]

Before the action is to be performed, the agent or the object must exist on
something. The entity on which the agent and the object exist is known as Adhikarana
Karaka (abode). It is indirectly involved in the main action through the agent and the

110
object. This is because the action is supposed to be located in the agent i.e. Karata or
otherwise in the Karma i.e. object.

Therefore, according to Mahamuni Panini, ‘Adhikarana’ is the place or thing,

which is the location of the action existing in the agent or the object. Thus by this rule the
term Adhikarana is assigned to the locus of the action i.e. Kriya. Adhikarana may
indicate the place at which the Kriya (the action) is taking place or the time at which the
Kriya is carried out [Jha, 1969]. Any action i.e. the Kriya is either bounded by space
(place) or by time. Thus, sentence and its constituent nominatives are related to space and
time. Considering this spatial or temporal relation among Adhikarana and Kriya, the
Adhikarana Karaka is classified into two types: ‘Kaladhikarana’ and ‘Deshadhikarana’
[Bharati et. al., 1994]

In this way, when the elaboration of all the Karakas is studied well then it can be
said that the karakas are mutually dependent at the same time they take in the part of the
action described by the verb. Therefore, Aakanksha (discussed in section 5.2.1) is the
base of the concept of Karaka. The phenomenon of the inter dependency (i.e. Aakanksha)
of Karaka theory is crucial when their role in the information representation scheme in
the Text Summarization process is to be studied.

5.4 Relation between Karaka and Sentence

From the above discussion of Karakas it became evident that the presence of
Karaka entirely rests in a sentence which is the basic element of a language. In other
words, the functions of Karaka can be defined only with reference to a sentence.

In a sentence, the Karakas appear with a sense of mutual expectancy i.e.

Aakanksha. From the sentence point of view, Aakanksha is the sense of mutual needs
among the Padas i.e. the words. With these mutual needs the Padas try to give a
complete meaning to the sentences. One Pada (or word) alone cannot give the sense of
complete sense. As discussed earlier the Karakas and the verb are bound with a sense of
mutual expectation in a sentence to complete. They mutually want to know one another.
If someone says ‘ate’, then the listener or the reader automatically will be curious about

111
‘Who ate?’, ‘What was eaten?’, ‘From where it was eaten?’, ‘When was it eaten?’ etc. As
an answer we may definitely get one, word i.e. Pada for each question. Hence, one can
say that the Karakas i.e. the participants and verb that come in a sentence are
undoubtedly linked with a need of mutual expectation. When a verb is heard, the
curiosity regarding the Karakas automatically arises. This is beautifully explained by the
following Sanskrit sutra:

B+या वणे कारक!य कारक वणे B+यायाः करण वणे इ त कत5यताया च ् िजJासा
वषय वात ् ।। [Dharmaraja Adhwarindra, 1963].

Therefore, from the linguistic, logical and grammatical point of view the Karakas
and the Kriya are equally important in a sentence and the Aakanksha is attached with
both of them. Hence, it became clear that the Karakas are the entities that add certain
semantic constraints during the analysis of a sentence. These concepts of Karakas can be
applied to English [Bharati, 1996]. If these concepts are mapped to English then they may
enrich the analysis of the English language text Syntacto-semantically. This Syntacto
semantic enrichment is quite useful for the applications like Text Summarization. For
this, a proper semantic representation is necessary. In the next section, the similarity
among the Karaka concept of Sanskrit and the case concept of English is discussed.
Further, we try to develop the Syntacto-semantic representation of the Karakas for
English.

5.5 The Semantic Model Representation of Karaka

Akshar Bharati in his work [Bharati et. al., 1996] has shown that, Paninian grammar
maps the Karaka and Vibhakti with each other and gives less importance to the position
of the words in the language. However as English is not inflectionally rich language, the
concept of Vibhakti is not present in English. In English, the concept of Karaka can be
studied from the point of view of ‘case’ and ‘case endings’. While defining the term case
different scholars have given the different meaning of the word ‘Case’ (Karaka). They
have also commented about their (nouns and pronouns) relation with other nominatives
or verbal used in a sentence.

112
 Fillmore’s theory of language structure states that, case relationships are important
terms in the basic component of each language. In this theory, the case relations are
defined as ‘syntactic relationships which involve certain semantically relevant
information in between various nouns and the different structure that contain them’. Most
of the European languages use the terminology of case to show certain in-flexional
category and also to show the set of semantic separation carried out by the forms of such
categories. Therefore, there is a systematic ambiguity while using the term Case.
However, one can differentiate them respectively as case-relations or case-functions
[Fillmore, 1968]. Similarly, [Quirk et. al., 1972] states that, ‘case is a grammatical
category that can express a number of different relationships between nominal elements’.
[Hockett, 1972] comments that, cases are inflected forms for nouns which fit them for
participation in key constructions relative to verbs. From these theories, actually it is
tough to propose a well-balanced as well as exact model of Karaka in terms of case by
keeping in view its function, meaning, usage, and other relevant aspects in relevance to
Sanskrit. Nevertheless, broadly it can be called as, ‘A Case is form of a nominative which
indicates its either direct or indirect relationship with the verb while carrying out the
action suggested by the verb and hence can be named as a Karaka.’

As discussed in the earlier sections, every sentence has Kriya i.e. an action and that
Kriya is shown by the verb. Thus, every verbal root denotes an action to be performed
and the result of that action, as shown in the figure (6.3). An activity (i.e. a Vyaapaara),
is carried out by the Karta and the result obtained through this activity is located in
Karma.

Figure 5.3 Kriya-Karya relations

The result is achieved when the action is complete. This complete result is obtained
by dividing the main action into several smaller sub actions. These sub actions when

113
work together with the help of Karaka, give the final complete result. For example, let us
revisit the example of wood cutting. In this activity, a person who wants to cut the wood
lifts the axe rings it down with certain pressure and then hits the wood with it. This action
of cutting the wood may get over with a single struck or the struck may have to be
repeated. The main action of cutting the wood is a bundle of various sub actions.

Let us try to study the discovery procedure of Karta that it an argument central to
the action indicated by the verb. The identification of Karaka Karaka role tries to use a
semantic definition along with certain morpho-syntactic information. Consider the
sentence, ‘The man cut the wood with an axe’. The above sentence can be uttered, written
or interpreted as,

I. The man cut the wood.

II. The axe cut the wood.

In above two sentences the action is of wood cutting but in the first sentence the
Karta of this action is ‘man’ while in the other case the Karta is ‘axe’. In the first
sentence, the speaker wants to emphasize the role of axe. Both the things are related to
the speaker’s Aakanksha, depending on the Aakanksha, the Karta is changed. Therefore,
Karta is not always assigned to and known as subject but it is known with the
nomenclature of agent. As the above example suggests, this is not mere a change in
nomenclature, but instead in general also the Karta and agent are different. However, the
same verb can be used to refer the main activity as well as the sub-parts of the complex.

Figure 5.4(a) Karta Karaka identification

114
 Figure 5.4 (b) Karta Karaka identification

Consider the example shown in figure (5.4 a, b); the action of cutting requires an
agentive participant to perform the action. Without the agent, the action of cutting is not
possible. Therefore, the sentence like the man cut the wood with the axe is correct.
However, many times the speaker or the writer does not want to express the role of the
agent, in this case the Karana i.e. instrument is used as an agent who is performing an
action. Hence, the sentence like ‘Axe cut the wood’ is also right. From Sanskrit grammar
point of view, this is referred as Vivaksha. The concept of Vivaksha emphasizes that a
sentence is not only a statement about activity but it also has certain information about
the speaker’s viewpoint [Sharma, 2004]. Karakas play an important role while exploring
this viewpoint.

e.g.: ‘Yesterday Shambhavi beat the dog with the stick in front of the shop.’ The Karaka
annotation of the above sentence can be given as:
Beat : verb (root)
Yesterday : Kala-Adhikarana (time)
Shambhavi : Agent i.e. Karta
Dog : Karma
Stick : Karana
House : Desh-Adhikarana (location i.e. Place)

These Karaka labels can also be reflected on the graph considering the central
element as verb. Let us take another example: ‘The child eats the fruits with the hands’.
The above type of annotation can be shown in the following form. The constraint graph
for the above sentence is shown in figure (6.5).

115
 Figure 5.5 Karaka dependency graph

By considering the Karaka relations, such sentence graphs can be formed for any
sentence by obeying verb centrality. When the graphs as shown in figure (5.5), are
created then it is emphasized that for every source there is only one inward arc. In case
there are many inward arcs for a source then it indicates that there exist more than one
parse for the sentence. In such cases the sentence can be ambiguous. In addition, when a
source has no inward arc then it is the indication of bad parse of the sentence and it can
be considered as an ‘ill formed’ sentence. This kind of graphical representation is closer
to the dependency representation as discussed in the fifth chapter. One more important
characteristic this representation shares with the dependency structure is that, both of
them are verb centric. This motivates us further to use this Karaka labeling for English
and make the sentence analysis semantically richer.

5.6 Karaka Roles for English

The cognitive scientists have elaborated the clear parallelism between the semantic
networks and classical Indian Grammarians frameworks, to carry out the task of Natural
Language processing (NLP). Because of its cognitively rich framework, researchers are
getting attracted towards using Sanskrit as a language for unambiguous knowledge
representation for Artificial Intelligence [Briggs, 1985; Jha, 2010; Nair and Sulochana
Devi, 2011].

As the Paninian grammatical framework is inspired by the inflectionally rich, free

order Sanskrit language; it is considered as well suited for the free order Indian languages
which follow the Sanskrit grammatical framework [Bharati and Sangal, 1993]. However,
[Bharati et. al., 1996] proved it very well that this grammatical framework is more
general than considered so far. It was discussed thoroughly that along with the free order

116
languages the Paninian Grammar can also be applied for the fixed order languages like
English. The Paninian grammar’s dependency framework tries to find out the relation
between the arguments and adjuncts in terms of the Karaka relations. On the same line,
[Begum et. al., 2008] discussed that from the semantics point of view the Dependency
Grammars are more useful than the regular syntax centered Phrase Structures Grammars,
provided that the dependencies are selected rationally. [Begum et. al., 2008] tried to map
the post position markers and the Karaka relations and thereby tried to develop the
dependency annotation strategy for the Paninian Grammar based languages in India.

[Bharati et. al., 1996], states that PG can be applied to English. Further, the Karaka
based annotation strategy for English is given by [Vaidya et. al., 2009]. In the preliminary
work, [Vaidya et. al., 2009] proposed a Karaka based (CPG based) annotation inspired
from the Paninian framework for English by [Bharati et. al., 1996]. They claimed that,
the Karaka concept would include the elements of local verb semantics by taking cue
information from surface level morphological and syntactic information. This annotation
framework adopted the computational Paninian grammar framework (CPG) proposed by
[Bharati et. al., 1996]. After that [Vaidya et. al., 2009] applied this scheme, which was
earlier developed for Hindi along with the other Indian languages, to English Grammar.
This work was further addressed by Chaudhary H. in [Chaudhary et. al., 2011]. These
approaches proved that mapping between the grammatical information processing
schemes of English and Sanskrit might help in the enrichment of the ‘Syntacto-semantic’
contents of the NLP processing applications. In view of this, we intend to make use of
Sanskrit Karakas, which are rich in ‘Syntacto-semantic’ information, for the task of Text
Summarization.

The Karaka theory concentrates upon the connections or dependencies between the
words. Form Karaka point of view the foremost important element of a sentence is verb
(closer to the view of Khandapaksha [Joshi, 1968; Arjunwadkar, 2008]). With the verb
centric view, this framework tries to find out the relation between the verb and the other
words in a sentence in terms of Karaka. As discussed earlier in the 6th chapter, there are
six different Karakas such as ‘Karta, Karma, Karana, Apadana, Sampradana,
Adhikarana’. Akshar Bharati in [Bharati and Sangal, 1993; Bharati et. al., 1996] shows

117
that Karaka relations capture some level of local semantics. While Rambow [Rambow et.
al., 2002] in his work states that, local semantics are closer to the verb semantics. From
Text Summarization point of view local semantics are more important. Hence, to capture
the semantically rich sentences we intended to use the Karaka relations for English text.

With Purva and Uttar Mimansa text processing principles and Karaka theory of
classical Sanskrit grammar it became quite evident that Sanskrit has well formed
grammatical and syntactical framework for the text to be processed. As discussed in the
5th chapter, the Stanford Dependency Grammar also concentrates upon the word
connectedness. Even though the Stanford parse considers the word connectedness it gives
less importance to the meaning i.e. semantics. One of the beneficial characteristic of
Stanford dependency framework is that, it tries to identify the dependency relation among
the words in the syntactic manner, by considering the ‘verb as the root element.

While as discussed in the sixth chapter, Karaka relations are ‘Syntacto-semantic’

in nature. They try to find out the role of a particular term (i.e. a word) with respect to the
verbal action. Thus, these are ‘Syntacto Semantic’ in nature. Thus, the Karaka
frameworks as well as the Stanford dependency framework both are verb centric; if these
two frameworks are mapped with each other properly then the semantically rich
characteristics of Karaka theory can be used for the syntactical analysis of English
sentences. Therefore, one can say that Karaka relations are semantically more salient
than Stanford dependencies. In this way, if we try to incorporate the ‘Syntacto-semantic’
nature of Karaka theory in the dependency annotation scheme of the English text i.e.
with Stanford parsing then it may help us to select the semantically rich sentences for the
task of Text Summarization.

When the Karakas are mapped to English dependency structures, then sometimes
these can be directly mapped to the thematic roles of English. However, the task is not
straightforward. Many times during this mapping, various dependencies get mapped to
same Karaka or various Karakas get mapped to the same dependency label. Such
‘Karaka – Dependency' mapping is discussed in the following section.

118
5.7 Mapping Issues of Stanford Dependencies and Karakas

English and Sanskrit are significantly divergent e.g. let us take an example of
motion verbs in English and Sanskrit. In English, the motion verbs are most often
accompanied with the preposition ‘into’. For instance, ‘He broke into the houses’.
Whereas, the above sentence in Sanskrit is written as ‘Saha balameva gruhe
apravishyat’. Here we can observe that there is no need of any preposition for the ‘Verb’
(Apravishyat).

One of the various divergences Paninian grammar based languages and English
Grammar is the use of expletive subjects. In English, the expletive subjects are used in
the absence of a logical subject to fill up the position of the subject. However, the Hindi /
Sanskrit / Marathi languages which in deed follow the Paninian grammar (PG) can drop
the expletive subject e.g. ‘raining outside’ may not be complete sentence in English
unless it is ‘it is raining outside’, if the expletive subject it is added. The other point of
divergence is copula verbs. The verbs like ‘is, am, are, was, are and were’ are present in
English. These are known as copula verbs. These copula verbs are not the actual verbs
instead these are used to link the subject with the predicate. However, in the PG
framework central binding element is always the verb [Vaidya et.al, 2009]. Therefore
while handling these cases the root is replaced by copula verb e.g. in sentence (1), it can
be easily observed from its Stanford parse [Marneffe and Manning, 2008], that the verb
(i.e. root) will be ‘is’ instead of ‘what’ as shown in sentence (1).

1. ‘What is a computer?’

root (ROOT-0, What-1)

cop(What-1, is-2)

det(computer-4, a-3)

nsubj(What-1, computer-4)

From these divergences, one can comment that the two sentence schemes diverge
syntactically. Hence, the Paninian Grammar cannot be applied to English language
constructs until a proper logical mapping is thought of in between the two grammatical

119
structures. During this mapping procedure, a number of problems may arise due to the
divergence between their language structures. Here our aim is not to build the annotation
corpora based on Karaka relations for English instead, we only intend to use it at the
surface level. Thereby we try to map the English dependency relations to Paninian
Karaka relations. To identify the Karaka relations, the annotation guidelines given by
[Bharati et. al., 2009] are used. While associating the dependency labels to the Karakas
the direct association became quite difficult. To carry out this task, initially one should
obtain the connection i.e. dependency between the words then a thought should be given
on how to map these dependencies to the Karaka. Figure (5.6) describes this process of
associating the Karakas with the dependency labels.

For the given text in English, the Stanford parser gives the dependency parse in
terms of typed dependencies. The annotation corpora guidelines and Stanford typed
dependency manuals are then referred to map these dependencies to the corresponding
Karaka relations. A Karaka mapping table is then prepared from these comparisons.

Figure 5.6 Identification of Karaka relations in an English sentence

In short, the following strategic approach is used to perform the Karaka mapping
for identifying Karaka relations in English:

Here a rule based approach is adopted to associate the Stanford dependencies to the
Karaka. In this case, the output of the Stanford parser is studied manually and rules are
created to map these dependencies under appropriate Karaka. The approach is devised
with following steps:

120
 1. In the first step the sentences are given to Stanford parser.

2. The generated Stanford parses for each sentence are mapped to a Karaka relation
for each dependency relation manually.

3. A rule set is then developed to facilitate the mapping from Stanford dependencies
to Karaka relations.

4. When the parse of a sentence is available all the dependencies are checked from the
rules table for the corresponding Karaka. If the corresponding relation is present
then it is mapped to its corresponding Karaka label. The process is continued for
all the dependencies in the given sentences.

5.8 Karaka Label Identification

The subject - object - verb relationships cannot be directly mapped to Karaka. In

some cases this direct mapping strategy works but in some cases it may not work. From
the DUC dataset almost 500 sentences are used for deciding the rules of mapping by the 5
human annotators who are postgraduate in English, so that a general rule set can be
developed. The human annotators were asked to map the English sentences to the Karaka
labels as per the guidelines given by Annocorra IIIT Hyderabad [Bharati, et. al., 2009].
The Stanford parser gives the direct relation between the words in sentences. While the
Karakas give the relation between the noun and the verbs. The rule set is formed from the
observation as the human experts map the karaka to English dependencies. From the rule
set the respective Karakas are mapped to the corresponding dependencies. While
explaining this association the notations used for the Karakas are as follows:

The six Karakas as declared earlier in chapter 6 are denoted as:

1. Karta (nominative) (k1)

2. Karma (accusative) (k2)

3. Karana (instrumental) (k3)

4. Sampradana (dative) (k4)

5. Apadana (ablative) (k5)

121
 6. Adhikarana (location) (k7)

As the Annocorra [Bharati et. al., 2009] guidelines have used the terms k1, k2, k3,
k4, k5 and k7 to represent the Karaka variables. The proposed model is also developed
by using these notions. Here, k6 notation is not present because in Annocorra [Bharati et.
al., 2009] k6 is not used to show any Karaka.

Some of the examples handled by the human experts to define the Karaka
dependency mapping are discussed in the following subsections. The Stanford
dependency graphs shown in the following section are drawn using the coreNLP
linguistic analysis tool available at http://nlp.stanford.edu:8080/corenlp/process. With the
integrated setup of the Stanford CoreNLP, it becomes quite easy to linguistically analyze
the text. The framework consists of the POS tagger, the dependency Parser along with the
typed basic and universal dependencies. Many more other tasks such as Named entity
recognizer are also available in the framework. The framework provides foundation for
higher level as well as domain-specific text understanding applications. We have used
here the dependency parser to elaborate the dependencies within the sentences

5.8.1 Karta Karaka (k1)

The Karta Karaka indicates the agent who is actually performing the action. When
we go through the Stanford typed dependency manual various dependencies can be
mapped to Karta. The instances those are mapped to Karta are explained here with an
example for each.

(a) ‘nsubj’ dependency

2. Tiger killed the goat in forest.

The Stanford typed dependencies of this sentence shown below are indicating that
for the root word i.e. verb; the nominal subject is ‘tiger’.

nsubj(Killed-2, Tiger-1)

root (ROOT-0, Killed-2)

122
det(goat-4, the-3)

dobj(Killed-2, goat-4)

case (forest-6, in-5)

nmod:in(Killed-2, forest-6)

Figure 5.7 nusbj dependency

Similarly, in figure 5.7 which is graphical dependency representation of the above

sentence, nsubj (killed-2, Tiger-1) is showing that Tiger is ‘Karta’ i.e. performing the
action and hence it can be mapped to subject.

(b) ‘preposition by’ and agent

Consider the following sentence (3),

3. Chocolates were finished by Shambhavi.

Figure 5.8 agent dependency

Here the case with ‘preposition by’ is showing that its noun modifier is the agent
i.e. performing the action. Hence it is mapped to (k1) Karaka i.e. Karta.

(c) csubj (clausal subject)

When the subject of clause itself is the clause then it is known as ‘csubj’. Consider
sentence (4),

4. What she did has no ground.

123
 Figure 5.9 csubj dependency

Here the clausal subject is indicating the phrase ‘what she did’ is the actual
performer of the action. Hence, ‘csubj’ dependencies are also mapped to Thus in short the
dependencies that can be mapped to Karta Karaka are given in the following table

Table 5.1 Dependency mapping for Karta karaka

Karakas Dependency Relations

(k1)(Karta) nsubj agent csubj ‘preposition by’ xsubj

5.8.2 Karma Karaka (k2)

The object or the patient of an action described by the verb is known as Karma
Karaka. The Stanford dependency instances those are mapped to Karma are explained
here with an example for each.

(a) dobj

The accusative object specified by the action is known as the direct object. Consider
the following sentences (6),

5. She delivered a baby boy.

Figure 5.10 dobj dependency

Here the ‘boy’ is the direct object on which the action is getting performed directly.
Therefore, the dependency ‘dobj’ gets mapped to (k2) Karma.

(b) nsubjpass

124
 It is the noun phrase, which is the subject of the passive clause. Let us consider the
following sentence (6).

6. A ball was hit by Shambhavi.

Figure 5.11 nsubjpass dependency

In figure (5.11) the dependency label ‘nsubjpass’ is connecting the verb ‘hit’ with
‘ball’ thereby showing that the object of the action is ‘ball’. Therefore, it is associated to
(k2) Karma.

(c) ccomp

The internal subject that works as the object is given the dependency label as
‘ccomp’. Consider the following sentence (7),

7. Shambhavi says that Asit likes to swim.

Figure 5.12 ccomp dependency

It can be observed in sentence (7) that, ‘likes’ which is the verb of the clausal
complement has the internal subject ‘Asit’. ‘Asit’ is the internal subject of the clause
‘likes to swim’. Hence ‘Asit’ is the object i.e. (k2) Karma.

Therefore, the words with dependency parse ‘ccomp + nsubj’ are assigned to
Karma. The complete universal dependencies generated for the above sentence (9) are
given below:

nsubj(says-2, Shambhavi-1)

root(ROOT-0, says-2)

125
mark(likes-5, that-3)

nsubj(likes-5, Asit-4)

nsubj(swim-7, Asit-4)

ccomp(says-2, likes-5)

mark(swim-7, to-6)

xcomp(likes-5, swim-7)

As can be seen from the universal dependencies, the subject of the open clausal
complement i.e. ‘xcomp’ is also the object. Hence, the dependency ‘xcomp + nsubj’
dependencies are assigned to Karma Karaka.

In the above examples, ‘dobj’, ‘nsubjpass’ are showing the terms on which the
action is getting performed hence they are mapped to (k2) Karma (object) while
‘preposition by’ is mapped to (k1) Karta i.e. subject because it is indicating the ‘doer’ of
the action. Table (5.2) shows the dependency mapping for Karma Karaka.

Table 5.2 Dependency mapping for Karma Karaka

Karakas Dependency Relations

xcomp +
(k2)(Karma) Dobj nsubjpass Csubjpass ccomp + nsubj
nsubj

6.8.3 Karana Karaka (Instrument) (k3)

The Karana Karaka is the instrument with which the action is carried out.

(a) Preposition with

8. Ram cuts the apple with knife.

Figure 5.13 ‘preposition with’ dependency

126
 As shown in figure (5.13), for sentence (8), with the instrument ‘knife’, which was
modified, by ‘preposition with’ the action has been carried out. Therefore, it is mapped to
(k3) Karana Karaka. Preposition with is modifying nouns hence it is Karana.

5.8.4 Sampradana Karaka (beneficiary) (k4)

Sampradana Karaka receives or gets benefited from the action. It can also be said
that, the person/object for which the Karma is intentional, is known as Sampradana. In
this regard, the Sampradana is the final destination of the action.

(a) ‘preposition to’

9. Dipti gave chocolates to Shambhavi.

nsubj(gave-2, Dipti-1)

root(ROOT-0, gave-2)

dobj(gave-2, chocolates-3)

case(Shambhavi-5, to-4)

nmod:to(gave-2, Shambhavi-5)

Figure 5.14 (a) ‘preposition to' dependency

In the sentence (9), the final destination of the action ‘gave’ is Shambhavi. The
word ‘Shambhavi’ is the beneficiary in the action. This relation is marked with
‘preposition to’ therefore, it is associated to the (k4) Sampradana Karaka.

10. He says that you like to swim.

Figure 5.14 (b) ‘preposition to’ dependencies

127
 The figure 5.14 (b) for sentence (10), also shows the ‘preposition to’ dependency
but in this case it cannot be mapped to Sampradana Karaka i.e. k4 because here ‘to’ is
part verb swim it is not modifying any noun hence it cannot be considered as a noun
which in turn can be Karaka.

11. Ram gave me a book.

The above example (11) shows that nominal subject of an open clausal complement
is object.

Figure 5.15 iobj dependency as Sampradana Karaka

The book is the ‘direct object’ while ‘me’ is the indirect object shown respectively
by ‘dobj’, ‘iobj’. However, both are ultimately the objects, in this case the indirect object
‘me’ is beneficiary hence it is mapped to Sampradana Karaka.

12. He gave flowers for Shambhavi.

In sentence (12), ‘preposition for’ is modifying the noun ‘Shambhavi’ which is

actual beneficiary of the action. Hence it can be mapped to Sampradana Karaka.

Figure 5.16 ‘preposition for’ dependency as Sampradana karaka

The table (5.3) gives the summarized table for Sampradana Karaka i.e. k4 mapping
with the dependency labels.

Table 5.3 Dependency mapping for Sampradana Karaka

Karakas Dependency Relations

(k4)(Sampradana) iobj ‘preposition to’ ‘preposition for’

128
5.8.5 Apadana Karaka (source) (k5)

Apadana is the source of the action. It is the point of departure. The word Apadana
shows the movement ‘away from' indicating the point of separation. In this task of
separation, the term on which Apadana Kriya (action) takes place (i.e. from which the
things are separated) remains stationary.

a) Preposition from

13. Shambhavi tore the page from the book with a scissor.

Figure 5.17 ‘preposition from’ dependency

Following are the universal dependencies for the above sentence.

nsubj(tore-2, Shambhavi-1)

root(ROOT-0, tore-2)

det(page-4, the-3)

dobj(tore-2, page-4)

case(book-7, from-5)

det(book-7, the-6)

nmod:from(tore-2, book-7): book remains stationary

case(scissor-10, with-8)

det(scissor-10, a-9)

nmod:with(tore-2, scissor-10)

In this case of sentence (13), the ‘book’ from which the pages are tore remains
stationary while ‘to tore’ the page the instrument used is ‘scissor’. Therefore, the

129
‘preposition from’ is mapped to ‘(k5) Apadana’ while the ‘preposition with’ is mapped to
‘(k3) Karana’ i.e. Instrument

b) Preposition of

14. The shoes are made of leather.'

det(shoes-3, The-2)

nsubjpass(made-5, shoes-3)

auxpass(made-5, are-4)

root(ROOT-0, made-5)

case(leather-7, of-6)

nmod:of(made-5, leather-7)

Here in sentence (14), for the dependency relation ‘nmod: of (made-5, leather-7)’ is
a case of Apadana i.e. k5. This assignment is done because of the fact that of conceptual
separation of ‘shoes’ from the original raw material ‘leather’.

Table 5.4 Dependency mapping for Apadana Karaka

Karakas Dependency Relations

(k5) ( Apadana) ‘preposition of’ ‘preposition from’

5.8.6 Karana and Apadana Conflict Resolution

15. Ram cuts the apple with knife.

root(ROOT-0, Ram-1)

det(apple-3, the-2)

dep(Ram-1, apple-3)

case(knife-5, with-4)

nmod:with(apple-3, knife-5)

130
 In sentence (15), ‘preposition with’ is modifying noun ‘apple’. It is indicating the
relations between two nouns such that to carry out the Kriya i.e. action of cutting on one
term ‘apple’ the other term ‘knife’ is used as the instrument. Hence the ‘preposition
with’ is mapped to (k3) Karana. Nevertheless, in the following example of (16), the case
is a bit conflicting.

16. Shambhvai is angry with Asit.

Figure 5.18 ‘preposition with’ dependency

The universal typed dependencies of the sentence (16) are given below:

nsubj(angry-3, Shambhvai-1)

cop(angry-3, is-2)

root(ROOT-0, angry-3)

case(Asit-5, with-4)

nmod:with(angry-3, Asit-5)

It can be observed from the figure (7.14) as well as from the typed dependencies of
the sentence (13) that in this case ‘preposition with’ is modifying the verb. ‘Asit’ is the
source of anger. Therefore ‘preposition with’ is assigned to (k5) Apadana instead of (k3)
i.e. Karana.

5.8.7 Conflict Resolution with ‘preposition to’

Consider the sentences (17) and (18) along with their universal typed dependencies,

17. What famous actress was married to Patudi?

det(actress-3, What-1)

amod(actress-3, famous-2)

131
nsubjpass(married-5, actress-3)

auxpass(married-5, was-4)

root(ROOT-0, married-5)

case(SaifAliKhan-7, to-6)

nmod:to(married-5, Patudi-7)

18. What district do the Kas Pathar belong to?

det (district-2, What-1)

preposition_to (belong-7, district-2)

‘Patudi’ is the beneficiary in sentence (17) so ‘preposition_to’ [nmod: to (married-

5, Patudi -7)] is mapped to (k4) and in sentence (18) ‘district’ is a place not beneficiary,
so cannot be (k4), which contradicts. However final destination of the ‘preposition to’ is
‘Patudi’ hence it is beneficiary i.e. (k4) Sampradana.

5.8.8 Adhikarana Karaka (time or place) (k7)

The place at which or the time during which the action takes place is known as
Adhikarana Karaka. The participant that denotes the location of the action is marked as
'k7p' indicating that it is ‘Deshadhikarana’ i.e. locative of place. While the participant
that denotes the time of the action is marked as 'k7t' indicating that it is ‘Kaladhikarana’
i.e. locative of place

(a) preposition on

19. I kept it on the table.

Figure 5.19 ‘preposition on’ dependency

132
 In figure (5.19) for the sentence (19) the ‘preposition on’ is indicating the location
of the Kriya i.e. action. Hence, it is mapped to (k7p) Adhikarana.

Consider the sentences (20), (21) and (22):

(b) tmod (temporal modifier)

For the sentences (20) and (21) the Stanford typed dependency ‘tmod’ is indicating the
time of action. Hence, it is mapped to (k7t) Kaladhikarana.
20. It rained yesterday.

nsubj(rained-3, It-2)

root(ROOT-0, rained-3)

nmod:tmod(rained-3, yesterday-4)

21. The cost of living was less those days.

det(cost-2, The-1)

nsubj(less-6, cost-2)

case(living-4, of-3)

nmod:of(cost-2, living-4)

cop(less-6, was-5)

root(ROOT-0, less-6)

det(days-8, those-7)

nmod:tmod(less-6, days-8)

(c) preposition in

22.Samvit was very naughty in his childhood.

nsubj(naughty-4, Samvit-1)

cop(naughty-4, was-2)

advmod(naughty-4, very-3)

root(ROOT-0, naughty-4)

133
case(childhood-7, in-5)

nmod:poss(childhood-7, his-6)

nmod:in(naughty-4, childhood-7): preposition in Adhikarana.

As given in sentence (22), the activity is taking place in the childhood which is the
locus i.e. (k7p) Adhikarana of the activity.

23. I am in India.

nsubj(India-4, I-1)

cop(India-4, am-2)

case(India-4, in-3)

root(ROOT-0, India-4)

In sentence (23) also the typed Stanford dependencies are showing that
‘preposition in’ is showing the place of the action. Hence it is associated with (k7p)
DeshAdhikarana.

(d) Preposition at

24. Shambhvi is not at home.

nsubj(home-5, Shambhvai-1)

cop(home-5, is-2)

neg(home-5, not-3)

case(home-5, at-4)

root(ROOT-0, home-5)

In the above sentence (24) ‘preposition at’ is showing the place where the action is
taking place. Hence, it is (k7p) DeshAdhikarana.

However, a case of conflict may arise for ‘preposition on’. In the sentence (19),
table is the location of books. So Preposition on can be mapped to k7 but consider the
following sentence (25):

134
25. On average, how many miles are there to the moon?

Most often ‘preposition on’ is mapped to locus (either time or place) however the
instances like on an average e.g. ‘On an average the rates are low.’ These are limitations,
which may hamper the performance.

Table 5.5 Dependency mapping for Adhikarana karaka

Karakas Dependency Relations

(k7) ( Adhikarana) ‘preposition on’ ‘preposition in’ ‘preposition at’ tmod

The Karaka mapping is further compared with [Gothi et. al., 2014] which was close
to their Karaka mapping also. Associating the Karakas with Stanford dependencies helps
in understanding the key trends and issues that are quite relevant while using grammatical
formalism of one language to other. This Karaka based dependency tagging can give the
considerably advantageous during the tasks of semantic role labeling which may further
help in the task of Text Summarization.

Here, we have tried to map the Stanford dependencies to Karaka labels but many of
the Stanford dependencies did not get mapped with any of the Karakas such as ‘complm,
conj, cop, nn, det, mark, poss, possive, preconj, predet, punct, reference [Marneffe and
Manning, 2008]’. Most of the not mapped dependencies are syntactical in nature.
Similarly, the propositions like ‘preposition on’ were showing cases of conflicts.

5.9 Conclusion

The Dependency Grammar structure of a sentence discussed in the earlier chapter

and the Karaka relations among the sentences both are verb dependent. Both of them
revolve around the concept of satisfying the mutual expectancy i.e. Aakanksha of the
words for each other in a sentence. From the graph structures observed and from the
Karaka structures studied it seems to be possible to map the dependency relations with
the Karaka relations. While the Karaka relation are mapped to English Stanford
dependencies, the verb centrality of Stanford dependency framework and Karaka makes
it quite possible to map the Karaka labels with some of the Stanford dependencies.

135
However, though the examples cover most of the generic cases, it can be observed that
many of the Stanford dependencies remained unmapped with Karakas. Typically, those
relations that are not much concerned with the argument structure of the verb; e.g.
adverb-verb relations are given less consideration. Here, for the task of Text
Summarization, the Karaka relations like Karta and Karma, which are most important
roles from the action execution point of view as well as linguistic point of view, are given
more emphasis. In addition, the main intention in the task of summarization was to
identify the important semantic roles in the sentence rather than the annotation. Hence,
much deeper analysis is not carried out.

136