Semantic Parsing Approach in Malayalam for Machine Translation

Semantic Parsing Approach in Malayalam for Machine Translation

Shabina Bhaskar

MPhil CS IIITMK-Trivandrum

Sandeep Chandran

Assistant Professor

LBS Center for Science and Technology Trivandrum

Abstract – Semantic parsing is the process of mapping a natural language sentence into a formal representation of its meaning. Shallow form of Semantic Parsing is proposed in this paper. Here Semantic roles are identified using Karaka theory in Paninian Grammar. It is useful for both the syntax analysis and semantic analysis of Malayalam sentences. And we proposed an algorithm for finding semantic relations using Karaka theory.

1. INTRODUCTION

   Semantic parsing maps text to formal meaning representations. There are two approaches in semantic parsing deep parsing and shallow parsing. Deeper semantic analysis means representation of a sentence in predicate logic or other formal representation of meaning.

   Shallow parsing means case role identification. The study of roles associated with specific verbs and across classes of verbs is called thematic role analysis or case role (karaka) analysis. The proposed approach is shallow parsing approach in Malayalam. We can consider a sentence compose of entities and interactions between entities. Here nouns and verbs take as entities and relation (karakas) between verbs and nouns as interactions. Here we identify abstract semantic roles or thematic relations such as Agent, Patient etc.

   According to Paninian perspective there are four levels in the understanding process of a sentence [4]. They are surface level (uttered sentence), vibakthi level, karaka level and semantic level. The karaka level has relationship to semantics on one side and to syntax on the other side. Karaka relation can be identified from post position markers after noun or surface case ending of noun. These markers and case endings are called vibhakthi. In Malayalam karaka relations are analyzed from vibakthi and post position markers. An algorithm is implemented for finding case roles of nouns and machine learning approach is used for finding the roles of adjectives, adverbs and postposition markers. So the system proposes a hybrid approach for Semantic Parsing.

2. RELATED WORKS

   Shallow semantic parsing is the process of assigning a simple WHO did WHAT to WHOM, WHEN, WHERE, WHY, HOW, etc. and this semantic parsing approach described by Gildea and Jurafsky[1].

   In Shallow Semantic Parsing using Support Vector Machines [2] first they placed their statistical classication algorithm with one that uses Support Vector Machines and then added to the existing feature set. Then evaluate results using both hand- corrected TreeBank syntactic parses, and actual parses from the Charniak parser.

   In Semantic role labeling using syntactic chunk by Kandri Hacogolu presented a semantic role labeler (or chunker) that groups syntactic chunks (i.e. base phrases) into the arguments of a predicate [3]. This is accomplished by casting the semantic labeling as the classication of syntactic chunks (e.g. NP-chunk, PP-chunk) into one of several classes such as the beginning of an argument (B- ARG), inside an argument (I-ARG) and outside an argument (O). This amounts to tagging syntactic chunks with semantic labels using the IOB representation. The chunker is realized using support vector machines as one- versus-all classiers.

3. METHODOLOGY

   The proposed system architecture contains two phases syntactic parsing and semantic parsing. In syntactic parsing the steps involved are tokenization, POS tagging and morphological analysis .In Semantic Parsing are Semantic role labeling and conceptual graph representation.

   Fig.1: System Architecture

   1. Syntactic parsing

      Syntactic parsing start with tokenization. As Malayalam is a highly inected and agglutinative language, tokenization cannot be done blindly. We used a supervised machine learning approach which classifies the tokens into two groups: the words to be split (compound word) and not to be split (Simple word).Compound word is splitted using sandhi rules. After tokenization we perform parts of speech tagging. For tagging we used a supervised machine learning approach which identifies the syntactic structure of the sentence. Next phase is Morphological analysis and it is the process of recognizing the root or stem and the categorical information of the items that may accompany the root or the stem. In the case of Noun, the root word and it categorical information such as person, number gender information and in the case of verb, tense, aspect and modality are included. And in this system we add an additional field which specifies the Semantic Property associated with nouns and verbs. And it is stored as

      • Noun {root, {suffix, vibakthi}, gender, number, person, semantic property}

      • Verb {root, {suffix, verb forms},semantic property}

      • For example The noun " " is stored as

        (radha,O:nirdeshika :prathigrahika, :samyojika,:udeshika:prayogika, :adharika,:sambandhika,female,singula r,third, {human})

        Semantic property of verbs such as Artistic Performance Verbs ( ,) Combine verbs( ,) and Nouns such as Animals(, ) Birds( ,)

        ,Buildings( ,) are used in this work.

   2. Semantic Parsing

      In Semantic Parsing we introduce Shallow Semantic Parsing approach in Malayalam. Here Semantic Roles of noun, adjectives, adverbs and postposition markers are considered. Semantic Roles of Nouns are identified using the Karaka Theory which has the basis in Paninian Grammar. Karaka theory specifies the relation between noun and verb. And the roles of adjectives, adverbs and postposition markers identified using a machine learning approach.

   3. Semantic Role Labeling

      Malayalam follows the system of marking grammatical relations and semantic roles through a set of case suffixes. According to Karaka Theory the Karaka Relations (Semantic relations) can be identified from vibakthi. In semantic parsing for semantic role (karaka) labeling is based on vibakthi prathyam. From morphological analysis the case markers of nouns and pronouns are obtained from which the vibakthi class is identified. In Malayalam there are 7 vibakthi.

      Table 1: Vibakthi list

   4. Semantic roles

      Semantic roles such as agent patient, instrument, beneficiary, experiencer, recipient etc are identified.

      • Agent: Deliberately performs the action

      • Experiencer: The entity that receives sensory or emotional input

      • Theme: Undergoes the action but does not change its state .

      • Patient: Undergoes the action and changes its state

      • Instrument: Used to carry out the action

      • Beneficiary: The entity for whose benefit the action occurs

        Karaka Relations in Malayalam

        1. Kartha Kararaka

           The activity actually resides in or springs from karta. The result of the verb is reflected in Kartha karakamVibakthi corresponding to kath karaka is Nirdeshika and prayogika.

           Nirdeshika (Nominative)

           Role can be agent , experiencer or causer.

           Here the sentence is subject+verb form and is Agent.

           If the verb is emotional verb then the role of is experiencer.

           Here the sentence is subject+object+verb form and if the verb is causative verb then the noun with nirdeshika vibakthi is take the role as Agent.

        2. Karma Karakam

           It denote the objet of the sentence.vibakthi is Prathigrahika (Accusative) or Nirdeshika

           (Agent) (Patient)

           The sentence is subject+object+ verb form then the Noun with Prathigrahika vibakthi is take the role of Patient. And if the verb is causative verb then the role is Experiencer.

           (Agent) (Experiencer)

           Nirdeshika(Nominative) The object denotes Patient.

           (Agent) (Patient)

        3. Karana Karakam

           It denotes the instrument for performing the action. It will be in the prayojika (Instrumental) vibakthi form.

           (Instrument)

        4. Hethu Karakam

           It denotes the reason for verb and it is in Prayogika vibakthi.The role is agent.

        5. Sakshi Karakam

           Sakshi Karakam will be in Samyojika(Sociative) vibakthi and it denote role as Participant.

           (Agent) (Participant)

        6. Swami Karakam

           Swami karakam denote the beneficiary of the activity and vibakthi is Udeshika(Dative).

           (Agent) (Beneficiary)

           (Patient)

        7. Adikarana Karakam

        Here the vibakthi is Adaraika(Locative) vibakthi and it denote location.

        (Location)

   5. Semantic Role Labeling Algorithm

      Based on the above Karaka Relations we developed an algorithm for Semantic Role Labeling.

      SRL Algorithm for a sentence

      Step1: identify the verb from the POS tagging and chunking.

      Step2: select the NP chunks and find the vibakthi from morphology.

      Step3: if the sentence is Noun+verb form

      Step3.1: if nirdeshika vibakthi then check the verb

      class

      Step3.2: if its causative then it is labeled as causer.

      Step3.3 : if it is emotional verb then it is labeled as experiencer .

      Step3.4: otherwise labeled as agent.

      Step3.2: if it is prathigrahika vibakthi it is labeld as Patient. Step3.3: if it is udeshika vibakthi then it is labeled as Beneficiary.

      Step3.4: if it in samyojika vibakthi it is labeled as patient. Step4: if the sentence is noun1+noun2+verb form

      Step4.1: one noun is nirdeshika

      Step4.1.1. if other is also nirdeshika and its semantic class is non human then nirdeshika noun is agent and other is patient.

      Step4.1.2: if other noun is prathigrahika or samyojika then check verb class then it is causative verb the nirdeshika noun is causer otherwise it is agent and other is patient.

      Step4.1.3: if other noun is udeshika verb class is beneficiary then udeshika noun is beneficiary other is agent. Otherwise udeshika noun is patient and other is agent.

      Step5: if the sentence is noun1+noun2+noun3+verb form Step5. 1:if noun is nirdeshika and its semantic class is non human then nirdeshika noun is agent. Step4.1.2: if other noun is prathigrahika or samyojika then check patient.

      Step4.1.3: if other noun is udeshika verb class is beneficiary then udeshika noun is beneficiary.

      Step6: Stop

   6. SRL using YamCha Toolkit

   If the above algorithm fails to find the semantic roles then we use machine learning approach for SRL. Here we used Yamcha for SRL. YamCha (Yet Another Multipurpose Chunk Annotator) is a generic, customizable, and open source text chunker oriented toward a lot of NLP tasks, such as POS tagging, Named Entity Recognition, base NP chunking, and Text chunking. YamCha is using the machine learning algorithm called Support Vector Machines and it is exactly the same system which performed the best in the CONLL2000 Shared Task, Chunking and BaseNP chunking task. The features of YamCha include its high performance chunker based on Support Vector Machines and partial chunking. It is independent from the given task, training/testing with any data which can be seen as a generic text chunking task. It uses PKE/PKI which make the classification (chunking) speed faster than the original SVMs. YamCha can also redefine feature sets (window-size), parsing-direction (forward/backward) and algorithms of multi-class problem (pair wise /one vs rest) and it also uses C/C++ library.

   Here we used three column format.1st column is 'word', second column is 'POS tag' third column is 'semantic roles'

   .The last column represents a true answer tag which is going to be trained by SVMs. First of all, run yamcha- config with –libexecdir option. The location of Makefile which is used for training is output. Please copy the Makefile to the local working directory. There are two mandatory parameters for training.

   CORPUS: The location of file which is written in the training/test format.

   MODEL: Prefix name of model file(s)

   D. Conceptual graph representation

   Conceptual Graph (CG) expresses meaning in a form that is logically precise, humanly readable, and computationally tractable. With their direct mapping to language, conceptual graphs serve as an intermediate language for translating computer-oriented formalisms to and from natural languages. With their graphic representation, they serve as a readable, but formal design and specification language. CGs have been implemented in a variety of projects for information retrieval, database design, expert systems, and natural language processing. In diagrammatic representation of CG rectangular box represents the concept node and oval represents the semantic relation.

   For example

   CG representation

   { ( : Agent) (: Patient) (: Beneficiary)}

4. CONCLUSION

The Proposed system is tested using 1000 sentence and we get 90% accuracy. The system can be utilized for the applications such as document summarization and natural language generation etc.

REFERENCES

1. Gildea and D. Jurafsky, Automatic labeling of semantic roles, In Proceedings of the 38th Annual Conference of the Association for Computational Linguistics (ACL-00), pages 512520, Hong Kong, October 2000.

2. Kadri Hacioglu, Sameer Pradhan, Wayne Ward, James Martin, and Dan Jurafsky, Shallow Semantic Parsing Using Support Vector Machines, CSLR Tech. Report, CSLR-TR-2003-1.

3. Kadri Hacioglu and Wayne Ward, Target word detection and semantic role chunking using support vector machines, In HLT- 03.

4. Akshar Bharati, Vineeth Chaithanya, Rajeev Sangal,Natural Language Processing: A Paninian Perspective, Prentice-Hall of India, New Delhi .

5. A.R.RajarajaVarma , Keralapanineeyam, Sahitya Pravarthaka ,C S Ltd., Kottayam,1968

6. Jisha P. Jayan and Rajeev R. R, Parts of Speech Tagger for Malayalam, IJCSIT International Journal of Computer Science and Information Technology, Vol.2, No.2, December 2009, pp.209-213.

7. Saranya S. Morphological Analyzer for Malayalam verbs, M.Tech Thesis, Amrita School of Engineering, Coimbatore, 2008.

8. A.R.RajarajaVarma, Keralapanineeyam ,SahityaPravarthaka,CSLtd.,

   Kottayam,1968

9. Daniel Jurasfky and James H. Martin, An introduction to Natural Language Processing, Computational Linguistics , and Speech Recognition, PearsonEducation , Inc.2000.