Skip to main content

Morphological Disambiguation of Classical Sanskrit

  • 270 Accesses

Part of the Communications in Computer and Information Science book series (CCIS,volume 537)


Sanskrit, the “sacred language” of Ancient India, is a morphologically rich Indo-Iranian language that has received some attention in NLP during the last decade. This paper describes a system for the tokenization and morphosyntactic analysis of Sanskrit. The system combines a fixed morphological rule base with a statistical selection of the most probable analysis of an input text. After an introduction into the research history and the linguistic peculiarities of Sanskrit that are relevant to the task, the paper describes the present architecture of the system and new extensions that increase its accuracy when analyzing morphologically ambiguous forms. The algorithms are tested on a gold-annotated data set of 3,587,000 words.


  • Word Order
  • Conditional Random Field
  • Parallel Corpus
  • Lexical Database
  • Noun Class

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

This is a preview of subscription content, access via your institution.

Buying options

USD   29.95
Price excludes VAT (USA)
  • DOI: 10.1007/978-3-319-23980-4_3
  • Chapter length: 19 pages
  • Instant PDF download
  • Readable on all devices
  • Own it forever
  • Exclusive offer for individuals only
  • Tax calculation will be finalised during checkout
USD   44.99
Price excludes VAT (USA)
  • ISBN: 978-3-319-23980-4
  • Instant PDF download
  • Readable on all devices
  • Own it forever
  • Exclusive offer for individuals only
  • Tax calculation will be finalised during checkout
Softcover Book
USD   59.99
Price excludes VAT (USA)


  1. 1.

    Bloch [2] gives an introduction into the linguistic history of Sanskrit. More details about the Vedic layer are found in Witzel [34].

  2. 2.

    In a recent research project, the Aṣṭādhyāyī has been fully annotated on the morphological, lexical and word-semantic level to make it easier accessible for Western researchers without knowledge of Sanskrit [25]. A web platform that gives access to this database is available at

  3. 3.

    The rules of the Aṣṭādhyāyī are not given in the order in which they need to be applied for generating a valid Sanskrit word. Instead, it is generally assumed that their order minimizes the resulting rule base. The Indian grammar uses the concept of anuvṛtti (“following”) rules for regulating the order in which rules and their elements are applied. These rules are not part of the text of the Aṣṭādhyāyī, but are recorded – and heavily discussed – in the commentary literature; refer to [4, 187ff.] for details about rule order in the Aṣṭādhyāyī, and to [26] for the proof of minimality in a subset of Pāṇinian rules.

  4. 4.

    Refer to page Subsect. 3.3 for the phonological phenomenon of Sandhi.

  5. 5.

    The GRETIL web repository ( contains less than 20 million strings. Several of the texts are not usable for automatic processing due to excessive formatting of their editors, as described in Sect. 3.6.

  6. 6.

    The following abbreviations are used in this paper: Nom.: nominative; Acc.: accusative; Ins.: instrumental; Dat.: dative; Gen.: genitive; Loc.: locative; Voc.: vocative; Co.: compound; Sg.: singular; Du.: dual; Pl.: plural; Msc.: masculine; Fem.: feminine; Neu.: neuter; Ind.: indeclinable; Pres.: present; Impf.: imperfect; Perf.: perfect tense; Proh.: prohibitive (a kind of imperative that is only used in negated phrases); PastPart.: past participle, frequently with a passive sense; PresPart.: present participle

    Ambiguities in a morphological analysis are expressed by a regex-style notation, with | denoting the operator OR and round brackets a set of options. So, (Nom.|Acc.|Voc.)Pl. Neu.means that a form is a neuter plural either in nominative or accusative or vocative.

    The plus operator + is used to separate elements of compounds, the ampersand sign & to indicate Sandhi at word boundaries (Sect. 3.3).

    Further abbreviations: tri: trigram based model for morphological disambiguation; crf: Conditional Random Fields; me: Maximum Entropy.

  7. 7.

    Note that the word bahuvrīhi is itself an example of a bahuvrīhi compound. In its “default interpretation” as a so-called tatpuruṣa (“his man”, an instance of relational compounding) compound, it means just “much rice.”.

  8. 8.

    From a purely grammatical point of view, the sentence can also be translated as “... destroyed by these bad actions.” Numerous references of the bahuvrīhi solution with unambiguous case endings (e.g., in Nom. Pl. Msc.) make the proposed interpretation much more plausible.

  9. 9.

    Though slightly outdated, the grammar of Stenzler still provides a good introduction into Sanskrit Sandhi rules [33, 3ff.].

  10. 10.

    Refer to [24, 1ff.] for a detailed linguistic description with several examples. Brockington locates the epics, especially the Mahābhārata, in a continuum “of dialects and language registers from classical or Pāṇinian Sanskrit at one end to colloquial MIA [Middle Indo-Aryan] at the other” [3, 83] and makes this linguistic situation responsible for the irregular application of Sandhi in epic texts.

  11. 11.

    Emeneau describes the basic parameters of the interaction between Indo-Iranian and Dravidian languages [5]. A quantitative overview of the major influences that is based on Mayrhofer’s etymological dictionary [20] is given in [9].

  12. 12.

    A quantitative evaluation of the reuse of Pāṇinian vocabulary is presented in [11].

  13. 13.

    A member of a low caste.

  14. 14.

  15. 15.

    As these data are only checked by one annotator and have not been adjudicated, they should rather be called semi-gold annotations.

  16. 16.

    The Mahābhārata and the Rāmāyaṇa are the two central epic texts written in Sanskrit. The term Purāṇa (“old (story)”) denotes a group of works dealing with virtually everything; refer to Rocher for an introduction [28].

  17. 17.

    The TTRs found in the third column of Table 1 are obtained by calculating the TTRs for each text, and then averaging these values over the topic levels. Because text lengths have not been used as normalizing factors, the TTRs of underrepresented topic levels such as śruti or Buddhist literature are most probably too high.

  18. 18.

    The one-solution case predicts the correct morphological category in about 99.8 % of all cases. The errors are caused by irregular word forms.

  19. 19.

    The parameter 3 for the window size was chosen after comparing disambiguation results for window sizes between 1 and 7. Window sizes above 3 did not consistently increase the accuracy, but required higher training times.

  20. 20.

    The final Sandhi  has been transformed into the pausa form m.

  21. 21.

    Used in the Java implementation of the OpenNLP package; settings: smoothing factor: 0.001, 100 iterations.

  22. 22.

    Used in the C++ implementation from; settings: L2 regularization: 2.0, one-dimensional architecture.


  1. Adler, M., Elhalad, M.: An unsupervised morpheme-based HMM for Hebrew morphological disambiguation. In: Proceedings of the 21st International Conference on Computational Linguistics, pp. 665–672 (2006)

    Google Scholar 

  2. Bloch, J.: Indo-Aryan from the Vedas to Modern Times. Librarie d’Amérique et d’Orient, Paris (1965)

    Google Scholar 

  3. Brockington, J.: The Sanskrit Epics. Brill, Leiden (1998)

    Google Scholar 

  4. Cardona, G.: . A Survey of Research. Mouton, The Hague - Paris (1976)

    Google Scholar 

  5. Emeneau, M.: Dravidian and indo-aryan: the indian linguistic area. In: Emeneau, M.B. (ed.) Language and Linguistic Area, pp. 167–196. Stanford University Press, Stanford (1980)

    Google Scholar 

  6. Gillon, B.S.: Review of “Natural Language Processing: A Paninian Perspective" by A. Bharati, V. Chaitanya, and R. Sangal. Prentice-Hall of India 1995. Computational Linguistics 21(3), 419–421 (1995)

    Google Scholar 

  7. Gillon, B.S.: Word order in classical Sanskrit. Indian Linguist. 57(1–4), 1–35 (1996)

    Google Scholar 

  8. Hellwig, O.: \(\mathtt{{SadnskritTagger}}\): a stochastic lexical and POS tagger for Sanskrit. In: Huet, G., Kulkarni, A., Scharf, P. (eds.) Sanskrit CL 2007/2008. LNCS, vol. 5402, pp. 266–277. Springer, Heidelberg (2009)

    CrossRef  Google Scholar 

  9. Hellwig, O.: Etymological trends in the Sanskrit vocabulary. Literary Linguist. Comput. 25(1), 105–118 (2010)

    CrossRef  Google Scholar 

  10. Hellwig, O.: Performance of a lexical and POS tagger for Sanskrit. In: Jha, G.N. (ed.) SCL. LNCS, vol. 6465, pp. 162–172. Springer, Heidelberg (2010)

    CrossRef  Google Scholar 

  11. Hellwig, O., Petersen, W.: What’s got to do with it? The use of -headers from the Aṣṭādhyāyī in Sanskrit literature from the perspective of corpus linguistics. In: Proceedings of the WCS 2015 (forthcoming)

    Google Scholar 

  12. Huet, G.: A functional toolkit for morphological and phonological processing, application to a Sanskrit tagger. J. Funct. Program. 15(04), 573–614 (2005)

    CrossRef  MATH  Google Scholar 

  13. Kiparsky, P.: On the architecture of grammar. In: Huet, G., Kulkarni, A., Scharf, P. (eds.) SCL. Lecture Notes in Computer Science, vol. 5402, pp. 33–94. Springer, Heidelberg (2009)

    Google Scholar 

  14. Knauth, J., Alfter, D.: A dictionary data processing environment and its application in algorithmic processing of Pali dictionary data for future NLP tasks. In: Proceedings of the 5th Workshop on South and Southeast Asian NLP, pp. 65–73 (2014)

    Google Scholar 

  15. Kneser, R., Ney, H.: Improved backing-off for m-gram language modeling. In: Proceedings of the International Conference on Acoustics, Speech, and Signal Processing, pp. 181–184 (1995)

    Google Scholar 

  16. Kulkarni, A., Shukla, D.: Sanskrit morphological analyser: some issues. Indian Linguist. 70(1–4), 169–177 (2009)

    Google Scholar 

  17. Kulkarni, M.: Phonological overgeneration in paninian system. In: Huet, G., Kulkarni, A., Scharf, P. (eds.) SCL. LNCS, vol. 5402, pp. 306–319. Springer, Heidelberg (2009)

    CrossRef  Google Scholar 

  18. Lafferty, J.D., McCallum, A., Pereira, F.C.N.: Conditional random fields: probabilistic models for segmenting and labeling sequence data. In: Proceedings of the Eighteenth International Conference on Machine Learning, pp. 282–289 (2001)

    Google Scholar 

  19. Lee, J., Naradowsky, J., Smith, D.A.: A discriminative model for joint morphological disambiguation and dependency parsing. In: Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, vol. 1, pp. 885–894 (2011)

    Google Scholar 

  20. Mayrhofer, M.: Kurzgefaßtes etymologisches Wörterbuch des Altindischen. Carl Winter Universitätsverlag, Heidelberg (1982)

    Google Scholar 

  21. Mishra, A.: Simulating the system of Sanskrit grammar. In: Huet, G., Kulkarni, A., Scharf, P. (eds.) SCL. LNCS, vol. 5402. Springer, Heidelberg (2009)

    Google Scholar 

  22. Mittal, V.: Automatic Sanskrit segmentizer using finite state transducers. In: Proceedings of the ACL 2010 Student Research Workshop, pp. 85–90. Association for Computational Linguistics, Stroudsburg (2010)

    Google Scholar 

  23. Monier-Williams, M.: -English Dictionary, 3rd edn. Munshiram Manoharlal Publishers Pvt. Ltd., New Delhi (1988)

    Google Scholar 

  24. Oberlies, T.: A Grammar of Epic Sanskrit. De Gruyter (2003)

    Google Scholar 

  25. Petersen, W., Soubusta, S.: Structure and implementation of a digital edition of the Aṣṭādhyāyī. In: Kulkarni, M. (ed.) Recent Researches in Sanskrit Computational Linguistics, pp. 84–103. D.K. Printworld, New Delhi (2013)

    Google Scholar 

  26. Petersen, W.: Zur Minimalität von Śivasūtras: eine Untersuchung mit Methoden der formalen Begriffsanalyse. Ph.D. thesis, Universität Düsseldorf (2008)

    Google Scholar 

  27. Ratnaparkhi, A.: Maximum Entropy Models for Natural Language Ambiguity Resolution. Ph.D. thesis, University of Pennsylvania (1998)

    Google Scholar 

  28. Rocher, L.: The , A History of Indian Literature, vol. II, Fasc. 3. Otto Harrassowitz, Wiesbaden (1986)

    Google Scholar 

  29. Scharfe, H.: Grammatical Literature. A History of Indian Literature, Volume 5, Fasc. 2, Otto Harrassowitz, Wiesbaden (1977)

    Google Scholar 

  30. Shacham, D., Wintner, S.: Morphological disambiguation of Hebrew: a case study in classifier combination. In: Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 439–447. Association for Computational Linguistics, Prague (2007)

    Google Scholar 

  31. Shukla, P., Kulkarni, A., Shukl, D.: Geeta: Gold standard annotated data, analysis and its application. In: Proceedings of ICON (2013)

    Google Scholar 

  32. Staal, J.: Word Order in Sanskrit and Universal Grammar. Foundations of Language, Supplementary Series, vol. 5. D. Reidel Publishing Company, Dordrecht (1967)

    CrossRef  Google Scholar 

  33. Stenzler, A.F.: Elementarbuch der Sanskrit-Sprache. Max Mälzer, Breslau (1872)

    Google Scholar 

  34. Witzel, M.: Early indian history: linguistic and textual parametres. In: Erdosy, G. (ed.) The Indo-Aryans of Ancient South Asia. Language, Material Culture and Ethnicity, vol. 1, pp. 85–125. Walter de Gruyter, Berlin (1995)

    Google Scholar 

  35. Yuret, D., Türe, F.: Learning morphological disambiguation rules for Turkish. In: Proceedings of HLT-NAACL (2006)

    Google Scholar 

Download references

Author information

Authors and Affiliations


Corresponding author

Correspondence to Oliver Hellwig .

Editor information

Editors and Affiliations

Rights and permissions

Reprints and Permissions

Copyright information

© 2015 Springer International Publishing Switzerland

About this paper

Cite this paper

Hellwig, O. (2015). Morphological Disambiguation of Classical Sanskrit. In: Mahlow, C., Piotrowski, M. (eds) Systems and Frameworks for Computational Morphology. SFCM 2015. Communications in Computer and Information Science, vol 537. Springer, Cham.

Download citation

  • DOI:

  • Published:

  • Publisher Name: Springer, Cham

  • Print ISBN: 978-3-319-23978-1

  • Online ISBN: 978-3-319-23980-4

  • eBook Packages: Computer ScienceComputer Science (R0)