The Sanskrit Heritage Site

Version 2.75 [2013-05-05] (fr)

Welcome to the Sanskrit Heritage site. It provides various services for the computational treatment of Sanskrit.

The first service is dictionary access. The dictionary is a hypertext structure giving access to the Sanskrit lexicon, given with grammatical information. There are currently two versions of the dictionary.
The first one is the original Heritage Sanskrit-French dictionary, which serves as morphology generator, and is thus fully equipped with grammatical tools. Furthermore it offers a rich encyclopedic contents about Indian culture. You may also download a printable pdf version of this dictionary, as explained below.

The second lexicon is a digital version of the Monier-Williams Sanskrit-English dictionary, a much more complete lexicon for the Sanskrit language. It is issued from Thomas Malten's digitalization of the Monier-Williams at Köln University, turned into an XML databank by Jim Funderburk, and finally adapted to the HTML Heritage look and feel by Pawan Goyal. The Sanskrit Heritage dictionary is thus mirrored in the Monier-Williams, which allows compatibility of the grammatical tools.

The choice of the dictionary is set to a default by the configuration of the server site. But each dictionary is accessible separately by its search page, respectively Sanskrit Heritage and Monier-Williams.

This site offers a number of linguistic services for the Sanskrit language, such as a Sanskrit Reader that parses Sanskrit transliterated text into Sanskrit banks of tagged hypertext. Various phonological and morphological tools are also provided, as explained below.

Sanskrit Heritage dictionary in book form

Multilingual hyper-text dictionary

Interactive browsing

The dictionary may be accessed through an indexing engine: Sanskrit Heritage or through its mirror Monier-Williams.

Your browser must be HTML5 compliant, and for proper viewing of Sanskrit text you must have installed on your system open type fonts for roman transliteration with diacritics, and for devanagari. For instance, install fonts IndUni, available from John Smith's site. A Unicode-compliant font for devanagari with proper ligatures is Apple's Devanagari MT for Macintosh OS X stations. For Windows users, installation of font 'Arial MS Unicode' is advised for proper rendering.

You may have to fiddle with the controls of your browser, so that the font declarations from the dictionary pages get precedence over the standard selection, and thus encoding is specified as Unicode compliant (UTF-8 encoding).

Remark that many words are given with their etymology as hypertext links. You may thus navigate from a word to its components, down to its roots. Also, the gender declarations of the main entries are mouse-sensitive, and give you direct access to the relevant declension table. Similarly, the present class mark of the verbal roots gives access to the conjugation schemes. Also for verb entries, preverbs lead you to the correspondingly prefixed derived verbs.

All these grammatical tools, originally developed for the Heritage dictionary, are being progressiveley extended to the Monier-Williams dictionary.

Sanskrit made easy

If you want to search for a Sanskrit word without knowing its exact transliteration, go to section "Sanskrit made easy" of the index page, which allows you to search for words without knowing precise diacritics usage. For instance, search Vishnou, Siva, or the grammarian Panini. This interface is limited for the moment to the Sanskrit Heritage dictionary.

Sanskrit Grammarian

This interface gives the declension tables for Sanskrit substantives. Try out this declension engine by submitting Sanskrit stems with intended gender. The same transliteration conventions as for the dictionary index apply. For instance, submit "deva" with gender Mas, or "devii" with gender Fem, or "brahman" with gender Neu. The fourth button, labeled "Any", may be used for the words which take their gender from the context, such as deictic pronouns ("aham", "tvad"), or numeral words such as "dva", "tri", etc.

A conjugation engine for roots is also available. It handles the full present system: present indicative, imperfect, imperative and optative, as well as the passive present, the perfect, the aorist and the future. Participial stems, absolutives and infinitives are listed as well. Some secondary conjugations (causative, intensive, desiderative) are also generated, for the full present and future systems. Try out this conjugation engine with data such as "bhuu" 1, "as" 2, "m.rj" 2, "han" 2, "haa" 3, "hu" 3, "daa" 4, "su" 5, "p.r" 6, "yuj" 7, "k.r" 8, "j~naa" 9, "namas" 10. In order to get the secondary conjugations of a root, enter code 0. You may cascade by generating declensions of the generated participial stems.

A word of caution is called for here. The only safe way to get correct inflected forms is to enter the stem and its morphological parameters consistently with their specification in the Heritage dictionary. This is specially true of roots, since thay appear with various names according to various Sanskrit grammars. For instance, root hū is called hū, hvā or hve according to various grammarians. Another problem is homophony. When two items have the same phonetic realisation, their respective lexemes are disambiguated by an integer index, which is specific to the lexicon. Thus there are three roots named mā in the Sanskrit Heritage dictionary. They are adressed respectively (in SH transliteration) as maa#1, maa#3 and maa#4. If you ask for the conjugated forms of maa in present classes 2 or 3, the system will guess you mean maa#1 (to measure). But if you mean maa#3 (to mow) or maa#4 (to exchange) you have to enter explicitly their stems maa#3 or maa#4. Entering an arbitrary stem and arbitrary morphology parameters may yield random results or error messages.

Lemmatizer

Conversely, a lemmatiser attempts to tag inflected words. Try for instance (in Velthuis transliteration format) "devaat", "jagmivaan", "a.s.tau" (clicking on Noun) or "apibat", "akaar.siit", "dudoha", "vaahyate" etc (clicking on Verb). This lemmatizer knows about inflected forms of derived stems in some secondary derivations. For instance, "darzayi.syati" is found as conjugated form: { ca. fut. a. sg. 3 }[dṛś_1], "dariid.rzyate" yields { int. pr. m. sg. 3 }[dṛś_1], "did.rk.sate" yields { des. pr. m. sg. 3 }[dṛś_1] and "bibhik.se" yields { des. pft. m. sg. 3 | des. pft. m. sg. 1 }[bhaj].

Morphology

Sanskrit Reader

Try our interactive Sanskrit Reader. It is able to segment simple sentences. Try for instance to segment "tryambaka.myajaamahesugandhi.mpu.s.tivardhanam" (we assume Velthuis transliteration here). Then push the "Tagging" button and get the fully tagged sentence. You will see two segmentations, one with an identified compound form "tri-ambakam", the second with a compounded segment "tryambakam". Note that each segment is indicated with a lemma giving its stem and the set of morphological parameters that may generate the segment form from its stem. The stem is hyperlinked to the dictionary of choice.

Note also that segments are separated by phonological information in the shape of a sandhi rule, justifying correct obtention of the original sentence by successive sandhi application. For instance, solution 1 explains the compound "tryambakam" as the sandhi of segments "tri" and "ambakam" by rule ‹i|a → ya›.

The reader may be helped by inserting blanks in the input at word junction. For instance, the above mantra may be entered as "tryambaka.m yajaamahe sugandhi.m pu.s.tivardhanam". But compounds should stay in one piece. Spaces are also needed for hiatus, in sentences such as: "tacchrutvaasa~njaya uvaaca".

Many options are provided in the menu of the Reader page. For instance, clicking on the Unsandhied button we may present text in padapāṭha form, where each chunk is in terminal sandhi form. For instance "tryambakam yajaamahe sugandhim pu.s.tivardhanam".

Two strengths of the Reader are provided. The Simplified mode, offered as a default, does not recognize vocatives. The Complete mode is more powerful, using the full range of participles of verbs, privative compounds, etc. It may however return so many solutions that listing all solutions is impractical, and other facilities must be used.

The grammar used to recognize sentences is explained as a local automaton state transition graph Lexer automaton. This is actually a simplification of the segmenter automaton control. A simpler one, close to the Simplified mode of the reader, is Simplified automaton. A fuller one, close to the Complete mode of the reader, is Complete automaton. The color codes of these diagrams explain the output conventions of the tags.

In these diagrams, transparent nodes are non generative, and colored nodes correspond to the lexical categories recognized by the lemmatizer. The category Auxi is the subset of Verb consisting of conjugated forms of roots "k.r", "as" and "bhuu" used as auxiliaries in periphrastic constructions. Pv denotes sequences of preverbs.

Sanskrit Parser

If in the reader you press the "Parsing" button, many irrelevant pseudo-solutions are eliminated. Try for instance example "pratilekhanenaak.saraa.nisundaraa.nibhavanti". In Simplified mode, it shows 80 potential segmentation solutions, but the parser keeps only 1.

Each solution returned with the parser is marked with a green check sign, which may be pressed to get the semantic analysis of the sentence in terms of roles (kāraka).

The parser recognizes sentences. It may be made to recognize nominal phrases, provided one presses the "Contextual topic" button with the intended gender. You may for instance analyze the compound: "pravaran.rpamuku.tama.nimariicima~njariicayacarcitacara.nayugala.h" as a masculine nominal. Alternatively, one can ask to recognize this form as a single word, by pressing "Word" rather than the default "Sentence" text category. When breaking the text with spaces, the Word mode allows to recognize texts given in padapāṭha fashion. It is also possible to recognize sequences of chunks in final sandhi form separated by spaces, where sandhi will be assumed to be undone between the chunks, by specifying the "Unsandhied" mode in the reader interface.

Sanskrit Tagger

The semantic analysis may be still ambiguous, since a given segment may be decorated by several morphological categories. All interpretations are presented under the role matrix, sorted by increasing penalty. Check for your favorite interpretation in this list, and select it by clicking on its green heart symbol. The system will return the corresponding unambiguously tagged sentence, as a page which you may save on your own station. Iterating this process allows you to progressively tag a Sanskrit text with the Sanskrit reader assistance.

Alternatively, you may select the ambiguous morphology choices, each being provided with a selection button. Selections are chosen by default at the first choice, but you may override this default and choose manually e.g. the genders of nominals. When your choice is finalized, just click on the "Submit" button and you will get the corresponding deterministically tagged sentence. This tool is useful for semi-manual corpus annotation.

Summary mode

Now that you are more familiar with using the various modes of the Reader on small Sanskrit sentences, it is time to try to analyse more complex sentences. Obviously the listing of all solutions is out of the question with long sentences in Complete mode. A new visual interface is offered for semi-automatic segmentation. This new Summary mode is actually now proposed as a default in the Reader page.

Try for instance "satya.mbruuyaatpriya.mbruuyaannabruuyaatsatyamapriya.mpriya.mcanaan.rtambruuyaade.sadharma.h sanaatana.h". The display presents a summary of the union of all solutions, as a chart of segments aligned on their respective input contribution. You see at the right end the segment sanātanas proposed first, on top of a forest of smaller words combinations. Click on the green check sign below it. The check sign becomes blue, and the forest of irrelevant combinations vanishes. Do the same under the satyam segments, then under apriyam, all segments presented as top candidates. Now choose the particle ca (and thus na). Now only one choice remains, between brūyāt and brūyām. Clicking on the first one will finish the job. Indeed only one solution remains, as may be checked by clicking on the "Show All Solutions" check sign, which may be invoked at any point on the process. You are now viewing the same output as given by the Reader in Tagging mode, but constrained to use only segments checked in the Summary. You may alternatively check the "Show Preferred Solutions" check sign, asking the Parser to further filter according to its system of penalties. If you make a mistake in the selection of segments, it is easy to backtrack using the "Undo" button.

Other Sanskrit Resources

We have on on-going cooperation with the Department of Sanskrit Studies of the University of Hyderabad on computational linguistics for Sanskrit. A joint research team has been formed, together with scholars from the Sanskrit Library. This team is actively developing cooperating multi-platforms Web services.

In october 2007 we organized the First International Sanskrit Computational Linguistics Symposium. Please visit the Symposium Site. This was followed by the Second Symposium in may 2008 at Brown University, by a third one in january 2009 at Hyderabad University, and by a fourth one in december 2010 at JNU. The next one just took place in january 2013 at IIT Bombay.

The Zen Library

This site reflects an ongoing project of Sanskrit processing on a comprehensive software platform. The project is based on a structured lexicographic database, compiled from the Sanskrit Heritage dictionary, and on the Zen computational linguistics toolkit. This toolkit is a library of programs implemented in Pidgin ML, functional core of the Objective Caml programming language. The Zen library and its documentation are available as free software under the Gnu Lesser General Public License (LGPL) from the Zen site.

The Sanskrit Portal

Artwork credits