ganapati

Sanskrit philology for a search index: verse boundaries, metre, lemmas, names and grammar

ganapati reads Sanskrit texts and describes them in ways a search index can use. It finds where each verse ends in a GRETIL, TEI, DCS or vedicreader file and cuts the text into chunks on those boundaries. It works out the metre of every verse from the weights of its syllables, and it records the lemmas (dictionary forms) and English glosses behind the words. It also maps an epithet such as Pārtha to the name a translation uses, and it can derive a word through Pāṇini’s grammar.

ganapati is built to work with litesearch, but only the search profiles and filters in ganapati.lemma need it. When both are installed, litesearch finds ganapati through a package entry point, so a Sanskrit file added to an index is chunked by verse and each chunk gets its metre as metadata, which by_meter can then filter on:

from litesearch import Index
from ganapati import Meter

ix = Index('vault.db')
ix.add('mahabharata.htm')
ix.search('dharma')[:3]
ix.db.by_meter(meter=Meter.ANUSTUBH)[:3]

/Users/71293/code/personal/orgs/ganapati/.venv/lib/python3.13/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm

[{'content': 'nārāyaṇaṃ namaskṛtya naraṃ caiva narottamam\ndevīṃ sarasvatīṃ caiva tato jayam udīrayet',
  'metadata': '{"meter": "anuṣṭubh", "variant": "pathyā", "gana": "ta_ra_ga_ga", "pada": "8"}',
  'node_id': '73a63025c7fc3d14#0',
  'doc_id': '73a63025c7fc3d14',
  'page': 0},
 {'content': 'samāsīnān abhyagacchad brahmarṣīn saṃśitavratān\nvinayāvanato bhūtvā kadā cit sūtanandanaḥ',
  'metadata': '{"meter": "anuṣṭubh", "variant": "pathyā ra-vipulā", "gana": "ya_ta_ga_ga", "pada": "8"}',
  'node_id': '73a63025c7fc3d14#0',
  'doc_id': '73a63025c7fc3d14',
  'page': 2},
 {'content': 'tam āśramam anuprāptaṃ naimiṣāraṇyavāsinaḥ\ncitrāḥ śrotuṃ kathās tatra parivavrus tapasvinaḥ',
  'metadata': '{"meter": "anuṣṭubh", "variant": "pathyā", "gana": "ja_sa_ga_ga", "pada": "8"}',
  'node_id': '73a63025c7fc3d14#0',
  'doc_id': '73a63025c7fc3d14',
  'page': 3}]

verse_meta writes that metadata. For Gītā 1.1 it finds an anuṣṭubh, the metre of the epics, in its regular form (pathyā), with eight syllables to the pāda (quarter-verse):

verse_meta('dharmakṣetre kurukṣetre samavetā yuyutsavaḥ '
           'māmakāḥ pāṇḍavāścaiva kimakurvata sañjaya')

{'meter': 'anuṣṭubh', 'variant': 'pathyā', 'gana': 'ma_ra_ga_ga', 'pada': '8'}

Behind it is the scansion. scan marks each syllable heavy (g, guru) or light (l, laghu), and detect_meter compares the whole pattern with the metres it knows. Here it recognises the first verse of the Meghadūta as mandākrāntā:

scan('dharmakṣetre kurukṣetre')     # g for heavy, l for light

'gggglggg'

detect_meter('kaścit kāntāvirahaguruṇā svādhikārātpramattaḥ '
             'śāpenāstaṃgamitamahimā varṣabhogyeṇa bhartuḥ '
             'yakṣaścakre janakatanayāsnānapuṇyodakeṣu '
             'snigdhacchāyātaruṣu vasatiṃ rāmagiryāśrameṣu')['name']

<Meter.MANDAKRANTA: 'mandākrāntā'>

A line at a time

Aligned recordings and speech recognition give you lines, not verses. match_pada names the metres a single pāda could belong to, and group_verses works out how many lines make a verse.

padas = ['kaścit kāntāvirahaguruṇā svādhikārātpramattaḥ',
         'śāpenāstaṃgamitamahimā varṣabhogyeṇa bhartuḥ',
         'yakṣaścakre janakatanayāsnānapuṇyodakeṣu',
         'snigdhacchāyātaruṣu vasatiṃ rāmagiryāśrameṣu']
match_pada(padas[0]), group_verses(padas).size

Names and terms

A reader may ask about Sri Ranganatha where the translation writes Viṣṇu, or about anumana where it says inference. canonical and term_gloss connect the two:

canonical('Sri Ranganatha'), canonical('Partha'), term_gloss('anumana')

(('viṣṇu', 'nārāyaṇa', 'hari'), ('arjuna',), ('inference',))

The modules

module	what is in it
ganapati.sanskrit	Vedic accent removal, an ASCII search key for a word in any script, citation ids
ganapati.text	verse boundaries, verse layout by pāda, the two chunkers, readers for GRETIL, TEI, vedicreader XML and JSON, and DCS
ganapati.metre	IAST transliteration, syllable weights, the gaṇas, 100 fixed-pattern metres, the āryā family, checks against recorded audio
ganapati.lemma	vidyut lemmas, Monier-Williams glosses, a source’s own word-by-word analysis and its repair, the litesearch profiles and search filters
ganapati.names	one key for every spelling of a name, epithets of deities and heroes, English for śāstra and jyotiṣa terms
ganapati.vyakarana	derivations through the Aṣṭādhyāyī, a word-by-word analysis matched to its line, the sandhi rule at a join, two words joined into their saṃhitā

from ganapati import * brings in every module except ganapati.sanskrit, which you import by name.

Install

pip install ganapati

Use pip install 'ganapati[search]' to install litesearch with it. Lemmas need vidyut’s data, an 81 MB download made the first time you ask for a lemma, and glosses need Monier-Williams, downloaded the first time you call mw_lexicon. Without either, an index still gets each verse’s metre, its audio timings, and any analysis the source itself carries.

Credits

The joiner in ganapati.vyakarana reads vidyut’s table of sandhi rules, shipped as ganapati/data/vidyut_sandhi_rules.csv. vidyut is by ambuda.org (https://github.com/ambuda-org/vidyut) and is MIT licensed. Its copyright notice and licence travel with the table in ganapati/data/LICENSE-vidyut.