Datasets:

SanskritVoyager
/

SanskritTravelogue

text_id int64 1 13.8k	collection stringclasses 9 values	title stringlengths 3 94k ⌀	author stringclasses 591 values	category stringclasses 117 values	word_count float64 1 2.76M ⌀	segment_count float64 1 742k ⌀	avg_segment_length float64 1 206k ⌀	source stringlengths 3 86 ⌀	notes null
1	DCS	Gheraṇḍasaṃhitā	null	Yoga	5,003	789	6.34	DCS	null
3	DCS	Nibandhasaṃgraha	Dalhaṇa	Ayurveda	7,794	1,034	7.54	DCS	null
4	DCS	Sūryasiddhānta	null	Jyotisha	1,561	278	5.62	DCS	null
5	DCS	Agastīyaratnaparīkṣā	null	Ratnashastra	559	87	6.43	DCS	null
8	DCS	Aṣṭāṅgahṛdayasaṃhitā	Vāgbhaṭa	Ayurveda	82,232	15,143	5.43	DCS	null
9	DCS	Sarvāṅgasundarā	Aruṇadatta	Ayurveda	6,868	938	7.32	DCS	null
10	DCS	Ayurvedarasāyana	Hemādri	Ayurveda	3,358	606	5.54	DCS	null
12	DCS	Mṛgendratantra	null	Tantra	2,671	480	5.56	DCS	null
13	DCS	Mṛgendraṭīkā	Nārāyaṇakaṇṭha	Tantra	16,680	1,837	9.08	DCS	null
14	DCS	Gaṇakārikā	Haradatta	Tantra	99	16	6.19	DCS	null
15	DCS	Ratnaṭīkā	Bhāsarvajña	Tantra	4,765	626	7.61	DCS	null
18	DCS	Tantrasāra	Abhinavagupta	Tantra	10,978	804	13.65	DCS	null
19	DCS	Maṇimāhātmya	null	Ratnashastra	740	127	5.83	DCS	null
20	DCS	Bhairavastava	Abhinavagupta	Tantra	138	21	6.57	DCS	null
21	DCS	Śivasūtra	null	Tantra	254	74	3.43	DCS	null
22	DCS	Śivasūtravārtika	Varadarāja	Tantra	3,973	757	5.25	DCS	null
23	DCS	Spandakārikā	Vasugupta	Tantra	593	106	5.59	DCS	null
24	DCS	Spandakārikānirṇaya	Kṣemarāja	Tantra	7,123	647	11.01	DCS	null
26	DCS	Yogasūtra	Patañjali	Yoga	843	194	4.35	DCS	null
27	DCS	Rasaratnākara	Nitya Nātha Siddha	Rasashastra	42,810	7,512	5.7	DCS	null
28	DCS	Carakasaṃhitā	null	Ayurveda	70,357	9,960	7.06	DCS	null
30	DCS	Commentary on Amaraughaśāsana	null	Tantra	510	60	8.5	DCS	null
31	DCS	Vātūlanāthasūtras	null	Tantra	46	13	3.54	DCS	null
32	DCS	Vātūlanāthasūtravṛtti	Anantaśaktipāda	Tantra	1,068	106	10.08	DCS	null
33	DCS	Kādambarīsvīkaraṇasūtramañjarī	Purūravas	Kamashastra	182	33	5.52	DCS	null
34	DCS	Commentary on the Kādambarīsvīkaraṇasūtramañjarī	null	Kamashastra	1,343	124	10.83	DCS	null
36	DCS	Janmamaraṇavicāra	Vāmadeva	Tantra	2,626	346	7.59	DCS	null
37	DCS	Kauśikasūtra	null	Grhyasutra	5,469	834	6.56	DCS	null
40	DCS	Gorakṣaśataka	Gorakhnath	Tantra	1,393	201	6.93	DCS	null
41	DCS	Amarakośa	Amara	Kosha	11,706	2,098	5.58	DCS	null
44	DCS	Pāśupatasūtra	null	Tantra	514	155	3.32	DCS	null
45	DCS	Pañcārthabhāṣya	Kauṇḍinya	Tantra	18,678	3,364	5.55	DCS	null
48	DCS	Nādabindūpaniṣat	null	Upanishad	269	41	6.56	DCS	null
49	DCS	Brahmabindūpaniṣat	null	Upanishad	310	44	7.05	DCS	null
50	DCS	Amṛtabindūpaniṣat	null	Upanishad	96	15	6.4	DCS	null
52	DCS	Bhāgavatapurāṇa	null	Purana	67,783	10,959	6.19	DCS	null
54	DCS	Vaikhānasadharmasūtra	null	Dharmasutra	3,539	53	66.77	DCS	null
55	DCS	Rasādhyāya	null	Rasashastra	5,630	973	5.79	DCS	null
56	DCS	Rasādhyāyaṭīkā	null	Rasashastra	6,193	687	9.01	DCS	null
59	DCS	Amaruśataka	Amaru	Kavya	2,197	212	10.36	DCS	null
60	DCS	Gobhilagṛhyasūtra	null	Grhyasutra	2,510	187	13.42	DCS	null
61	DCS	Vṛddhayamasmṛti	Yama	Dharmashastra	879	140	6.28	DCS	null
62	DCS	Rasārṇavakalpa	null	Rasashastra	3,957	708	5.59	DCS	null
63	DCS	Skandapurāṇa	null	Purana	13,421	2,408	5.57	DCS	null
64	DCS	Rasārṇava	null	Rasashastra	26,249	4,706	5.58	DCS	null
65	DCS	Rāmāyaṇa	Valmīki	Ramayana	242,803	38,038	6.38	DCS	null
66	DCS	Sarvadarśanasaṃgraha	Mādhava	Darshana	719	116	6.2	DCS	null
67	DCS	Āyurvedadīpikā	Cakrapāṇidatta	Ayurveda	25,870	2,966	8.72	DCS	null
68	DCS	Rasaratnasamuccaya	Vāgbhaṭa	Rasashastra	20,928	3,841	5.45	DCS	null
69	DCS	Mahābhārata	null	Mahabharata	1,057,598	168,034	6.29	DCS	null
70	DCS	Rasahṛdayatantra	Govindbhagavatpāda	Rasashastra	5,903	1,035	5.7	DCS	null
71	DCS	Mugdhāvabodhinī	Caturbhuja Miśra	Rasashastra	21,933	2,340	9.37	DCS	null
72	DCS	Carakatattvapradīpikā	Śivadāsasena	Ayurveda	66	7	9.43	DCS	null
73	DCS	Bhāvaprakāśa	Bhāvamiśra	Ayurveda	10,190	1,841	5.54	DCS	null
74	DCS	Toḍalatantra	null	Tantra	4,743	849	5.59	DCS	null
75	DCS	Paramānandīyanāmamālā	Makarānanda	Kosha	175	36	4.86	DCS	null
76	DCS	Padārthacandrikā	Candranandana	Buddhist	25	8	3.12	DCS	null
77	DCS	Matsyapurāṇa	null	Purana	97,558	18,149	5.38	DCS	null
78	DCS	Rājamārtaṇḍa	Bhoja	Yoga	822	105	7.83	DCS	null
80	DCS	Kāvyādarśa	Daṇḍin	Alamkarashastra	5,690	947	6.01	DCS	null
85	DCS	Rasakāmadhenu	Śrīcūḍāmaṇimiśra	Rasashastra	6,058	1,107	5.47	DCS	null
86	DCS	Rasendracūḍāmaṇi	Somadeva	Rasashastra	13,344	2,469	5.4	DCS	null
87	DCS	Kāvyālaṃkāra	Bhāmaha	Alamkarashastra	4,188	801	5.23	DCS	null
88	DCS	Kāvyālaṃkāravṛtti	Vāmana	Alamkarashastra	60	17	3.53	DCS	null
89	DCS	Comm. on the Kāvyālaṃkāravṛtti	null	null	281	58	4.84	DCS	null
90	DCS	Saṅghabhedavastu	null	Buddhist	7,223	226	31.96	DCS	null
91	DCS	Rasaratnasamuccayabodhinī	Āśubodhavidyābhūṣaṇa, Nityabodhavidyāratna	Rasashastra	4,383	587	7.47	DCS	null
92	DCS	Bhramarāṣṭaka	null	Kavya	164	16	10.25	DCS	null
93	DCS	Haṃsadūta	Rūpa Goswāmin	Kavya	2,555	285	8.96	DCS	null
94	DCS	Rasikasaṃjīvanī	Arjunavarmadeva	Rasashastra	116	17	6.82	DCS	null
95	DCS	Madanapālanighaṇṭu	Vinoda	Kosha	5,032	977	5.15	DCS	null
96	DCS	Vaikhānasagṛhyasūtra	null	Grhyasutra	5,020	63	79.68	DCS	null
98	DCS	Rasikapriyā	Kumbha	Rasashastra	221	35	6.31	DCS	null
99	DCS	Liṅgapurāṇa	null	Purana	106,186	18,791	5.65	DCS	null
100	DCS	Meghadūta	Kālidāsa	Kavya	2,468	244	10.11	DCS	null
101	DCS	Bhadrabāhucarita	null	null	177	32	5.53	DCS	null
102	DCS	Kauśikasūtradārilabhāṣya	null	null	593	121	4.9	DCS	null
103	DCS	Cakra (?) on Suśr	null	Ayurveda	93	12	7.75	DCS	null
104	DCS	Hārāṇacandara on Suśr	null	Ayurveda	120	12	10	DCS	null
105	DCS	Laṅkāvatārasūtra	null	Buddhist	8,396	1,071	7.84	DCS	null
107	DCS	Jaiminigṛhyasūtra	null	Grhyasutra	5,685	187	30.4	DCS	null
108	DCS	Aṣṭāṅgasaṃgraha	null	Ayurveda	1,402	246	5.7	DCS	null
109	DCS	Śārṅgadharasaṃhitā	Śārṅgadhara	Ayurveda	4,375	812	5.39	DCS	null
110	DCS	Śārṅgadharasaṃhitādīpikā	Āḍhamalla	Ayurveda	11,211	1,832	6.12	DCS	null
111	DCS	Gūḍhārthadīpikā	Kāśirāma	Ayurveda	3,971	619	6.42	DCS	null
112	DCS	Nāḍīparīkṣā	Rāvaṇa	Ayurveda	1,224	195	6.28	DCS	null
113	DCS	Nāḍīvijñāna	Kaṇāda	Ayurveda	113	15	7.53	DCS	null
114	DCS	Ānandakanda	null	Rasashastra	64,264	12,162	5.28	DCS	null
115	DCS	Prasannapadā	Candrakīrti	Buddhist	731	130	5.62	DCS	null
117	DCS	Viṃśatikākārikā	Vasubandhu	Buddhist	245	45	5.44	DCS	null
118	DCS	Viṃśatikāvṛtti	Vasubandhu	Buddhist	1,309	184	7.11	DCS	null
119	DCS	Kālikāpurāṇa	null	Purana	3,547	601	5.9	DCS	null
120	DCS	Ratnadīpikā	null	Ratnashastra	1,440	256	5.62	DCS	null
121	DCS	Pāraskaragṛhyasūtra	null	Grhyasutra	3,112	322	9.66	DCS	null
122	DCS	Hiraṇyakeśigṛhyasūtra	null	Grhyasutra	3,925	359	10.93	DCS	null
123	DCS	Rasaprakāśasudhākara	Yaśodhara	Rasashastra	12,294	2,121	5.8	DCS	null
125	DCS	Khādiragṛhyasūtra	null	Grhyasutra	1,830	361	5.07	DCS	null
126	DCS	Khādiragṛhyasūtrarudraskandavyākhyā	Rudraskanda	null	121	15	8.07	DCS	null
127	DCS	Āpastambagṛhyasūtra	null	Grhyasutra	3,029	405	7.48	DCS	null
128	DCS	Āśvalāyanagṛhyasūtra	null	Grhyasutra	989	125	7.91	DCS	null

End of preview. Expand in Data Studio

Sanskrit Travelogue: Unified Sanskrit Text Corpus

Unified, deduplicated, and morphologically annotated Sanskrit corpus, aggregating 13,010 texts (~182 million words, ~15.7 million segments) from 8 major digital Sanskrit libraries. All texts are normalized to IAST (International Alphabet of Sanskrit Transliteration).

Note:

Some annotations are still missing; In a follow up release, there will be added paragraph level metadata and translations for the GRETIL and SARIT corpus, plus the corpus in a format for LLM training;

Dataset Summary

Metric	Value
Unique texts (deduplicated)	13,010
Total segments	15,731,645
Total words	181,745,216
Morphologically annotated segments	11,705,716 (74.4%)
Source collections	8

Source Collections

Collection	Texts	Words	Description
GRETIL	1,544	76.6M	Gottingen Register of Electronic Texts in Indian Languages (TEI XML)
GRETIL_htm	408	29.3M	GRETIL HTML-only texts not in TEI format
Dharmanexus	1,129	29.7M	Buddhist and Hindu philosophical texts (Dharmamitra project)
SARIT	151	18.4M	Search and Retrieval of Indic Texts (TEI-encoded)
Muktabodha	495	11.7M	Muktabodha Indological Research Institute digital library
SanskritDocuments	8,573	8.8M	Community-contributed texts from sanskritdocuments.org
DCS	203	3.3M	Digital Corpus of Sanskrit (Oliver Hellwig)
DSBC	471	2.6M	Digital Sanskrit Buddhist Canon (University of the West)
YogaVaisaradi	36	1.3M	Yoga-related classical texts

Dataset Structure

Configurations

The dataset is available in multiple configurations:

default -- Merged deduplicated corpus (metadata + segments + annotated segments)
DCS, DSBC, Dharmanexus, GRETIL, GRETIL_htm, Muktabodha, SARIT, SanskritDocuments, YogaVaisaradi -- Per-collection subsets with full segments (pre-dedup, overlaps may exist)

Files

sanskrit_travelogue_metadata.parquet     # 13,010 texts with enriched metadata
sanskrit_travelogue.parquet              # 15,731,645 deduplicated segments (raw text)
sanskrit_travelogue_annotated.parquet    # 15,731,645 segments with morphological annotations
per_collection/
  {collection}_metadata.parquet          # Per-collection enriched metadata
  {collection}_segments.parquet          # Per-collection full segments

Metadata Schema (`sanskrit_travelogue_metadata.parquet`)

Column	Type	Description
`text_id`	int	Unique identifier
`collection`	string	Source collection name
`title`	string	Text title (IAST)
`author`	string	Author name (15.1% coverage)
`category`	string	Genre/category (33.9% coverage)
`word_count`	int	Number of words in this text
`segment_count`	int	Number of segments
`avg_segment_length`	float	Average words per segment
`source`	string	Source filename
`notes`	string	Editorial notes

Segments Schema (`sanskrit_travelogue.parquet`)

Column	Type	Description
`segment_id`	string	Unique ID (`{text_id}_{segment_number}`)
`text_id`	int	Foreign key to metadata
`segment_number`	string	Position within text
`text`	string	Sanskrit text in IAST transliteration
`type`	string	Segment type (verse, prose, note, etc.)
`chapter`	string	Chapter identifier
`section`	string	Section identifier
`verse_number`	string	Verse number
`page_number`	string	Page reference

Annotated Segments Schema (`segments_annotated.parquet`)

Contains all columns from segments.parquet plus:

Column	Type	Description
`original_segment`	string	Original text before cleaning
`cleaned_segment`	string	Text after cleaning/normalization
`analysed_segment`	string	Morphologically analyzed (ByT5), IAST
`analysed_segment_no_diacritics`	string	Analyzed text without diacritics
`stemmed_segment`	string	Lemmatized/stemmed text
`morphology`	string	Morphosyntactic tags

74.4% of segments have morphological annotations. Unannotated segments (recently recovered data, GRETIL_htm) have original_segment populated from the raw text, with other annotation columns empty.

Annotation Pipeline

Morphological annotations were produced by a ByT5-based neural pipeline (chronbmm/sanskrit5-multitask) that performs:

Segmentation: Splitting sandhi-joined compounds
Lemmatization: Extracting word stems
Morphosyntactic analysis: Case, number, gender, tense, etc.

The pipeline processes texts in chunks with GPU acceleration (optimized for A100) and supports checkpointing for fault tolerance.

Deduplication

The corpus was deduplicated in two passes:

Title-based: Fuzzy title matching across collections (90% similarity threshold) with collection priority ordering
LLM-verified: Content-level comparison of sampled segments using Claude Haiku agents for cases where titles differ but content overlaps

Combined deduplication removed 1,474 duplicate texts from the original 14,484.

Category Taxonomy

Categories are derived from multiple sources:

Dharmanexus: 77 hierarchical codes covering Buddhist (K/T prefix) and Hindu (G prefix) traditions
DCS: Subject classifications (Yoga, Ayurveda, Tantra, etc.)
SanskritDocuments: Genre inferred from folder structure (Stotra, Purana, Upanishad, etc.)
Muktabodha: Default "Tantra (Shaiva)" based on collection focus

Transliteration

All texts are normalized to IAST (International Alphabet of Sanskrit Transliteration). Source texts in Devanagari, Harvard-Kyoto, SLP1, ITRANS, or other schemes were automatically detected and converted. The corpus uses Unicode NFC normalization.

Usage

from datasets import load_dataset

# Load the full merged corpus
ds = load_dataset("SanskritVoyager/SanskritTravelogue")

# Load metadata
metadata = ds["metadata"]

# Load segments
segments = ds["segments"]

# Load annotated segments
annotated = ds["segments_annotated"]

# Filter by collection
gretil = metadata.filter(lambda x: x["collection"] == "GRETIL")

# Filter by category
yoga_texts = metadata.filter(lambda x: x["category"] == "Yoga")

Limitations

Author metadata: Only 15.1% of texts have author information. The largest collection (SanskritDocuments, 66% of texts) has no author data.
Category coverage: 33.9% of texts have genre/category. Coverage varies by collection.
Annotation gaps: 25.6% of segments lack morphological annotations, primarily from recently recovered Dharmanexus segments and GRETIL HTML texts.
Dating: Composition dates are available for only ~200 texts (from DCS). Not included in the published metadata due to low coverage.
OCR quality: Some texts (particularly Muktabodha) were digitized via OCR and may contain errors.

Citation

If you use this dataset, please cite:

@inproceedings{luca-etal-2026-sanskrit,
  title = {Sanskrit Travelogue: A Large-Scale Unified and Annotated Corpus of Sanskrit Texts},
  author = {Luca, Giacomo De and Croce, Danilo and Basili, Roberto},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)},
  month = {May},
  year = {2026},
  pages = {6722--6730},
  address = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy and Krek, Simon and Toral, Antonio},
  doi = {10.63317/5ibfaw5j9br6},
  abstract = {We present Sanskrit Travelogue, to our knowledge the largest open, unified and richly annotated Sanskrit corpus. Aggregating eight digital libraries, it comprises 12,394 texts, 73.1M tokens and 9M segments after de-duplication. A reproducible pipeline standardizes transliteration to IAST, reconciles heterogeneous metadata, preserves structural semantics (verse markers, chapter hierarchies, textual apparatus) and adds automatic annotations. We provide corpus-scale morphosyntactic annotation combining two systems: the BYT-5 Sanskrit model for compound and sandhi splitting, and the process-sanskrit library for inflection removal and morphological tagging through a hybrid deterministic-statistical cascade. For each segment we materialize synchronized representations: cleaned, analyzed (sandhi/compound split), stemmed, diacritic-normalized and morphologically tagged. These representations are indexed jointly for retrieval. Both approaches achieve high accuracy (84.61% sentence-level exact matches for BYT-5 segmentation, 92.37% correct root extraction for compounds, 95.94% on the Yoga Sūtra). Manual evaluation on the Yoga Sūtra showed 98% correct root extraction when combining both methods, outperforming individual approaches. These annotations enable searching across orthographic sandhi and within compounds, robust lemma-level retrieval despite rich inflectional variation, and provide training material for segmentation and lemmatization while maintaining ambiguity for downstream modeling. We release the annotated corpus as TSV shards, code for corpus acquisition, processing and annotation, a query normalizer, all under a Creative Commons non-commercial license.}
}

Acknowledgments

This corpus aggregates texts from multiple digital Sanskrit libraries. All credit for the original texts goes to:

GRETIL (Gottingen Register of Electronic Texts in Indian Languages) -- University of Gottingen
Dharmanexus -- Dharmamitra project
SARIT (Search and Retrieval of Indic Texts) -- Collaborative project
Muktabodha Indological Research Institute -- Digital library of Sanskrit texts
Digital Sanskrit Buddhist Canon (DSBC) -- University of the West
Digital Corpus of Sanskrit (DCS) -- Oliver Hellwig
sanskritdocuments.org -- Community-contributed texts
Yoga Vaisharadi -- Foundation dedicated to Krishnamacarya

License

CC-BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0 International)

Downloads last month: 239