@inproceedings{de-luca-etal-2026-sanskrit,
title = "{S}anskrit Travelogue: A Large-Scale Unified and Annotated Corpus of {S}anskrit Texts",
author = "De Luca, Giacomo and
Croce, Danilo and
Basili, Roberto",
editor = "Piperidis, Stelios and
Bel, N{\'u}ria and
van den Heuvel, Henk and
Ide, Nancy and
Krek, Simon and
Toral, Antonio",
booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference",
month = may,
year = "2026",
address = "Palma de Mallorca, Spain",
publisher = "ELRA Language Resource Association",
url = "https://aclanthology.org/2026.lrec-1.535/",
doi = "10.63317/5ibfaw5j9br6",
pages = "6722--6730",
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 S{\={u}}tra). Manual evaluation on the Yoga S{\={u}}tra 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."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="de-luca-etal-2026-sanskrit">
<titleInfo>
<title>Sanskrit Travelogue: A Large-Scale Unified and Annotated Corpus of Sanskrit Texts</title>
</titleInfo>
<name type="personal">
<namePart type="given">Giacomo</namePart>
<namePart type="family">De Luca</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Danilo</namePart>
<namePart type="family">Croce</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roberto</namePart>
<namePart type="family">Basili</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Fifteenth Language Resources and Evaluation Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Stelios</namePart>
<namePart type="family">Piperidis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Núria</namePart>
<namePart type="family">Bel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Henk</namePart>
<namePart type="family">van den Heuvel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nancy</namePart>
<namePart type="family">Ide</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Simon</namePart>
<namePart type="family">Krek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Antonio</namePart>
<namePart type="family">Toral</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resource Association</publisher>
<place>
<placeTerm type="text">Palma de Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<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.</abstract>
<identifier type="citekey">de-luca-etal-2026-sanskrit</identifier>
<identifier type="doi">10.63317/5ibfaw5j9br6</identifier>
<location>
<url>https://aclanthology.org/2026.lrec-1.535/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>6722</start>
<end>6730</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Sanskrit Travelogue: A Large-Scale Unified and Annotated Corpus of Sanskrit Texts
%A De Luca, Giacomo
%A Croce, Danilo
%A Basili, Roberto
%Y Piperidis, Stelios
%Y Bel, Núria
%Y van den Heuvel, Henk
%Y Ide, Nancy
%Y Krek, Simon
%Y Toral, Antonio
%S Proceedings of the Fifteenth Language Resources and Evaluation Conference
%D 2026
%8 May
%I ELRA Language Resource Association
%C Palma de Mallorca, Spain
%F de-luca-etal-2026-sanskrit
%X 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.
%R 10.63317/5ibfaw5j9br6
%U https://aclanthology.org/2026.lrec-1.535/
%U https://doi.org/10.63317/5ibfaw5j9br6
%P 6722-6730
Markdown (Informal)
[Sanskrit Travelogue: A Large-Scale Unified and Annotated Corpus of Sanskrit Texts](https://aclanthology.org/2026.lrec-1.535/) (De Luca et al., LREC 2026)
ACL