CarakaBinary: A binary-labelled dataset from early Āyurvedic texts for the identification of compound words in classical Sanskrit Authors: Shriganesh Devaru Bhat, Premjith B․ Published: 2026 ABSTRACT Charaka-Binary, the presented domain-specific Ayurvedic textual dataset in this article, is intended to support natural language processing (NLP) tasks in the Sanskrit language, particularly in analyzing compound words. Charaka-Binary includes 31,179 instances comprising 23,224 non-compound words and 7,955 compound words in Devanāgarī script along with the relevant numerical labels, including “0” indicating non-compound words and “1” denoting compound words to support future computational tasks. The dataset provides a resource for extracting compound words from Sanskrit text. We utilized the publicly accessible ancient Ayurvedic literature, Carakasaṃhitā. Among the 7 chapters of Carakasaṃhitā, we employed the first chapter along with its 30 lessons and fourth chapters’ two lessons as the foundation for dataset collection. The first phase of the dataset collection was to convert 228 pages of Carakasaṃhitā into individual .png files suitable for text extraction, while the second phase included the use of the Optical Character Recognition (OCR) technique to extract the text, followed by proofreading. Then the sentence segmentation step was carried out by word-level labeling into compound and non-compound word categories. The preprocessing steps, including proofreading, segmentation, and labeling, are carried out with the help of domain experts. Caraka-Binary can be used in both computational tasks, which include training models to identify Sanskrit compound words, as well as non-computational purposes, which encompass classroom teaching on Sanskrit compound words, comprehending their types, and analyzing compound patterns. Specifications TableSubjectComputer SciencesSpecific subject areaSanskrit compound word analysis, Sanskrit Computational Linguistics and Natural Language Processing in SanskritType of dataRaw Text (Segmented phrases into words and labelled); available in CSV file formatData collectionCarakaBinary was sourced from the Carakasaṃhitā book, which was in .pdf format. To convert the PDF into editable text, we utilized pdftoppm from the Poppler library to transform the targeted PDF's 228 pages into high-quality .png files to use the Optical Character Recognition (OCR) process. The SanskritOCR platform was employed to extract the text from the .png files, which was then considered for proofreading, followed by the manual segmentation, labeling, and validation process under expert supervision. The segmentation was carried out as per Pāṇinian grammatical notions.Data source locationAmrita School of Artificial Intelligence, Amrita Vishwa Vidyapeetham, Coimbatore, Tamil Nadu, India, 641112Data accessibilityRepository name: CarakaBinaryData identification number: https://doi.org/10.5281/zenodo.20431856Data Availability Statement:The dataset included with this work deposited in Zenodo and is currently under embargo to align with the publication of the manuscript. The dataset will be made publicly available upon acceptance of the article.For peer review purposes only: The dataset can be accessed using the following private direct link:https://zenodo.org/records/20431856?token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjE5NTMzYTAwLWMyMTYtNDE1Yy1hMTNmLTI2NjY4ZThiNmRlOSIsImRhdGEiOnt9LCJyYW5kb20iOiIzYjYxMDlkMjFkNWUzODg5NDMyY2U3YTRlYzhkYmE0OSJ9.JClje3hjdOlBc90XUwvtkIXIdw1W4s4zcaoWb6yUsItliZKtaQrHcnKJYd9znxvMPu-TRp5Fu1IF29jJsLf9JwFor other researchers, access to the dataset during the embargo period can be requested from the authors upon reasonable request.Related research articleN/A SubjectComputer SciencesSpecific subject areaSanskrit compound word analysis, Sanskrit Computational Linguistics and Natural Language Processing in SanskritType of dataRaw Text (Segmented phrases into words and labelled); available in CSV file formatData collectionCarakaBinary was sourced from the Carakasaṃhitā book, which was in .pdf format. To convert the PDF into editable text, we utilized pdftoppm from the Poppler library to transform the targeted PDF's 228 pages into high-quality .png files to use the Optical Character Recognition (OCR) process. The SanskritOCR platform was employed to extract the text from the .png files, which was then considered for proofreading, followed by the manual segmentation, labeling, and validation process under expert supervision. The segmentation was carried out as per Pāṇinian grammatical notions.Data source locationAmrita School of Artificial Intelligence, Amrita Vishwa Vidyapeetham, Coimbatore, Tamil Nadu, India, 641112Data accessibilityRepository name: CarakaBinaryData identification number: https://doi.org/10.5281/zenodo.20431856Data Availability Statement:The dataset included with this work deposited in Zenodo and is currently under embargo to align with the publication of the manuscript. The dataset will be made publicly available upon acceptance of the article.For peer review purposes only: The dataset can be accessed using the following private direct link:https://zenodo.org/records/20431856?token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjE5NTMzYTAwLWMyMTYtNDE1Yy1hMTNmLTI2NjY4ZThiNmRlOSIsImRhdGEiOnt9LCJyYW5kb20iOiIzYjYxMDlkMjFkNWUzODg5NDMyY2U3YTRlYzhkYmE0OSJ9.JClje3hjdOlBc90XUwvtkIXIdw1W4s4zcaoWb6yUsItliZKtaQrHcnKJYd9znxvMPu-TRp5Fu1IF29jJsLf9JwFor other researchers, access to the dataset during the embargo period can be requested from the authors upon reasonable request.Related research articleN/A Value of the Data •CarakaBinary includes annotated Ayurvedic Sanskrit terms classified into compound and non-compound classes, thereby providing a systematised linguistic resource for Sanskrit NLP and Ayurvedic computational research.•It can be re-employed for the binary text classification purposes including compound word identification by training machine learning and deep learning models. In addition to this, dataset’s class weight imbalance nature provides the realistic conditions in evaluating models’ robustness.•It can support teaching linguistic and grammatical notions pertaining to Sanskrit compound words and may help second language learners and Ayurvedic practitioners.•Alongside supporting digitized resource and digital humanities, the dataset supports domain specific computational tasks encompassing fine-tuning, modelling Sanskrit language oriented tasks, and linguistic analysis that is pertaining to Ayurvedic domain.•This dataset may serve as a foundational resource for future benchmark-oriented research in Sanskrit compound analysis, including compound splitting, parsing, and compound-type identification.•The dataset encompasses domain specific contextual Ayurvedic Sanskrit terms and linguistic characteristics, thereby may help in interdisciplinary research in digital humanities. •CarakaBinary includes annotated Ayurvedic Sanskrit terms classified into compound and non-compound classes, thereby providing a systematised linguistic resource for Sanskrit NLP and Ayurvedic computational research. CarakaBinary includes annotated Ayurvedic Sanskrit terms classified into compound and non-compound classes, thereby providing a systematised linguistic resource for Sanskrit NLP and Ayurvedic computational research. •It can be re-employed for the binary text classification purposes including compound word identification by training machine learning and deep learning models. In addition to this, dataset’s class weight imbalance nature provides the realistic conditions in evaluating models’ robustness. It can be re-employed for the binary text classification purposes including compound word identification by training machine learning and deep learning models. In addition to this, dataset’s class weight imbalance nature provides the realistic conditions in evaluating models’ robustness. •It can support teaching linguistic and grammatical notions pertaining to Sanskrit compound words and may help second language learners and Ayurvedic practitioners. It can support teaching linguistic and grammatical notions pertaining to Sanskrit compound words and may help second language learners and Ayurvedic practitioners. •Alongside supporting digitized resource and digital humanities, the dataset supports domain specific computational tasks encompassing fine-tuning, modelling Sanskrit language oriented tasks, and linguistic analysis that is pertaining to Ayurvedic domain. Alongside supporting digitized resource and digital humanities, the dataset supports domain specific computational tasks encompassing fine-tuning, modelling Sanskrit language oriented tasks, and linguistic analysis that is pertaining to Ayurvedic domain. •This dataset may serve as a foundational resource for future benchmark-oriented research in Sanskrit compound analysis, including compound splitting, parsing, and compound-type identification. This dataset may serve as a foundational resource for future benchmark-oriented research in Sanskrit compound analysis, including compound splitting, parsing, and compound-type identification. •The dataset encompasses domain specific contextual Ayurvedic Sanskrit terms and linguistic characteristics, thereby may help in interdisciplinary research in digital humanities. The dataset encompasses domain specific contextual Ayurvedic Sanskrit terms and linguistic characteristics, thereby may help in interdisciplinary research in digital humanities. Background Ayurveda, a branch of Indian knowledge systems (IKS), includes texts pertaining to health and disease management that is documented in the Sanskrit language. In the 2019 World Health Organization report, it was noted that 93 member states reported the use of Ayurveda in their countries, based on 2012 statistics [1]. Many individuals involved in the preservation and dissemination of Ayurveda, either may be second language (L2) learners or unfamiliar with traditional approaches to learning Sanskrit grammar [2] and linguistic theory. In Ayurveda, the identification of compound words and a thorough understanding of their meanings are may help an accurate interpretation. Prior research in compound word analysis employed generic data instead of focusing on the Ayurvedic domain. Therefore, CarakaBinary provides the resource to differentiate compound and non-compound words in the Sanskrit language, specifically within the Ayurvedic domain, to support Sanskrit NLP tasks and to assist L2 learners and beginners. Data Description The repository titled CarakaBinary encompasses a folder that includes the following files:•CarakaBinary-DIB-v3.csv- This is the primary dataset file that has 31,180 rows where each row relates to Ayurvedic Sanskrit term with categorical and numerical binary class labels.•pre-proofread-txt.zip – This includes the sample of pre-proofread texts that were extracted through OCR engine.•sample-png-images.zip - This includes the sample .png images that were employed to produce editable texts.•ReadMe.txt - This text file elucidates the detail of the dataset, meaning of the columns, connotations of each label, fundamental directions for its use and other details closely related to the files that are uploaded in the data repository. •CarakaBinary-DIB-v3.csv- This is the primary dataset file that has 31,180 rows where each row relates to Ayurvedic Sanskrit term with categorical and numerical binary class labels. CarakaBinary-DIB-v3.csv- This is the primary dataset file that has 31,180 rows where each row relates to Ayurvedic Sanskrit term with categorical and numerical binary class labels. •pre-proofread-txt.zip – This includes the sample of pre-proofread texts that were extracted through OCR engine. pre-proofread-txt.zip – This includes the sample of pre-proofread texts that were extracted through OCR engine. •sample-png-images.zip - This includes the sample .png images that were employed to produce editable texts. sample-png-images.zip - This includes the sample .png images that were employed to produce editable texts. •ReadMe.txt - This text file elucidates the detail of the dataset, meaning of the columns, connotations of each label, fundamental directions for its use and other details closely related to the files that are uploaded in the data repository. ReadMe.txt - This text file elucidates the detail of the dataset, meaning of the columns, connotations of each label, fundamental directions for its use and other details closely related to the files that are uploaded in the data repository. CarakaBinary does not contain any additional raw or processed subfolders and it only contains clean tokens. To clarify, CarakaBinary does not include special characters. All samples are encoded using UTF-8 to ensure proper rendering of Devanāgarī characters across platforms. Table 1 outlines the overview of CarakaBinary.Table 1Overview of CarakaBinary.Table 1 dummy alt textItemValue / ExplanationNumber rows31,180Number of columns4Name of Column 1Words (Sanskrit terms from Ayurvedic text)Name of Column 2English TranslationName of Column 3Labels (Categorical labels)Name of Column 4Numerical Labels (0/1 encoding of categorical labels)Number of compound words7,955Number of non-compound words23,224 Table 1Overview of CarakaBinary.Table 1 dummy alt textItemValue / ExplanationNumber rows31,180Number of columns4Name of Column 1Words (Sanskrit terms from Ayurvedic text)Name of Column 2English TranslationName of Column 3Labels (Categorical labels)Name of Column 4Numerical Labels (0/1 encoding of categorical labels)Number of compound words7,955Number of non-compound words23,224 Overview of CarakaBinary. To analyze the distribution of compound and non-compound lexical terms, we randomly selected and included two lessons from the chapter Śārīra Sthāna, comprising 2,628 instances, which consist of 2,063 non-compound words and 565 compound words. Nevertheless, the relative distribution noted in the Sutrasthāna remained predominantly constant, even after the inclusion of two random lessons, indicating that the found class weight imbalance is an intrinsic linguistic feature of Carakasamhitā. Table 2 outlines the statistic of both versions of dataset.Table 2Statistics of original and expanded datasets.Table 2 dummy alt textDataset VersionTotalNon-compoundCompoundRatioOriginal28,55121,1617,3902.86:1Expanded31,17923,2247,9552.92:1 Table 2Statistics of original and expanded datasets.Table 2 dummy alt textDataset VersionTotalNon-compoundCompoundRatioOriginal28,55121,1617,3902.86:1Expanded31,17923,2247,9552.92:1 Statistics of original and expanded datasets. Experimental Design, Materials and Methods The construction of the dataset included the following steps: 1. Source of the dataset The dataset is sourced from Carakasaṃhitā [3], a Sanskrit Ayurvedic text that includes domain-specific terminologies and compound words suitable for a binary text classification task, in which we focused only on the 30 lessons of the initial chapter and 2 lessons of the fourth chapter. 2. Converting pdf into editable text Due to the targeted material was in PDF format, we utilized the SanskritOCR [4] tool a web-based SanskritOCR engine1 to convert the PDF into editable text. A total of 228 pages were evaluated for the conversion process. During conversion, we also utilized “pdftoppm” from the “Poppler” library (version 25.11.0) [5] to transform the target PDF into high-quality .png files to facilitate the Optical Character Recognition (OCR) process, given that the SanskritOCR platform exclusively accepts images as input. The “pdftoppm” provided the conversion of pages into distinct .png files, ensuring perfect layout and script clarity. We executed the command “pdftoppm -png -r 300 Caraka.pdf caraka.png.” We converted the image format to .png to prevent loss during compression and set the resolution to 300 DPI to ensure premium quality for the OCR process. The “pdftoppm” generated pages as separate picture in .png format, utilizing the default grayscale color mode and an upright orientation. The output file names were uniform (caraka-1.png, caraka-2.png) to facilitate the input photos for the OCR process workflow. Fig. 1 illustrates the sample .png file extracted from “pdftoppm.”Fig. 1An example output produced by pdftoppm.Fig 1 dummy alt text An example output produced by pdftoppm. After turning pages of the target PDF into separate .png files, we commenced the transformation of each .png file into editable text using the OCR process. The resulting editable text contained extraneous content in addition to the primary text. The upper margin of the left page included the page number, book title, and abbreviation of 'adhyāya,' (lesson) while the right page displayed the chapter number, book title, and page number in a left-to-right orientation, as shown in Fig. 2. Footnote numbers in the main text indicate multiple variant readings in the primary text, accompanied by corresponding footnotes at the bottom margin.Fig. 2An example output generated by SanskritOCR when a .png file was provided as input.Fig 2 dummy alt text An example output generated by SanskritOCR when a .png file was provided as input. We deemed the elements in the top and bottom margins superfluous, as our attention was only directed towards the main text. Consequently, we commenced the removal of superfluous elements from the editable text and accurately matched the characteristics of each phrase sequentially to ensure their order corresponds with that of the .png file as shown in Fig. 3. During this phase, we permitted unnecessary elements, including full stops and verse numbers, to coexist with the primary text, which were subsequently removed in later phases. During this phase, the punctuation at the end of each verse prevented a merger of content with other verses, while the verse numbers facilitated the accurate representation of all verses in the editable text within the PDF or PNG file.Fig. 3An arrangement of phrases derived from SanskritOCR in accordance with the original order in the editable text.Fig 3 dummy alt text An arrangement of phrases derived from SanskritOCR in accordance with the original order in the editable text. The Sanskrit OCR engine produced very accurate editable text that closely matched the original source material. The output produced by the OCR engine was manually examined in the proofreading phase, where only the minor character-level errors were identified, primarily involving Anusvāra (a dot above a Devanāgarī character) and Visarga (two dots, one below the other) representations in the Devanāgarī script, yielding a CER of 3.58%. Table 3 illustrates the translation of the Sanskrit verse mentioned in the Figs. 1, 2, and 3.Table 3Translation of the Sanskrit verses mentioned in the Fig. 1, Fig. 2, Fig. 3.Table 3 dummy alt textFigNoVerse NoSanskrit verseTranslation1,2,312-15kāṅkāyanaḥ kaikaśeyo dhaumyo mārīcakāśyapau1| śarkarākṣo hiraṇyākṣo lokākṣaḥ2 paiṅgireva ca || 12 ||śaunakaḥ śākuneyaśca maitreyo maimatāyaniḥ | vaikhānasā bālakhilyāstathā cānye maharṣayaḥ || 13 ||brahmajñānasya nidhayo damasya niyamasya5 ca |tapasastejasā dīptā hūyamānā ivāgnayaḥ ||14||sukhopaviṣṭāste tatra puṇyāṃ cakruḥ kathāmimām | dharmārthakāmamokṣāṇāmārogyaṃ mūlamuttamam || 15 ||The revered scholars, including Kāṅkāyana, Kaikaśeya, Dhaumya, Mārīca, Kāśyapa, Śarkarākṣa, Hiraṇyākṣa, Lokākṣa, Paiṅgi, Śaunaka, Śākuneya, Maitreya, Maimatāyani, Vaikhānasā, Bālakhilyas, and other sages, are comfortably seated, involved in a deeply philosophical debate on the subject that “health is the foundational factor in performing virtuous actions, acquiring wealth, satisfying desire, and attaining liberation.” These sages were regarded as the real source of supreme knowledge (Brahma-jñāna), self-control, and discipline. Furthermore, they were radiant with the brilliance of austerity, like blazing sacrificial fires being kindled.12 - 15.1,2,316rogāstasyāpahartāraḥ śreyaso jīvitasya ca | prādurbhūto manuṣyāṇāmantarāyo mahānayam || 16 ||Diseases are considered a great obstacle for human beings due to the fact that they destroy health, well-being, and life.161,2,317kaḥ syātteṣāṃ śamopāya ityuktvā dhyānamāsthitāḥ | atha te śaraṇaṃ śakraṃ dadṛśurthyānacakṣuṣā || 17||Thus, after engaging in noble discussions about health and debating potential permanent solutions, the sages began to practice meditation. Through the meditation, they found the lord Indra can be their shelter.171,2,318sa vakṣyati śamopāyaṃ yathāvadamaraprabhuḥ | kaḥ sahasrākṣabhavanaṃ gacchet praṣṭuṃ śacīpatim || 18 ||In addition to that, they perceived that Indra, the Lord of all Gods, will elucidate the ways of health improvement. But who will approach Lord Indra, who is thousand-eyed (Indra), the husband of Śacī, to inquire about these matters?181,2,319ahamarthe niyujye'yamatreti prathamaṃ vacaḥ | bharadvājo'bravīttasmādṛṣibhiḥ sa niyojitaḥ || 19||"I will go to Lord Indra to inquire about health-related subjects," Bharadwaja said, which was the very first response given among the crowd of illustrious sages. Thus, all sages nominated him for the task .191,2,320sa śakrabhavanaṃ gatvā surarṣigaṇamadhyagam | dadarśa balahantāraṃ6 dīpyamānamivānalam || 20 ||Sage Bharadwāja approached Lord Indra and saw him seated in the midst of the gods and other sages, shining like a blazing fire. 201,2,321so'bhigamya jayāśīrbhirabhinandya sureśvaram | provāca vinayāddhīmānṛṣīṇāṃ vākyamuttamam || 21||Sage Bharadwāja approached and offered the blessing of victory and revered greetings, followed by communicating the message from the sages.211,2,322vyādhayo hi samutpannāḥ sarvaprāṇibhayaṅkarāḥ | tad brūhi me śamopāyaṃ yathāvadamaraprabho || 22||Hey Lord Indra, there are many diseases that have arisen, threatening all living beings; therefore, kindly suggest an appropriate way to control them.221,2,323tasmai provāca bhagavānāyurvedaṃ śatakratuḥ | padairalpairmatiṃ buddhvā vipulāṃ paramarṣaye || 23 ||Having received the messages from Bharadwāja, the Lord Indra taught Ayurveda using concise words, recognizing the esteemed sage’s vast intelligence.231,2,324hetuliṅgauṣadhajñānaṃ svasthāturaparāyaṇam | trisūtraṃ śāśvataṃ puṇyaṃ bubudhe yaṃ pitāmahaḥ || 24||Brahmā, the grandfather of all science, comprehended the holy and existing forever three-tiered science, including knowledge of causes, symptoms, and treatments, which are related to the well-being of the healthy and the unhealthy. 241,225so'nantapāraṃ triskandhamāyurvedaṃ mahāmatiḥ | yathāvadacirāt sarvaṃ bubudhe tanmanā muniḥ || 25 ||The sage Bharadwāja, having a one-pointed mind, quickly understood the three tiers of science through Ayurveda. 251,226tenāyuramitaṃ lebhe bharadvājaḥ sukhānvitam1 | ṛṣibhyo'nadhikaṃ tacca śaśaṃsānavaśeṣayan || 26 ||Through the knowledge of Ayurveda, the sage Bharadwāja lived a long and immeasurable life, including well-being. Then, he passed that complete knowledge without holding anything back with him. 261,227ṛṣayaśca bharadvājājjagṛhustaṃ prajāhitam | dīrghamāyuścikīrṣanto vedaṃ vardhanamāyuṣaḥ || 27 ||The sage Bharadwāja imparted the entire knowledge of Ayurveda, dedicated to the wellness of all beings and long life, to other sages. 271,228maharṣayaste2 dadṛśuryathāvajjñānacakṣuṣā | sāmānyaṃ ca viśeṣaṃ ca guṇān dravyāṇi karma ca || 28 ||With this knowledge, all the great sages accurately perceived universals, particulars, qualities, substances, and actions through their eye of knowledge. 281,229samavāyaṃ ca tajjñātvā tantroktaṃ vidhimāsthitāḥ | lebhire paramaṃ śarma jīvitaṃ cāpyanitvaram3 || 29 ||In addition to this, the sages accurately perceived inherentness and followed ways explained in the Ayurveda texts in order to attain supreme well-being and long life. 291,230-31atha maitrīparaḥ puṇyamāyurvedaṃ punarvasuḥ | śiṣyebhyo dattavān ṣaḍbhyaḥ sarva bhūtānukampayā || 30||agniveśaśca bhela (ḍa)śca jatūkarṇaḥ parāśaraḥ | hārītaḥ kṣārapāṇiśca jagṛhustanmunervacaḥ || 31 ||Then, the sage Punarvasu taught Ayurveda to his six pupils, including Agniveśa, Bhela, Jatūkarṇa, Parāśara, Hārīta, and Kṣārapāṇi, full of compassion and friendliness. 30 – 311,232buddherviśeṣastatrāsīnnopadeśāntaraṃ muneḥ | tantrasya kartā prathamamagniveśo yato'bhavat || 32 ||Though there were no differences in the teaching of Punarvasu, the pupil Agniveśa became the first composer of the treatises called Tantra due to his noble intelligence. 321,233atha bhelā (ḍā)dayaścakruḥ svaṃ svaṃ tantraṃ kṛtāni ca | śrāvayāmāsurātreyaṃ sarṣisaṅghaṃ sumedhasaḥ || 33||Later, other pupils of Punarvasu, including Bhela and others, composed their own treatise and presented it in front of all sages and Punarvasu.331,234śrutvā sūtraṇamarthānāmṛṣayaḥ puṇyakarmaṇām | yathāvatsūtritamiti prahṛṣṭāste'numenire || 34||Having seen the presentation, all the sages approved by saying it has been properly composed with systematic arrangement. 341,235sarva evāstuvaṃstāṃśca sarvabhūtahitaiṣiṇaḥ | sādhu4 bhūteṣvanukrośa ityuccairabruvan samam || 35||Then all the sages appreciated the authors who composed the treatises dedicated to everyone's well-being and said, ``Excellent! Compassion towards all beings is truly noble.'' 351,236taṃ puṇyaṃ śuśruvuḥ śabdaṃ divi devarṣayaḥ sthitāḥ | sāmarāḥ paramarṣīṇāṃ śrutvā mumudire param || 36 ||Divine sages and gods in heaven heard that auspicious sound; upon hearing the words of the great sages, they rejoiced immensely. 36 Table 3Translation of the Sanskrit verses mentioned in the Fig. 1, Fig. 2, Fig. 3.Table 3 dummy alt textFigNoVerse NoSanskrit verseTranslation1,2,312-15kāṅkāyanaḥ kaikaśeyo dhaumyo mārīcakāśyapau1| śarkarākṣo hiraṇyākṣo lokākṣaḥ2 paiṅgireva ca || 12 ||śaunakaḥ śākuneyaśca maitreyo maimatāyaniḥ | vaikhānasā bālakhilyāstathā cānye maharṣayaḥ || 13 ||brahmajñānasya nidhayo damasya niyamasya5 ca |tapasastejasā dīptā hūyamānā ivāgnayaḥ ||14||sukhopaviṣṭāste tatra puṇyāṃ cakruḥ kathāmimām | dharmārthakāmamokṣāṇāmārogyaṃ mūlamuttamam || 15 ||The revered scholars, including Kāṅkāyana, Kaikaśeya, Dhaumya, Mārīca, Kāśyapa, Śarkarākṣa, Hiraṇyākṣa, Lokākṣa, Paiṅgi, Śaunaka, Śākuneya, Maitreya, Maimatāyani, Vaikhānasā, Bālakhilyas, and other sages, are comfortably seated, involved in a deeply philosophical debate on the subject that “health is the foundational factor in performing virtuous actions, acquiring wealth, satisfying desire, and attaining liberation.” These sages were regarded as the real source of supreme knowledge (Brahma-jñāna), self-control, and discipline. Furthermore, they were radiant with the brilliance of austerity, like blazing sacrificial fires being kindled.12 - 15.1,2,316rogāstasyāpahartāraḥ śreyaso jīvitasya ca | prādurbhūto manuṣyāṇāmantarāyo mahānayam || 16 ||Diseases are considered a great obstacle for human beings due to the fact that they destroy health, well-being, and life.161,2,317kaḥ syātteṣāṃ śamopāya ityuktvā dhyānamāsthitāḥ | atha te śaraṇaṃ śakraṃ dadṛśurthyānacakṣuṣā || 17||Thus, after engaging in noble discussions about health and debating potential permanent solutions, the sages began to practice meditation. Through the meditation, they found the lord Indra can be their shelter.171,2,318sa vakṣyati śamopāyaṃ yathāvadamaraprabhuḥ | kaḥ sahasrākṣabhavanaṃ gacchet praṣṭuṃ śacīpatim || 18 ||In addition to that, they perceived that Indra, the Lord of all Gods, will elucidate the ways of health improvement. But who will approach Lord Indra, who is thousand-eyed (Indra), the husband of Śacī, to inquire about these matters?181,2,319ahamarthe niyujye'yamatreti prathamaṃ vacaḥ | bharadvājo'bravīttasmādṛṣibhiḥ sa niyojitaḥ || 19||"I will go to Lord Indra to inquire about health-related subjects," Bharadwaja said, which was the very first response given among the crowd of illustrious sages. Thus, all sages nominated him for the task .191,2,320sa śakrabhavanaṃ gatvā surarṣigaṇamadhyagam | dadarśa balahantāraṃ6 dīpyamānamivānalam || 20 ||Sage Bharadwāja approached Lord Indra and saw him seated in the midst of the gods and other sages, shining like a blazing fire. 201,2,321so'bhigamya jayāśīrbhirabhinandya sureśvaram | provāca vinayāddhīmānṛṣīṇāṃ vākyamuttamam || 21||Sage Bharadwāja approached and offered the blessing of victory and revered greetings, followed by communicating the message from the sages.211,2,322vyādhayo hi samutpannāḥ sarvaprāṇibhayaṅkarāḥ | tad brūhi me śamopāyaṃ yathāvadamaraprabho || 22||Hey Lord Indra, there are many diseases that have arisen, threatening all living beings; therefore, kindly suggest an appropriate way to control them.221,2,323tasmai provāca bhagavānāyurvedaṃ śatakratuḥ | padairalpairmatiṃ buddhvā vipulāṃ paramarṣaye || 23 ||Having received the messages from Bharadwāja, the Lord Indra taught Ayurveda using concise words, recognizing the esteemed sage’s vast intelligence.231,2,324hetuliṅgauṣadhajñānaṃ svasthāturaparāyaṇam | trisūtraṃ śāśvataṃ puṇyaṃ bubudhe yaṃ pitāmahaḥ || 24||Brahmā, the grandfather of all science, comprehended the holy and existing forever three-tiered science, including knowledge of causes, symptoms, and treatments, which are related to the well-being of the healthy and the unhealthy. 241,225so'nantapāraṃ triskandhamāyurvedaṃ mahāmatiḥ | yathāvadacirāt sarvaṃ bubudhe tanmanā muniḥ || 25 ||The sage Bharadwāja, having a one-pointed mind, quickly understood the three tiers of science through Ayurveda. 251,226tenāyuramitaṃ lebhe bharadvājaḥ sukhānvitam1 | ṛṣibhyo'nadhikaṃ tacca śaśaṃsānavaśeṣayan || 26 ||Through the knowledge of Ayurveda, the sage Bharadwāja lived a long and immeasurable life, including well-being. Then, he passed that complete knowledge without holding anything back with him. 261,227ṛṣayaśca bharadvājājjagṛhustaṃ prajāhitam | dīrghamāyuścikīrṣanto vedaṃ vardhanamāyuṣaḥ || 27 ||The sage Bharadwāja imparted the entire knowledge of Ayurveda, dedicated to the wellness of all beings and long life, to other sages. 271,228maharṣayaste2 dadṛśuryathāvajjñānacakṣuṣā | sāmānyaṃ ca viśeṣaṃ ca guṇān dravyāṇi karma ca || 28 ||With this knowledge, all the great sages accurately perceived universals, particulars, qualities, substances, and actions through their eye of knowledge. 281,229samavāyaṃ ca tajjñātvā tantroktaṃ vidhimāsthitāḥ | lebhire paramaṃ śarma jīvitaṃ cāpyanitvaram3 || 29 ||In addition to this, the sages accurately perceived inherentness and followed ways explained in the Ayurveda texts in order to attain supreme well-being and long life. 291,230-31atha maitrīparaḥ puṇyamāyurvedaṃ punarvasuḥ | śiṣyebhyo dattavān ṣaḍbhyaḥ sarva bhūtānukampayā || 30||agniveśaśca bhela (ḍa)śca jatūkarṇaḥ parāśaraḥ | hārītaḥ kṣārapāṇiśca jagṛhustanmunervacaḥ || 31 ||Then, the sage Punarvasu taught Ayurveda to his six pupils, including Agniveśa, Bhela, Jatūkarṇa, Parāśara, Hārīta, and Kṣārapāṇi, full of compassion and friendliness. 30 – 311,232buddherviśeṣastatrāsīnnopadeśāntaraṃ muneḥ | tantrasya kartā prathamamagniveśo yato'bhavat || 32 ||Though there were no differences in the teaching of Punarvasu, the pupil Agniveśa became the first composer of the treatises called Tantra due to his noble intelligence. 321,233atha bhelā (ḍā)dayaścakruḥ svaṃ svaṃ tantraṃ kṛtāni ca | śrāvayāmāsurātreyaṃ sarṣisaṅghaṃ sumedhasaḥ || 33||Later, other pupils of Punarvasu, including Bhela and others, composed their own treatise and presented it in front of all sages and Punarvasu.331,234śrutvā sūtraṇamarthānāmṛṣayaḥ puṇyakarmaṇām | yathāvatsūtritamiti prahṛṣṭāste'numenire || 34||Having seen the presentation, all the sages approved by saying it has been properly composed with systematic arrangement. 341,235sarva evāstuvaṃstāṃśca sarvabhūtahitaiṣiṇaḥ | sādhu4 bhūteṣvanukrośa ityuccairabruvan samam || 35||Then all the sages appreciated the authors who composed the treatises dedicated to everyone's well-being and said, ``Excellent! Compassion towards all beings is truly noble.'' 351,236taṃ puṇyaṃ śuśruvuḥ śabdaṃ divi devarṣayaḥ sthitāḥ | sāmarāḥ paramarṣīṇāṃ śrutvā mumudire param || 36 ||Divine sages and gods in heaven heard that auspicious sound; upon hearing the words of the great sages, they rejoiced immensely. 36 Translation of the Sanskrit verses mentioned in the Fig. 1, Fig. 2, Fig. 3. 3. Editing for accuracy After aligning the phrases according to the serial number, we commenced the proofreading phase, which involved reading each verse to verify the sequence of the phrases, analyzing textual problems resulting from the OCR process, and rectifying them accordingly. 4. Segmenting phrases into individual words For novices and second language learners, text segmentation helps in recognizing compound words. The segmentation of a text in Sanskrit involves three essential aspects. The preliminary stage of segmentation will utilize the whitespace-based segmentation procedure (WBSP). WBSP involves replacing each whitespace in the sentence with newlines. The second approach is segmentation according to phonetic principles, known as the euphonic-based segmentation procedure (EBSP). Euphonic rules are grammatical laws governing substitutions between two alphabets that display textual juxtaposition. Both approaches WBSP and EBSP enable segmentation as practical techniques, whereas the compound-based segmentation process (CBSP) functions as a cognitive principle in this work. It concerns the segmentation of nouns that are semantically unified within a compound word. CBSP should not be employed in the preparation of the dataset that facilitates the compound word and non-compound word categorization, as CBSP diminishes semantic information and disrupts the language relationship that underlies their semantic integration, and it leads to a single non-compound word category. Compound-based segmentation helps in excluding compound words during the segmentation process, thereby preserving semantic information. Consequently, we incorporated sub-phases in the proofreading stage, each dedicated to identifying euphonically altered combinations and semantic combinations independently. Euphonically modified combinations were digitally emphasized for inclusion in the segmentation process, while semantic combinations were indicated for exclusion. The succeeding subsection illustrates the digital highlighting of euphonically altered combinations in red and while semantic combinations in blue, along with the sample of original text and the outputs of WBSP. Table 3 illustrates the translation of the sample of original text, the outputs of WBSP, and EBSP. a. A sample of the original text: b. Output produced by WBSP kāṅkāyanaḥ, kaikaśeyo, dhaumyo, mārīcakāśyapau. śarkarākṣo, hiraṇyākṣo, lokākṣaḥ, paiṅgireva, ca||12||. śaunakaḥ, śākuneyaśca, maitreyo, maimatāyaniḥ. vaikhānasā, bālakhilyāstathā, cānye, maharṣayaḥ ||13||. brahmajñānasya, nidhayo, damasya, niyamasya, ca. tapasastejasā, dīptā, hūyamānā, ivāgnayaḥ ||14||. The output of the WBSP supports syntactic analysis; however, as the current challenge involves word-level analysis, some outputs from the WBSP are grammatically erroneous and include euphonically altered non-semantic combinations that require EBSP intervention. In the WBSP, 26 outputs were created in total. The output of EBSP is generated exclusively by considering the euphonically modified non-semantic combinations, which comprise a total of 32 outputs, as illustrated below. c. Output produced by EBSP kāṅkāyanaḥ, kaikaśeyo, dhaumyo, mārīcakāśyapau śarkarākṣo, hiraṇyākṣo, lokākṣaḥ, paiṅgir, eva, ca ||12|| śaunakaḥ, śākuneyaś, ca, maitreyo, maimatāyaniḥ vaikhānasā, bālakhilyās, tathā, ca, anye, maharṣayaḥ ||13|| brahmajñānasya, nidhayo, damasya, niyamasya, ca tapasas, tejasā, dīptāḥ, hūyamānāḥ, Iva, agnayaḥ ||14|| Subsequently, each comma is substituted with a newline, while full stops and verse numbers are removed. The outputs comprised compound and non-compound words that exhibit both grammatically valid and invalid word boundaries. 5. Establishing grammatically correct word boundaries In Sanskrit, each word simultaneously conveys core and grammatical meaning, with word boundaries aiding in the determination of grammatical meaning. In summary, the case markers illustrate the grammatical role of the core meaning of the word within the sentence. Retaining grammatically improper terms may distort the learning phase of novices and L2 learners in the concept of pedagogy, while in the concept of computational linguistics, it may mislead the model's training phase. Consequently, we conducted a distinct phase in which we concentrated on the assignment of grammatically valid word boundaries for each sample. 6. Manual annotation Throughout the manual labeling procedure, we examined each instance in the output from the preceding step to ensure grammatical and linguistic correctness and integrity. Each instance was labeled as either ``NC'' for non-compound words or ``C'' for compound words. The annotator for CarakaBinary studied Sanskrit grammar in a conventional context and holds advanced degrees. Before the manual annotation of the Sanskrit terms, we commenced an examination of the ideas of compound words and their semantic categories with specialized characters in accordance with Pāṇinian perspectives. This process enacted the head bridge for categorizing the Sanskrit Ayurvedic terms into predefined classes. In Pāṇini’s Aṣṭādhyāyī tradition, there are five coarse-grained semantic classes explained under two categories, which all are considered under the compound class during the annotation process. The compound class includes two major classes namely, Kevala (Uncommon) and Viśeṣa (Specail) where Viśeṣa encompasses avyayībhāva (adverbial), tatpuruṣa (determinative), bahuvrihi (exocentric), dvandva (coordinative), and karmadhāraya (oppositional determinative). Fig. 4 showcases the tree structure of the semantic classes according to the Sanskrit grammar perspective along with their modern terms.Fig. 4Tree structure of the Sanskrit Compound types.Fig 4 dummy alt text Tree structure of the Sanskrit Compound types. Through the linguistically and grammatically oriented analysis of Aṣṭādhyāyī, a seminal work of Pāṇini, we figured out the following pointers for each type of compound words to set as parameters in determining compound class during the manual annotation phase.1.Avyayībhāva (Adverbial) – A left-headed compound word.a.An avyayībhāva compound word is an indeclinable term.b.An avyayībhāva will consistently be in the neuter gender, regardless of the gender of its constituents.c.An avyayībhāva is always in singular form, regardless of whether its constituents are dual or plural.d.Avyayībhāva compounding is a binary process. To elucidate, only two nouns can be compounded simultaneously.e.An avyayībhāva consistently concludes with a short vowel.2.Tatpuruṣa (Determinative) A right-headed compound word.a.The formulation of the tatpuruṣa represents a binary process, indicating that, at any given moment, only two constituents are capable of undergoing semantic amalgamation.b.The gender of a compound word is determined by the gender of its head component. In a straightforward manner, it can be stated that a compound word assumes the gender of the head noun within the compound structure.c.The tatpuruṣa compound takes over the grammatical number inherent to the head it possesses.3.Bahuvrīhi (Exocentric) An adjectival form.a.In contrast to other classes, bahuvrīhi transforms its components into an adjectival form.b.Given that the constituents possess an adjectival form, the bahuvrīhi conveys a meaning that is markedly different from that of its constituents.c.Given that bahuvrīhi becomes an adjective, through the core entity in the sentence, it connects with the main verb of the sentence.4.Dvandva (Coordinative) – a multi-headed compound word.a.Dvandva structures can only be formed when the constituents meet the criteria for inclusion in either `mutual conjunction' or `aggregation'.b.In both categories, the constituent characterised by a lesser number of vowels will be positioned as the initial element in the sequence within the compound word.c.In instances where all terms exhibit an equal frequency of vowels, the term that both commences and concludes with vowels shall be designated as the primary constituent.5.Karmadhāraya (Oppositional Determinative) A right-headed compound word.a.It is a binary CW, equivalent to tatpuruṣa.b.Identification of a CW's grammatical number and gender indistinguishable from the tatpuruṣa class.c.In contrast to the tatpuruṣa class, constituents of the karmadhāraya denote identical objects, therefore, all components within the compound shares the same morphological case marker.d.Like tatpuruṣa, it is a right-headed compound word. 1.Avyayībhāva (Adverbial) – A left-headed compound word.a.An avyayībhāva compound word is an indeclinable term.b.An avyayībhāva will consistently be in the neuter gender, regardless of the gender of its constituents.c.An avyayībhāva is always in singular form, regardless of whether its constituents are dual or plural.d.Avyayībhāva compounding is a binary process. To elucidate, only two nouns can be compounded simultaneously.e.An avyayībhāva consistently concludes with a short vowel. Avyayībhāva (Adverbial) – A left-headed compound word.a.An avyayībhāva compound word is an indeclinable term.b.An avyayībhāva will consistently be in the neuter gender, regardless of the gender of its constituents.c.An avyayībhāva is always in singular form, regardless of whether its constituents are dual or plural.d.Avyayībhāva compounding is a binary process. To elucidate, only two nouns can be compounded simultaneously.e.An avyayībhāva consistently concludes with a short vowel. a.An avyayībhāva compound word is an indeclinable term. An avyayībhāva compound word is an indeclinable term. b.An avyayībhāva will consistently be in the neuter gender, regardless of the gender of its constituents. An avyayībhāva will consistently be in the neuter gender, regardless of the gender of its constituents. c.An avyayībhāva is always in singular form, regardless of whether its constituents are dual or plural. An avyayībhāva is always in singular form, regardless of whether its constituents are dual or plural. d.Avyayībhāva compounding is a binary process. To elucidate, only two nouns can be compounded simultaneously. Avyayībhāva compounding is a binary process. To elucidate, only two nouns can be compounded simultaneously. e.An avyayībhāva consistently concludes with a short vowel. An avyayībhāva consistently concludes with a short vowel. 2.Tatpuruṣa (Determinative) A right-headed compound word.a.The formulation of the tatpuruṣa represents a binary process, indicating that, at any given moment, only two constituents are capable of undergoing semantic amalgamation.b.The gender of a compound word is determined by the gender of its head component. In a straightforward manner, it can be stated that a compound word assumes the gender of the head noun within the compound structure.c.The tatpuruṣa compound takes over the grammatical number inherent to the head it possesses. Tatpuruṣa (Determinative) A right-headed compound word.a.The formulation of the tatpuruṣa represents a binary process, indicating that, at any given moment, only two constituents are capable of undergoing semantic amalgamation.b.The gender of a compound word is determined by the gender of its head component. In a straightforward manner, it can be stated that a compound word assumes the gender of the head noun within the compound structure.c.The tatpuruṣa compound takes over the grammatical number inherent to the head it possesses. a.The formulation of the tatpuruṣa represents a binary process, indicating that, at any given moment, only two constituents are capable of undergoing semantic amalgamation. The formulation of the tatpuruṣa represents a binary process, indicating that, at any given moment, only two constituents are capable of undergoing semantic amalgamation. b.The gender of a compound word is determined by the gender of its head component. In a straightforward manner, it can be stated that a compound word assumes the gender of the head noun within the compound structure. The gender of a compound word is determined by the gender of its head component. In a straightforward manner, it can be stated that a compound word assumes the gender of the head noun within the compound structure. c.The tatpuruṣa compound takes over the grammatical number inherent to the head it possesses. The tatpuruṣa compound takes over the grammatical number inherent to the head it possesses. 3.Bahuvrīhi (Exocentric) An adjectival form.a.In contrast to other classes, bahuvrīhi transforms its components into an adjectival form.b.Given that the constituents possess an adjectival form, the bahuvrīhi conveys a meaning that is markedly different from that of its constituents.c.Given that bahuvrīhi becomes an adjective, through the core entity in the sentence, it connects with the main verb of the sentence. Bahuvrīhi (Exocentric) An adjectival form.a.In contrast to other classes, bahuvrīhi transforms its components into an adjectival form.b.Given that the constituents possess an adjectival form, the bahuvrīhi conveys a meaning that is markedly different from that of its constituents.c.Given that bahuvrīhi becomes an adjective, through the core entity in the sentence, it connects with the main verb of the sentence. a.In contrast to other classes, bahuvrīhi transforms its components into an adjectival form. In contrast to other classes, bahuvrīhi transforms its components into an adjectival form. b.Given that the constituents possess an adjectival form, the bahuvrīhi conveys a meaning that is markedly different from that of its constituents. Given that the constituents possess an adjectival form, the bahuvrīhi conveys a meaning that is markedly different from that of its constituents. c.Given that bahuvrīhi becomes an adjective, through the core entity in the sentence, it connects with the main verb of the sentence. Given that bahuvrīhi becomes an adjective, through the core entity in the sentence, it connects with the main verb of the sentence. 4.Dvandva (Coordinative) – a multi-headed compound word.a.Dvandva structures can only be formed when the constituents meet the criteria for inclusion in either `mutual conjunction' or `aggregation'.b.In both categories, the constituent characterised by a lesser number of vowels will be positioned as the initial element in the sequence within the compound word.c.In instances where all terms exhibit an equal frequency of vowels, the term that both commences and concludes with vowels shall be designated as the primary constituent. Dvandva (Coordinative) – a multi-headed compound word.a.Dvandva structures can only be formed when the constituents meet the criteria for inclusion in either `mutual conjunction' or `aggregation'.b.In both categories, the constituent characterised by a lesser number of vowels will be positioned as the initial element in the sequence within the compound word.c.In instances where all terms exhibit an equal frequency of vowels, the term that both commences and concludes with vowels shall be designated as the primary constituent. a.Dvandva structures can only be formed when the constituents meet the criteria for inclusion in either `mutual conjunction' or `aggregation'. Dvandva structures can only be formed when the constituents meet the criteria for inclusion in either `mutual conjunction' or `aggregation'. b.In both categories, the constituent characterised by a lesser number of vowels will be positioned as the initial element in the sequence within the compound word. In both categories, the constituent characterised by a lesser number of vowels will be positioned as the initial element in the sequence within the compound word. c.In instances where all terms exhibit an equal frequency of vowels, the term that both commences and concludes with vowels shall be designated as the primary constituent. In instances where all terms exhibit an equal frequency of vowels, the term that both commences and concludes with vowels shall be designated as the primary constituent. 5.Karmadhāraya (Oppositional Determinative) A right-headed compound word.a.It is a binary CW, equivalent to tatpuruṣa.b.Identification of a CW's grammatical number and gender indistinguishable from the tatpuruṣa class.c.In contrast to the tatpuruṣa class, constituents of the karmadhāraya denote identical objects, therefore, all components within the compound shares the same morphological case marker.d.Like tatpuruṣa, it is a right-headed compound word. Karmadhāraya (Oppositional Determinative) A right-headed compound word.a.It is a binary CW, equivalent to tatpuruṣa.b.Identification of a CW's grammatical number and gender indistinguishable from the tatpuruṣa class.c.In contrast to the tatpuruṣa class, constituents of the karmadhāraya denote identical objects, therefore, all components within the compound shares the same morphological case marker.d.Like tatpuruṣa, it is a right-headed compound word. a.It is a binary CW, equivalent to tatpuruṣa. It is a binary CW, equivalent to tatpuruṣa. b.Identification of a CW's grammatical number and gender indistinguishable from the tatpuruṣa class. Identification of a CW's grammatical number and gender indistinguishable from the tatpuruṣa class. c.In contrast to the tatpuruṣa class, constituents of the karmadhāraya denote identical objects, therefore, all components within the compound shares the same morphological case marker. In contrast to the tatpuruṣa class, constituents of the karmadhāraya denote identical objects, therefore, all components within the compound shares the same morphological case marker. d.Like tatpuruṣa, it is a right-headed compound word. Like tatpuruṣa, it is a right-headed compound word. Having considered the aforementioned pointers, we categorized each term into compound and non-compound words. Table 4 displays the example of each category with explanation.Table 4Examples of each categorical data with explanation.Table 4 dummy alt textCategoryTermExplanationCompound wordpakvāśayaḥThe term ``pakvāśayaḥ'' is an example of a compound word. It is made up of pakva and āśaya, where pakva indicates the completely digested food while āśaya marks the place. When the terms ``pakva'' and ``āśaya'' undergo compounding under the Tatpuruṣa (Determinative) rule, the term ``pakvāśayaḥ'' stands for the place where the food fully becomes digested; more precisely, ``pakvāśayaḥ'' stands for the large intestine.Non-compound wordyathāThe term ``yathā'' is an independent word and considered an indeclinable in Sanskrit grammar. It means ``as,'' “just as," or “according to.” Table 4Examples of each categorical data with explanation.Table 4 dummy alt textCategoryTermExplanationCompound wordpakvāśayaḥThe term ``pakvāśayaḥ'' is an example of a compound word. It is made up of pakva and āśaya, where pakva indicates the completely digested food while āśaya marks the place. When the terms ``pakva'' and ``āśaya'' undergo compounding under the Tatpuruṣa (Determinative) rule, the term ``pakvāśayaḥ'' stands for the place where the food fully becomes digested; more precisely, ``pakvāśayaḥ'' stands for the large intestine.Non-compound wordyathāThe term ``yathā'' is an independent word and considered an indeclinable in Sanskrit grammar. It means ``as,'' “just as," or “according to.” Examples of each categorical data with explanation. 7. Manual verification Given that CarakaBinary was designed to support development of language models and Ayurvedic pedagogy, we adhered to a reverse manual validation procedure. To clarify, we commenced assessing each instance in reverse order to assure validity based on grammar, linguistics, and the labelling process. Table 5 includes the representative samples of CarakaBianry. In Table 5 the column named “Words” includes the Sanskrit terms from the ancient Ayurvedic text Carakasamhitā, while the column English-Translation outlines the rough translation of the corresponding terms. Labels NC and C refer to non-compound word that is a single noun without semantic amalgamation with other nouns, and compound word that is the semantic amalgamation of two or more nouns. For the code execution purpose, we additionally added numerical labels including 0 that refers to non-compound word while 1 stands for compound words.Table 5Sample representatives of CarakaBinary.Table 5 dummy alt text Table 5Sample representatives of CarakaBinary.Table 5 dummy alt text Sample representatives of CarakaBinary. Additionally, we assessed inter-annotator agreement employing Cohen’s Kappa coefficient on independently annotated CarakaBinary samples. The three annotators achieved a Cohen’s Kappa score of 0.8600, thereby demonstrating excellent agreement and consistency in the labeling process. All three annotators involved in the manual annotation of CarakaBinary were trained on Sanskrit grammar in a traditional setup, and they hold advanced academic qualifications, including MPhil and PhD. Fig. 5 summarizes the end-to-end dataset preparation pipeline.Fig. 5End-to-end pipeline of the dataset preparation process.Fig 5 dummy alt text End-to-end pipeline of the dataset preparation process. We addressed the following points throughout the manual segmentation phase, taking into account three basic factors—WBSP, EBSP, and CBSP—in the Sanskrit segmentation process.•In cases when terms lack semantic proximity but their phonemes qualify for the euphonic alteration, the outcome will be regarded as a euphonically altered non-semantic combination, with segmentation executed using the EBSP approach.•In instances where both terms and their phonemes qualify for compounding and the euphonic alteration, the outcome will be designated as a euphonically altered semantic combination, and CBSP will not be employed to maintain the compound word and its original meaning.•In instances where the phrases can be compounded but their phonemes do not adhere to the euphonic alterations, the outcome will be regarded as a euphonically unaltered semantic combination, and no segmentation method is employed.•The euphonically altered non-semantic combinations resemble CW, yet they are not regarded as one or the opposite.•Although euphonic alterations may be present in certain compound words, it is not requisite for all compound words to contain euphonic alterations, nor do euphonic alterations exclusively occur within all compound words. •In cases when terms lack semantic proximity but their phonemes qualify for the euphonic alteration, the outcome will be regarded as a euphonically altered non-semantic combination, with segmentation executed using the EBSP approach. In cases when terms lack semantic proximity but their phonemes qualify for the euphonic alteration, the outcome will be regarded as a euphonically altered non-semantic combination, with segmentation executed using the EBSP approach. •In instances where both terms and their phonemes qualify for compounding and the euphonic alteration, the outcome will be designated as a euphonically altered semantic combination, and CBSP will not be employed to maintain the compound word and its original meaning. In instances where both terms and their phonemes qualify for compounding and the euphonic alteration, the outcome will be designated as a euphonically altered semantic combination, and CBSP will not be employed to maintain the compound word and its original meaning. •In instances where the phrases can be compounded but their phonemes do not adhere to the euphonic alterations, the outcome will be regarded as a euphonically unaltered semantic combination, and no segmentation method is employed. In instances where the phrases can be compounded but their phonemes do not adhere to the euphonic alterations, the outcome will be regarded as a euphonically unaltered semantic combination, and no segmentation method is employed. •The euphonically altered non-semantic combinations resemble CW, yet they are not regarded as one or the opposite. The euphonically altered non-semantic combinations resemble CW, yet they are not regarded as one or the opposite. •Although euphonic alterations may be present in certain compound words, it is not requisite for all compound words to contain euphonic alterations, nor do euphonic alterations exclusively occur within all compound words. Although euphonic alterations may be present in certain compound words, it is not requisite for all compound words to contain euphonic alterations, nor do euphonic alterations exclusively occur within all compound words. In summary, the segmentation procedure utilized the first two approaches, WBSP and EBSP, to extract non-compound terms, while the omission of CBSP was implemented to maintain the integrity of compound words for the study. Thus, the construction of CarakaBinary primarily included manual linguistic processing and expert-led verification methods. More precisely, the conversion of PDF pages into .png files and .png files into editable text through OCR was carried out automatically, whereas all other remaining phases were performed manually. 8. Computational evaluation of dataset imbalance In order to examine the implications with respect to the computational standpoint for the class weight imbalance, additionally, we executed classification experiments employing Support Vector Classification (SVC) with MuRIL and XLM-R multilingual embedding models, including both balanced and imbalanced class-weight settings. The models’ consistent performance across all settings in terms of Macro F1 Score demonstrated that the dataset preserves meaningful discriminative linguistic patterns without being affected by imbalanced class weight. Table 6 showcases the Macro F1 Score across all settings.Table 6Macro F1 Score across all settings.Table 6 dummy alt textModelClass WeightMacro F1MuRiL + SVCBalanced0.8824MuRiL + SVCImbalanced0.8857XLM-R + SVCBalanced0.8291XLM-R + SVCImbalanced0.8323 Table 6Macro F1 Score across all settings.Table 6 dummy alt textModelClass WeightMacro F1MuRiL + SVCBalanced0.8824MuRiL + SVCImbalanced0.8857XLM-R + SVCBalanced0.8291XLM-R + SVCImbalanced0.8323 Macro F1 Score across all settings. Moreover, the dataset includes the duplicate entries, which are intentionally preserved to safeguard the authentic lexical recurrence and natural frequency distribution marks that exist in Carakasamhitā. This may aid the tasks, including computational and frequency-based linguistic analysis, in future. However, the initial experiment conducted stated that the class weight imbalance and duplicate entries did not significantly impact the models’ performance in a negative manner. Limitations CarakaBinary is limited for binary classification of compound and non-compound Sanskrit Ayurvedic terms. Apart from categorical and numerical labels, it does not include further linguistic representations, including part-of-speech tags and others. Despite the distinguished Sanskrit scholar's annotation and validation of the dataset for compound and non-compound classifications, human errors cannot be completely eliminated. The dataset also possesses the imbalanced nature and includes duplicate entries. Anticipated Challenges •The dataset belongs to the Sanskrit language, which possesses morphological complexity and unclear word boundaries that may possibly lead to ambiguity in determining between compound and non-compound words.•The compound word often includes euphonic alterations at the junction of its components that might complicate the identification of those components.•The dataset, prepared for word-level analysis, lacks the syntactic features necessary to categorize specific terms as compound words.•The meaning of the compound word is influenced by variations in linguistic interpretation, as identical components might produce compound words that appear similar yet differ in meaning. Consequently, specific compound terms may induce ambiguity, thereby lead complexity in compound type identification. However, the compound and non-compound categorisation will not affect by this ambiguity. •The dataset belongs to the Sanskrit language, which possesses morphological complexity and unclear word boundaries that may possibly lead to ambiguity in determining between compound and non-compound words. The dataset belongs to the Sanskrit language, which possesses morphological complexity and unclear word boundaries that may possibly lead to ambiguity in determining between compound and non-compound words. •The compound word often includes euphonic alterations at the junction of its components that might complicate the identification of those components. The compound word often includes euphonic alterations at the junction of its components that might complicate the identification of those components. •The dataset, prepared for word-level analysis, lacks the syntactic features necessary to categorize specific terms as compound words. The dataset, prepared for word-level analysis, lacks the syntactic features necessary to categorize specific terms as compound words. •The meaning of the compound word is influenced by variations in linguistic interpretation, as identical components might produce compound words that appear similar yet differ in meaning. Consequently, specific compound terms may induce ambiguity, thereby lead complexity in compound type identification. However, the compound and non-compound categorisation will not affect by this ambiguity. The meaning of the compound word is influenced by variations in linguistic interpretation, as identical components might produce compound words that appear similar yet differ in meaning. Consequently, specific compound terms may induce ambiguity, thereby lead complexity in compound type identification. However, the compound and non-compound categorisation will not affect by this ambiguity. Ethics Statement The authors have read and follow the ethical requirements for publication in Data in Brief and confirm that the current work does not involve human subjects, animal experiments, or any data collected from social media platforms. CRediT Author Statement Shriganesh Devaru Bhat: Conceptualization, Methodology, Data curation, Writing, Original draft preparation, and Image preparation; Premjith B: Conceptualization, Methodology, Supervision, Reviewing and Editing.