glossAPI/libiep

LIBIEP - Historical Greek School Textbooks

Dataset Language: Greek (with extensive coverage of historical written varieties: Καθαρεύουσα, Απλή Καθαρεύουσα, Δημοτική, plus a long tail of Αρχαία Ελληνικά, Λατινικά, Γαλλικά, Γερμανικά, Τουρκικά, Αγγλικά, Καραμανλίδικα).

Dataset Info: This dataset is a structured snapshot of the Historical Collection of School Textbooks of the Institute of Educational Policy (Ινστιτούτο Εκπαιδευτικής Πολιτικής - IEP), the public-law body that advises the Greek Ministry of Education. The collection is surfaced at e-library.iep.edu.gr and was assembled under a project co-funded by the European Regional Development Fund (ERDF) and Greek national funds.

The dataset contains 6,020 records spanning printed school textbooks, curricula and educational-policy proceedings dated 1808 – 1998, with peak coverage between the 1880s and the early 1980s. Each record corresponds to a single physical volume that IEP (or one of its partner-institutions) has digitised and made publicly available as a downloadable PDF. The collection aggregates volumes held by 33 distinct holding institutions, the largest of which are the Institute of Educational Policy itself (2,807 volumes), the Library of the Hellenic Parliament (1,296), the Greek Literary and Historical Archive - ELIA (942), the Centre for the Research and Study of Modern Greek History - EKEDISY, the University of Ioannina, the National Library of Greece, the Academy of Athens, the Moraitis School, the Eugenides Foundation, the Aristotle University of Thessaloniki and several municipal / public libraries.

The corpus splits across three sub-collections by publication type:

Sub-collection	Records	Description
books	5,813	School textbooks (Σχολικά Εγχειρίδια), pupil aids (Μαθητικά Βοηθήματα), teacher aids (Παιδαγωγικά Βοηθήματα) and assorted other educational books
curricula	105	Curricula and syllabus documents (ΦΕΚ Προγραμμάτων Σπουδών, Αναλυτικά Προγράμματα)
minutes	102	Proceedings of historical pedagogical councils (Πρακτικά ΑΕΣ / ΚΕΜΕ / ΣΣ)

Twelve type categories appear in total, dominated by Σχολικό Εγχειρίδιο (4,289 / 71.2 %), Μαθητικό Βοήθημα (1,124) and Παιδαγωγικό Βοήθημα (227).

Metadata Info: The parquet has one row per volume. All 22 columns are stored as Arrow large_string. The dataset has zero nulls in every column - cataloguer-missing fields are encoded as the empty string "", not as null. The content column carries the OCR / text extraction of the underlying PDF and is the bulk of the file (≈ 2.83 GB uncompressed, mean 281 K characters per record, max 6.97 M characters).

Metadata Field Descriptions:

Field	Type	Description
title	string	Title of the volume, as catalogued by IEP. Length 6 – 374 chars (median 63). 4,700 distinct values; common multi-volume / series titles (e.g. Πρακτικά Ολομέλειας Εκπαιδευτικού Συμβουλίου, Τμήμα Μέσης) repeat across editions. Always populated.
content	string	Full text extraction of the underlying PDF. UTF-8 plain text with line breaks preserved. Mean length 281,886 chars, max 6,967,728 chars. See Known Data Quality Notes for OCR / extractor caveats.
abstract	string	Short editorial blurb supplied by the IEP cataloguer summarising the volume's pedagogical content (e.g. "Αναγνωστικό Ε΄ Δημοτικού που περιέχει κείμενα εμπνευσμένα από την εθνική, κοινωνική και οικογενειακή ζωή..."). 196 records carry an empty abstract at the source.
author	string	Author(s) of the volume, comma-separated, in the form Surname, Given Name(s) (Greek script). 256 records (≈ 4.3 %) carry an empty author field, mostly anonymous or committee-authored volumes (proceedings, ministerial curricula).
illustrator	string	Illustrator(s) of the volume. 5,332 records (≈ 88.6 %) carry an empty value at the source - illustrator credits were not systematically catalogued, especially for the older holdings.
grade	string	Target school grade (e.g. Α' Δημοτικού, Γ' Γυμνασίου, Δ' Λυκείου). May combine multiple grades or be empty (1,181 records ≈ 19.6 %) for non-grade-specific volumes (curricula, teacher aids, proceedings).
subject	string	Academic subject the volume teaches (e.g. Νέα Ελληνικά, Αρχαία Ελληνικά, Ιστορία, Μαθηματικά, Φυσική). 65 distinct values; 395 records empty.
collection	string	One of three values - books (5,813), curricula (105), minutes (102). Encodes the IEP sub-collection the volume belongs to.
type	string	More granular pedagogical-type taxonomy. Twelve values, dominated by Σχολικό Εγχειρίδιο (4,289), Μαθητικό Βοήθημα (1,124), Παιδαγωγικό Βοήθημα (227).
keywords	string	Free-text descriptor list (comma-separated) describing the textbook's topical coverage, e.g. υποκείμενο, αντικείμενο, ρήματα, προσδιορισμοί, προθέσεις. 272 records empty.
education_level	string	One of Πρωτοβάθμια (primary, 1,545), Δευτεροβάθμια (secondary, 3,921), Τριτοβάθμια (tertiary, 7), Προσχολική (pre-school, 4), Πρωτοβάθμια,Δευτεροβάθμια (60), or empty (483).
language	string	Written variety of the volume's main text. 25 distinct values capturing the diglossic history of Modern Greek. Most-populous: Καθαρεύουσα (2,736), Δημοτική (1,350), Απλή καθαρεύουσα (1,333), Δημοτική με στοιχεία καθαρεύουσας (126). A long tail records bilingual or foreign-language textbooks (Γαλλικά, Αρχαία Ελληνικά, Τουρκικά, Λατινικά, Γερμανικά, Αγγλικά, Καραμανλίδικα). 214 records empty.
publisher	string	Publisher of the volume. 901 distinct publishers; the most-populous are the state textbook organisations Οργανισμός Εκδόσεως Διδακτικών Βιβλίων (1,596) and Οργανισμός Εκδόσεως Σχολικών Βιβλίων (605). 430 records empty.
year	string	Year of publication, as a four-digit string (e.g. 1925). Spans 1808 – 1998 (median around 1935). 392 records (≈ 6.5 %) empty at the source. Parseable with pandas.to_datetime(..., errors='coerce').
place	string	Place of publication, in Greek. 49 distinct values; Αθήνα dominates (5,511), with substantial historical contributions from Κωνσταντινούπολη (51), Πάτρα, Θεσσαλονίκη, Σύρος, etc. 284 records empty.
edition	string	Edition number, in Greek ordinal form (e.g. 1η, 5η, 13η). 2,627 records empty - especially common for first-edition volumes and for proceedings / curricula.
ministerial_decision	string	Identifier of the Ministry of Education decree that approved the volume for use in Greek schools, in the form <protocol-number>/<gazette-date> (e.g. 2577/28-4-1975, 51231,51232/20-8-1934). 3,083 records carry an empty value - the field is only populated for textbooks that went through a formal approval procedure, not for curricula, proceedings, or many of the older free-market textbooks.
approval_year	string	Year the volume was approved for school use by the Ministry. Spans 1812 – 1983. 3,212 records empty (same logic as ministerial_decision). Note: approval_year and year (publication year) are distinct - approval should normally precede or match publication. See Known Data Quality Notes.
original_location	string	Holding institution / library that contributed the physical volume to IEP for digitisation. 33 distinct values. The largest contributors are Ινστιτούτο Εκπαιδευτικής Πολιτικής (2,807), Βιβλιοθήκη της Βουλής των Ελλήνων (1,296), Ελληνικό Λογοτεχνικό και Ιστορικό Αρχείο (942).
size_format	string	Free-text physical-description string supplied by the cataloguer (e.g. Σχολικό εγχειρίδιο, μαλακό εξώφυλλο, 184 σελίδες. Σχέδια (έγχρωμα).). 4,374 distinct strings. Always populated.
url	string	Direct HTTPS link to the IEP e-library landing page for the record, e.g. https://e-library.iep.edu.gr/iep/collection/browse/item.html?code=01-17000. Always populated, always under e-library.iep.edu.gr, and unique per row (6,020 distinct URLs).
pdf_url	string	Direct HTTPS link to the downloadable PDF of the digitised volume, e.g. https://e-library.iep.edu.gr/iep/collection/browse/view.html?action=download&code=01-17000. Always populated, always under e-library.iep.edu.gr, and unique per row (6,020 distinct URLs).

Dataset Statistics: Records: 6,020

Words: 263,670,832

Tokens: 510,785,207 (WordPiece tokens from nlpaueb/bert-base-greek-uncased-v1, computed on title + abstract + content combined. The three columns split as: title 98,188 tokens, abstract 185,690 tokens, content 510,501,329 tokens - i.e. content carries > 99.94 % of the corpus.)

Size: 1.36 GB (parquet on disk) - ≈ 2.83 GB of plain text once decoded.

Known Data Quality Notes: The source documents are PDFs digitised from physical books spanning almost two centuries of Greek educational publishing (1808 – 1998), so the content column carries a substantial OCR signal-to-noise floor that is inherent to the source material. Born-digital or well-scanned modern volumes extract cleanly; nineteenth- and early-twentieth-century scans of polytonic Καθαρεύουσα with worn typefaces extract less cleanly. Our policy is to publish the extracted text as faithfully as possible to the underlying PDFs rather than aggressively post-process it: this preserves reproducibility and lets downstream users apply their own filtering, re-OCR, or normalisation strategy as appropriate to their task.

• Empty bibliographic fields at the source. Most metadata columns carry empty strings (not nulls) where the IEP cataloguer did not supply that field - most notably illustrator (≈ 88.6 % empty), approval_year (≈ 53.4 %), ministerial_decision (≈ 51.2 %), edition (≈ 43.6 %), grade (≈ 19.6 %), year (≈ 6.5 %), publisher (≈ 7.1 %), education_level (≈ 8.0 %), subject (≈ 6.6 %), keywords (≈ 4.5 %), author (≈ 4.3 %), abstract (≈ 3.3 %), language (≈ 3.6 %), place (≈ 4.7 %). These reflect the IEP catalogue as supplied; we do not synthesise placeholders.
• PDF-extraction stubs in content. 16 records (≈ 0.27 % of the dataset) carry the same 55-character placeholder "Ψηφιοποιήθηκε από το Ινστιτούτο Εκπαιδευτικής Πολιτικής" in their content field, where the PDF was either an image-only scan our extractor could not OCR or was otherwise unreadable. These rows still carry valid bibliographic metadata and a working pdf_url. They can be filtered with len(content) > 60 or by SHA1-hashing the content column.
• year vs approval_year anomalies. For 12 records the publication year precedes the approval year (year < approval_year). This is taken as supplied by the IEP catalogue - typically a transcription error or a re-printing of an older textbook whose original approval decree post-dates the printed copy in the collection.
• Generated <table> markup inside content. The PDF-to-text extractor emits <table> / <tr> / <th> / <td> tags (with HTML entities such as >, <) when it detects a table on a page. This affects 2,390 of 6,020 records and is structured output, not corruption: it captures pedagogical tables (multiplication tables, declension paradigms, periodic-table fragments, etc.). Strip or parse it depending on whether your downstream pipeline wants structured tables. The embedded HTML entities account for the 218 "encoding-artifact" hits in the QA scan.
• Trailing whitespace in content. 1,850 of 6,020 content values end in trailing whitespace (rogue newlines from the extractor). Easy to .strip() if desired, but left as-is in the published file to preserve byte-for-byte fidelity to the extraction step.
• Unicode replacement characters. 114 content rows contain U+FFFD (the replacement character), typically clustered on hard-to-OCR pages of older Καθαρεύουσα scans. These are extraction defects at the source and are not silently removed.
• Long non-alphanumeric runs. 3,674 records contain runs of 8+ consecutive non-alphanumeric characters somewhere in content. Inspection shows the vast majority are legitimate mathematical formulas, dot-leader tables of contents, dingbat separators between chapters, or polytonic accent clusters in pre-monotonic Greek - i.e. expected features of nineteenth- / early-twentieth-century textbooks, not extraction noise.
• Duplicate content hashes are limited and explained. The dataset is fully deduplicated at the row level (no two rows are byte-for-byte identical across all 22 columns). Only one content value repeats - the 55-char PDF-extraction stub described above, occurring 16 times. Every other content row is byte-unique. No url or pdf_url value is duplicated either.
• Duplicate non-content cells are expected. Bibliographic fields like grade, language, education_level, place, subject, publisher, collection, type legitimately repeat across rows by definition (a controlled-vocabulary value can recur on thousands of textbooks).
• year date coverage is not uniform. Records are sparsely scattered between 1808 and the 1860s (1 – 5 records / year), gradually densify from the 1880s onward, and peak in the 1970s (105 – 125 records / year). The biggest gap is 1819 → 1828 (3,287 days with no recorded textbook in the collection). Consider this when using year for diachronic / temporal slicing.
• Mixed orthographic conventions. The corpus spans the diglossia transition: Καθαρεύουσα predominates pre-1976, Δημοτική post-1976. Polytonic and monotonic Greek both appear, and pre-1955 volumes routinely use the polytonic Greek that the OCR-trained models of today are least fluent in. Plan tokenisation budgets accordingly.

License: The dataset compilation, metadata schema, text-extraction pipeline and accompanying tooling are released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

The underlying texts (each individual textbook, curriculum, or proceedings volume) remain the intellectual property of their original authors, publishers and/or holding institutions, as established under and the corresponding EU directives. According to that statute, intellectual property in literary works lapses 70 years after the death of the author; many of the nineteenth- and early-twentieth-century volumes in this collection are therefore likely in the public domain in their own right, while later twentieth-century textbooks remain copyrighted. We do not provide a per-record copyright assessment.

The Institute of Educational Policy publishes the e-library under terms that explicitly permit personal, educational and non-commercial use with source attribution, and explicitly prohibit commercial reproduction or republication without prior written permission (see Όροι Χρήσης of iep.edu.gr). The dataset is therefore offered for non-commercial academic and research use only.

Downstream users who wish to use any individual textbook commercially, to redistribute it outside research contexts, or who are rightholders of any specific work in the collection and object to its inclusion, should independently contact the Institute of Educational Policy (iep.edu.gr) or the relevant holding institution. We will honour any takedown request from a verified rightholder.

When using this dataset, please attribute both the original repository and this derivative compilation:

LIBIEP - Historical Greek School Textbooks, derived from the Historical Collection of School Textbooks of the Greek Institute of Educational Policy (e-library.iep.edu.gr), licensed under CC BY-NC-SA 4.0.

Source: Historical Collection of School Textbooks - Ινστιτούτο Εκπαιδευτικής Πολιτικής (Institute of Educational Policy), e-library.iep.edu.gr. The digitisation project was co-funded by the European Regional Development Fund (ERDF) and Greek national funds, and aggregates volumes contributed by 33 holding institutions including the Library of the Hellenic Parliament, the Greek Literary and Historical Archive (ELIA), EKEDISY, the Academy of Athens, the National Library of Greece, the University of Ioannina, the Aristotle University of Thessaloniki, the Moraitis School, the Eugenides Foundation and several municipal libraries.