ಅಕ್ಷರAkshara

Akshara No. 50 · Surveys and Gazetteers

The Early History of the Tea Industry

in North-East India

Harold H. Mann · Calcutta, 1918

1815–1856 · Assam, Bengal and North-East IndiaTea Industry · Plantation Economy · Assam Company · Colonial Botany

44pages
2chapters
3tables
97of 100 in automatic checks

Read by machine from the photographs, and not yet checked by a person. What that means

How it begins

For ten years these materials and notes have remained with me unused.

Harold H. Mann, The Early History of the Tea Industry, page 3

Contents

Pages as printed in 1918
  1. Title & Imprint1
  2. i— Government Experiments (1834–1840)3
  3. ii— The Assam Company (1839–1856)19

From its tables

All the tables

Three tables in the book, each set as it was printed and given as data. Here is the first. See it in its chapter.

As printed on page 29
Yield of Tea.Expenditure.
1841.10,505 lbs.£12,984
1842.31,398 „£16,560
1843.87,705 „£13,146
1844.120,422 „£7,284

This table as data, CSV: as printed · tidy

How this text was read

Read by machine

This book was read from photographs of its pages by a vision-language model, an AI that looks at the image of a page and writes out its text, then run through Akshara’s automatic checks, which flag passages to be read again against the scan. A model reads far more cleanly than ordinary OCR, the plain machine text that usually sits beside a scan online. In this book the checks found little, so what errors remain should be few and small, and should not spoil the reading. But no machine reading is perfect, and until a person has read a page against its photograph, some errors will remain.

When it does go wrong, its mistakes are not the ones older OCR made. OCR leaves garbled letters that catch the eye. A language model reads for sense, so when it goes wrong it usually writes something fluent and plausible: its errors look like the book.

Automated reading (1 open review item); 4 of 40 pages had flagged passages re-read against the scan; no pages yet checked by a person.

The quirks of a machine reading

Each is uncommon; these are the kinds a careful reader may meet.

  • It tidies what it should copy. An old spelling modernised; the author’s own slip quietly corrected; a romanisation as the author printed it, Bhagavad Gita, given modern IAST, Bhagavad Gītā.
  • Where the print is faint, it guesses. A word, a figure, now and then a phrase, supplied because it fits rather than because it is printed. A line may be dropped, or read twice. This has grown rare, and it is the hardest kind to catch.
  • Accents and other scripts. Diacritics dropped, or added where the author had none; one name spelt two ways in two chapters. Words in Indian scripts, in Urdu, Persian or Arabic, and quotations in Greek or Latin are all read less surely than English, and a right-to-left line can come out in the wrong order.
  • It reads one page at a time. So a paragraph that runs over a page turn may be broken in two, or two paragraphs joined; the running head and page number may slip into the text; a heading may be set at the wrong level, or a title broken; a page in two columns read across instead of down.
  • Notes, tables and pictures. A footnote parted from its mark, or joined to the wrong one; a table’s columns shifted, or a figure misread, and a wrong number in a table looks exactly like a right one; an illustration, map or plate left with no marker where it stood.
  • Small things. Punctuation missing or changed; italics lost; a line-end hyphen kept, or a hyphenated word run together; library stamps, bookplates and handwriting in the front matter read as if they were printed.

97of 100

from the automatic checks of structure, notes, tables and spelling

  • Structure25/25 Nothing found
  • Footnotes22/25 1 footnote not matched to its mark
  • Tables and figures25/25 Nothing found
  • Spelling and accents25/25 Nothing found

The checks find what a machine can find. A book can score 100 and still hold misreadings that only a reader, with the scan beside the text, will catch.

Machines have brought these books this far; only people can finish them. That is why Akshara is open. It is meant to become a community of readers who care for India’s written heritage: people who will sit with a page and its photograph, and put it right. Every page of this book is on the proofreading desk with its scan beside it, and one page takes about five minutes.

Proofread a page of this book

This edition

Open for proofreading

Colophon

Printed in Calcutta by The Calcutta General Printing Co. (Reprinted from the Bengal Economic Journal), 1918. This edition gives all 44 printed pages, 4 of them read a second time against the scan, with its 45 footnotes kept where they were printed. It was read from the copy in the Internet Archive, cu31924013772441.

Akshara No. 50, on the shelf of Surveys and Gazetteers, and in the collection Indian Economic History & Political Economy.

Released under CC BY 4.0: share and adapt it freely, crediting Akshara. The book itself is in the public domain.

A note on this edition

This digital edition was produced by transcribing a scanned copy of the original printed book. This note records where the scan came from, what we normalized from the printed page, and what we know still awaits human eyes, so that you can judge the text for yourself and consult the original scan where it matters.
The full record of this edition

1. The Copy We Worked From

  • Title: The Early History of the Tea Industry in North-East India (1918)
  • Author(s): Harold H. Mann
  • Imprint: Calcutta: The Calcutta General Printing Co. (Reprinted from the Bengal Economic Journal)
  • Internet Archive Scan: cu31924013772441
  • Source File: early-history-tea-industry.originals.pdf
  • Source SHA-256: 8eb4249d4d28687becbe062a27ff2758a0aa54b77e60ce84d8a4b6ec93c2c170
  • Leaf Accounting: 44 PDF leaves — 40 content leaves transcribed, 4 blank/tissue leaves (40, 41, 42, 43), 0 failed leaves.
  • Edition Status: community (97.0 / 100 Rubric Score)

2. Editorial Policy: Presentation vs. Historical Text

  • Modernized (Presentation Only): Running heads and institutional library stamps (8 flyleaf/stamp block(s)) are excluded from reading prose; 28 paragraphs running across page breaks were rejoined (preserving [p. N] folio markers); 0 cross-page split footnotes and 0 multi-page tables were stitched; and 32 end-of-line hyphens were resolved against intra-book attestation.
  • Preserved Verbatim: Period spellings, nineteenth-century romanization and diacritics (3 distinct non-ASCII characters across 6 leaves), and printed statistical figures (Rs. a. p., £ s. d., lakhs/crores). We never alter printed totals to force column arithmetic.

3. Transcription Provenance & Remaining Open Checks

  • Models: gemini-3.8-flash (40 pp.).
  • Reader: Read by gemini-3.8-flash on 40 of 40 pages. Declared reader profile: english_prose (new readings use gemini-3.8-flash, with the book spelling note on prose pages).
  • Human Proofing: 7 replayed correction(s). Automated reading (1 open review item); 4 of 40 pages had flagged passages re-read against the scan; no pages yet checked by a person.

Rather than pretend machine transcription is infallible, here is what our validators still flag:

  • 1 finding(s) across 1 leaf(s) (pp. 36): pages where a footnote marker and note body could not be paired automatically.
  • 1 finding(s) across 1 leaf(s) (pp. 15): rare character.

If you spot an error, the scan identified above is the final authority. We would rather be corrected than merely trusted.