Close reading
Reading a book you cannot read
Much archival material is not in a language its researcher reads fluently, and the usual answer — paste a page into a translation box — throws away the thing that makes it evidence. Archēglyph keeps the original beside the translation, says what produced every line it derived, and names the romanisation it used.
Four columns, and you choose which
Hide the columns you do not need and the rest widen. Point at a passage and the same passage lights up in every other column, because each one is cut at the same place in the text rather than merely running alongside it.
| The page image, as scanned | Original transcribed, in its own script | Romanisation · IAST machine-written, signed | Translation · English machine-written, signed |
|---|---|---|---|
| আমার সোনার বাংলা, আমি তোমায় ভালোবাসি। | আমার সোনার বাংলা, আমি তোমায় ভালোবাসি। | āmāra sonāra bāṃlā, āmi tomāẏa bhālobāsi. | My golden Bengal, I love you. |
| চিরদিন তোমার আকাশ, তোমার বাতাস, আমার প্রাণে | চিরদিন তোমার আকাশ, তোমার বাতাস, আমার প্রাণে | ciradina tomāra ākāśa, tomāra bātāsa, āmāra prāṇe | Forever your skies, your air, in my heart — |
Rabindranath Tagore, Āmār sonār bāṃlā — opening lines, first published in Bangadarshan, 1905.
Hover or tab to a line to see it in every column.
- The page
- The page image, straightened and zoomable. Everything else on the screen is drawn from it, and it stays in view so you can check.
- The original
- The text as transcribed, in its own script and reading direction, with a note of what read it.
- The romanisation
- A Latin transliteration in a named scheme — IAST, ALA-LC, Hepburn, Pinyin — so it can be cited and compared, not just sounded out.
- The translation
- Into whichever language you work in, not only English. Passage for passage with the original, so the columns line up.
The romanisation has a name
The same word, romanised under three conventions, gives three different strings. A transliteration whose convention is not stated cannot be compared with other scholarship or cited in it — so the scheme is chosen for each book, named in the instruction the machine is given, and shown beside the text it produced. Change the scheme and the text is written again, not relabelled.
| Word | IAST | Harvard-Kyoto | ITRANS |
|---|---|---|---|
| कृष्ण Kṛṣṇa (Sanskrit) | kṛṣṇa | kRSNa | kRRiShNa |
| বাংলা Bengal, Bengali | bāṃlā | bAMlA | bAMlA |
IAST, with its diacritics, is the scholarly default for Indic scripts; the other two are keyboard-friendly schemes common in older digital texts. None is wrong — but a reader has to know which one they are looking at.
Chapters, then people
An anthology read as one block confuses the characters of unrelated stories. So chapters are found first — from the table of contents, with each title located by searching the text for where it actually begins, because front matter means printed page 37 is rarely the thirty-seventh page of the scan.
People, places, organisations and dates are then listed chapter by chapter. A machine decides what they are; it is never trusted to say where they appear, because a machine asked for positions produces plausible ones that are wrong. Every mention is found by searching the text itself, so clicking one takes you to the exact passage — and a name that appears nowhere in the text is discarded rather than shown.
Scripts it reads
Where sentences end, where words break, which way the text runs and how it is set are handled language by language. Bengali ends a sentence with a danda, Chinese has no spaces between words, Arabic runs right to left — none of which a tool shaped around English gets right by accident.
- Indic
- Bengali, Hindi, Marathi, Sanskrit, Tamil, Punjabi, Odia, Gujarati
- Right-to-left
- Arabic, Persian, Urdu, Hebrew
- East Asian
- Chinese (Simplified and Traditional), Japanese, Korean, Thai
- Other scripts
- Cyrillic, Greek, Armenian, Georgian, Ethiopic
- Latin
- English, French, German, Spanish, Italian, Portuguese, Dutch, Latin, Turkish and more
Two honest caveats. Chinese, Japanese and Thai have no spaces between words, so searching them by keyword works on runs of characters rather than on words — serviceable, but not what a dedicated tool would give you. And a translation is a machine's reading of a text, not the text: it is labelled with what produced it precisely so you can weigh it rather than take it.
Reading your own material
Upload a PDF to a collection, let Archēglyph read it, then open that file as a book and choose its language. The reader uses the text already read for the collection — it does not read the scans a second time — so the book and the collection never disagree about what a page says.