11 · The Facing Page

In 1911 James Loeb laid down a single rule of page layout: the Greek or Latin original on the left-hand page, the English translation on the right-hand page facing it. The first twenty volumes appeared in 1912, published simultaneously by Heinemann in Britain and Macmillan in America. The format has a name of its own: en-face.

Three more publishers picked up the same format over the century that followed. The I Tatti Renaissance Library began in 2001, Latin on the left, English on the right. The Clay Sanskrit Library released its first twelve titles in February 2005, setting Sanskrit in Roman transliteration on the left-hand page, and closed the series at 56 volumes in 2009.

The Murty Classical Library of India began in 2015 and stated its reasoning on its own web page — "the page spread is designed so text and translation reflect each other rather than allowing one to decide the other's fate." The companion faces for all six Indic scripts were commissioned from John Hudson and Fiona Ross at Tiro Typeworks. One studio, one brief.

These four publishers did not invent the same format four separate times. Clay states that it was modelled on Loeb, and Harvard University Press says that every facing-page series it has built since follows Loeb's model. This is inheritance, not four independent discoveries. What the evidence does establish is something else: en-face was carried onto scripts from entirely different worlds — Greek, Latin, Sanskrit in transliteration, seven Indic scripts — and it worked every time. Along the way the trim size changed, the typefaces changed, the cover colours changed. What nobody changed was the facing page. The thing that survives while everything around it is altered is the thing carrying the load.

Thai makes a page like that more expensive, because Thai puts no space between words. The W3C's Thai Layout Requirements of 30 April 2024, edited by Richard Ishida, states it plainly: "Spaces separate phrases, rather than words." The consequence is that UAX #14, Unicode's line-breaking algorithm, hands Thai, Lao, Khmer and Burmese off to a separate dictionary-based word segmenter. ICU does this work with its DictionaryBasedBreakIterator. Ordinary page-layout software does not do it for you, so Thai copy has to pass through a segmenter first.

The ambiguity is a property of the language itself, not a defect in the software. Wirote Aroonmanakun of Chulalongkorn University measured it in 2007: he asked twelve Thai speakers to mark, on the same page of text, where each sentence ended. Of 59 candidate positions, only 15 drew agreement from 10 to 12 of them, and 24 drew agreement from only 1 to 3. His conclusion was that "the notion of Thai sentence is fuzzy" — for native readers, not merely for machines. Where the dictionary breaks the text wrongly, the compositor fixes it by hand, with ZWSP (U+200B) and WJ (U+2060).

The second problem is vertical. Thai sits on the same baseline as Latin; Ishida writes that "Thai uses the so-called 'alphabetic' baseline, which is the same as for Latin and many other scripts." The difficulty is therefore not aligning the two languages' lines but the space above them. A composite Thai vowel can involve as many as four glyphs surrounding a base consonant on three sides, a tone mark stacks above that again, and it rides higher still when a vowel sign already sits beneath it. Ishida's summary is short: "Thai tends to add more interline spacing than Latin text does."

This is the point at which this book has to stop and say it does not know. Two Thai line-height ratios circulate on the web: one says at least 1.55, the other says 1.8 or more. Neither traces back to a foundry, a named type designer, or an attributable study; neither appears in the W3C or Ishida documentation; and they contradict each other by a wide margin. One of the two sources announces itself as "the definitive English-language resource for Thai graphic design" while carrying the marks of AI-generated SEO content. So this book prints neither figure. A book about what cannot cross does not get to fake a number about setting two scripts together.

Two things survive. The first is a direction that has a source. Noto splits its fonts into two families: the standard one, with the generous line height that continuous reading requires, and the UI one, compressed for screens. The generous family is the default. The second is the one figure with a name attached to it. Ishii Genta, DTP Director at Human Science Corporation, reports that InDesign's auto-leading default differs by the application's own language — 175% in the Japanese version against 120% in the English one. That is Japanese, not Thai. For Thai, nobody has yet written anything at the level of the page: the "Page & book layout" section of that same W3C document contains one word, "tbd."

The volume you are holding is not set en-face. It is set in alternating paragraphs — one Thai paragraph, one English paragraph, in a single column. I chose it because our production line could already do it, not because it is better. The cost is that a reader cannot read either language as continuous prose. Every time a paragraph ends, the other language gets in the way.

I cannot tell you how large that cost is. No study comparing alternating paragraphs against the other two layouts was found. The one comparison that does exist set interlinear against facing-page — 38 Korean university students read both — and found comprehension at roughly the same level, which means en-face does not trade comprehension away for elegance. As for the read-so-many-percent-faster figures circulating in advertising copy, they trace back to no study at all.

This is the closing volume of one generation of books from this workshop. The previous generation wrote English first and rendered Thai from it, and the result was Thai that was correct in every word and lifeless. This generation reversed the direction on purpose: Thai first, English as the render. The reason is not love of the language, it is instrumentation. This workshop has an instrument for measuring the quality of Thai and does not yet have one for English, so the language we cannot yet measure was put on the render side. That explanation expires the moment the second instrument exists.

The next generation is to be beautiful in both languages at once. That sentence carries no weight as a wish. A specification differs from a wish in saying what will be built, with what numbers, and who can check that it was done.

1. Layout. Use en-face, Thai always verso, English always recto, throughout the volume; open any spread and you can check it. Sourced, from the four publishers above. Rejecting alternating paragraphs is a judgment call, not an experimental result.
2. The typeface pair. Thai and Latin must come from one brief. Either use a family whose foundry has published its own pairing rationale, such as Cadson Demak's Prompt, or commission a new pair from a single studio on the Murty model. The failing side is already documented: in Angsana New the Latin x-height falls below the height of the Thai ก, and Thai users have been setting Latin at 12 pt against Thai at 16 pt by hand for 23 years.
3. Point size. Set Thai roughly 10 to 20% larger than English, then correct by eye on the actual pair, because the base height of Thai letters falls between Latin's x-height and cap-height. That range is a judgment call, not a figure anyone looked up.
4. Line height. Let the Thai column set the shared grid and the English column absorb the surplus. The direction is sourced, from both Ishida and Noto. The figure of 150 to 160% is a judgment call, not a figure anyone looked up, and must be tested against copy carrying stacked tone marks before it is locked.
5. Thai line-breaking. All Thai copy must pass through a dictionary-based segmenter, ICU or equivalent, before it reaches the layout program, with wrong breaks corrected by hand using ZWSP and WJ. Justify by stretching the space between phrases, not between words. Machine-checkable.
6. Language tagging. Every paragraph must carry the Lang value of its own language, per PDF19 and H58, to satisfy WCAG 3.1.2. Never one tag across a spread. The file must conform to PDF/UA. Machine-checkable.

Items 1, 5 and 6 can be checked by machine today. Items 2, 3 and 4 cannot: a sample spread has to be printed and put in front of Thai readers first.

This book does not fully do what it argues for. It argues about what grammar obliges, and then sets its two languages so that they interrupt each other on every page. The six items above are a debt this volume incurred and cannot pay itself. The next volume pays it, and will know whether it has paid in full, because somebody wrote down in advance what was owed — which items have numbers, and which do not.

What could not cross: Line height could not cross, because Thai orthography obliges a tone mark to stack above a vowel sign that is already stacked on a consonant — vertical space Latin never had to ask for. Nor could the line break, because Thai does not write word boundaries into the text. What Latin gets free from its spaces, Thai has to buy with a dictionary.

Overturned if: Someone sets the same bilingual text both ways — en-face and alternating paragraphs — measures comprehension and reading speed with Thai readers, and finds no difference; item 1 falls immediately. And if a foundry or a standards body publishes a Thai-to-Latin line-height ratio with a name and a method attached to it, items 3 and 4 stop being judgment calls and become figures that can be looked up.

Sources: Loeb Classical Library, publisher history https://www.loebclassics.com/page/history · I Tatti Renaissance Library https://itatti.harvard.edu/publications/book-series/i-tatti-renaissance-library · Clay Sanskrit Library https://en.wikipedia.org/wiki/Clay_Sanskrit_Library · Murty Classical Library of India, design and typography https://www.murtylibrary.com/about/design-and-typography · W3C, Thai Layout Requirements, Group Draft Note, 30 April 2024 https://www.w3.org/TR/2024/DNOTE-thai-lreq-20240430/ · Richard Ishida, Thai Orthography Notes https://r12a.github.io/scripts/thai/th.html · UAX #14, Unicode Line Breaking Algorithm https://unicode.org/reports/tr14/tr14-51.html · ICU, Boundary Analysis https://unicode-org.github.io/icu/userguide/boundaryanalysis/ · Wirote Aroonmanakun, "Thoughts on Word and Sentence Segmentation in Thai," SNLP 2007 http://pioneer.netserv.chula.ac.th/~awirote/ling/snlp2007-wirote.pdf · Microsoft typography-issues #1116 on Angsana New, January 2024 https://github.com/MicrosoftDocs/typography-issues/issues/1116 · Cadson Demak, Prompt https://github.com/cadsondemak/prompt · Type Together, Beginner's Guide to Multiscript Typography https://www.type-together.com/multiscript-typography-guide · Noto usage guidance https://notofonts.github.io/noto-docs/website/use/ · Ishii Genta, Human Science Corporation https://www.science.co.jp/en/nmt/blog/27655/ · the interlinear-versus-parallel comparison with 38 Korean students https://www.researchgate.net/publication/337704420_The_Effects_of_Dual-language_L2_Texts_on_English_Reading_Comprehension_and_Text-type_Preferences · WCAG 2.0 Understanding SC 3.1.2 Language of Parts https://www.w3.org/TR/UNDERSTANDING-WCAG20/meaning-other-lang-id.html · WCAG technique PDF19 https://www.w3.org/TR/2016/NOTE-WCAG20-TECHS-20161007/PDF19 · WCAG technique H58 https://www.w3.org/TR/2016/NOTE-WCAG20-TECHS-20161007/H58 · the two sources whose figures this book does not accept, named so they can be checked https://www.samui-infotech.com/the-importance-of-thai-typography-in-web-design/ and https://thaigraph.com/faq/