Encoding Multilingualism: Technical Affordances of Multilingual Publication from Manuscripts to Unicode and OpenType

From Manuscripts to the Computer

“Multilingual publishing” is capacious through its ambiguity, potentially referring to the publication of documents that contain multiple languages, or to publications that may be individually monolingual but collectively represent norms and practices connected with many languages. The latter meaning of “multilingual publishing” is easier to grapple with, given the thousands of years of written attestation for several scripts and their descendants. One can select case studies from different times and places, juxtaposing them and drawing out common threads, and call the result a “history of multilingual publishing.” The former meaning, referencing individual documents that contain multiple languages, offers another lens that allows us to examine how the creator(s) grappled with the challenges of working multilingually.³ In this article, we move between these definitions as needed to illustrate different challenges across the history of text technologies.

The shortcomings of manuscripts come to mind most readily in a primarily digital age: they are time consuming to produce, and—at least prior to lithography—this labor resulted in the creation of only a single copy. They are physical objects that can exist in only one place at one time, unless digitized as an image, which then falls short of digital text in terms of search and retrieval. It is easy to take the constraints of the digital for granted, as our baseline, and fail to consider the affordances manuscripts offer as another framework for multilingual text creation, and one we could aspire to support with our current technologies. What if multilingual writers in digital spaces—from blogs to monographs to social media—could shift between languages and scripts in text as easily as they do so in speech, or effortlessly arrange text within a canvas in ways that made sense semantically and aesthetically? In written manuscripts we see examples of how writers have done this. Some examples include Mark Dickens (2020) examining multilingual manuscript fragments in Syriac, Sogdian, and Uyghur Turkic from the Turfan Oasis in western China; Alpo Honkapohja (2011) looking at Middle English and Latin in a 15th-century medical handbook; and Galina Miškinienė (2009) studying the 18th-century Kitab of Ivan Lutskevich, which is a Lithuanian Tatar manuscript written in Arabic script with a mix of Turkic, Slavic, and Arabic text. Documents from the Parsi Zoroastrian community in India grapple with the challenges of combining left-to-right (Gujarati, the spoken language) and right-to-left (Avestan, the liturgical language) by writing Avestan upside-down relative to Gujarati (Cereti 1996). A Qing dynasty pentaglot dictionary from the end of the 18th century (Chinese, Manchu, Mongolian, Tibetan, Chagatai) involves four writing directions in a single work (Corff 2016). Even a document with unknown language(s) and a novel script such as the famous Voynich manuscript does not pose a technical problem for a human scribe, who can simply write as they see fit.

The shift from handwritten manuscripts to print ushered in tremendous change in the affordances for multilingual publication. While in practice a set of culturally and linguistically contingent norms constrained the creative range of scribes, we find sparks of the manuscript’s creative potential around the edges—sometimes literally, in the case of marginalia. Handwriting is vastly flexible, enabling the writer to freely vary size, shape, position, and stylistic flourishes, all within a single word or across a whole document. The text technologies that followed, from woodblock printing and the printing press to the typewriter, varied considerably in their uptake across different linguistic contexts. However, with the exception of lithography, which facilitated scaling up production of multiple copies of handwritten documents (see Perkins 2017 for a discussion of its use for Perso-Arabic printing), they imposed much stricter constraints on the representation of text—constraints that we have since inherited in the age of the computer. The fixed inventories of characters, tied to sets of professionally produced type (printing press) and possibly keyboards (typewriter), set the precedent for the challenges that writers in multiple languages still face in a computing environment. There has been an uptick in recent scholarship focused on the multilingual history of these text technologies, particularly the typewriter, as they were adopted and adapted across the globe (see Deuchar 2023 [Arabic]; Mullaney 2017 [Chinese]; Lindia 2022 [Amharic]). Nonetheless, much work remains to be done on the global documentation of the hacks, workarounds, and problematic compromises that typewriter and printing press advocates undertook in an effort to force non-Latin scripts into text technology models established for the Latin script.

One finds a similar narrative in the early history of multilingual computing: a new technology was designed, assuming that people would be inputting text in English. Kurt Hensch (2005) describes efforts in the 1960s to adapt the IBM 1401 Data Processing System (a popular mainframe computer) for use in Japan, by only partially implementing one of the three writing systems (katakana) used for Japanese. Using the corresponding IBM 1403 printer to print digital text in Thai gives us an egregious example of the contortions the digital has forced upon those who would write in a non-Latin script. Writers would have to undertake complex script layout, using punched cards for data entry, to record “top vowels and tone marks, followed by consonants, followed by bottom vowels” (Hensch 2005, 25). There is non-trivial mental effort required to visually decompose the text you want to write based on the relative vertical location of the characters’ component parts and then enter the text in that manner. Imagine, for instance, having to begin writing this sentence with all the dots on the i’s, the lines on the t’s, the ascending lines on letters like l, b, and f, and all the apostrophes before moving on to the parts of the letters that appear towards the middle of the line and then entering the commas and descending lines. It’s difficult to imagine a way of writing more divorced from the fluidity of the handwritten manuscript and a major hurdle in the process of expression, which is the purpose of writing and sharing text to begin with.

The mainframe computer was poorly suited for multilingual text input, but with the following generation of hardware—the word processor—we begin to see greater attention being paid to multilingual publishing as a market whose needs were worth meaningfully addressing.

Word Processing

The transition from electronic typewriters to “word processors” (a dedicated piece of hardware that performed a similar task, with additional functionality) spanned from the 1960s into the 1980s. A 1971 New York Times write-up of a trade show labeled “word processing” a “buzz word” for that year’s events, though it glossed the term as “use of electronic equivalent, such as typewriters, procedures, and trained personnel to maximize office efficiency” (Smith 1971).⁴ One major corporation in the transition from typewriters to stand-alone word processor hardware was Wang Laboratories, which was founded by An Wang in 1951 and filed for bankruptcy in 1992, as its major product had been thoroughly eclipsed by the more multi-functional personal computer. Despite Wang’s background as a Chinese immigrant, the word processor was not a device that centered multilingual support. Wang’s vision of being an American entrepreneur running an American company meant an unbending attitude about the primacy of English (Kenney 1992).⁵

A 1989 Popular Mechanics article waxed rhapsodic about word processors, writing off the personal computer as an intimidating new technology that “scared off many potential buyers with its fearsome notion of ‘computing,’ ” whereas “[w]hat millions wanted was a better typewriter” (Scibilia 1989, 72). As usual in Anglophone tech writing for an Anglophone audience, there is no acknowledgment of the fact that the systems profiled in the article and their affordances, including “competent spelling-checkers, with dictionaries ranging from 50,000 to 77,000 words” (72), are tied to a particular language: English. This is particularly ironic given the article’s dramatic introduction: “It might have been an ancient Sumerian scribe, etching his cuneiform onto a clay tablet, who first dreamt of a writing engine that would take the drudgery out of moving thoughts from mind to a more permanent medium” (71). In fact, it would take another 15 years for a Unicode proposal to be submitted for representing the Sumero-Akkadian Cuneiform script in a standard way digitally (Everson, Feuerherm, and Tinney 2004). Prior to that, Sumerian scholars would often transcribe characters by hand or use a custom font.

Word processor hardware was poised to be the successor to the typewriter, representing an opportunity for small companies to address linguistic needs and expectations that had remained unmet due to the lack of an adequately large market to attract attention from large corporations. This also had downsides: a 1988 workshop write-up noted that at the small companies manufacturing these niche systems, “changes in personnel, or in the status of companies could lead to software and hardware being suddenly withdrawn—for example the Ferranti scholar. Some computer systems were thought to be somewhat crude by some users, lacking the friendliness of more mainstream software” (Tyler 1988).

Some companies were more adroit in handling the shift from stand-alone hardware to PC add-ons. Beginning in 1988, Applied Electro-Magnetics sold a stand-alone Indian Scripts Video Terminal with two working modes: English-only and multilingual. The unique Indic script features were later repackaged for use with an existing PC as a “video co-processor plugin card which … enables you to interact in any Indian language or script at the touch of a key,” fully compatible with MS-DOS and text-oriented applications such as word processing software and with a port for plugging in a dot matrix printer. The card could also format the scripts, including bold, italics, shadow, and underlining. It implemented the Indian Standard Code for Information Interchange (ISCII) standard endorsed by the Indian government (discussed below), and language support was sold on a per-module basis with each module supporting English Latin script, Devanagari, and one other “regional script” (Applied Electro-Magnetics, n.d.).⁶

Anglophone trade magazines and newsletters from the late 1980s point to a brief-lived flourishing of word processing software designed for multilingual needs, although not all packages were developed with both user experience (UX) expertise and specialist linguistic knowledge.⁷ Users also faced the frustration of packages only working on particular hardware. The hardware of greatest interest among multilingual computer users at a 1988 workshop was PC-compatible desktops by a large margin, followed by the Apple Macintosh and Amstrad PCW. However, the LocoScript 2 software which boasted support “in every European language! And soon in Arabic, Farsi and Urdu too” was only available for the Amstrad PCW (LocoScript 1988), as was Qalam (1988), a text system for South Asian scripts. INTEXT (1988) for the PC supported several Western European languages, along with Russian, Hebrew, and Arabic. The fact that not all software ads in these newsletters specified hardware compatibility added to users’ confusion; for instance, an ad for the ARROW Publishing System claimed to be “capable of producing countless foreign accents, characters, and symbols commonly encountered in foreign languages” (Arrow 1988). Nonetheless, these freewheeling years saw some degree of font-based support for a tremendous range of both modern and historic scripts, including Mayan hieroglyphs and Glagolitic (the earliest Slavic script). These years saw a less robust computing culture in the USSR, but the rollout of a mandatory computer literacy course on “Principles of Information Science and Computers” caught the attention of Western education researchers (Judy and Lommel 1986).

The high development costs and minimal market for massively multilingual word processing systems had largely kept major tech companies from engaging, but this changed with Xerox’s 1984 multi-million dollar contract with Voice of America (VOA) for a version of their Star workstation that could support other languages (such as a 10,000-character Chinese implementation and Arabic), including physical keyboards (Xerox 1985).⁸ While the VOA’s parent organization in the US government had initially planned for a English-only computer implementation, Chris Kern and a colleague persuaded management of the importance of multilingual support, leading to a multi-year partnership with Xerox, the first time a company had attempted to support such a diverse range of scripts (Kern 2007). Xerox had already positioned its Star workstation as a leader in multilingual computing on the basis of its support for several Western European languages, Cyrillic, and Japanese (Xerox, n.d.)

Market consolidation happened quickly in the early 1990s. By 1994, a set of recommendations around “Practical Choices for Hardware and Software” for linguists and translators only mentioned WordPerfect (PC) and Word (PC and Mac) among its recommendations, although not without caveats.⁹ The author looked to Windows 95 optimistically, which was slated to be released in 27 languages with more to follow (Ball 1994). Even as Word was positioned as a leading option in the mid-1990s, its support for right-to-left languages remained thoroughly broken until 2016 (Stanton 2021). Even better-supported scripts such as Japanese were the result of developing a fundamentally separate version of Word for that script, at the cost of significant developer time and delays, until Microsoft adopted Unicode, discussed below.

The Need for Encoding Standards

Using text, even monolingual text, on computers requires the computer to have a model of how to represent that text in computer memory and how to display that text to the user. A choice of mapping between in-memory representation and abstract text is known as an “encoding,” and it gets displayed using a “digital font.” Early encodings include ASCII and Extended Binary-Coded Decimal Interchange Code (EBCDIC), which primarily represented the basic unaccented Latin alphabet, as well as JIS X 0201 for Japanese.

As more and more documents were being composed on computers, people developed more encodings for their usage, occasionally publishing them with fonts (see Rockwell 1986). For personal word processing, picking an encoding/font for a document would be sufficient, especially if the end product was printed. However, the moment one needed to share documents in a digital format, or write software that displays text, the sender and receiver of the document or software would need to agree on how their computers would interpret the text. To facilitate this, more effort was put into standardizing encodings, conventions which could be made available on most computer systems and selected as needed. This was a better solution than the common workaround—shared custom fonts that mapped other characters to the standard Latin script letters—offered.

The custom font approach was quite functional within scholarly communities, in which a small group of people could agree on an informal standard that text in a given script would be written in a particular font, and that font would be shared throughout the network, passed between professors and students, and among students, on diskettes. For a language such as ancient Sumerian, where text in that language is produced by scholars for other scholars, this approach is adequate. But for living languages, there are considerations beyond simply producing text for an in-group. Ideally, the language should be incorporated into software interfaces. Writers in a language community have different aesthetic needs and desires that call for the use of multiple different fonts. Standards are essential here so that changing the font will not change the text. A language community that can freely use a wide range of digital fonts for different expressive purposes is on its way to reclaiming some of the graphic flexibility lost in the shift from manuscripts to the printing press and typewriters.

The need for encoding standards became more pressing with the growth of the internet and, later, the World Wide Web, as more and more people were sharing documents across computers. Systems would typically track the encoding of a document in some form of metadata, either a part of the document format itself or served separately.

Having to choose an encoding was not the only stumbling block for using text on computers. Some writing systems required complicated rendering support to accurately render ligated text, others required transliterating input methods to allow for input of their large character set on regular-sized keyboards, and yet others required careful processing of bidirectional text (Becker 1984). Multilingual documents suffered greatly from the need to agree on encodings: most encodings supported at most two scripts, typically Latin and some other script. If one wished to use characters from more than one encoding, they were typically unable to, unless they were working with a format like RTF that supported switching encodings on the fly.¹⁰ It was common to encounter gibberish, dubbed mojibake from a Japanese word meaning “character transformation,” resulting from text being read using the wrong encoding, and users would have to manually switch encodings until they found one that made the text legible.¹¹

Unicode

In 1988, Joseph D. Becker, a computer scientist working at Xerox, put forward a proposal for “an international/multilingual text character encoding system, tentatively called Unicode,” described as a “ ‘wide-body ASCII’ that has been stretched to 16 bits [from 7, supporting only 128 characters] to encompass all the characters of all the world’s living languages” (Becker 1988).¹² It was the culmination of work begun in 1987 at a meeting between Apple and Xerox.

The financial logic of large corporations around priorities is visible in this initial Unicode proposal; to make a good case for the encoding, it had to be economically viable and not only a service to the world. Section 1.2 of the overview includes a table that “ranks the world’s writing systems roughly in order of commercial importance, as measured by the total GNP of countries using each system” (Becker 1988). Latin is pegged at 68% of the world GNP, CJK and Cyrillic at 14% each, Arabic at 3%, “Devanagari family” (North Indian Brahmic scripts) and Korean at 1%, and the remaining are negligible. The list is hardly comprehensive, but the point is clear: a cost-benefit analysis would not work out in favor of supporting most of the world’s scripts on a one-by-one basis. The technical structure of Unicode encoding introduced the possibility of a different calculus: implementing Unicode would lay significant technical groundwork for making any living language work.

This initial proposal described Unicode as a “multilingual generalization of ASCII codes,” where “each individual Unicode code is an absolute and unambiguous assignment of a 16-bit number to a distinct character [in one of the world’s writing systems]” (Becker 1988). It lays out five major principles for the system:

• Drawing a distinction between characters (“abstract text content-bearing entities”) and glyphs (“visible graphic forms”).

• Unifying “tens of thousands of equivalent ideographs” that may look different as used in China, Japan, and Korea; this Han unification Unicode initiative quickly led to a dedicated sub-committee and no small amount of controversy.

• Public vs. private: an area of the Unicode code space would be set aside and not used for assigning characters from the world’s writing system. This is important to support experimentation, vendor-specific implementations, and some of the flexibility that users enjoyed through custom fonts prior to the introduction of Unicode.

• Plain vs. fancy text: plain text is defined as purely a sequence of Unicode, in contrast to “fancy text” which contains more information in some manner.

• Process-based design: the Unicode proposal frames text encoding as “exist[ing] solely to support the various processes that act upon text,” including “rendering, filtering, and so on.”

OpenType and Digital Fonts

Unicode made it technically possible, for the first time since the manuscript, to engage in multilingual publishing that could fairly easily juxtapose any arbitrary set of scripts. The migration to Unicode was a serious technical undertaking, but one with considerable value for phasing out separate software versions with different language support, as reported by Microsoft at the 1999 Unicode conference (Pratley 1999). Script encoding by itself, however, is not enough: fonts are necessary for actually rendering the text on a screen or in print. Furthermore, a font that simply has a glyph for each character in the writing system is not enough: some writing systems require complex ligatures, positional forms, or combining rules that must be implemented in order for the text to be legible, let alone natural looking.²³ To add to the complexity involved in producing legible text in many major scripts, the software used to enter and display the text needs to use a shaping engine to turn the input keystrokes into the right combination of ligatures and forms that exist in the font—or into reasonable fallback options if the optimal form did not exist.

Modern fonts use a set of formats descended from PostScript, developed at Adobe in the 1980s with a focus on high-fidelity print representation of glyphs. Jason E. Lewis and Bruno Nadeau (2009) criticize PostScript and its ubiquitous descendant formats, TrueType and OpenType, for continuing to prioritize print as a paradigm, at the expense of emerging 20th- and 21st-century media including 3D environments.²⁴ While the most common narrative of OpenType is centered on the “font wars” of the tech industry in the 1980s, Anushah Hossain (2024) offers a nuanced history that centers the impact on what the software industry designates “complex scripts” (namely Arabic and Indic scripts). While script complexity is in the eye of the beholder, the technical “solutions” for addressing the ligatures and variants in these scripts were indeed complex, particularly prior to the debut of OpenType. Two aspects of OpenType’s feature set—data tables for glyph positioning and glyph substitution—streamlined what had been an awkward set of workarounds, although adoption across South Asia took time due to delays in the full implementation of Microsoft’s Indic shaping engine (Hossain 2024).

With the transition to digital typography, new possibilities also emerged for Nastaliq printing. The first digital Nastaliq typeface, Noori Nastaliq, was developed in 1981 by Monotype and Ahmad Mirza Jamil. Digital typography still operated on some of the assumptions of physical type, including those of the horizontal baseline. To support the large variety of contextual letterforms as well as the diagonally stacked nature of words in Nastaliq, Noori Nastaliq required over 20,000 separate glyphs. The majority of these glyphs were ligatures and therefore could not be mapped onto Unicode. The font could only be used on Monotype’s Lasercomp and was not available to average users. This meant that despite the breakthrough introduction of a Nastaliq font, the majority of Urdu publishing relied on handwritten lithography until the late 1990s.

The development of OpenType allowed for complex, algorithmic glyph shaping and positioning. For Nastaliq, this meant fonts required far fewer glyphs, relying on the font and layout technology to properly lay out the glyphs vertically within the line.²⁵ By inputting individual Urdu letters, the user can receive properly laid-out Nastaliq writing with the positioning and shaping of the glyphs handled automatically by the technology.

However, it should be noted that even with OpenType’s advancements, the underlying model of digital typography is still highly Latin centric. There is still an assumption of a horizontal baseline, and the technology is in some ways “hacked” to display a writing style that breaks those assumptions. As such, while it is now possible to create attractive Nastaliq typography, the technological skill required to engineer a Nastaliq font is prohibitively high, even for accomplished font engineers. In particular, the placement of nuqtas in a Nastaliq font so that they do not clash with one another and appear balanced within a diagonally stacked word remains a key challenge for font engineers. As such, there are very few Nastaliq fonts available compared to other Arabic script styles such as Naskh or Kufic.

Given the crucial role that fonts play in Unicode text readability, advances in technologies that underpin the representation of text on the internet have come a long way in addressing earlier challenges, such as the need to have a font with the appropriate script coverage installed on the display device. The introduction of the @font-face attribute in CSS2 enabled embedding custom fonts for display on a website, to ensure universal legibility even for text in obscure or historical scripts. Still, this required thoughtful planning on the part of web developers, to identify and embed fonts that together cover all the scripts used in a given web page. Google’s OpenType Noto font family has closed this gap by providing open-licensed coverage of most of the scripts encoded in Unicode, in a uniform weight and style for aesthetically pleasing script juxtaposition.²⁶ Despite minor criticisms of the font quality for historical scripts (which have increasingly been handled as outsourced projects rather than developed in-house), and the legibility when displayed at a small size on mobile devices (Park 2023), having a relatively comprehensive, uniformly designed cross-script font family is a significant step forward for multilingual publishing, particularly online and in an ecosystem of apps.

Social Media

What is “publishing” in the 2010s and 2020s? The media and device landscape has shifted once again, such that interacting with digital media (and multimedia) is nearly ubiquitous among adults worldwide, but primarily through mobile devices. Where earlier days of the internet saw content spread across a huge swath of diverse websites—including traditional news, lifestyle, and informational publications moving online, as well as blogs that anyone could write and anyone could subscribe to updates from using RSS—control of the content landscape has moved into the hands of a small number of tech companies through the use of mobile apps. On a website, with enough technical skill, an individual creator could make and implement their own decisions about how to display multilingual text, including using embedded custom fonts mapped to the Unicode private use area, if needed. The level of graphical freedom and flexibility that building one’s own website affords is remarkable and has been increasing as web standards more broadly evolve to support complex and creative layout. A custom-coded website in the 2020s comes closer to the affordances of handwritten manuscripts than anything that preceded it digitally. However, when publishing takes place in the walled garden of an app, users have little recourse if the app does not support their script due to a lack of fonts (now less common thanks to the Noto font family) or mangles the display (still common, especially for lesser-known “complex scripts” such as Javanese).²⁷ While a computer with a keyboard and word processing software are still indispensable for people who regularly produce digital text, much casual communication (social media, but also comments on articles, which in some linguistic contexts involve a polyglot discourse that is by itself “multilingual publishing”) is done using mobile devices. If a script does not have a reasonably easy-to-use mobile keyboard, it increases the likelihood that users will resort to a simpler text input method—which often looks like transliterating into Latin, or choosing to communicate in another language. At the 2023 Face/Interface conference, typographer and typography instructor Hrant Papazian noted that young people are bombarded with images and text on social media associating English with wealth, power, and cultural capital. Cultivating youth pride in their own identity, language, and script—with long-term consequences including language maintenance, particularly in the diaspora—is made much harder when that script is displayed in an unevenly mangled way across operating systems and apps (Papazian 2023; Syauqi 2023).

In a world where publishing is so often mediated by tech companies’ apps, we are dependent on those companies to take an interest—either for their own profit-driven market plays or, less common, as a charity project—in investing in better infrastructure for multilingual publishing on their platforms, particularly for scripts with a smaller user base. In practice, however, the choices that companies make around script support cascade down into other choices about things like content moderation. Facebook recognized that many users in Myanmar were not using the Unicode implementation of the Burmese script but rather one traceable back to a free font that mapped Burmese characters onto different code points within the assigned Burmese Unicode space, using a different logic for how to define characters—meaning no one-to-one mapping with the official Unicode Standard was possible (Hotchkiss 2016). Facebook implemented support for this manner of writing Burmese in its app, automatically analyzing character frequencies to determine whether a post was using Zawgyi or Unicode and picking the display font accordingly. As of 2019, when the government declared an official switch to Unicode with a two-year implementation window (AFP-JIJI 2019), around 90% of Burmese internet users used Zawgyi. But it is difficult to celebrate Facebook’s accommodation of Zawgyi (see also LaGrow and Pruzan 2019) as a win for multilingual publishing in the context of the genocide of the Rohingya people fueled by hate speech that spread like wildfire through Burmese-language Facebook as a result of inadequate Burmese-language moderation, as documented by Erin Kissane (2023).

Infrastructure Futures for Multilingual Publishing

Multilingual publishing is digital. Even alternative publications such as zines are likely to draw upon digital tools at some point in their production.²⁸ While work remains to be done, decades of labor on standards bodies and within the tech companies tasked with implementing those standards means that we finally have technical infrastructure for doing multilingual publishing across almost any set of languages that can, much of the time, simply work.

Who shapes the future of Unicode? Looking at the members list and seeing that nine of the 10 full voting members are large tech companies might cause some alarm, that Silicon Valley is controlling the digital future of the world’s scripts. In practice, the corporations active in Unicode are more interested in ensuring that the organization’s work happens, rather than shaping individual decisions. Scripts and new emoji need to continue to be encoded based on careful review and expert opinion, the Unicode libraries need to be maintained, and the databases need to be up to date. Because the membership dues particularly at the full member level help pay the modest administrative costs of the Unicode consortium, they have recently engaged in outreach to tech companies to encourage them to join—and many have, not least because Unicode runs a project called Unicode Common Locale Data Repository (CLDR) that lowers the bar to software localization.

The difference between voting members (mostly tech companies and a couple government organizations) and non-voting members (a much broader group, including universities, companies outside the United States, and individuals) is mitigated by a very strong culture of consensus and taking in feedback from experts—to the point where voting on anything other than organizational processes almost never happens.²⁹ There is a broader critique one can make about the risks of relying on corporations for so many pieces of the publishing ecosystem: from the hardware we use to compose and read text; to implementing the text stack of encoding standards, fonts, and shaping engines needed to make that text legible regardless of script; to managing social media platforms so publishing isn’t turned into a vehicle for violence. The transformation of Twitter into X following its acquisition by Elon Musk may serve as an object lesson in these risks and a spur to change.

If we can break out of the walled gardens of social media and turn our attention towards creative use of a broader digital publishing stack—including pushing forward with the development of HTML5, CSS, and JavaScript—there are opportunities to imagine a future for digital multilingual publishing that values aesthetics as well as information, reincorporating visual affordances offered by manuscripts that have been sidelined in the evolution of text technologies.

Looking ahead, there is much to be hopeful about: Unicode has moved, and continues to move, towards much greater diversity in gender, background, and life experiences, which will shape the future development of the standard. Likewise, a new generation of typographers and typography researchers—from Ariq Syauqi’s work on Javanese and Balinese, to Niteesh Yadav’s efforts to address typographic shortcomings in augmented reality/virtual reality (AR/VR) environments, to students of Hrant Papazian and Yara Khoury designing creative, beautiful typefaces for non-Latin scripts—are laying the foundation for beautiful, thoughtfully designed digital multilingual publishing.

Multilingual publishing inherently brings with it challenges not faced by monolingual publishing—even more so if that language is English, as so often it is. But the difficulty gap is smaller than it has been for centuries. The best way to continue to push the standards and technical specifications forward is to use them and report on where they fall short. The standards may feel distant and impersonal, but they are written and maintained and revised by people who care deeply about linguistic diversity. Your next moment of technological frustration may be the use case they need to move forward with changes that will make the road easier for the next person.³⁰