How Not to Localize Your Game: a Case Study

Released in 2019 in South Korea and 2022 in other parts of the world, Lost Ark is an action MMORPG in the tradition of World of Warcraft and Diablo, offering a massive virtual world filled to the brim with quests, characters, and monsters. Since its global release, the game has ranked consistently among the most played games on Steam and has received favorable reviews across the board.

However, Lost Ark is also a prime example of how game localization matters in upholding the quality of the game—and how failing to localize your game properly can significantly deter player immersion. Despite being one of the most played games on Steam, Lost Ark is host to a number of translation and localization errors.

Today, we’re taking a look at some of the mistranslations and localization failures in Lost Ark as a case study of how such errors can inhibit player immersion in the game and thus impact the gamer community’s attitudes toward the game and its publisher/developer.


Cinematic Subtitles

In a cinematic short released for Lost Ark’s South Bern World Quest, a dialogue between a battalion and their commander has been mistranslated, catching the attention of some Lost Ark fans. Here is a screenshot of the mistranslation:

The sentence “We are the window of Bern!” makes little sense. What does it mean to be the window of something? In fact, it seems counterintuitive for a troop of soldiers to call themselves a window: an object designed to let things through and penetrate. The original dialogue in Korean is “우리는 베른의 창!”, which, when translated correctly, means “We are the spear of Bern!” The mistranslation is due to the fact that the word 창 (chang) is a homophone meaning both spear and window.

The word for spear was actually translated correctly a mere two sentences ago, when the commander tells the troop, “Knights of the Sun, hold the spear!” Of course, this sentence comes with its own translation problems; given the context, spear should be in the plural form, given there are dozens of knights. Furthermore, “hold” isn’t the right word here; perhaps “grasp your spears” or “hold up your spears” might’ve been a smoother-sounding translation.


In-Game Dialogue

If cinematics—perhaps the most carefully crafted representations of any given game—are riddled with translation errors, it’s fair to assume that the actual game itself will be host to many, many more. That is certainly the case with Lost Ark, whose in-game dialogue falls short at times, like in this mistranslation:

Here, the dialogue isn’t actually referring to a cow. The original dialogue for this scene is “소, 솔라스님!”, in which the word 소 (so, meaning cow) doesn’t actually signify cow, but rather is used to express that the speaker is, in fact, stuttering. The correct translation, then, would be “So… Solas!” (Note that the name Solas has also been mistranslated as Solaras.)


UI Localization Errors

What Lost Ark fans have most complaints with is the game’s general UI localization, which has sparked intense debates and beef sessions on the game’s online forums, like this massive thread on localization errors, as well as this thread on translation and localization errors.

Game UI localization is an especially difficult task to accomplish, not only because game UIs are encoded and displayed in very different settings and through different programs, but also because testing them and checking for errors take up a great deal of time.


There are issues with consistency in orthography, seen in the pictures below:

Image credits: Ragestyles, Lost Ark Forums



Image credits: Faranim, Lost Ark Forums


On the same page, the same name (벨렌) is spelled Bellen and Velen; on another page, it’s Selen and Celene. It’s clear that the localization has been automated, or that different translators were involved in translating different parts of the game interface. Such inconsistencies can confuse the player—are Celene and Selen different character, or are they different spellings for the same person? What if a quest directs the player to Selen, but the player can only locate Celene?

Also note that, in the former example, the land of 베른 is spelled Vern, not Bern, as was the case in the cinematic. Turns out, Vern is the official spelling of the location for most versions of the game that has been published in most North American and European countries. Such inconsistencies inhibit player immersion and interest in the game, leading gamers to doubt the game and its developers’ QA methods and their sincerity to the game.

Inconsistencies are also spotted in item names:

Item Name 1 Item Name 2
Job Offer Wallpaper Help Wanted Poster
Memoirs of Avesta Avesta Journal
Timeworn Fighter’s Blade Timeworn Gladiator Blade
Amalone’s Diary Amalone’s Journal
Heavy Walker Report Report on Heavy Walker

As was the case with inconsistencies in character names, inconsistencies within item names also confuse players. It’s difficult to discern whether Timeworn Fighter’s Blade is the same item as Timeworn Gladiator Blade (surprise: they are different names for the same item!) These errors betray the lack of proper localization on the part of Lost Ark. It seems that there were no glossaries in use to ensure that different translators could render item names in standardized, consistent manners.


In some cases, errors in localization directly impact user gameplay in rather detrimental ways:

Image credits: Ragestyles, Lost Ark Forums

In omitting a minus sign in front of 25%, the Cursed Doll item seems to be an optimal choice, as it provides 25% additional health at level 3. However, the healing is actually reduced by 25%. Major errors like these directly impact gameplay, as they pertain less with the storyline and more with the actual mechanics of the gameplay itself.


Quality of Translation

Lost Ark also employs a number of ungrammatical or awkward translations that detract from the storyline and gameplay. A good example of this is the abbreviation of the word “level”—an integral component of in-game characters—as “LV”, not “LVL”. The former is a Koreanized abbreviation used on the peninsula, and the latter is what the Anglosphere is more familiar with.

Other seemingly minor problems, yet with major ramifications, include frequent spacing errors (spaces where there should be none, and no spaces where there should be one) and capitalization issues.


Note that the button says “raid Quit”, instead of “Quit raid” or “Exit raid”. Assuming from the word order, the mistranslation seems to be a direct, word-by-word translation from Korean. Image credits: Ragestyles, Lost Ark Forums


An example of user feedback regarding improperly localized gameplay. Image credits: Wilczasty, Lost Ark Forums


An example of a missing spacing that hinders legibility and gameplay immersion. Image credits: Ragestyles, Lost Ark Forums


More grammatical mistakes spotted by users. Image credits: Ragestyles, Lost Ark Forums


An example of a UI error that failed to take into account how word length expands when translating from Korean into English. Image credits: Ragestyles, Lost Ark Forums


Many of these issues have now been fixed, but only after leagues of users pleaded the developers with pages of screenshots taken of all sorts of mistranslated strings and poorly localized interface components. While Lost Ark should be commended for its willingness to admit to its mistakes and reflect user feedback in its localization process, this is an inefficient way to correct its mistakes. It’d be much better not only for its developers but also its player base to solve these problems before the game even went public.


Here at Sprok DTS, we offer consultations for game developers and publishers to ensure that localization is taken into account from the early stages of game development. Our linguists work directly with companies so that their player base can immerse themselves in the games without being distracted by mistranslations or localizations errors, big or small.

Expand your possibilities with Sprok DTS’s professional, experienced localization services.


Desktop Publishing: on the Technical Formatting of Translation

One of the main differences between translation and interpretation is that each medium faces a different constraint to their art. The interpreter speaks into open air as if to conjure their wildest imaginations, but are able to do so only within the strictest temporal constraints. On the other hand, translators are fettered by the confines of the page: word limits, page counts, characters.

The problems therein are countless. When translating from a compact, agglutinative language such as Finnish to an expansive, analytic language such as English, the sheer volume of text can expand to up to 40%. Even between languages with similar roots—German and English, for example—text lengths can expand anywhere from 5 to 40%, depending on the domain of the text and the translational style.¹

The problems manifest themselves on paper. PowerPoint documents are suddenly disjointed as words spill out of their designated boxes. In formal documents, annotations are left asunder, and page numbers are rendered meaningless.


Image credits: Etienne Girardet on Unsplash

Introducing desktop publishing

To fix these problems, Sprok DTS has implemented a crucial step in its translation process: desktop publishing. Sprok DTS employs professional desktop publishers whose main role is to review the physical contents of a translated file—no matter what kind of file format it may be—and make sure that the target text exactly matches the formatting of the source text.

The process of desktop publishing can take place on a number of different programs. For documents, our publishers may use the classic array of word processors and platforms: Microsoft Word, PowerPoint, Excel, and Adobe Acrobat. For more technical, visually laden translation files, programs such as InDesign, Photoshop, and Illustrator may be used.

With the recent rise in the popularity of audiovisual content, yet other software such as After Effects, Premiere Pro, and a host of subtitling editors may be used.


The implications of desktop publishing

While desktop publishing is not a scene one imagines when thinking of the typical translation process, it is an indispensable part of translation. Many language service providers do not offer desktop publishing services, but Sprok DTS employs a number of professional desktop publishers whose services may be utilized upon customer request.

The desktop publishers handle everything from the slightest formatting problems to larger structural changes that may be required to retain the format of the source text. For example, some fonts may not display characters unique to specific languages, such as umlauts (ä, ö, ü) used in German, Hungarian, and other languages; cedillas (ç, etc.) of various kinds in French, Hebrew, Latvian, and other languages; and so many other diacritics applied to the Latin alphabet used around the world.

On the other hand of the orthographic spectrum are languages that use entirely different scripts, such as Arabic (and its various variations), Chinese (Mandarin and otherwise), Japanese, Korean, and yet many other languages. These alphabets function in fundamentally different manners than Latin-based alphabets. Arabic, for one, is written right to left, entailing extensive changes to the target text so that it works in the same way as the source text. Characters East Asian scripts may be formatted differently due to orthographic rules or differences in coding.

Aside from linguistic issues, there remain technical problems. PDF files, for example, are notoriously difficult to manage and render. Graphs, images, and other visual content may have to be shifted around to accommodate changes to the text. All this is done to ensure that the resulting final product resembles, in all ways minus the language, the source text.


Image credits: Martin Faure on Unsplash


Where does desktop publishing fit?

When exactly does desktop publishing happen in the grand scheme of things? While slight variations may exist in the order of translation processes, here at Sprok DTS, we follow a strict four-step regimen consisting of 1) pre-processing, 2) language processing, 3) postprocessing, and 4) linguistic sign-off.

In the pre-processing phase, source texts are obtained from our customers and optimized for convenience and efficiency. Files are neatly rendered into more manageable sections and formats our linguists can access and edit. The pre-processing phase includes optical character recognition processes as part of an overall character data optimization scheme not only to benefit our linguists hard at work, but also to improve the eventual quality of the target translation.

The second phase—language processing—is the most recognizable step in the translation process: namely, translation. Our linguists work on the text, after which the text is reviewed and revised by another proficient bilingual linguist for any issues that may have occurred during the initial translation process. Finally, the revised draft is proofread for clarity, style, and overall quality in a monolingual proofreading process. Here at Sprok DTS, we call this three-step process the “TRP process”: an effective method at minimizing translation costs while maximizing translation quality and efficiency.

The third phase is postprocessing, in which the translated file is formatted to match the source text. Desktop publishing falls into this category, equivalent in importance as any other step. Here at Sprok DTS, we believe in the importance of desktop publishing in ensuring our customers the maximum quality of our translation services. For this reason, we duly note and pay attention to the rigors and difficulties that postprocessing procedures may face, so that we can resolve any textual, technical issues present in the visual layout of the document.

Lastly, we have the linguistic sign-off, at which point the translated, reviewed, proofread, and postprocessed document is reviewed in its entirety before it is submitted back to our customers.

Should you choose to work with Sprok DTS, desktop publishing will be part of the third postprocessing step, after our TRP process and before the linguistic sign-off.


Image credits: Markus Spiske on Unsplash

What’s next?

Here at Sprok DTS, we proffer translation services that not only adhere to the strictest of international standards, but also cater to your most immediate needs.

If you’re searching for a language service provider for your translation needs, visit for more information on our services and philosophy.





The Challenges of Medical Translation

In 2004 to 2005, France saw one of the most major incidents of medical mistranslation in history. A hospital in the eastern commune of Épinal accidentally provided massive overdoses of radiation to patients with prostate cancer, resulting in several deaths and dozens of affected patients. The cause, according to an article by Afaf Steiert and Matthias Steiert, was a “dose defining software used for cancer therapy,” which ran only on English without a French user guide. “The hospital’s administration relied on bilingual staff members who used the software,” writes the Steierts, leading to mistranslations that ultimately ended in great tragedy.

This is but one of many instances of mistranslations in the medical field. Unlike other modes of translation—legal, patent, science—medical translation is more closely connected to the livelihoods of people. Sure, other forms of specialized translation are also prone to mistranslations that result in tragedies. Mistranslated legal texts could massively hinder legal proceedings and possibly and wrongfully convict an innocent person; mistranslated technical documents could possibly harm a worker by providing wrong instructions for a piece of heavy equipment; mistranslated commerce documents could possibly result in large monetary losses for an enterprise. Medical translation, however, almost always affects human lives at some level, sometimes to fatal degrees as can be seen in the preceding example.

Today’s blog deals with the nuances, history, and difficulties of medical translation. We hope that the contents of this blog serve to enlighten you in the perils and importance of medical translations and provide a deeper understanding of the important role it has in our lives.


History of medical translation

Like all forms of interlingual exchanges, medical translation has been a part of humankind for as long as it can be documented. In the textbook Medical Translation Step by Step, authors Vicent Montalt Resurrecció and Maria González Davies traces the origin of medical translation to Ancient Mesopotamia, where “medical, chemical mathematical, and astrological knowledge was gathered, organized and stored in cuneiform symbols written on clay tablets, some of which contained information in different languages such as Ugaritic, Akkadian, Sumerian, hittite, and Hurrian.” Later, in the 5th century BCE, comes Hippocrates, known for his role in drafting the Corpus Hippocraticum and founding the tenets of ancient medical practices. Montalt Resucció and González Davies charts the history of medical translation thereafter, to the Greek city of Pergamon (an “important center of scholarship” between Arabic, Greek, and Latin medical scholars), the School of Toledo (featuring “collaboration between Islamic, Christian, and Jewish scholars” leading to translations in Romance Castilian and other European languages), the Middle Ages, and the Renaissance (when Latin established itself as the lingua franca of medical knowledge).

This is only in the Western hemisphere; other civilizations elsewhere—such as Asia, Africa, the Americas, etc.—have developed and continued their own histories of medical practices, charted and codified in numerous documents spanning centuries and millennia.


Fields of medicine

One of the reasons why medical translation is so difficult is the sheer variance and range of  specialities and studies involved in medicine. “Medical translation involves the communication of knowledge generated and needed in various specialties,” writes the authors. Following are some examples of medical specialties that compound the field of medical writing:

  • Internal medicine
  • Obstetrics and gynecology
  • Orthopedics
  • Pediatrics
  • Psychiatry
  • Surgery
  • Pharmacology

The authors also note that medicine doesn’t exist in isolation; often enough, medical texts overlap with other areas of knowledge and expertise, such as “anthropology, psychology, sociology, economics and law,” for example. The interconnected nature of medicine makes it all the more difficult for translators to equip themselves with all the knowledge necessary to translate a particular medical text, as the required knowledge doesn’t just stop at a certain medical field, or even the field of medicine.



Perhaps the most difficult of all factors involved in medical translation is the complex, historical, and confounding nomenclature and terminology utilized by medicine. In the English language, medicine derives largely from both Greek and Latin terminology. Furthermore, medical language—like other specialized forms of translation—utilize grammatical processes such as nominalization and distillation to create non-colloquial sentence and word structures, which further add to the difficulty of medical translation. 

Languages that share these Greek and Latin roots are often prone to miscommunication when being translated among themselves. For example, French, Spanish, and English have striking similarities in their medical terminology, as can be seen in the following example provided by the authors:

English French Spanish
Embryonal Embryonnaire Embrionario
Germinal Germinative Germinativo
Carpal Carpien Carpiano
Aortal Aortique Aórtico
Migrant Migrateur Migrador
Depressant Dépressif Depresivo
Coronary Coronarien Coronario
Flagellate Flagellé Flagelado
Calmative Calmant Calmante
Ulcerative Ulcéreux Ulcerativo, ulceroso
Appendiceal Appendiculaire Apendicular
Embolism Embolie Embolia
Olfactory Olfactif Olfatorio

However, English isn’t Latinate at its root; due to its complex Germanic and Norman linguistic history, English has a “double-layered medical vocabulary,” having two words for a single concept where other languages would only have one. Take the following for example:

English French Spanish
Coagulation, clotting Coagulation Coagulación
Myopia, short-sightedness Myopie Miopía
Cicatrization, scarring Cicatrization Cicatrización

The authors also note examples in which concepts that only have one term in English may have more than one terms in other languages:

Danish English
Blindtarmsbetændelse, appendicitis Appendicitis

Often, two synonymous terms for a single concept differ in registers; the Latinate term is considered to be of a “high register,” whereas the Germanic/Norman term is “low register.” This might not be the case, however, in other languages, leading to confusion:

English French Spanish
Birth defect Malformation congénitale Malformación congénita
One-egg twins Jumeaux univitellins Gemelos univitelinos
Growth of germs Prolifération microbienne Proliferación microbiana
Sweat canal Canal sudorifère Canal sudoríparo
Caked breast Engorgement mammaire Obstrucción mamaria
Taste buds Papilles gustatives Papilas gustativas

Then, of course, there are cases in which similar-sounding terms with similar origins may refer to completely different concepts across languages, such as “schizophrenia, chronic bronchitis, and peptic ulcer, [which] have different meanings in German, French and English.” Another example is the English “amygdala,” which “does not coincide semantically with the Spanish ‘amígdala’.” In some cases, certain languages might not even differentiate certain anatomical features; “French and German have no terms for knuckle, and Russian has no distinction between hand and arm.”

Another difficulty in medical terminology is the overlapping usage of Greek and Latin roots in nomenclature. The authors provide the following example of “etymologically synonymous morphemes” that are present in the English language:

English Greek Latin
vein phléps phlebós (phleb-, phlebo-) vena (ven-, vene-)
blood vessel aggeion (angi-, angio-) vas (vas-, vasculo-)
breast mastós (mast-, masto-) mamma (mammo-)
mouth stoma stómatos (stom-, stoma-, stomat-) os (or-, oro-)
face prosópon (prosop-, prosopo-) facies (facio-)
eye ophthalmós (ophthalm-, ophthalmo-) oculus (oculo-)
nose rhís rhinós (rhin-) nasus (nas-)
ear oûs otós (ot-, oto-) auris (auri-)
tongue glóssa (gloss-, glossa-) lingua (linguo-)
spine rháchis (rachi-, rachio-) spina (spin-, spino)
white leukós (leuk-, leuko-) albus (alb-, albi-)
milk gála gálaktos (gal-) lac lactis (lac-)
sugar glykys (gluc-, gluco-) sácharon (sacc-)

In some cases, there are numerous names associated with a single affliction, such as “Basedow’s disease, Flajani’s disease, Graves’ disease, Parry’s disases, all of which refer to exophtalmic goiter.” Sometimes, languages might employ names based on people where others do not; the English terms atrophic arthritis, chronic infectious arthritis, proliferative arthritis, and rheumatoid arthritis are referred to as maladie de Heine-Médin, paralysie spinale infantile, and polyomyélite anterieure aigüe in French.

Other times, intralinguistic differences pose problems for translators. For example, the chemical term “phosphoglyceride” is now referred to as “glycerophospholipid” in modern times; some texts will feature both terms, which confuse not only readers but translators as well. Speaking of chemical terminology in medicine, different cultures and regions use different names to describe the same chemical or pharmaceutical object. An example of this is “the international non-proprietary name for Nolotil®, one of the most common analgesics in Spain” which is metamizole. In the Anglophone world, metamizole is called “dipyrone”; in India, China, or Russia, it is called “analgin”; France calls it “noramidopyrine”; and Hungary and post-Yugoslav nations might call it “noraminophenazone.”

Not all is difficult, however. Due to the prevalence of the English language and its influence in the medical sector, there exists loanwords and calques, accepted by non-English languages. Examples include: stress, bypass, feedback, shock, test, borderline, etc. 


All these factors, among many not mentioned, present difficulties for translators seeking to accurately translate from their source language to their target language of choice. Modern developments in medical translation and translation technology have made it easier for translators to carry out more accurate, proper translations in hopes of decreasing the amount of mishaps and miscommunications that might possibly prove fatal to patients.

However, translators still must undergo rigorous education and studies to make sure that their techniques and knowledge are sound; medicine develops with each passing day, and medical translators must keep up to date with changes in the field. 


González Davies, M. & Montalt, V. (2015). Medical translation step by step: Learning by drafting. Routledge.
Steiert, A., & Steiert, M. (2011). Medical translation basics. MultiLingual, (July/August), 27–28.


An Introduction to Legal Translation

Although our daily lives seem so far away from the high echelons of judicial courts, laws are still very much an integral part of our laws. From the moment we are born, our existence is documented, legalized, and codified by the laws of our country. Our daily interactions all take place under the regulations of a pre-existing law. Buying a drink from a local cafe? We have zoning laws, business regulations, and laws on commercial transactions to thank. Getting married? Nuptial law’s got your back. 

When these laws and regulations exist within a country in which a single, de facto or “official” language is spoken, not many problems exist—of course, aside from the difficulties of interpreting the notoriously difficult language of laws, complex and illegible in all languages. However, in countries that have more than one spoken/written languages or in cross-linguistic interactions, these legal texts must be translated. In today’s globalized world, legal translation is more needed than ever so as to facilitate the very legal actions that we partake in every day. This is all the more true for businesses, governments, and other entities that deal with international commerce, trade, and other forms of interactions.

Like any other form of specialized translation, legal translation is difficult and faces its own unique set of challenges. From traditional modes of writing to cultural differences in legal procedures, legal translation is one of the most difficult specialized modes of translation, as it requires thorough understanding of the legal systems of not only the country of the source text but also that of the target language. This blog post seeks to explain the features, challenges, and methods of legal translation as it is used in the world today.


What is legal translation?

Legal translation is, in the barest of terms, the translation of one legal text from one language to another. It sounds simple, but nothing about the process is simple. For one, language used in legal texts are highly specialized forms of languages that require in-depth understanding of the legal system. In that sense, legal translation is not only a word-by-word, sentence-by-sentence rendering of the source text, but essentially a transposition of the structure of the legal text into a wholly different legal structure. Many factors, such as cultural differences in legal proceedings and linguistic differences in terminology, inhibit translators from carrying out straightforward translations, as would be expected from other forms of translation.


What are the features of legal texts?

Before understanding the specific challenges of legal translation, we must first understand how exactly difficult legal texts are, as well as the specific features of legal texts that make them so difficult to translate. In his comprehensive article, “Linguistic Features of Legal Texts: Translation Issues,” linguistics scholar Maurizio Gotti explores the treacherous depths of legal texts and the various ways they confound readers.

For starters, legal texts must maintain an “avoidance of ambiguity and precision of interpretation.” It is critical that legal texts be as concise and accurate as possible so as to minimize misunderstandings, which have detrimental effects on those affected by such failures. How such ambiguity is prevented, however, is the main problem; Gotti writes that, in legal texts, “old formulae are preferred to newly coined words because of their centuries-old history and highly codified, universally accepted interpretations”. Hence, the fancy, uptight words we find in legal texts: hereto, in forma pauperis, etc. Such a strict adherence to tradition, Gotti notes, “also reflects [the legal texts’] close link with the ancient practice of using special formulae for oaths or appointments, for drafting edicts and statutes, for issuing laws, conferring honours, or assigning property”. Gotti provides two examples of how such old-fashioned language exists in legal texts, both in English and Italian:

salvo che la legge disponga altrimenti > unless otherwise provided by law

salvo quanto previsto dall’art. […] > without prejudice to the provisions of Article […]

As a result of this adherence to tradition, legal texts suffer from a “high level of redundancy” in the form of long-winded, pleonastic descriptions and “use of lexical items.” Here, we come to a paradox: legal texts are supposed to be as concise and precise as possible so as to avoid misunderstanding, but in doing so, legal texts resort to difficult linguistic structures and phrases that end up befuddling readers. Gotti explains how, specifically in English, “legal drafters often employ two interchangeable terms for the same concept”, examples being “new and novel, false and untrue, made and signed, terms and conditions, able and willing”. Such usage derives from the age of the Norman Invasion, a period in English history where both English and Norman French were spoken on the island. Such linguistic obstacles make it difficult for translators to properly translate between English and other languages.

All this leads to the notoriety of legal texts, of their “great length and complexity.” Gotti provides the following examples:

This Agreement, effective as of the first day of April 2011 between Dale Johnson Ryder Warren, an Association organized and existing under the laws of Switzerland (‘Grantor’), its successors and assigns, and DJRW Johnson Ryder Simpson & C., its successors and assigns (‘Member Firm’) […] 

The Tenant will […] pay for all gas and electric light and power which shall be consumed or supplied on or to the Property during the tenancy and the amount of the water rate charged in respect of the Property during the tenancy and the amount of all charges made for the use of the telephone on the Property during the tenancy or a proper proportion of the rental or other recurring charges to be assessed according to the duration of the tenancy 

In the first example, a single noun phrase (this agreement) is “specified by 42 other words providing information about its validity, the names of parties involved, their legal address, the names assigned to them later in the contract and mention of the successors to the two signatories”. The second example show cases how legal texts “exhibit a high number of postmodifiers and relative clauses, in contrast to other kinds of specialized discourse, which instead prefer pre-modification.” If you have read our previous post on technical translation, you will know that technical texts exhibit a tendency to utilize pre-modifying nominalization to elaborate and specify, such as in the example depth-control drill operation. All this is to say that legal texts do not grant the benefit of the doubt; everything must be expressed in full so as to avoid ambiguity, and in doing so, legal texts are laden with modifying clauses.

In the end, despite the legal systems’ need to be both unambiguous and precise, “compliance with tradition is stronger in legal discourse than is the search for conciseness.” Legal drafters forgo precision, rather sticking to formal, antiquated writing. However, there are some ways that legal drafters work around this tendency, namely by using adverbials to refer to parts of the text itself. Gotti notes how, in the following example provided, “textual-mapping” adverbials “hereto, herein, hereof, and thereto” are used for clarification:

Whereas, Johnson Ryder Archer & C., Johnson Ryder Chester & C., Dale Johnson Nelson & C., Dale Johnson Stokes & C., Grantor, Johnson Ryder International a partnership, and Dale Ryder Warren an association, have entered into the Component License Agreement, effective as of 1 April 2011 (‘Component License Agreement’), a copy of which is attached hereto as Appendix B (without Appendices A and B attached thereto which are Appendix A hereto and a form of this Agreement) and made a part hereof as if fully recited herein and to which the Member Firm agrees to be fully bound as if originally a party thereto […] 


Challenges of legal translation

Gotti notes how recent decades have seen a great demand for legal translation due to “globalization and increased contact and exchange between people and states.” To facilitate the ever-growing need for accurate legal translation, legal entities and drafters have come up with the “principle of legal equivalence… which underlines the consideration of the legal effects that a translated text will have in the target culture.”

Why is equivalence important in legal translation? It’s because legal discourse is so “culture-laden” and because there must exist some “formal correspondence between equally authoritative versions of the same text.” Not only must the target text be as equally authoritative in nature, but it must also be accurate and abide by the legal infrastructure of the target culture and government. This requires the legal translator to “undertake a process of conceptual analysis by means of which he is able to identify and assess the most important differences between the source and target legal systems as they are expressed in the text to be translated.” In simpler words: translators not only translate, but assess the most important parts of the source text and transpose it into pre-existing (or not) terms and concepts in the target language. 

However, there are numerous difficulties that lie in achieving these goals in legal translation. Firstly, there are linguistic constraints; the sheer differences between language inhibit translators from easily making a transposition of one word/concept/sentence to another. Gotti provides a particularly illuminating example of such difficulties when translating between Chinese and English. Mandarin Chinese, like many languages, do not make clear distinctions in inflection (singular/plural nouns and verbs, for example) and grammatically significant elements (such as definite and indefinite articles). 

According to Gotti, a literally translated Chinese phrase might look like this: “cause serious environmental pollution accident.” This can be translated into either

  • cause a serious environmental pollution accident
  • cause serious environmental pollution accidents

Such ambiguity in linguistic differences poses problems: “does a person have to cause more than one such accident to be criminally liable, or just one accident is sufficient?”

Another example is Mandarin Chinese’s omission of certain conjuctions, such as the literally translated phrase “resulting in serious loss to public private property.” Does this mean “to public and private property” or “to public or private property?” The same applies to “personal injury and death or personal injury or death” both of which are rendered as “personal injury death” in the transliterated Mandarin original.

Another difficulty of legal translation is the “drafting traditional and… stylistic conventions” of a particular legal system. The legal systems of different countries have very different backgrounds, traditions, and methodologies in dealing with legal issues. One of the most notable distinctions within legal systems are the differences between systems based on civil law and those based on common law, the former used more around the world and the latter used mostly in Anglophone nations. 

An example of such differences in tradition is, Gotti notes, the way the English legal system utilizes the modal “shall,” whose purpose is to signal juridical obligation. The following example shows this in action:

States Parties shall respect and ensure the rights set forth in the present Convention to each child within their jurisdiction without discrimina- tion of any kind, irrespective of the child’s or his or her parent’s or legal guardian’s race, colour, sex, language, religion, political or other opinion, national, ethnic or social origin, property, disability, birth or other status.

Gotti notes that, in other languages such as French and Italian, “legal discourse often adopts a present indicative to state legal provisions.” Here is an example:

Gli Stati parti si impegnanoa rispettare i diritti enunciati nella presente Convenzione ed a garantirli ad ogni fanciullo che dipende dalla loro giurisdizione, senza distinzione di sorta ed a prescindere da ogni consid- erazione di razza, di colore, di sesso, di lingua, di religione, di opinione politica o altra del fanciullo o dei suoi genitori o rappresentanti legali, dalla loro origine nazionale, etnica o sociale, dalla loro situazione finanzi- aria, dalla loro incapacità, dalla loro nascita o da ogni altra circostanza.

The effect of the present indicative is that it “emphasiz[es] the actuallity and applicability of the legal provision and also impl[ies] that the law draws its force from the natural order of things rather than expressing an order imposed by human agents.” Such differences already show widely varying perspectives as to how each tradition treats and thinks of the power and position of their own legal systems.

Finally, the last challenge faced by legal translators is the difficulty of translating legal terminology. Gotti quotes René David, who writes: 

To translate into English technical words used by lawyers in France, in Spain, or in Germany is in many cases an impossible task, and conversely there are no words in the languages of the continent to express the most elementary notions of English law. The words common law and equity are the best examples thereof; we have to keep the English words […] because no words in French or in any other language are adequate to convey the meaning of these words, clearly linked as they are to the specific history of English law alone. 

In other words, there really isn’t a one-on-one correspondence between legal terminology across languages. Gotti notes that, “even in legal systems that are closely related and even share the same language, such as Austria and Germany,” one concept is not necessarily recognized as its exact equivalent in another language. 

Gotti provides a number of examples. For one, English legal discourse loves to use the adjective “reasonable,” found in phrases such as “reasonable steps, reasonable measures, reasonable person, and proof beyond a reasonable doubt.” The word “reasonable” itself, however, can’t just be translated into its counterparts, such as “ragionevole” (Italian), “raisonnable” (French), or “vernünftig” (German). Or take the “culturally specific French collocations” such as “acteurs sociaux, acteurs économiques, acteurs institutionnels, acteurs publics, acteurs politiques” none of which have direct translations in English. 

As a result, legal translators must be equipped with ample knowledge of the legal systems in both the languages of their source text and target text. Gotti cites Marta Chromá, who illustrates how translators must find the legal equivalents despite seemingly different structures and concepts. For example, the “arraignment” in English legal discourse is translated into the Czech phrase zahájení hlavního líčení, which means “commencement of criminal trial.” In reverse, the Czech phrase hmotná odpovědnost, meaning “to be found in employment law” is equivalent to the English legal phrase “liability to indemnify.”


Translation methods and strategies

As in the example above, translators either find linguistic/legal equivalents or resort to other methods so as to retain the unambiguity, precision, and authority of the source text in the target text. This requires legal translators to be creative, resourceful, and cunning. One method is the usage of calques, or loan words, in which translators borrow directly from the source text. For example, the phrase visible minorities (minorités visibles), found commonly in Canadian political discourse, has been borrowed into the Italian language as minoranze visibili, referring to minority groups whose traits are visually identifiable (skin color, etc.) 

Another method is explication, such as the French word âgism, corresponding with the English word ageism. Due to the fact that the word for “age” is different in other languages—not to mention different linguistic constructions of nouns—the Italian translation explains the concept (la discriminzione per età). 

There are a number of other translation methods and strategies available to the most proficient, experienced legal translators. A single word may be translated into very different words depending on the context, for example. Whatever the case may be, legal translators must be prepared to handle any and all linguistic, legal, and cultural differences that exist between the source and target languages. 


Gotti, M. (2014). Linguistic Features of Legal Texts: Translation Issues. Statute Law Review, 37(2), 144–155. doi:10.1093/slr/hmu027


Understanding Technical Translation

When thinking of translation, one most probably thinks first of literary translation, revered for all the creativity and artistry that goes into translating a work of literature. Some might think of more specialized translations, such as legal or medical translation—well-known areas in which interlingual communication is of the utmost importance. These are all equally valid and important areas of translation that exist in the world; at heart, translation fulfills its purpose by facilitating communication between people and entities, bridging a gap that only translators—capable of navigating the difficult nuances of various languages—can bridge.

Technical translation is often left out when discussing fields of translation. It gets a pretty bad rap, as it is deemed inferior to, say, literary translation, on the basis that technical translation does not require as much deliberation and work. On the surface, it seems like an easy task translating a rigidly written, formal document into a different language. Equipped with the proper terminology, anyone could perhaps engage in technical translation.

Nothing could be further from the truth. In today’s blog post, we discuss technical translation and why, despite its near-dominant presence in the translation market, it is still derided for being a mindless task. 


What is technical translation?

What ideas do you conjure upon hearing the term “technical translation”? Science, mathematics, technical manuals—these are all parts of technical translation, though in the realm of translation, boundaries are drawn as to where technical translation starts and ends. In his book Technical Translation: Usability Strategies for Translating Technical Documentation, Jody Byrne discusses the precise definition of technical translation as it is commonly accepted in the field today. “Simply because a field or subject area has unique or specialised terminology does not make it technical,” says Byrne, noting the difference between specialized translation and technical translation. “Technical translation deals with technological texts” in the strictest sense of the word; “more specifically, technical translation deals with texts on subjects based on applied knowledge from the natural sciences.”

This distinction and delineation is critical in understanding technical translation. Medical, economic, and legal translation all are specialized forms of translation in the same way that technical translation is specialized, but the aforementioned forms of translation are markedly different from technical translation in a number of ways. They, like technical translation, come with specific constraints in language, grammar, and terminology, and must be seen as different activities. Technical translation, as a result, deals strictly with “texts on subjects based on applied knowledge from the natural sciences,” because these subjects and texts abide by common, distinguishable traits that both bring them together and set them apart from other types of translation.

Byrne also notes that technical translation is often confused with scientific translation, which are connected yet different ideas. In his book, Byrne notes that “scientific translation relates to pure science in all of its theoretical, esoteric and cerebral glory while technical translation relates to how scientific knowledge is actually put to practical use, dirty fingernails and all.” In other words, technical translation deals with subjects and texts that are applications of science, not descriptions of science itself. Distinguishing this difference is critical in understanding technical translation; the application of science lies at the heart of technical translation, and this focus on facilitating the application or usage of science in the real world (via manuals, directives, etc.) is what makes technical translation different yet special. Unlike legal, medical, or literary translation—all of which have their own purposes, objectives, and processes—technical translation is closely related to technological advancements and technical communication.

An example of technical writing. Image credits: Russ Ward, Unsplash


Why is technical translation so important?

There are a number of reasons why technical translation is important, the most prominent of which is that it is simply the most widely used form of translation. Byrne cites a 2002 research in his book that has found that “technical translation accounts for some 90% of the world’s total translation output each year.” Byrne attributes such numbers to “increasing international cooperation in scientific, technological and industrial activity,” as well as “various laws, directives and regulations across the world that require the provision of comprehensive, accurate and effective technical documentation in a variety of languages.” In an increasingly connected world, technical translation is more important and relevant than ever, as it helps businesses conduct trade and share technical ideas that foster economic and scientific growth and trade.

Byrne, an ardent and well-known advocate for technical translation, is cited in Mahmoud Altarabin’s The Routledge Course on Media, Legal and Technical Translation: 


Virtually every aspect of our lives from education and work to entertainment, shopping and travel has been swept along by a seemingly unstoppable wave of new inventions and technological advances. What many people do not realize is that these inventions and advances are accompanied at almost every step of the way by translation in its capacity as a vehicle for disseminating scientific and technical knowledge.


Likewise, Altarabin notes that technical translation “promotes the most significant technological advances, which remarkably change our lives… the advances would not be possible without translation, the key role of which is sharing technical knowledge.” And while all this seems obvious and irrefutable, the truth is that technical translation is so often misunderstood, misrepresented, and disregarded. Byrne cites a 2004 report by Franco Aixelá, which reveals that, “out of 20,495 publications listed in the [Bibliography of Interpreting and Translation, BITRA] multilingual bibliography of translation research only 1,905 or 9.3% addressed technical translation. Literary translation, on the other hand, is the subject of some 4,314 entries accounting for 21% of the total number of entries despite its niche status in professional practice.”

Byrne attributes this dire lack of interest and research into technical translation to the fact that research on technical translation is “restricted to terminological issues or technical issues.” The general population—and even academic and professional circles—think of technical translation only in terms of terminology and technicality, despite the fact that technical translation is so much more than that. Not only do these statistics prove that technical translation is important, but they also prove that research and interest in technical translation is also sorely needed.


Features of technical translation

Given the importance of technical translation, it is helpful to assess how exactly technical translation differs from other modes of translation and note its peculiarities. In his Routledge course, Altarabin notes three major features of technical English, citing research by scholar Isadore Pinchuk. While these features have to do with technical language and not technical translation, per se, we must realize that technical translation in itself is technical writing—a process of rewriting a technical source, a process that is, at its core, a fundamental act of writing in technical language.

  1. Technical language is a specialized language and, as opposed to ordinary language, tends to become more specialized.
  2. Technical language seeks to be economic in terms of using linguistic means.
  3. Technical language avoids ordinary language associations and defines terms accurately.

The above features apply generally to all forms of specialized languages, but it is useful to note that technical language—technical translation—has its own set of grammatical, stylistic, and vocabulary-related features that mark it different from other types of specialized languages.

Another example of technical writing. Image credits: Brett Jordan, Unsplash


Style and metaphors

For one, Altarabin notes that technical writing features “simple and informative language.” Language used in manuals, user guides, and other kinds of technical writing are very different from what we see in novels or daily conversations in that the style is markedly impersonal and informative in nature. Furthermore, technical language tends to use metaphors to “put a concrete name to an abstract concept.” Altarabin gives examples such as Black Hole and Greenhouse Effect to reveal how technical translators must work with extended metaphors to aid readers as they attempt to understand difficult scientific concepts.


Terminology and facts

Perhaps one of the most recognizable features of technical language is the terminology. Words—even those used in everyday language—have very set and strict definitions in technical writing, and sometimes their definitions veer from their commonly accepted meanings. Altarabin cites scholar Peter Newmark, who found that “about 5 to 10% of a given text contains specialized terms.” While this is not a large proportion—a fact to be discussed later—the presence of specialized terminology also means that translators must understand and be familiar with specialized terms when translating.

Another major feature of technical writing is, of course, the “presentation of facts.” Such a factual, fact-heavy style of writing lends technical writing its informative, formal voice.


Syntactic features

Due to the fact that technical writing must deal with such terminology, the grammar of technical writing is rendered more malleable and prone to change, resulting in the fact that “the grammar of scientific language is complex to the layman” according to Christopher Tylor, as cited by Altarabin. There are a number of grammatical differences between normal speech/writing and technical writing; for example, technical language “has a higher proportion of complex noun phrases and a few simple noun phrases as clause subject” (Tylor). Altarabin gives an example:

Two or more atoms joined to forma molecule are represented by…


Abstract subjects and passive formations

Drawing on various research, Altarabin notes that “subjects in technical texts tend to be abstract… the pronoun I is replaced by we or passive forms.” Oftentimes, the pronoun “I” is omitted so as to direct the focus of the text onto the subject at hand or to make the writing less personal, overall maintaining the impersonal, informative diction of the text.

Related to this phenomenon is the extensive usage of passive structures in technical language. Altarabin cites Ana Fernández Guerra, who explains that “the passive voice is normally used to emphasize the importance of the message.” By omitting the subject and rendering the text passive to important information, the text can retain a razor-sharp focus on the subject at hand.


Connectors and simple sentences

Guerra also notes that “the use of connectors and repetition of key words (mainly nouns) is common in technical texts.” Connectors here means conjunctions, adverbs, and other grammatical components that direct the reader’s attention. These include words such as on the other hand, secondly, therefore, etc.: words that are used as signposts to guide the reader through the difficult text and provide referential points.

At the same time, the difficulty of the subject also means that the sentence structure must be as simple and straightforward as possible; otherwise, readers will struggle with not just the difficulty of the subject, but also of the content and writing style. Byrne points out that “simplicity aims at reducing the work readers need to do and reducing the risk of misunderstanding.” Utilizing “simple and declarative sentences can improve the simplicity of technical texts,” writes Byrne, giving the following example of how simplified structures work in technical sources:

The detector automatically checks the condition of the batteries.



Perhaps one of the most defining features of technical writing is nominalization, which is defined by Hervey and Higgins as “the use of a noun which, in the same language or [translation type], could be replaced by an expression not containing a noun.” In other words, technical writing features the usage of nouns in place that nouns normally wouldn’t go. According to Pinchuk, the reason for utilizing nominalization is because it “is easier to write and its impersonality avoids commitment to tense, unlike the controversial style.” An example of this given by Byrne is something you would see very often in technical writing: “the flywheel housing installation position must be ensured.” Nominalization heavily slows down reading speed as much information is given in shorter spans, leading to decreased legibility. However, nominalization also helps translators maintain an impersonal, formal style of writing necessary for technical writing and understanding.


Altarabin, M. (2022). The routledge course in Arabic Business Translation: Arabic-English-arabic. Routledge, Taylor & Francis Group.
Byrne, J. (2010). Technical translation: Usability Strategies For Translating technical documentation. Springer.


Microsoft Style Guide: Differences Between the British and American English Styles

With the global rise and domination of the United Kingdom and the United States of America, English has situated itself as the lingua franca of the world. Though not spoken throughout the world, English remains the closest thing to a de facto universal language of communication between nations and people. For this reason, English is a necessary tool in today’s international business and is taught all over the world in grade schools as part of mandatory school curriculums. 

However, English is not the same around the world. There are the two major strains of English—British and American—which are considered the most “standard.” However, there are a plethora of other equally valid English dialects spoken in all parts of the world—South African, Singaporean, Irish, Kenyan, Canadian, etc. For the purpose of today’s blog post, we will limit ourselves to the comparison between the two major modes of English: British and American. 

A table listing examples of the differences between English dialects spoken in the United States, Canada, the United Kingdom, and Australia. Image credits: Wikipedia

There is a good reason why the two Englishes are different. American English diverged from its British roots when American settlers broke off from the empire in 1776, when the original thirteen colonies announced their independence from their motherland. This was before there existed standardized spelling, and as such, the differences between the two languages are present not only in speech and usage, but also in written text. The two Englishes have each evolved in their separates ways on either side of the pond, complete with their own nuances, conjugations, and vocabulary.

Thankfully, the two Englishes are still very much mutually intelligible (save for particularly difficult dialects). In fact, with the ever-influential scope of American media, American English has slowly seeped into British English and other dialects to the point that the American dialect is often very easily understood more than other dialects. On the flip side, this also means that speakers of American English are less familiar with how other dialects sound and seem, leading to difficulty in communication.

Today’s blog post hopes to ameliorate some of those communication issues by pointing out the differences in British and American English, using the Microsoft Style Guide as a reference.



Abbreviations commonly used in the United States take on different forms in other dialects and must be localized as such. For example, the US Department of Education is commonly abbreviated as ED; the UK’s version of the institution is the UK Department for Education, abbreviated as DfE. Other US English abbreviations are not well-recognized outside the country, and must be spelled out to avoid confusion, such as the EPA, which would be written out as Environmental Protection Agency (EPA).

One thing to note here is the difference between acronyms and initialisms; the previous examples are instances of initialisms, which is a form of abbreviation in which individual letters are pronounced separately (EPA, GPS, etc.) Acronyms, on the other hand, are abbreviations that are pronounced as if it were its own word (LASER, RADAR, etc.)



Given the separate standardization processes the Englishes have undergone, there are various differences in spelling. The chart below offers some examples of how certain words are spelled differently and the grammatical components that cause such differences:

Type of change US English UK English
hyphenation antialiasing anti-aliasing
e-mail email
cohosting co-hosting
word separation anymore any more
-l / -ll labeled labelled
-ize / -ise localize localise
-yze / -yse analyze analyse
-or / -our watercolor watercolour
-er / -re center centre
milliliter millilitre
air- / aer- airfoil aerofoil (but not airport)
-e- / -ae- anesthesiologist anaesthesiologist
-ey / -y flakey flaky
-a- / -au- balk baulk
-i- / -y- cipher cypher
-e / -é coupe Coupé

One of the most prominent differences is the doubling of some consonants preceding grammatical endings (such as labeled vs. labelled) as well as the usage of z and s in -ize/ise and -yze/ise (localize/localise, analyze, analyse). Other differences include the British English’s addition of the vowel “u” in words such as colour and flavour, as well as the tendency to spell some -er endings as -re, such as centre and mililitre.

The list below offers more differences that you might find to be useful. Learning such differences is crucial, especially for people who work across different Englishes. Despite the similarities of the two dialects, speakers of a certain dialect might find their own to be more familiar and comfortable.



Word substitution

In some cases, the two dialects use completely different words to refer to the same thing, not just a different spelling. Such words are much more important to recognize and remember, as these words can create confusion.

US English UK English
parenthesized bracketed
buddy mate
ZIP code postcode
soccer football


Tricky issues

In British English, two forms of a word are used depending on the context.

US English UK English Comment
program program
computer program
TV programme
fall autumn
to fall
meter meter
draft draught
of wind, or beer
of a document



In American English, the preposition “through” is used to denote the duration between days, as in the phrase “Monday through Wednesday.” In UK English, however, “through” is never used in this exact sense; speakers of UK English say instead “Monday to Wednesday.” Other differences in prepositions include: finish up (AmE) vs. finish (BrE) and waiting on (AmE) vs. waiting for (BrE).



Commas are ever the fickle punctuation, and its usage varies from case to case, dialect to dialect. One of the major differences—and the root of quite a heated debate—is the usage of the Oxford comma, or the comma before the final “and” in a list-type construction. In American English, the Oxford comma is often opted in, whereas British English leaves it out.

US English UK English

Check for available updates to the Software, such as bug fixes, patches, and enhanced functions.

Check for available updates to the Software, such as bug fixes, patches and enhanced functions.

The same applies for the final “or” in a list-type construction.

US English UK English

Check for available updates to the Software, such as bugfixes, patches, or enhanced functions.

Check for available updates to the Software, such as bug fixes, patches or enhanced functions.

Another case in which comma usage differs is before the conjunction “but.” In US English, a comma is often place before a “but” when it acts as a conjunction, but in UK English, the comma is omitted. This only pertains, however, when the sentence features a dependent clause. When the clauses form independent sentences, a comma is used for both varieties of English.

US English UK English

This user will be able to see your photos and documents on SkyDrive, but can’t make changes to them.

This user will be able to see your photos and documents on SkyDrive but won’t be able to make changes to them.

Finally, there are differences in comma usage before the abbreviations “etc.” and “i.e.” In US English, a comma is used after “i.e.” (id est, meaning “in other words”) and before “etc.” (et cetera).

US English UK English

Select the “Date Range” for your report by clicking the pull-down menu and choosing the time span (i.e., “Last seven days,” “Last thirty days,” etc.).

Select the “Date Range” for your report by clicking the pull-down menu and choosing the time span (i.e. “Last seven days”, “Last thirty days” etc.).



For both varieties of English, dashes are also difficult and fickle. There are three types of dashes in the English language: the hyphen (-), the en dash (–), and the em dash (—). One major difference in hyphen usage between the US and UK dialects are word breaks, which are situations in which a word is broken into two to straddle over different lines when the line is too long. In US English, the break occurs at syllable breaks; in UK English, the break occurs at morphological breaks.

US English UK English

In regard to en dashes and em breaks, US English favors the em dash over the en dash, whereas UK English favors en dashes. In some cases, US English might use two en dashes in place of an em dash; UK English tends to use a single en dash.

US English UK English

This is an example — and must be taken into account — when localizing into UK English.

This is an example – and must be taken into account – when localising into UK English.


These are all the differences listed in the Microsoft Style Guides for UK and US English dialects. However, there are much more present—differences that must be actively recognized and remembered if you work across dialects. There are numerous differences—some of them very nuanced—in terms of tenses, verbs, verbal auxiliaries, subject-verb agreement, transitivity, and other miscellaneous grammatical elements. It is always useful to keep in mind that, despite being the same language, the two dialects vary somewhat in their usage.


English (UK) Style Guide – Download Center › download › eng-…


A Beginner’s Guide to Software Localization

In today’s smartphone- and computer-dominated world, the language of coding and algorithms affect our lives in ways more intimate than we can imagine. We use smartphone apps for everyday tasks—apps that were developed in countries whose languages we don’t speak. Our electronics have been manufactured and programmed in yet other countries we’ve never been to and whose languages we are not familiar with. And yet, these electronic devices and applications play such integral roles in our lives, from helping us cook and clean to setting up doctor’s appointments and lubricating our social lives.

We have software localization to thank for that. For a long time—and still now—many people thought of, and think of, translation as a straightforward, simple process: the methodical transcription of a text into a target language. But that era has long been passed; translation has been replaced by localization, which is a more nuanced form of translation that takes into account the various linguistic, cultural, and pragmatic differences between the source and target languages. And in a world of applications, gadgets, and electronics galore, this translation/localization process has expanded its reach to include software, which, despite being written in the language of computers, must still be localized for it to function properly.


What does it mean to localize software?

Given that the states and regulations of software development varies from country to country, software cannot be fully translated into the language and/or culture of another nation. Phrase, a localization software company that recently teamed up with Memsource, defines software localization as “the process of adapting a web or mobile app to the culture and language of users in a specific market.” Localizing one’s software well means to flawlessly and seamlessly transfer the full functionalities and effects of a software into another language; in doing so, the software will earn the trust and familiarity necessary for users in a different country to use the said product.

Eccentric circles of globalization. Image credits: Phrase

Localization can be divided into three general steps, as defined by Phrase: UI/UX design, content translation, and testing and quality assurance. UI/UX design refers to all forms visible on the screen—the tangible manifestations of the software that must translate into a new language with its many variables. There is also content translation, which has more to do with the content or meaning of the source material; this is the more traditional notion of translation we are familiar with. Finally, localization must also take into account testing and quality assurance. The lattermost is a necessary step; not only must software be designed and translated well, but it must also be tested and assured for quality in the target language.

An important step in localization, first and foremost, is internationalization, also known as i18n (given the 18 characters between “i” and “n”). Internationalization means, as the word suggests, adapting the language of a software so that it fits the linguistic, regional specifications and technical requirements of the target region; in other words, it is the process in which software is edited to be better adapted by other countries. Phrase defines it as “prepar[ing] your source code for localization to remove any barriers to deploying your software internationally.”

In their blog post on software localization, Phrase lists some elements that require careful oversight and editing for the sake of internalization, including the following:

  • user interface elements (“mov[ing] all translatable strings into a separate resource file… so that localization can be performed without touching the source code”)
  • text length (“ensur[ing] that translated text stays visible in software interfaces… by accounting for the differences in length between the various languages while writing software code”)
  • currency symbols and numbers (“software should be prepared to accept all these formats [of writing numbers and currency”)
  • date and time
  • character encoding (“software should be able to process various scripts, including non-Latin ones”)
  • language direction

An example of character encoding that must be taken into account during localization. Image credits: W3C


An example of different language directions. Image credits: W3C


An example of the different word orders between English and Hindi. Image credits: W3C


An example of the different word orders between English and Finnish. Image credits: W3C


Why is software localization so important and necessary?

At first, software localization doesn’t seem like much of a deal. Any bilingual person with enough experience handling software products seem capable for the job. Some companies might not even consider localizing their software, mainly for the reason that their product is in a lingua franca commonly accepted around the world. However, software localization has a direct impact on company revenue and the success of the software in the countries to which they are exported.

Most important is product presence. For one, it is difficult to introduce a product to a new region when its software has not been properly localized. Without proper localization processes, it is difficult to create brand awareness; while possible, unlocalized or poorly localized products may take much more budget and effort to achieve the same amount of attention and attraction from possible customers compared to products that have been localized well. Phrase notes that localization “helps build brand equity, which makes your product attractive to potential customers and investors.” The blog post also introduces the term symbolic capital, “a sociological term… [referring] to the resources available to a group or an individual on the basis of prestige or recognition.” A product must be properly recognized for its value, prestige, and function for it to be deemed worthy and useful in a society. Only proper localization can help a product garner that level of “symbolic capital” in a new region, even if the product enjoys that symbolic capital in their country of origin.

Another important benefit of localization is the cultivation of a loyal base of clients. “Brand loyalty is oftentimes built on cultural affinity,” says Phrase. This has much to do with the fact that non-native speakers of English prefer to see advertisement written in their native tongue; research by CSA Research suggests that about 40% of internet users won’t buy from sources that are not written in their mother tongue. With a well-localized product, customers are much more likely to purchase the product, and by extension, companies may see more loyalty associated with their brand and product names. Phrase also notes that software localization helps with “faster customer acquisition” and “increased conversion rates.”


What are the processes of software localization?

So how what exactly does software localization look like? What are the steps involved, and how exactly does the process pan out? The blog post by Phrase gives great examples on the various kinds of workflows that exist: waterfall, agile, and continuous.

The waterfall workflow is the traditional mode of software localization, typical in its structure. The waterfall workflow has six steps—planning, design, development, localization, delivery, and maintenance—in that exact order. Phrase defines the workflow as “a process whereby the localization team only starts working on translating the software after the developers have completed the development stage.” This poses a problem, however, as localization cannot happen until previous steps have taken place, making the entire workflow quite inefficient. 

A visual of the waterfall workflow. Image credits: Phrase

Phrase posits a much better, more efficient localization workflow, called the “agile workflow” which combines “software development and localization into one process so that software developers work in parallel with the localization team from the beginning.” The localization team would work in “iterations” of one or two weeks.

A visual of the agile workflow. Image credits: Phrase

Phrase goes one step further to posit the “continuous workflow” paradigm in which “the localization and development teams work agile in tandem and feedback loops are established in a continuous flow.” This allows “every stakeholder [to] provide invaluable insights before the next cycle starts.” The six steps remain the same, but such reforms to the traditional workflow allow localization teams to work in tandem with software development departments to create a more efficient, logical workflow that gets the product localized much faster. Paul Jakovlev from Lokalise calls the continuous localization process “a step closer to making localization automated and seamless.” In the Lokalise blog post, Jakovlev quotes Miguel Sepulveda, Global Localization Manager at King: “In continuous localization, the content is always ready for a release. In agile localization, the content is not always ready to be release; we need to wait until the sprint is completed.”

A visual of the continuous workflow. Image credits: Phrase

In their blog post on software localization, Lionbridge also shares some areas of application that such localization processes can take place in:

  • companies producing desktop applications
  • eLearning and Learning Experience Designers
  • mobile app developers
  • webapp developers
  • metaverse and AR/VR application developers
  • Software as a Service (SaaS) companies
  • system integrators
  • content providers, localization as a feature

Each of these areas require software localization for their work to be fully integrated into the web or mobile environments of a target country or language. Software is used nearly ubiquitously all around the world, and software localization is a necessary step in branching out into unexplored regions in hopes of spreading the reach of a product or brand.


What are some specific examples of good software localization?

If you don’t have much experience with software localization, it might be hard to imagine what the process actually looks like in action. Thankfully, Paul Jakovlev of Lokalise has a list of 7 software localization best practices that might help you picture the process a bit more clearly in your mind.

  1. Use separate resource files
  2. Manage your code to handle multiple languages
  3. Build in space as language length will change
  4. Verify your images and symbols make sense
  5. Be as precise as possible with the locale
  6. Create a style guide
  7. Use a software localization tool

As you can see, these best practices are not really difficult tasks to pull off. Rather, they are small, conscious decisions and guidelines that, once implemented, will only aid the globalization process of a product. For example, it is not a difficult implementation to make, allowing for graphics that adjust to differences in text length—a simple design implementation can make or break the product and the user’s interactions with it.

A visual portraying the importance of adjustable graphics that correspond to text length. Image credits: Lokalise




On the Importance (and Difficulties) of Subtitles

It was a historic moment for non-English filmmakers and aficionados everywhere, Korean director Bong Joon Ho stepping onto the stage at the Golden Globes to receive a Foreign Language Film award. “Once you overcome the 1-inch-tall barrier of subtitles, you will be introduced to so many more amazing films,” Bong says, or rather, his translator Sharon Choi. 

The way he said the phrase in Korean was slightly fumbled; the way he said it originally goes something more like this: “Those… subtitles, that barrier of subtitles (in English), though not really a barrier, but if you overcome that roughly 1-inch-tall barrier of subtitles, you will enjoy so many more movies.” The way Choi interpreted it on spot is remarkably impressive and put-together, capturing the core message of what Bong was trying to convey and presented in a way that the sentence alone carries so much power. A proverb. A proverbialization, if you will.

A scene from Parasite(2019) featuring English subtitles. Image credits: Korea Herald


The Importance of Subtitles

This is the power of subtitles: they allow viewers to experience whole new cinematic realms in their native language (or language of preference), effectively breaking down cultural and linguistic barriers that would have otherwise prevented people from enjoying works of cinema (or TV) from other parts of the world. Subtitles connect viewers to works of visual media from places around the globe they could never have imagined existed or mattered, and in doing so, attempts to make relevant cultures and languages that were previously not deemed so by prevailing cultures and languages (primarily Western).

Subtitles are by no means the only form of translation available for audiovisual media. There is, of course, dubbing and voice-overs, which are the preferred form of audiovisual translation in various parts of the world and more accessible for those that are visually impaired. 

A map of Europe indicating the usage of subtitles in films. Countries in red prioritize dubbed films over subtitled ones, whereas countries in blue prioritize subtitled films. Image credits: Wikipedia


The Difficulties of Subtitles

People who work professionally to create subtitles are, indeed, translators, as they are tasked with translating the audiovisual content of a film in a source language into another. However, subtitles occupy an ambiguous space within translation as well. Unlike more institutional forms of translation—technical, literary, legal, patent, etc.—subtitles come with very specific constraints. This isn’t to say other modes of translation are not constrained; technical, legal, and patent translation have their own corpora and regulations that mark it different from average translation. More than these translational modes, however, subtitles are constrained by space and time. They must be legibly displayed in extremely constrained spaces: a time span of mere seconds, and the space of a couple lines. Bong may think of subtitles as a mere “1-inch-tall barrier,” but this barrier presents significant difficulties for translators working with audiovisual media.

Furthermore, subtitles are different from other translations in that the medium of the source language is spoken language, and the target medium of the translation is written. Subtitling, then, is not merely a transition between languages, but also form as well. This proves to be difficult: spoken language and written language are dissimilar, and translators must always look for ways to present their subtitles as natural and legible while, at the same time, remaining mindful of the differences in form and language.

To achieve this tremendous task, translators often abide by certain rules, regulations that the translation theorist Lawrence Venuti calls “instrumental” in his essay “The Trouble with Subtitles.” Such instrumental rules can include: being faithful to the general “message” or “point” of a sentence rather than the actual language; reducing the source sentences and condensing them into their ideas so as to be accommodated into their limited spaces; or opting for standard dialects and forms of the target language to improve legibility. These guidelines serve to clarify and facilitate the translation process: translators can thus spend less time worrying about linguistic choices and more time on actually getting the work done.

A scene from Charade(1963) featuring subtitles. Image credits: Wikipedia


Lost in Translation

For ordinary, unassuming subtitling tasks, the aforementioned “instrumental” methods suffice. However, Venuti disapproves of such methods of reduction and standardization, as so much meaning is lost in the process. He provides the following example from Alfred Hitchcock’s 1960 film Psycho: specifically, a scene featuring a salesman saying “One thing people never oughta be when they’re buyin’ used cars and that’s in a hurry.”

A scene from Psycho(1960). Image credits: TMDB

Venuti compares the original line to three translated subtitles in French, Italian, and German:

Non si dovrebbe mai andare di fretta
quando si compra una macchina.
[One should never be in a hurry
when one buys a car.] 

On ne devrait pas être pressé
quand on achète une voiture d’occasion.
[One should not be hurried
when one buys a used car.]

Beim Gebrauchtwagenkauf
sollte man es nie eilig haben.
[When buying a used car
one should never be in a hurry.] 

Venuti points out that, in each one, the familiar subject “people” is replaced by a more formal, abstract “One”: “non si dovrebbe mai” (“one should never”), “on ne devrait pas” (“one should not”), and “sollte man… nie” (“one should never”). Furthermore, other omissions are present, such as the omission of the “used” state of the car in the Italian; the Italian and French translations also modify the structure of the sentence so that it is more generic and fitting with the standardized structures of their language. However, these are very acceptable translations; such standardized subtitles are the majority of the subtitles we see today when watching foreign films. So why is this a problem for Venuti? 

The main flaw that such standardized subtitles have are that they are reductive, both in meaning and in interpretation. In reducing the original language, meaning is erased and the original film or video is rendered impossible to interpret. Good subtitles should be open for interpretation, Venuti argues. In his words: “A hermeneutic model takes for granted that translation is transformation, even when a semantic correspondence is strictly maintained or a stylistic approximation is established.” Good subtitles must actively engage with the viewers’ expectations and experiences, as well as the standards of the receiving viewer’s cultures. In brief, such an interpretive, hermeneutic process of subtitling can only be possible when not only the translator, but also the viewers, are cognizant of such cultural, linguistic factors in play. 


Subtitling in the Technological Age

In a world where machine translation and AI-based language processing is consistently reforming our relationship to language and communication, human translators are always fighting for a place and situating their relevance and importance. Can the subtitling process ever be fully automated? 

Given all the information above—the deliberation that goes into making the right choices in subtitle translation—it seems highly unlikely that an AI-based model could ever rival the intentional choices made by human translators. Under Venuti’s framework of hermeneutic subtitling, the subtitling subject would always have to question their choices to create subtitles that are most open to interpretation, negotiation, and meaning-making. Even under instrumental, standard guidelines subtitling translators use today, there is still a difficult, deliberate process that goes into translating the audiovisual content of a film (or any other medium) into the constraints of the written subtitle. 

One of the main areas in subtitle translation in which AI-based models fail is the realm of puns and wordplay, abundant in films and TV shows by nature of the puns’ oral and creative presence in language. Venuti provides an example from the 1970 Woody Allen film Annie Hall, in which a conversation involves the play of words between “Jew” and “did you”: very similar in pronunciation, but difficult to translate. Venuti offers a Spanish translation by film critic José Luis Guarner, who renders the scene beautifully by using the Iberian Spanish word for green beans (“judías”) and the Spanish word for Jewish women (“judías”). Not only would an AI-based model not be able to create such corresponding translations, but it would also fail to take into account the nuances and ramifications of anti-Semitism presented by the wordplay or the cultural significance and difference presented by Guarner’s translation (as judías—green beans—are not called that in Latin America). 

A scene from Annie Hall(1970). Image credits: TMDB


The Subtitle Industry Today

Rest assured, translators—and most of all, subtitling translators, do not have to fear the loss of their jobs. The more constrained and deliberate a translation task is, the more human involvement it takes. This is one of the reasons that the subtitling industry is thriving today; there are more content available than ever before, yet the subtitles for such films around the world can only be mediated and translated by professional translators who are able to create such deliberate, intentional renditions.

According to Valuates Reports, the captioning and subtitling solution market was worth USD282 million in 2021 and is expected to reach USD476.9 million by 2028, at a CAGR of 7.7%. The report cites “the increase in demand for streaming of content from media platforms such as Netflix, Amazon, Youtube, etc.” as well as “the rise in demand for captions and subtitles in the media industry, production industry” as reasons for such growth in the market. 

A chart showing the growth of the captioning and subtitling solution market. Image credits: Valuates Reports

Things aren’t all rosy in the market, however; with the recent limelight on subtitles and film translators comes news of quite appalling conditions translators are subject to. Gavin J. Blair of The Hollywood Reporter notes that “there is a widespread lack of appreciation in the [film and TV] industry for just how challenging the work of a subtitler can be” and speaks to a Korean-to-English subtitler who “reported payment of $255 for a 110-minute film for a local streaming service, and that such low pay, often accompanied by short deadlines, can result in a shoddy final product.” Furthermore, less frequent language pairs are often indirectly translated, Blair notes, meaning the translation into the final target language is done from an intermediary language, primarily English, hence further lowering the accuracy of a translation.

A report by UNESCO also sheds light on the plight of subtitlers; according to Roshanak Taghavi, a DC-based journalist, “while 50 per cent or more of most films’ revenue is earned from their foreign translated versions, only 0.01 per cent to 0.1 per cent of budget is spent on them.” Furthermore, “subtitlers are generally paid per minute of content rather than per subtitle… this per minute rate has been gradually falling over the past thirty years.” The problem? According to Taghavi, “there is no standardized process for assignments, contracts or payment, with rates and methods for contracting subtitling services varying vastly by region.”

All that being said, the more attention and spotlight subtitlers receive will hopefully force governments and relevant institutions to implement necessary regulations that protect the livelihoods of subtitlers and other translation-based vocations related to the film and TV industries. And in doing so, the so-called “1-inch-barrier” will no longer be a barrier, not just for viewers all over the world, but also for translators.


Venuti, L. (2019). The Trouble with Subtitles. In Contra instrumentalism: A translation polemic. essay, University of Nebraska press. 


The Age of the Multilingual Metaverse

Metaverse: it’s a word you hear frequently tossed around on the Internet, the metaverse is the next evolution, the metaverse is a hypothetical iteration of the Internet. The term metaverse has taken a life of its own in the past few years, spurred on by the likes of Mark Zuckerberg, founder of Meta (PKA Facebook), who has renamed his company in honor of this new, futuristic concept that he believes will profoundly influence the world as we know it.

But to this day, most people struggle to define what exactly a metaverse is. Older generations might be familiar with the term via Neal Stephenson, the science fiction writer who coined the term in his 1992 novel Snow Crash, in which the denizens of a futuristic world have access to a virtual urban environment where they can portray themselves to be anyone they want. The concept has since been replicated and made familiar to younger generations through the likes of the Spielberg film Ready Player One, the Kosinski film Tron: Legacy, or popular online games such as Fornite or Roblox. The general consensus regarding the definition of a metaverse seems to be a virtual reality immersive enough for its users to take on entire personages and lives disparate from the ones they live in their immediate reality. One of the main components of this immersion is the very tangible social relationships users form through these metaverse platforms.

But just how accurate is this understanding of the metaverse? Is there a singular definition of the metaverse that we can all agree to? The notion of the metaverse isn’t as straightforward as one would like it to be, and for a good reason: it’s not actually real. In an article for The Verge, reporters Adi Robertson and Jay Peters note that the metaverse “doesn’t necessarily exist,” and is “partly a dream for the future of the internet and partly a neat way to encapsulate some current trends in online infrastructure, including the growth of real-time 3D worlds.” Others beg to differ, one example being Matthew Ball, the author of the Metaverse Primer, who defines the metaverse as “an expansive network of persistent, real-time rendered 3D worlds and simulations that support continuity of identity, objects, history, payments, and entitlements, and can be experienced synchronously by an effectively unlimited number of users, each with an individual sense of presence.” Though more articulated, this definition is similar to our general consensus of the metaverse. 

Meta’s Horizon Worlds, a metaverse platform. Image credits: Meta

The company Meta goes even further and sums up the metaverse in a single sentence: “the metaverse is the next evolution in social connection and the successor to the mobile internet.” Meta describes the metaverse as “a set of digital spaces that you can move seamlessly between,” a world that will help “connect with people when you aren’t physically in the same place.” Meta’s definition retains the social aspect of the metaverse that we are familiar with; on the other hand, it introduces the metaverse as the successor—and replacement—for the internet. 

Overall, it’s helpful to think of the metaverse as a collection of loosely stringed-together ideas and desires that paint individual images of a vague notion of the metaverse. Robertson and Peters provide a list of features that contemporary tech figures usually refer to when speaking of the metaverse:

  • Feature sets that overlap with older web services or real-world activities
  • Real-time 3D computer graphics and personalized avatars
  • A variety of person-to-person social interactions that are less competitive and goal-oriented than stereotypical games
  • Support for users creating their own virtual items and environments
  • Links with outside economic systems so people can profit from virtual goods
  • Designs that seem well-suited to virtual and augmented reality headsets, even if they usually support other hardware as well

Definitions aside, how exactly big of a market does the metaverse boast? A report by Grand View Research on the global metaverse market roughly estimates the value of the market to be USD 38.85 billion in 2021. The authors expect the market to grow at a CAGR of 39.4% between 2022 to 2030. The growth is attributed to “a growing focus on integrating digital and physical worlds using the Internet, increasing momentum and popularity of Mixed Reality (MR), Augmented Reality (AR), and Virtual Reality (VR), and the outbreak of COVID-19, as well as the situation’s subsequent developments and outcomes.” The report also notes that, according to industry experts, the metaverse market could one day reach more than USD 1 trillion in yearly revenues. 

Image credits: Grand View Research

The GVR report also cites two other major drivers of growth for the metaverse market, the first of which is the “growing demand for metaverse to purchase digital assets using cryptocurrencies.” Every metaverse has a specific cryptocurrency used to purchase virtual items and is integral to connecting the physical and virtual realms. Aspects of commercialism, business, economics, and trade are naturally reflected in the hyper-realistic worlds configured by the metaverse and tie into the tangible, realistic economics of the real world. The second major driver is “expanded opportunities for Business-to-Consumer (B2C) and Business-to-Business (B2B) enterprises.” The metaverse is rich in opportunities that will help businesses more effectively reach their consumers or facilitate their operations. For example, the metaverse can one day host “trade exhibitions, product demos, client meetings, customer service, and commercials” or aid “workers from low-income countries… find work in western corporations without emigrating.” Virtual reality will also “enhance educational options, as they are a low-cost and effective way to learn.”

Given the immersive nature of the metaverse—and how communication-heavy it all seems to be—translation is a necessary step in building a metaverse that is linguistically and culturally varied enough to cater to the majority of its users. After all, it’s hard to feel immersed in a location, virtual or otherwise, when you don’t speak the language that is spoken in that locale. But the metaverse is an ambiguous, burgeoning field of technology, and for this reason, it is difficult to tell how pre-existing translation methods and tools will be implemented in the realm of the metaverse.

Image credits: Grand View Research

However, there are already researchers, specialists, and translation agencies mapping the frontiers of this new era. In introducing a multilingual AI-based translation system, Meta CEO Mark Zuckerberg confesses that such instantaneous, large-scale translation models will be “especially important when people teleporting across virtual worlds and experiencing things with people from different backgrounds… Now, we have the chance to improve the internet and set a new standard where we can all communicate with one another, no matter what language we speak, or where we come from. And if we get this right, this is just one example of how AI can help bring people together on a global scale.” In other words, Zuckerberg’s plans of expanding the reach and possibilities of the metaverse are to be facilitated by the complementary development of an all-purpose translation model.

There are others that believe that AI-based machine translation will be the solution to the metaverse’s language problem. In a short paper titled “Evaluation of Language Translator Module for Metaverse Virtual Assistant”, presented at the 2022 Korean Institute of Communications and Information Sciences Summer Conference, authors Cosmas Ifeanyi Nwakanma et al. configure a basic translation module for the metaverse, examining the possibilities of using front-end translation models such as DeepL, Google, and Rozetta in the new field of technology. It’s not a perfect fit, however; there are problems that remain. The authors pose important questions necessary for the implementation of machine translation in the metaverse: how many languages will be supported? Will various metaverse platforms be interoperable? Will apps and interfaces be capable of being integrated? 

Image credits: Grand View Research

These cases paint a grim picture for human translators and translation agencies: what kinds of roles do human translators play in the localization and translation of the metaverse? As much as the looming notion of a metaverse seems threatening—bringing the world ever so closer to the extinction of human translators—quite the opposite is true. Just like translation and localization in non-metaverse spaces, the metaverse is populated by real-life people (in their avatar forms, presumably) communicating via languages grounded in real-life situations. That means that communication in the metaverse carries connotations, nuances, and deliberation that only human translators can get across; AI-based machine translation still has a long way to go in developing the kind of finesse required to carry out cross-lingual, cross-cultural translations.

There are already localization agencies that offer services catered to the metaverse. Stepes, for example, provides “metaverse translation services and solutions… across linguistic and cultural barriers.” The company offers translation services that help in the localization of “software GUI strings, product documents, training videos, or marketing websites; their translators are experts at translating “the latest metaverse technology products, utilizing systems to allow “professional human linguists to translate at efficiency.” 

Another example is Bureau Works, which, in their blog post, note the specific kinds of linguistic differences that require careful human oversight and supervision. The company notes how “Arabic screen applications are read from the top right-hand corner” as opposed to many other languages that don’t, and thus require deliberate localization efforts to make sure translated texts and content are formatted in the correct way in the metaverse. The company also explains how “symbols also have their own specific cultural contexts” and how “colors are also heavily influenced by culture.” These are the kinds of cultural and linguistic differences that only human translators and localization experts can offer; these differences persist in the metaverse and must be dealt with by experts.


Here at Sprok DTS, our localization experts provide tailored support for clients seeking translation and localization services relating to the metaverse. Covering a wide array of languages and domains, our experts offer top-tier services, aided by numerous machine translation technologies employed here at Sprok DTS. Visit our website and start your journey with Sprok DTS today.




To Localize or Not to Localize: Global Expansion and the Benefits of Localization

As of 2022, more than 6,000 languages are being spoken around the world. It’s hard to imagine such diversity when most of us grow up speaking one or two—perhaps even three—languages at most. 5.03 billion people currently have access to the Internet; the sheer amount of linguistic diversity one can experience online is impressive. However, despite the prevalence of English, only 25.9% of the 5.03 billion internet users are English speakers (Data Reportal, Statista); in other words, business done in a single language is only partially effective at best, even when carried out in English. Linguistic barriers stand in the way between people as they attempt to communicate with one another.

An overview of global internet user statistics. Image credits: DataReportal

The same goes for other languages. Trailing closely behind the 1.132 billion English speakers of the world is Mandarin at 1.117 billion. A little ways behind, in third place, is Hindi—spoken as both a native tongue and lingua franca mainly in India—at 615 million, and then Spanish at 534 million speakers. After that is French, Arabic, Bengali, Russian, Portuguese, and Indonesian. The data shows that the world exists in these linguistic spheres; facilitated by the ease of communication, companies and businesses within these spheres can freely collaborate with each other and satisfy the needs of that linguistic population, but have a hard time reaching outside of their respective spheres.

A map illustrating the ten most spoken languages in the world. Image credits: Babbel

And there is much, much to be gained by looking beyond one’s linguistic barriers. An article in the American Express notes some of the many benefits to be gained by trading internationally across linguistic boundaries. Companies can benefit from increased revenues; the 2016 FedEx Trade Index surveyed 1,004 small business leaders, revealing that “business leaders engaged in global trade say they’re growing faster and hiring more employees than small businesses who stay stateside.”  Furthermore, companies that trade internationally can enjoy decreased competition—not having to compete for limited space within their linguistic zone—and enhance their reputation in other parts of the world. Other benefits of international trade include easier cash-flow management, better risk management, benefiting from currency exchange, and access to export financing, among others (American Express).

There are other reasons why companies should look to open up their horizons beyond their boundaries. For one, 51.8% of the world’s internet users are in Asia. More than 50% of all searches on Google are in a non-English language. China—not the United States—is the world’s largest app market (Phrase). Perhaps most shocking of all: “about 40% of internet users will never buy from websites in other than their native language” (CSA Research). To reach this 40% of users, localization is a must, especially if you’re operating your business or company in a non-English language. In other words, localization opens up one’s business and enterprise to other major international markets. 

According to the same research carried out by CSA Research, around 75% of online shoppers claim to be open to purchasing more from a website if the “aftersales care is in their language.” Almost 76% of internet users “prefer to read product information in their native language.” This is the case even in Sweden—a country boasting one of the world’s best non-native English speakers—“over 80% of online shoppers [in Sweden] prefer to make a purchase in their own language.” The familiarity and comfort of encountering a foreign product in one’s own language lie at the heart of international business.

An infographic of the benefits of localization. Image credits: OneSky

This is why localization is so necessary. To integrate one’s business into the global network of supply and demand, localizing one’s products and services into the language and culture of a target region is imperative; localization is no longer a choice, but a necessity. Localization is a seemingly easy process—nothing more than the cultural and linguistic conformity to a certain culture and language, but there are layers to the process, each requiring much expertise, knowledge, and consideration that only localization experts can provide at agencies such as Sprok DTS. One must go about localizing their product with a strategy and course of action in mind; simply diving into a foreign market will automatically lead to increased revenue and facilitate trade.

An article on Phrase explains the benefits and results of localization in greater detail. First, localization allows companies to enter new markets with ease. Extending one’s product range and services to a different region entails “legal issues, logistical hurdles, and also cultural and communication challenges,” but localizations can help ease the stress and burden of these issues by helping the company integrate their product or service better into the flow of the new market. Second, localization helps companies beat out their competitors. Companies that enter new markets can beat out local markets with increased revenue from said new markets, all the while competing head-on with global competitors who might not have the same expertise or the specific kinds of products or services you might happen to offer.

Phrase also notes that localization helps boost customer satisfaction. After all, the purpose of localization is to communicate one’s product and service—and as a result, the company’s ethos and message—to a new market. By catering to the cultural and linguistic expectations of the consumers in a new market, companies can better attract new customers, who realize that the company in question is dedicated to communicating with them. Another related benefit is an increase in brand loyalty: “If you provide your buyers with a satisfying user experience you will start to gain their trust… Communicating with your users in their native language and tailoring your product will help you gain their trust” (Phrase). Most importantly, localization will lead to an increase in revenue. We already mentioned this before—that branching out one’s enterprise into international markets will increase revenue—but the only way to successfully do so is to localize one’s products and services.

An infographic on specific benefits of globalization and localization. Image credits: Phrase

All this being said, companies and industries dealing with highly technical products or services run into additional problems. Unlike normal localization strategies—which are more geared toward marketing—technical companies must deal with more specific localization techniques that will ensure that their technological product can function properly in another language. These technologies pertain to a wide range of applications, from industrial products to computer software; companies working with such products must work with localization experts to make sure everything translates, both in language and in science, into another culture.

A good example of such technological localization—closely associated with the notion of internationalization, or localization to fit the technological standards of another region is explained in this article by OneSky. Taking an app as an example, the article explains that “you must enable your app to support various languages and symbols across different markets and cultures (e.g. left to right script vs. right to left, $ vs. €)” (OneSky). Websites are another example; translation and localization of online sources must be carried out in a way that does not affect or damage the infrastructure of the website, abiding by certain coding standards and rules. The localization expert, then, must be skillful in not only the target language, but also retain basic knowledge regarding coding. Furthermore, the specialist utilizes the following practices to ensure their job is carried out efficiently: encoding website content in Unicode (UTF-8), developing to remove barriers that might impede localization or international deployment (text resizing, etc.), incorporating predefine localization data and features (e.g. date, time, currency, etc.), among others (Phrase). 

Thankfully, there are systems and tools that help localization experts do their jobs more efficiently and accurately. Phrase briefly touches on the several tools localization specialists use in their jobs: translation management systems (TMS), computer-assisted translation (CAT) tools, machine translation (MT), linguistic quality assurance (QA) tools, and terminology management tools (Phrase). These tools help not just with localizing apps and websites, but also with technical manuals and instructional leaflets. Technical manuals require stricter adherence to given terminology, and the above features help localization specialists ensure conformity in their word choices and terminology use.

An infographic describing the benefits of localization. Image credits: LinguaSol

Here at Sprok Document Translation Services, we utilize the most up-to-date technology and translation systems to ensure that our services cater to our customers beyond their greatest expectations. Our translators and localization experts are equipped with the experience and online infrastructure to properly carry out their tasks—from translating technical industrial manuals to localizing marketing content, among numerous other jobs. 




The Future of the Symbiotic OTT-Subtitling Industrial Ecosystem

The past decade has seen the meteoric rise of OTT services. It’s hard to think of what we subsisted on before: video rentals like Blockbuster, movie theaters (still clinging on), DVDs, VCRs, and television. Most of that is unfamiliar now; we have effectively entered the era of OTTs—over-the-top services—subscription-based or on-demand media streaming services that now reign supreme. Spurred on by the development and ubiquity of the internet and its related peripherals such as cell phones, smart TVs, and laptops, OTT services have effectively replaced the role of television and its many channels.

A list of the types of streaming services available now, alongside the platforms on which they can be viewed. Image credits: Conviva

Where did it all start? In an article for matterkind, Jason Han recounts the history of the OTT and its rise to success over the past few years. Netflix, perhaps the most emblematic of all OTT services, dates back to 1997 when it was founded as a competitor of Blockbuster; unlike Blockbuster, which utilized its stores to rent DVDs and VCRs to its customers, Netflix instead utilized the postal service to reach its customers. In 1999, Netflix transitioned to a subscription service, fully instating the monthly subscription plan at $19.95—Han attributes “the main reason of Netflix success to their efforts of delivering better customer experience.” 

A timeline of the development of TV and OTT systems. Image credits: Conviva

In 2007, Netflix implemented what we now know it to be: a streaming service, although it still retained its mail-order rentals. Not many people know, but in the same year, Blockbuster also started a streaming business—several, in fact—but nowhere near as successful as Netflix. Also in the same year was Apple’s launch of Apple TV, which allowed anyone with a Mac or PC to enjoy music, podcasts, and photos on the big screen. A year later—in 2008, Samsung introduced the Smart TV (the name came later in 2010), which is that familiar, TV-centric format we still enjoy today. While all this was happening, Amazon was figuring out its relationship to the streaming video format: from Amazon Unbox in 2006 to Amazon Video On Demand, then Amazon Instant Video, until it finally settled on Amazon Prime Video in 2018. Its growth, in turn, was driven by its development of the Fire stick, alongside other Fire TV devices.

As of now, in 2022, the OTT industry has grown to immense proportions. A report by Mordor Intelligence gives a coherent overview of the present state of the OTT market. Mordor Intelligence notes that the increasing adoption of OTT “can be attributed to the narrow genre choices, packaging flexibility, wider device availability, internet penetration, and overall lower costs,” alongside customized content. The report also notes that OTT platform services are “no longer interested in being viewed as platforms just for accessing movies and TV shows but are also investing in the production and licensing of their content,” resulting in “direct competition with traditional TV and among the OTT industry,” which is further intensified by “the deployment of advanced technologies within the platforms.” In short, Netflix and other OTTs are no longer mere means of streaming movies and visual content; rather, it has grown large enough to encompass video production and technological production, which serve to not only incentivize users to enjoy their products, but also to effectively dominate the realm of media as we know it.

The report also notes a very critical factor in the development and growth of the OTT landscape: COVID-19. Citing Zuora’s Subscription Economy Index Report, Mordor notes that “trends of the COVID-19 impact on subscriber acquisition rates from March 1-31, 2020, in comparison to the last 12 months, suggested that the global subscription growth rate for OTT Video Streaming companies grew 7x in March 2020, as compared to the growth rate over the previous 12 months.” Data gathered by the British Association for Screen Entertainment (BASE) also suggests that “consumer spending on digital movie purchases grew by more than 87% during the COVID-19 pandemic lockdown period through June 30 to reach GBP 113 million (USD 148 million).”

The OTT market is massive and doesn’t seem like it’s fading away anytime soon. Amir Shahzeidi of Uscreen has collected some important statistics for the OTT industry for 2022 in this article here, and the numbers are impressive. The revenue OTT services rake up will soon pass $210 billion by 2026 according to Digital TV Research; this number is almost twice the revenue generated by the OTT industry in 2020 ($106 billion). Statista, on the other hand, places the projected revenue at $272 billion by 2025. Of this revenue, 51.58% comes from advertising video-on-demand (AVOD), and 40.16% from subscription video-on-demand (SVOD). And while advertising takes up 51% of the total revenue, subscription customers are reportedly responsible for generating nearly double the money per user compared to those using ad-supported streaming services (Statista). 

At the individual level, the average revenue garnered per user (for SVOD services) was estimated to be around $200, according to Statista. Deloitte Insights points out that 82% of US consumers are subscribed to a video streaming service—with an average of 4 subscriptions per person. How does all this translate into the actual hours people spend watching streamed content? According to Domo, Netflix users all over the world watched a total of 452,000 hours of content in 2021. DataAI calculates that a single person, on average, consumes 38 hours of content each month, Netflix or otherwise. 

OTT user projections. Image credits: Uscreen

A graph indicating the market concentration of the OTT market. Image credits: Mordor Intelligence

As a localization company, we are always curious as to how these visible trends in major industries pertain to the work we do as localization experts and translators. The OTT industry, in particular, has much to offer localization agencies, in that the international flux of multilingual media always requires translation in the form of subtitles and dubbing, the former of which receives much of the spotlight in the language industry. And international content, by which we mean visual media in languages other than English, is becoming more and more relevant, exposing the Anglophone sphere to whole new worlds outside its borders. 

Recent years have seen a rising tide of international content flooding through OTT services; we can attribute this phenomenon to the relative ease at which content can be shared across the globe. The most-watched show ever on Netflix, for example, is the first season of Squid Game, which has seen critical acclaim and has been watched for a total of 1.65 billion hours: Squid Game is a South Korean series, shot entirely in the Korean language. Other top-rated non-English series include Money Heist—a Spanish thriller show whose fifth season ranks third in the number of hours watched at 792.2 million hours—All of Us Are Dead, a Korean zombie thriller (560.8 million hours), the first season of Lupin, a French heist show (316.8 million hours) inspired by the eponymous fictitious robber, and the third season of Elite, a Spanish teen drama (275.3 million hours). While no non-English language feature films remain in the top rankings for Netflix, some come pretty close, such as Blood Red Sky, a German action horror film (110.5 million hours), The Platform, a Spanish horror flick (108.1 million hours), and Black Crab (94.1 million hours) (CNET). 

For each and every show, translations and captioning services are necessary; with the rise of non-English content comes the resulting growth of the captioning and subtitling industry. According to a report by Valuates Reports, the captioning and subtitling solution market size was recorded to be USD 282-million big in 2021. Valuates names key drivers in the industry as “the rise in media streaming platforms like Netflix, Amazon Prime, and the surging popularity of video content”; the company also emphasizes the “advent of artificial intelligence,” which some key vendors are integrating into their captioning and subtitling solutions, helping translators by facilitating the “process of editing and adding subtitles for end users.” In other words, the subtitling and captioning industries are not only affected by the growth of the OTT industry, but also by the AI and technology sectors as well. 

Manik Gupta, corporate vice president for Teams Consumer, Skype, and GroupMe at Microsoft, speaks more on the ramifications of technological advancement in the subtitling and captioning industry in an article for BroadcastProME. “As video service providers look to globalise their content,” writes Gupta, “subtitling and audio dubbing are becoming even more crucial for SVOD services.” But with so many languages around the world, the success of international features—English or otherwise—depends on how well and comprehensively they are translated so that they can be disseminated around the globe. Recently, the streaming industry is “seeing a major shift toward the use of artificial intelligence (AI) and machine learning (ML) technologies, to minimise captioning and subtitling costs and maximise efficiency,” writes Gupta, “with AI- and ML-based QC solutions, video service providers can ensure that OTT content delivered to different geographies maintains outstanding quality.”

The bond between subtitling and technology doesn’t just stop there, however. Gupta notices another trend in video streaming: the increasing adoption of cloud technologies. More and more OTTs are utilizing cloud services to facilitate and maximize the efficiency of video streaming; “this shift to the cloud by OTT video service providers is apparent across the entire media workflow, from encoding to quality control (QC). Using a cloud-based [automatic speech recognition] system, video service providers can reap all the benefits of the cloud to create captions and subtitles with increased flexibility, scalability, and cost efficiencies.” Gupta ends his article by noting that, in the future, OTT services must embrace the possibilities of AI/ML- and cloud-based QC solutions, so that these services can focus their time on more creative jobs, leaving other tasks for capable AI systems. The subtitling, OTT, and AI industries are inextricably intertwined in their workings; growth and development in one inevitably affect others to grow and develop, with positive ramifications for all.

With all that being said, the subtitling market is looking brighter than ever. According to Valuates, the captioning and subtitling solution market is projected to grow to $476.9 million by 2028, at a CAGR of 7.7%. While this is not as steep as the OTT industry, it is still an impressive growth projection, propelled by the ever-present need for streaming content. And the demand for subtitling and captioning is not going away anytime soon; a survey by Stagetext revealed that 80% of respondents between the ages of 18 and 24 used “subtitles some or all of the time watching TV on any device,” followed closely by respondents between 26 and 35 with 64%. With younger generations more familiar and comfortable with utilizing subtitles, the industry seems like it’s sailing smoothly.

The projected growth of the captioning and subtitling solution market. Image credits: Valuates Reports

A table of the percentage of people that utilize subtitles, classified by age. Image credits: BBC

In short, the OTT and subtitling industries are co-dependent on the growth of each other, stuck—quite happily so—in a symbiotic relationship. International content on OTT services cannot be translated and spread to other regions without subtitling solutions; on the other hand, subtitling and captioning agencies rely on the constant output of content on the part of the OTT services to survive. And with impressive growth looming in the future, these two industries seem as if they have nowhere else to go but up. Without subtitling services, the world would not have the great masterpieces that have propelled Netflix and its competitors to the forefront—Squid Game, Money Heist, Lupin. And without the plethora of content produced by OTT services, the subtitling industry would be nowhere as vibrant and thriving as it is today.

Furthermore, growth on both sides is heavily affected by the development and implementation of new technology in the AI, ML, and cloud sectors. Advancements in these fields help OTT providers and subtitling agencies to better make use of their time and resource, effectively  allowing them to produce more content at higher quality. It makes sense, then, to call this interconnected web of industries an ecosystem. One industry dependent on another, the sectors move forward into the future.




Translation as a Variable Process: Venuti and the Hermeneutic Model of Translation

In recent years, the world has been enraptured by the dark, grisly, comical, and cynical realm of Korean-language media. For years, K-drama has flourished in parts of the world, followed by the meteoric rise of Korean literature (by the likes of International Booker Prize winner Han Kang and recent nominees Park Sang-young and Bora Chung). Squid Game is still Netflix’s top-watched show of all time, and Parasite has helped cineastes rediscover the trove of delectable Korean cinema, spearheaded by Bong Joon-ho, Park Chan-wook, among others.

Each time a work of Korean creativity (if it can be labeled as such) comes to the forefront, it undergoes a now-standardized process: the initial acclaim, followed by intense criticism and scrutiny regarding the translation of the works. When translator Deborah Smith first translated Han Kang’s darkly enigmatic novel The Vegetarian, the work was met with rapturous awe and amazement, only to be marred by an onslaught of critics and Korean-language aficionados, nitpicking every single sentence in an attempt to disprove Smith’s understanding of the language. Squid Game’s English subtitles came under scrutiny for its mistranslations, some going as far as to say that “if you don’t understand korean [sic] you didn’t really watch the same show.” Likewise, Parasite has had its own share of acclaim (for being the first Korean film to win the Palme d’Or and the first non-English language film to win the Academy Award for Best Picture, alongside ranking first on Letterboxd’s prestigious list of the 250 best narrative feature films). The main thing to take away from all this is that works of translation undergo a high level of criticism, not for their content, but for the way in which it is translated from one language to another. But oftentimes, we are left wondering: on what basis do people—many of whom aren’t acquainted with the art and techniques of translation—criticize translations? And what does that say about how people view the tenuous work of translation?

Renowned TikToker, podcaster, and content creator Youngmi Mayer tweeted the following upon watching Squid Game:

not to sound snobby but i’m fluent in korean and i watched squid game with english subtitles and if you don’t understand korean you didn’t really watch the same show. translation was so bad. the dialogue was written so well and zero of it was preserved

It’s a sentiment many agree with; the translated subtitles of Squid Game failed to capture the true meaning of the original work, almost to the point that the subtitles were a show of their own. Whether the fault is on the part of the translator or the sheer difficulty of translation processes, the months following Squid Game enunciated a widespread thought: so much is lost in translation.

A couple years earlier, a similar incident happened with Deborah Smith’s translation of Han Kang’s The Vegetarian. A 2017 article on The JoongAng—a Korean newspaper—cites Cho Jaeryong, a professor of French language and literature at Korea University, saying that Smith’s Korean skills are subpar, hence leading to a critical mistranslation in which a subject pronoun omitted in the Korean original (as is characteristic of the language) reappeared, mistranslated. In an article for Korea Exposé, Charse Yun was also critical—perhaps not as much as Cho—of the translation, which he called “flawed yet remarkable.” 

Cho has done his research; he cites a paper—an actual research paper—presented at a conference in which Smith’s translation of the novel is analyzed for flaws. According to the research, “10.9 percent of the first part of the novel was mistranslated. Another 5.7 percent of the original text was omitted.” However, Cho is more accepting of the kind of flaws that exist in the process of translation: “it would serve us well to remember that “unfaithfulness to the original” doesn’t necessarily mean betrayal, as if the translator carried out willful acts of mistranslation. For one thing, it presumes a lack of sincerity and respect for the source material.” Other critics were more respectful with their criticism, such as an article by the Hankook Ilbo, in which the deviations arising from the translation were examined and applauded for the ways in which it “recreated the effects of the original by transfiguring word nuances and sentences.”

Deborah Smith and Han Kang. Image credits: The Guardian

By examining the critical discourse of Squid Game and The Vegetarian, we observe a couple of trends running through the criticism offered for the two works in translation. First, some critics believe in an underlying or comprehensive “message” of the original work, which must be “faithfully” translated. Second—and this perhaps positions itself in opposition to the previous point—translated works must be faithful to the semantic, grammatical, word-for-word meaning of the original work. The dichotomy between “word-for-word” translation and “sense-for-sense” translation is a debate that comes up frequently in translation discourse, though most of these points are made by critics who aren’t acquainted with other translation paradigms.

In his essay “Genealogies of Translation Theory: Jerome,” theorist and translator Lawrence Venuti provides a much more nuanced dichotomy in which to analyze and make sense of translation. For Venuti, translation can be seen as either “instrumental”; the instrumental model “treats translation as the reproduction or transfer of an invariant which the source text contains or causes, typically described as its form, its meaning, or its effect” (483). On the other hand is the “hermeneutic model,” which “treats translation as an interpretation of the source text whose form, meaning, and effect are seen as variable, subject to inevitable transformation during the translating process” (483). 

Simply put, the instrumental model is what most people think of when they think of translation: the translator takes the original work then reproduces it—either word-by-word or sense-by-sense, capturing the message—into a different language. When people argue about “mistranslations” in The Vegetarian or debate whether Squid Game’s subtitles fully encapsulate the message of the show, they are referring to an instrumental model of translation. Venuti actively protests against the instrumental model, instead advocating for the hermeneutic model, which is not only “paradigmatic and generative,” but also “will lead to a productive investigation into the conditions of the translation process.”

Most of Venuti’s essay deals with the legacy of Jerome, a theologian who was tremendously influential in the ancient era for his translations. In short, Venuti notes that Jerome’s work is unwittingly instrumental in how it deals with translation; Jerome’s advocacy for instrumental translation has profoundly affected how the contemporary world views translation. And the results are harmful: by pressuring translators into staying “faithful” to a certain intrinsic message present in the original work, the translators commit the error of succumbing to certain political and cultural biases in a fatal oversight of these necessary forces that, whether we like it or not, affect and influence all that we read and write.

In Jerome’s case, it was Christian values: 

Jerome’s effort to rehabilitate the word-for-word strategy proved to be no more than an initial feint before his triumphant valorization of sense-for-sense translation, the strategy that he inherited from Roman authors like Cicero, Horace, and Quintilian and deployed in the construction of a Christian tradition dating back to the Septuagint. Jerome’s assimilation of the two traditions reflected his conformity to the Latinized Christianity that came to dominate the empire during the fourth century. This conformity associated him with a cultural elite, so that although he had adopted the asceticism of a monastic intellectual, in which self- mortification took the form of working with languages he considered unrefined like Biblical Hebrew and Greek, he nonetheless shared Damasus’s Roman aristocratic view of the early Christian church. (501)

Simply put, Jerome’s conscious choice to advocate for word-for-word or sense-for-sense translation has resulted in his inability to acknowledge to confess and point out the political, religious ways in which his translations have benefited him—a “conformity” that “associated him with a cultural elite.” 

Painting of Saint Jerome by Jacques Blanchard. Source: Wikipedia

A hermeneutic model of translation takes into account these intertextual factors. Instead of relying solely on the words and messages of the original text, a hermeneutic translation instead reveals the fact that translations never really exist in a vacuum of their own; every single meaning, word, and sense is mediated by the world and its real conditions and ideologies. A translator that submits themself to such a hermeneutic translation relies on these factors—something Venuti refers to as “interpretants”—effectively “interrogating and changing the instrumentalist theories that have prevailed for millennia.” Or as Venuti says:

The hermeneutic model can avoid the potentially questionable ethics of instrumentalist theories like Jerome’s because of its capacity to expose the various determinations at work in any translation. This model allows for the possibility that different yet equally effective interpretants might be applied by translators and translation scholars, and that in any translating or in any analysis of a translation another set of interpretants will always lie outside the ken of the ones that have been applied. An instrumentalist understanding of translation, in contrast, is driven by a metaphysics that assumes the existence of an invariant, and that assumption must necessarily exclude any notion of variability, so much so that it can powerfully rewrite its linguistic, cultural, and social conditions (502).

In Yun’s critique of Deborah Smith’s translation of The Vegetarian, he notes that, unlike the “lack of agency” present in the original work—as is the general literary pathos for Korean literature—Smith’s version has “a heightened defiance, a quasi-agency.” While others point out Smith’s mistranslations and failure to stay “faithful,” perhaps Smith understood that a word-for-word or sense-for-sense translation of the novel could potentially be misrepresented and misunderstood in the context of Western readers, thereby saving the novel from being mischaracterized as passive. Smith’s understanding of the differences in these power dynamics allows the translated version of The Vegetarian to be more subversive in its effect, compared to what might have transpired if she opted for an instrumentalist translation of the work.

What is important, then, in translation—at least, according to Venuti and, to an extent, Smith—are the external factors that influence the original text and translated work. Only by critically examining these factors and reflecting such deliberation in the way we translate—forgoing “faithfulness”—can we fully grasp the scope of the ramifications of what we do.

As is the case with all literary translation theories we encounter, it is helpful to ask how this pertains to the realm of technical translation. Given its status as the most commercially available kind of translation—yet derided as being secondary to literary translation—technical translation must examine its position and how these theories can be applied to its particular situation. Venuti prefaces the essay by saying that this notion of instrumental and hermeneutic translation “encompasses technical translation as well, in which terminologies with a high degree of standardization are transposed in legal, commercial, and scientific documents. A standardized terminology would seem to be a formal and semantic invariant that might lend cogency to the instrumental model: jargon with a stable meaning. But this sort of language can certainly be varied when translated – through rewording, for instance, or replacement by explanatory renderings. The decision to transpose standardized terms from the source text to the translation is an interpretive choice made by the translator in fulfilling a client’s commission and determined by the precisely defined function that is assigned to the technical text” (483-4).

In this sense, technical translators are also tasked with abiding by a hermeneutic mode of translation; even in the most technical of translations, the translator must still be cognizant of the external forces—interpretants—at work, being deliberate in the way our works will take effect in the minds of our readers.


Venuti, Lawrence. “Genealogies of Translation Theory: Jerome.” The Translation Studies Reader, Routledge, London, 2021.


The Invisible Translator of the Localization Industry

On September 10, 2021, translator Jennifer Croft—best known for her achievement of translating the works of Polish Nobel laureate Olga Tokarczuk—wrote an article in The Guardian, succinctly titled “Why translators should be named on book covers.” In the article, she speaks about a strange phenomenon in the realm of literature: translators are simply never acknowledged in any significant way. “Since the 2016 launch of the [International Man Booker Prize],” she notes, “not one of the six winning works of fiction has displayed the translator’s name on the front. Granta didn’t name Deborah Smith there; Jonathan Cape didn’t name Jessica Cohen; Fitzcarraldo didn’t name me; Sandstone Press didn’t name Marilyn Booth; Faber & Faber didn’t name Michele Hutchison. At Night All Blood is Black by David Diop, 2021’s winner from Pushkin Press, doesn’t name Anna Moschovakis on its cover, although its cover does display quotes from three named sources. Four names, in other words, on the cover of a book Moschovakis wrote every word of. But her name would have been too much.”

Croft points out the lamentable experience of the translator, whose name is omitted at every single chance. To those unfamiliar with literature in translation, such omission might make sense. After all, the translator isn’t the author, per se. It’s the author’s book, not the translator’s. One might add that the translator is just a medium, through which the original essence of the author’s work might carry through. 

To what extent should the translator’s presence be visible on the page? Should the translator enjoy as much recognition as the original author, or should they be tucked away somewhere hidden, lest their presence serves to undermine the “authenticity” of the author’s work? These are questions translators—and the publishing world—have been asking for a long time, though nowhere near enough. 

In the first chapter of his groundbreaking book of theory The Translator’s Invisibility, writer, translator, and theorist Laurence Venuti struggles to make sense of these questions, examining the historical and political forces that have shaped the world of literature and translation to see if there are any other ways translators can imagine their role as. The book is aptly titled, and so is the first chapter—“Invisibility”; the cold, hard truth is that translators are effectively effaced from their work, that is how it has always been, at least, in the Anglophone hemisphere. 

Venuti first starts by noting how historically grounded these questions are: that the invisibility of the translator is not a natural human inclination, but rather, a politically induced one. The world of Anglophone literature, Venuti claims, is grounded in a historical movement towards erasing the presence of the translator in literature. Such erasure comes hand-in-hand with the readers’ expectations for fluency. Venuti offers a “selection of excerpts… from various British and American periodicals” that contextualizes the Anglophone love for fluency and naturalness:

It is not easy, in translating French, to render qualities of sharpness or vividness, but the prose of Mr. Gilbert is always natural, brilliant, and crisp. (Wilson 1946: 100)

Rabassa’s translation is a triumph of fluent, gravid momentum, all stylishness and commonsensical virtuosity. (West 1970: 4) 

Helen Lane’s translation of the title of this book is faithful to Mario Vargas Llosa’s – “Elogio de la Madrastra” – but not quite idiomatic. (Burgess 1990: 11) 

The translation by Frank Wynne is fluent and natural-sounding, though I notice that Wynne has now and then clouded the clarity of the cranky ideas. (Berman 2000: 28) 

The English language adheres so much to this sense of naturalness and fluency that we have devised new terms to describe writing we associate with translation—“translationese”—which we use to describe clunky, unnatural, unidiomatically written sentences. 

Why is this? Is it a natural human desire to weed out foreign-sounding phrases in our speech and language, or have we been socially conditioned to do so? Venuti advocates for the latter: “the dominance of fluency in English-language translation reflects comparable trends in other cultural forms, including other forms of writing… [Western] developments have affected every medium, both print and electronic, by valorizing a purely instrumental use of language and other means of representation and thus emphasizing immediate intelligibility and the appearance of factuality” (5). Contemporary English-language translation stems from his history of international development and the U.S. and U.K’s position as world powers, exerting a uniform, imperialistic language. Any form of unnatural English is a subversion of that ideological notion of America, and thus, must be quashed. 

There is also another reason why fluency is valued so much: British and American cultures place a heavy emphasis on “the individualistic conception of authorship” in which “the author freely expresses his thoughts and feelings in writing, which is thus viewed as an original and transparent self-representation” (6). Compare this to the translator’s work: “translation is defined as a second-order representation: only the foreign text can be original, an authentic copy… whereas the translation is derivative, fake, potentially a false copy” (6). Translators approach their work, hence, with a degree of suspicion: am I a fake? Is what I do less important than the author’s work? 

What is worse is that all these antiquated worldviews on translation versus original are backed by the legal system; our “shadowy existence in British and American cultures in further registered, and maintained, in the ambiguous and unfavorable legal status of translation, both in copyright law and in actual contractual arrangements.” In the face of the law, translators are secondary to the author, who retains all copyright of all transmutations of their work in translation. According to the Berne Convention, the author is the only one who can “enjoy the exclusive right of making and of authorising the translation,” and any kind of translations are adoptions, “protected as original works without prejudice to the copyright in the original work”—held by the author.

Such lack of protection leaves translators in an abysmal place; their visibility, or lack thereof, within legal and cultural spaces, actively drives them out. Not protected by law, translators are subject to subpar conditions: “below-subsistence fees force them either to translate sporadically, while working at other jobs… or to undertake multiple translation projects simultaneously” (10). And even then, the fees are abhorrent: according to research, freelance literary translators can’t even make enough to make the cut for the poverty level. Because such a situation “drives freelance translators to turn out translations as quickly as humanly possible… it inevitably limits the literary invention and critical reflection applied to a project, while pitting translators against each other—often unwittingly—in the competition for projects and the negotiation of fees” (10).

Other countries have it slightly better. According to Venuti’s research, Italian publishers published a total of 12,531 works of translation in 2002, which makes up 22.9% of the entire body of published books of that year. German publishers published 5,406 works of translation in 2004, at 7.3%. American publishers, however, published a measly 4,040 translations in 2004 at 2.07%, which is still better than the British in 2001: 1.4%.

Venuti is seriously concerned about this: these statistics not only show the failings of the American and British publishing world to fully embrace translated works as valid forms of literature. Rather, there’s a more sinister force at work, as Venuti claims that “these translation patterns point to a trade imbalance with serious cultural ramifications” (11). The U.S. and the U.K. rarely buy rights to publish English-language translations but sell many translation rights for English-language books, and the results are astonishing:

By routinely translating large numbers of the most varied English-language books, foreign publishers have exploited the global drift towards American political and economic hegemony since World War II, actively supporting the international expansion of British and American cultures. British and American publishers, in turn, have reaped the financial benefits of successfully imposing English- language cultural values on a vast foreign readership, while producing cultures in the United Kingdom and the United States that are aggressively monolingual, unreceptive to foreign literatures, accustomed to fluent translations that invisibly inscribe foreign texts with British and American values and provide readers with the narcissistic experience of recognizing their own culture in a cultural other. (12)

The hegemony of English literature all over the world—and the reactionary movement towards domesticated work and fluency—is directly connected to the cultural, military, political, and economic power of the Anglophone sphere. Everything must be translated, not just in words, but in message and meaning, into a certain Anglophone structure and philosophy. The dissemination of English-language works is universally accepted and widespread, whereas the influx of foreign literature into the Anglophone world is heavily policed and gatekept. Venuti puts it more succinctly when he says: “the translator’s invisibility is symptomatic of a complacency in British and American relations with cultural others, a complacency that can be described—without too much exaggeration—as imperialistic abroad and xenophobic at home” (13).

Venuti speaks more in the chapter about how translators can aim to combat such forces in their work by utilizing, namely, foreignization, which Venuti explains to be “a form of resistance against ethnocentrism and racism, cultural narcissism and imperialism, in the interests of democratic geopolitical relations” (16). For Venuti, foreignizing translation doesn’t mean that the translation is necessarily unreadable; rather, it’s a more deliberate, thoughtful process of translating in which—in the spirit of resistancy—it tries to avoid the narrow kinds of fluency that have long dominated English-language translation” and “challenges the receiving culture even as it enacts its own ethnocentric violence on the foreign text” (18). In doing so, the translator can not only explore outside the bounds of what we consider to be fluent English; moreover, we can formulate new ways in which we can balance legibility and foreignization, making for a richer, more unfamiliar text.

Going back to Jennifer Croft’s article, we better understand the anger of translators in their erasure. The invisibility of the translator is largely a systemic, structural issue, a subjugation, of sorts, of the translator’s autonomy and subjectivity by legal, cultural, political, and historical forces in the making. When Croft says that translators must be recognized and rendered visible, they are actively fighting against this history and system of subjugation, which ultimately affects not only the translators’ livelihoods, but even more so the countries outside of the United States, caught in a hierarchy and imbalance of linguistic trade. 

How does any of this relate to the work we do as in-house or freelance translators working with localization agencies? Unlike literary translators, this notion of the invisibility of the translator affects us only partially; localization specialists and linguists work with both literary (or at least, creative) texts and technical ones. Venuti clarifies that the invisibility of the translator affects mainly literary or “humanistic translation,” not because it is more important, but because “it has long set the standard applied in technical translation (viz. fluency), and, most importantly for present purposes, it has traditionally been the field where innovative theories and practices emerge” (34). It’s not that literary genres are better, or that literary translators have it worse and are more affected by these issues mentioned above. Simply put, technical translation is, in itself, a byproduct of the kind of military, scientific, political forces and is simultaneously constrained by them: technical translation is directly related to “multinational corporations” that “seek the expansion of foreign markets and the creation of overseas workforces and thus increasingly require fluent, immediately intelligible translations of international treaties, legal contracts, technical information, and instruction manuals” (34).

In that sense, non-literary translators—or rather, translators that don’t work with literature directly—are exempt, when working on technical translations, from this work of resistancy. However, it’s difficult to deny that even technical translations—save for the most scientific, legal, or technical of them—offer a little leeway for creativity, provided it doesn’t deter the accuracy of the information to be translated. Everything, in a way, is imbued with the element of politics, culture, and history. Computer manuals, for example, still utilize the master-slave dichotomy; international legal documents still reference colonial terminology. These kinds of matters require the translator to make a conscious, deliberate choice about how they translate. And these kinds of matter—these are the questions that we, as translators, must face and answer.


Venuti, L. (2018). The translator’s invisibility: A history of translation. Routledge, Taylor & Francis Group.


Does Where We Work Influence How We Translate?: Translation Policies in the EU Parliament and UNESCO

Any person remotely acquainted with translation—literary or technical, theory or praxis—will at one point come across the scholar Lawrence Venuti and his dichotomy of domestication vs. foreignization. In his 1998 book The Scandals of Translation: Towards an Ethics of Difference, Venuti, a prominent translation theorist and expert translator, classifies translation on a spectrum: on one side is domestication, which is translation aimed to achieve a certain sense of “authenticity” and “naturalness” in the target language, and on the other is foreignization—a translation that employs the characteristics of the source language, remaining faithful to the linguistic constructions of the source text without assimilating into the target language.

Venuti, whose focus is largely on literary translation, argues for the foreignization of texts in translation. Crudely put, foreignization reminds the readers (of the translated text) that the text comes from somewhere else, effectively transporting the reader to the origin of the text and thereby opening up the reader to the inherent multiplicity of the world. Domestication, on the other hand, erases foreignness from the text, thus upholding a monolithic, centralized idea of a single target language, and in the process effaces the existence of the translator as a subjective medium of ideas and words. 

In most realms of translation—technical, political, literary—domestication is still the preferred mode of translation. A majority of institutions that employ translation services codifies “authenticity” in the target language as a general rule to which translators must adhere. Venuti laments such domesticated translations, pointing out that, in English, domestication has been and still remains the primary mode of translation.

Despite current trends of domestication in English-language translations, translators are always faced with the difficult (not to mention unanswerable) decision of whether to stay faithful to the grammatical and linguistic structure of the source text or to domesticate it so that it sounds more natural in the target language. No work of translation is perfectly domesticated or foreignized; there is a spectrum in which a translator must place their work, domesticating and foreignizing as necessary to best convey the message—if that is their primary purpose or skopos.

A large factor in a translator’s decision to take certain steps toward domestication or foreignization is the institution one belongs to. Many, if not most, localization and translation agencies already standardize certain modes of writing (e.g. the Microsoft style guide, the Google developer documentation style guide), as do academic institutions (e.g. MLA, Chicago, etc.) The same remains for large corporations and major organizations.

In his essay “Text Parameters in Translation: Transitivity and Institutional Cultures,” linguistics professor Ian Mason writes about such ingrained tenets of translation guidelines in institutions and how they affect translators as they make conscious and unconscious decisions as they do their work. Mason uses the European Union and UNESCO as his main examples, focusing on English-French-Spanish translations of speeches and discussions. The questions he asks has less to do with the specific kinds of linguistic guidelines (most institutions are lax about this given the fluidity of language) and more to do with the linguistic phenomena or differences that occur while translating under certain guidelines in place.

Given that most translators work under institutions, answering this question of how much institutions affect translators as they work is an important, pertaining to nearly all translation practices happening around the world. Mason is primarily occupied with the notion of “transitivity,” which “pertains to the way processes are viewed and presented” (400); in other words, transitivity shows “how speakers encode in language their mental picture of reality and how they account for their experience of the world around them” (Simpson 1993: 88). How to translators make sense of a certain text in the source language and configure it to work in a target language? How are the messages in the source text mediated to be conveyed in a certain way in the target language, and in what ways are these messages mediated? These are the kinds of questions that pertain to transitivity, which lies at the heart of translation practices in institutions. By questioning such acts of transitivity, one can examine the effect institutions have on translators; at the same time, transitivity allows us to look at how translators themselves position themselves in terms of their relationship to the discourse of the text at large.

In short, Mason narrows his inquiry down to two major questions: “what evidence is there, if any, of a uniformity of approach across different language sections, consistent with the professed aims of the institution?” Second, “to what extent to actual shifts of transitivity contribute to signalling significantly different values at the level of text and discourse in translated documents?” Mason notes that as of now, there aren’t definitive answers to the questions above, and that his inquiry is small-scale. However, even small-scale studies such as this reveal much about the nature of contemporary translation practices.

So what kind of guidelines or policies to institutions implement for translators? Mason provides several examples. First, the Canadian federal government has a “translation doctrine” that states that a translator should render “not the words or the structures of the source-text but rather the message or, in other words, the author’s intention” (Translation Bureau 1984: 3; emphasis by Mason). Furthermore, translators working for the Canadian government should aim for authenticity, in which “authenticity is the impression conveyed by a translation that it is not, in fact, a translation, that it was composed in the target language from the outset, that it is an original piece of writing” (Translation Bureau 1984: 6, cited in Mossop 1990: 347n.)

Such policies obviously advocate for domestication, per Venuti’s dichotomy, and thus propagate the notion of the “invisibility of the translator” popular to “Western (and especially Anglo-American) culture.” This rationale is echoed by the European Union Commission, whose Translation Service is characterized by “a clear, albeit unwritten, preference for surface-level similarity, which is assumed to guarantee that readers of the various translations all get the same message” (401). Translations are completed “as if texts were drafted in all languages simultaneously, as if no source text existed” (401). 

Mason notes that the UNESCO, on the other hand, aims for accuracy as “the very first requirement” for translators, although UNESCO periodicals still insist on “a ‘readable text.’” But all these words thrown around—readable, authentic, clear, similar—are very vague terminology in the fields of linguistics and translation. By what standard do we measure similarity, authenticity, and readability? How do we judge that a translation is similar/authentic/readable when every language is, despite their lexical and linguistic similarities, markedly different in their configuration of ideas and messages?

To approach these questions as directly as possible, Mason compares source and target texts, examining possible trends and patterns in translation techniques to see if there are any correlations to the guidelines mentioned above. However, Mason notes that such an examination is extremely hard to carry out, as it is difficult to determine if certain techniques (“shifts” in transitivity) are natural, indispensable processes when translating from language A to language B or are decisions influenced by the institution. An example Mason provides is “the pronominal verbal construction in Spanish, whereby a process can be presented in an agentless way and the translator has to find an alternative structure in English (often the passive)” as is the case in:

De no hacerse así — y no se ha hecho así — el riesgo que se corre […] (Calzada Pérez 1997: 153)

which is translated into “if it is not done this way — and it has not in fact been done in this way — the risk that is run […]” (403). Here, it is difficult to tell if such a shift in agency has been made as an individual choice by the translator or because all Spanish pronominal processes are treated this way in English as a result of certain linguistic constructions.

Mason introduces other problems he ran into while carrying out his study, such as the fuzzy “boundary between disallowed structures and those which are (more or less strongly) dispreferred” (403). The French sentence “Cela permet d’évacuer rapidement le personnel” can be translated as “This allows rapid evacuation of staff,” although “in this way staff can be evacuated quickly” is a shift in transitivity that would be more preferable. Another problem deals with “the issue of the relative significance of shifts,” such as in the translation of the French phrase “Ma pensée va aux victimes,” which is shifted to become “My thoughts are with the victims” in English: shifts that are conventionally necessary yet insignificant.

Having abandoned quantitative research for these reasons, Mason undertakes a brief analysis of translations, making general yet non-normative observations of translation practices within the institutions of the EU and UNESCO, focusing on “the translator’s creativity and the limits which translators themselves impose on this” (404).

The EU Parliament. Image credits: DW

First, Mason observes that “the translations of the speeches to the European Parliament stay relatively close to the transitivity patterns observable in the source texts (STs)” (404). In other words, translators don’t deviate too far from the source text, making a conscious decision to remain faithful to the source text; Mason deduces that a motivating factor for this can be “the sensitivity of pronouncements by prominent politicians and the need to avoid misrepresenting not only intended meanings but the words” (404), which many translators will empathize with and have experienced at least once when translating official documentation and the like. On the other hand, in the UNESCO Courier, the periodical that Mason is researching, translators “display greater latitude, as befits the field of journalism where ease of processing by the reader of the translations may be seen as a high priority” (405). Already, there is difference in translation practices based on the institution and medium under which translators work.

Mason then notes that, with translations for European Union Parliament, shifts occur “for the sake of idiomatic preference.” For example, “English material processes frequently become French nominalisations; French active processes become English passives; Spanish ‘se hace’, etc. becomes French ‘on fait’, and so on” (405). Furthermore, “in English, there is often personalisation of actors in material processes, where in French and Spanish the actor is not made explicit” (405). So far, all these shifts in transitivity are routine and normal, used by experienced translators to make for “natural expression in the languages concerned” (405). Mason also notes, however, that the reverse processes also happen: “personalisations may be added in translations from English to French and English nominalisations become French material action processes”: for example, the English “adoption” becomes “l’adopte” and the English “implementation” becomes “mettre en œuvre.” It is safe to assume, then, that “a heterodox range of approaches to the task co-exist in both institutions” (405).

At the same time that these shifts take place, there is also “a high incidence of calques,” or loan translations, which Mason describes as “characteristic of a widespread strategy—evinced in both institutions—of adhering as closely as possible to the formal arrangement of the [source text]” (405). This happens in between English and Spanish, as seen below:

Source text: By destroying accumulated wealth and the sources of future production, total war has sharply increased the pressure of existing populations upon their resources and has thereby sharply curtailed the liberties of vast numbers of men and women, belonging not only to the vanquished nations, but also to those which were supposed to be victorious.

Target text: Al destruir la riqueza acumulada y las fuentes de la producción futura, la guerra mundial ha aumentado intensamente la presión de la [sic] poblaciones existentes sobre sus recursos, y, por lo mismo, ha mutilado gravemente las libertades de un vasto número de hombres y mujeres pertenencientes no sólo a las naciones vencidas, sino también a aquellas que se suponían victoriosas.

Such calques also happen between English and French:

Source text: It [the accident rendered a large number of houses uninhabitable and affected the electricity distribution system.

Target text: Il a par ailleurs rendu inhabitables de nombreuses maisons et affecté le système de distribution électrique.

Mason observes that calques occur most frequently between French and Spanish, given the “syntactic similarities of the two languages”:

Source text: La orientación de la PAC ha favorecido la aparición de ciertos problemas. La búsqueda de la competitividad a cualquier precio favorece la introducción de métodos y técnicas cuyas consecuencias a largo plazo se desconocen.

Target text: L’orientation de la PAC a favorisé l’apparition de certains problème. La recherche de la compétitivité à tout prix favorise l’introduction de méthodes et de techniques dont les conséquences à long terme ne sont pas connues.

Of more interest, however, are shifts in transitivity that betray a more conscious decision by the translator to deviate from the source text for the purpose of pursuing a certain discourse. Mason notes that, in some “French-to-English translations of one parliamentary debate, there are instances of a move towards increased directness affecting process, participants or circumstances,” which consequently “serve to intensify some aspect of the overall process” (406). Here are some examples:

Source text: La responsabilité du trust TotalFinaElf […] est entière.

[The responsibility of the TotalFinaElf trust… is complete.]

Target text: The TotalFinaElf corporation […] is fully responsible.

Source text: Le groupe TotalFinaElf récidive, de manière tragique

[The TotalFinaElf group is committing another offence, in a tragic way]

Target text: The totalFinaElf group […] has acted criminally, once again

The above examples are examples of intensification via a wide array of moves such as re-lexicalizations and shift of attitudes, among others. “What is most striking, however,” remarks Mason, “is a general tendency in these translations to move further in the direction of perceived intended meanings.” In other words, the translator has understood the discourse of the French source text to be one of blame, and thus justifies their translations to “intensify the blame or signal dissent,” given the “plenty of discoursal signals in the co-text” (407). With such different strategies exhibited—from linguistic deviation for discoursal similarity to near-equivalent calques—Mason comes to the conclusion that “the treatment of transitivity patterns varies widely within each institution and within each language pair” (407).

What is most interesting, however, is what Mason discovers in the UNESCO Courier as a trend in which “active processes [become] passive ones”; as it happens no fewer than 16 times, we can assume that “an overall trend is established in which processes may be viewed as happening independently of agents [the translators] or at least the dynamism of actors in processes is reduced” (408). Mason quotes from the December 2001 issue of the Courier—the final edition:

Source text: L’Egypte conquit […] son autonomie.

[Egypt won… her autonomy.]

Target text: Egypt came into her own.

Source text: la mystérieuse présence par laquelle les œuvres de l’Egypte s’unissent aux statues de nos cathédrales

[the mysterious presence whereby the works of Egypt unite with the statues of our cathedrals]

Target text: the inexplicable quality which brings the Egyptian masterpieces into communion with the statues of our own cathedrals

In these examples, Egypt is turned from a subject with agency into an object, something Simpson (1993: 89) calls “an action supervention process, that is where the process may occur independently of the volition of the actor” (408). This phenomenon is also exhibited in a speech by a Spanish MEP who is “critical of the British government’s handling of the crisis over BSE (Bovine Spongiform Encephalopathy or, more popularly, Mad Cow Disease)” (408):

Source text: La supeditación de las decisiones políticas a las presiones económicas en el Reino Unido está en el orígen de la problemática inherente a la EEB.

[The subordination of political decisions to economic pressures in the UK is at the root of the inherent problem of BSE.]

Target text: The underlying problem with BSE is that political decisions have been subordinated to economic pressures in the United Kingdom.

Source text: La enfermedad se originó con la introducción de harinas de carne.

[The disease began with the introduction of bone meal.]

Target text: Cette maladie est due à l’introduction de farines de viande…

[The disease is due to the introduction of bone meal.]

Note that the English translation veers towards attenuation, whereas the French translation heads toward intensification. Mason directs attention to how the English translation “concentrates on the results of [the] subordination… merely [claiming] the subordination and [relegating] the UK from being (part of the Actor to a Circumstantial” (409). On the other hand, Mason analyzes how the French translation “introduces direct causation (est due à/is due to) and thus enhances the accusatory illocutionary force, its force as a speech act that performs the action of accusation, by making it explicit” (410).

All this is to say that “the treatment of transitivity patterns varies widely within each institution and within each language pair” (407). The European Parliament translations tend to remain faithful to the source texts, where as the UNESCO Courier translations vary a bit more with their shifts in transitivity. Calques and radical shifts “co-exist in both sets of data,” eventually pointing to the conclusion that “there is… little uniformity of practice or evidence of influence of institutional guidelines on translator behaviour” (410). Mason contrasts this, however, by saying that translators, in all their adherence and deviation from the source text, “[displays] traces of other discourses, fain echoes of ideological stances which are present in the environment (and which, by their very nature, are transindividual)” (410).

Mason’s inquiry into the relationship between the institution and the translator is both thought-provoking and not fully explored; there is simply too much to cover in a study that examines institutional policies and their influence on translation practices. However, Mason’s analysis does provide some evidence that necessary translation practices and individual inclination toward discourse (justified, we should note, by the source text) are more prevalent than actual influence by institutions. 


Venuti, Lawrence, and Ian Mason. “Text Parameters in Translation: Transitivity and Institutional Cultures.” The Translation Studies Reader, Routledge, London, 2021.


Michael Cronin’s “The Translation Age” and the Role of Localization Agencies Today

There are not many vocations out there in which one must prove their necessity quite as much as a translator. A farmer, for example, is a noble position that directly affects the physical well-being and national security of a country—perhaps many countries. A businessman might provide jobs for other people, all the while facilitating commerce and trade both in and out of the country. But translators are often asked the question—both in their public and private spheres—“what is the point of translating for a living when there are machine translations available for free?”

The death of the translator is a tenuous, worn-out discourse. In a world where technological advancements in the form of Google Translate and DeepL have bridged linguistic gaps, it’s obvious that people should ask such questions. If anything, contesting the existence and necessity of professional translators is important, if anything, to goad translators into contemplating more on why their vocations are as important as others. Before computers, translators might’ve had an easier time positing themselves as requisite figures in society, but now, translators often find themselves as the butt of their friends’ jokes in reunions.

In his essay, “The Translation Age: Translation, Technology, and the New Instrumentalism,” Michael Cronin pleads with his readers to think of translation in a more nuanced light. Cronin recounts the general history of translation and its role, at once subverting the tired notion of translation as a mere substitution of words and redefining translation beyond its usual understanding. And the conclusion he comes to is that a human translator is no longer a scribe—never was—to be ousted by machines. Rather, human translators—aided and informed by machines—stand at the gates of information as it passes from one country to another, from one language to the next. In fact, all form of understanding and meaning-making—whether it be programmers coding, musicians singing, scientists experimenting—is a form of translation. But what is that even supposed to mean? 

Cronin starts his essay by positing that “human presence in the world can only be understood through and in the context of the made objects that mediate human existence” (470). In other words, we can perhaps say that we are the product of the tools we employ; Cronin mentions twelfth-century scholar and translator Adelard of Bath who, in his treatise On the Use of the Astrolabe, expresses the idea that “knowledge is inexpressible outside the language of artefacts” (469). 

Tools are important in the human effort to understand what it means to be human, Cronin says. Citing the archaeologist Timothy Taylor, Cronin makes the argument that “evolution for humans is, in a sense, both biological and cultural.” After all, aided by our tools and structures, humans do not need physical attributes to help us survive in the wild. Humans depend on the inanimate—our tools, buildings—so much for our survival that they may as well be a part of us. Or as Cronin says, “we need to take account of the intrinsic and not simply extrinsic involvement of technē. It is a question of ontology rather than utility. We evolve or are defined by the artefacts we use. The tools shape us as much as we shape them” (470-471).

What does any of this have to do with translation? Well, translation, in short, is a tool we employ to evolve. Cronin would say that “there is no transcendence without translation.” Like the famous painting The Tower of Babel by Pieter Brueghel, humans aim for higher worlds—for a united language and front against regression—through the use of tools, such as the “ladders, levers, pulleys, scaffolding” present in the painting. Without this tool of translation, there is no transcendence; Cronin cites nineteenth-century writer Samuel Butler, who defines writing as follows: “The written symbol extends infinitely, as regards time and space, the range within which one mind can communicate with another; it gives the writer’s mind a life limited by the duration of ink, paper, and readers, as against that of his flesh and blood body” (Butler 198). Translation is how we extend our ideas to different areas and societies (English to other languages) and through time (Old and Middle English to Modern English).

That still doesn’t give readers any clear idea of how modern-day translation as we think of it—localization agencies, freelance translators—prove their necessity in the era of the internet. So Cronin, in this next section, brings out the big guns, namely the printing press. Cronin quotes Francis Bacon, who once claimed that “no empire, no sect, no star seems to have exercised a greater power and influence in human affairs” than the movable-type printing press. And it certainly sounds true. Writing, which was once the private pleasure and activity of the ruling elite soon became a commonplace medium of information, available in markets and in average households. And while the Gutenberg press is venerated for having spread translations of the Bible, what’s really more important than the content is the medium itself: the role of the printing press to democratize information. 

The Gutenberg Press. Image Credits: The Britannica Encyclopedia

And these mediums shape us, in the same way tools shape humans. Quoting the American writer Nicholas Carr, “every intellectual technology, to put it another way, embodies an intellectual ethic, a set of assumptions about how the mind works or should work. The map and clock shared a similar ethic. Both placed a new stress on measurement and abstraction, on perceiving and defining forms and processes beyond those apparent to the senses” (Carr 2010: 45). The map and clock have renegotiated the relationship of the human to the world; tools—technology, overall—have the power to reposition the paradigm we utilize to think about the world. 

For the printing press, the advent of such a revolutionary technology “implied an intellectual ethic of mobility which would be hugely significant for the role of translation in religious and political history.” For example, Bibles translated into vernacular English were forbidden in England, but the notion that such a translated Bible could now be published—in England or abroad—via the printing press is, itself, a great subversion to political and religious ideology. Furthermore, there is the “pressure to standardize spelling,” a further consolidation of a certain English “self-consciousness” that the printing press brings, so much so that translation, in a sense, is a “culture-technology hybrid,” invented by man yet shaping so much of the intangible culture humans exist in. “Through print, translations colonize the space of the everyday, menacing in their accessibility,” Cronin writes, telling the story of how translations of Italian tales spread virally through late sixteenth-century England, much to the dismay of the ruling elites who prioritized a certain sense of English, religious pureness in print (Cronin 474).

Fast forward a few centuries. The world has changed so much, yet the way in which tools influence our very lives has not changed at all. 2022 is the heyday of ubiquitous computing—computers everywhere, in our hands, our surroundings, our homes, our furniture, even our bodies—to the point that “computing capacity dissolves into the physical surroundings, architectures and infrastructures” (475).

In such a “trans-architectural” world, the multilingual is impossible to omit. No longer do we simply read books in a uniform language; through technology such as smartphones and laptops with their constant notifications and ease of access, we have moved from “the static and serial presentation of information in a limited number of languages” to a “customized interaction with the possibility for continuous expansion in languages and information offered.” Most of the information we need, we access in the language of our choice, in the comfort of our homes and relative solaces. 

And in such a world, translation takes on a new form. Translation is still, in many ways, a necessary and important tool that humans undertake to disseminate information and shape society; it’s hard to argue the sheer importance translation holds in modern society when all major scientific developments and technological advancements required some form of cross-linguistic, cross-cultural, cross-temporal translation. But now, the agents of translation have shifted. Before Google Translate and the general internationalization of English, professional translators and bilingual scholars have taken pride in being the only ones to disseminate new ideas from abroad. Now, translation is a communal, collaborative experience, evidenced by the “proliferation of crowdsourced translation or open translation projects such as Project Lingua, Worldwide Lexicon, Wiki Project Echo, TED Open Translation Project, and Cucumis,” alongside Facebook which has “used a crowdsourcing model to translate content into languages other than English” as well as fan translations for “everything from Japanese anime to Korean soap operas” (476). Such is the state of translation in the era of the internet and Web 2.0, characterized by “interactive, user-generated content” (476).

Michael Cronin sums it up into three major ideas. First, translation is now a mode of prosumption; in other words, the consumer of a translation is also an active producer. The audience is at once the producer, and this phenomenon redefines the diametric dichotomies that translation theorists have spent decades pondering over. 

Second is post-print translation literacy; the internet has dramatically transformed the way people read, “from steady, cumulative, linear reading to a form of accelerated power browsing” (477). Cronin cites a report by the Israeli company Clicktale that found that “in most countries people spend between nineteen and twenty-seven seconds looking at a web page before moving on to the next one” (477). And with such transformation to the “paradigms of literacy… we must expect translation to change in nature” (477). No longer is quality the priority for many; fast information—one can’t help but think of the recent debacle over poorly-paid, overtime-working translators for Netflix’s Squid Game—is king. 

Third, such collaborative and communal translation processes represent a “reinvestment of translation technology by the human,” which effectively contests the “conception of machine-human interaction in translation as fundamentally dehumanizing” (477). Cronin writes that “a tendency in localization discourse has been to accentuate the role of automation in translation activity and to minimize the intervention of the human agent,” to which one can give the example of the recent rise of MTPE—machine translation post-editing—as a feasible, cost-efficient solution to traditional translation practices. And in this transformation, translation moves from “the monadic subject of traditional translation agency… to a pluri-subjectivity of interaction” (477).

These ideas—or rather, the central idea that translation is no longer a professional, solitary job and more of a crowdsourced, collaborative one—seem to disprove, if anything, the necessity of the localization industry. But that is not what Cronin argues for; up to this point, Cronin has argued for a wider understanding of translation, one that regards the practice as expansively transformational and utterly indispensable to the human experience and evolution. “The global expansion of business demanded that translation carry messages from one point to another, leading to the development of the localization industry,” writes Cronin, suggesting the notion that the localization industry is still key to that integral process of cultural dissemination. Cronin goes on to quote Reinhard Schäler, a scholar of translation theory: “Localization can be defined as the linguistic and cultural adaptation of digital content to the requirements and locale of a foreign market, and the provision of services and technologies for the management of multilingualism across the digital global information flow.” (Schäler 2009: 157)

From here, Cronin goes one step further and argues that “when we talk about the Information Age, Information Technology, and the Information Society, we should be talking about the Translation Age, Translation Technology, and the Translation Society.” In other words, translation is information; information is not possible without translation, not the opposite, as we so often think. When Samuel Morse designed his code—to be used for the telegraph—all he did was replace signs (alphabet) with other signs (Morse code), according to Cronin (480). This process of “replacing signs with other signs, mapping one set of objects onto another—it might be argued that this is precisely what translators do. They are continually engaged with forms of encoding, moving from one symbolic level or system to another,” argues Cronin (480). In this sense, translation can be interpreted to include “the ultimate translatability of all content to the binary code of machine language.” Our very own relationship with machines is that of translation. We shouldn’t be thinking of all the ways translation fails us (the so-called notion of “lost in translation”), but rather, thinking of all the ways in which something is a translation: a passage of ideas from one sign to another.

Samuel Morse, c. 1857. Image Credits: Wikipedia

No one can deny the “transformative impact of information technology.” But what Cronin is ultimately trying to prove is that “the history of information and information technologies is if anything a history of forms of translation.” All science comes from the Greeks, who got it from the Egyptians, who took it from the Hebrews, says sixteenth-century English translator John Florio.

And this is the power of Cronin’s “The Translation Age: Translation, Technology, and the New Instrumentalism”; by recounting the history of translation and its tremendous influence on humankind, Cronin reconfigures our relationship to the world around us, helping his readers view the world in terms of translation. In that aspect, translators are harbingers of ideas, messengers that enable human evolution. Our lives consist of translations, are borne of translations. 

With such a general, expansive definition of translation, it is worth asking what role localization experts and professional translators play. If translation is any form of meaning-making via a transfer of signs, aren’t computer programmers translators? Aren’t musicians translators? Aren’t farmers translators, taking the signs of land and agriculture and transforming them into the signs of food and commodity? In a sense, all these are true. According to Cronin, translation is effectively technology, that process through which humans extend functions and ideas to different modes of understanding.

However, despite Cronin’s argument that the medium is more important than content (or at least, of equal relevance), even within such a one-size-fits-all definition of translation, professional translators occupy a special role of translating one human language to another. Natural language, unlike other signs, is compounded by a variety of elements—history, culture, technology, politics. Natural language is a code that, unlike most codes, is fraught with contexts. And given that most people in the world speak at least one natural language, the implications of natural language translation are endless. The results of professional translation affect many, and the circumstances that have made natural languages so heavily contextual and fraught with risk are beyond our scope of control.

Professional translators, unlike other forms of translation, differ fundamentally in that their job is to facilitate communication directly between one person to another. Cronin’s paradigm of a universalized translation only goes to prove that translation is an ever-present, ever-necessary function of human life and experiences, the most exemplary of which is that process of making ourselves understood, human-to-human, culture-to-culture, language-to-language. That is the kind of professional translation that Sprok DTS stands for. 


Venuti, Lawrence, and Michael Cronin. “The Translation Age: Translation, Technology, and the New Instrumentalism.” The Translation Studies Reader, Routledge, London, 2021.


Meta Makes a Breakthrough with a 200-Language MT Model

Last week, Meta announced that they had made their first breakthrough in their No Language Left Behind project, which is their effort to “develop high-quality machine translation capabilities for most of the world’s languages. Their first breakthrough, which is explained in full detail in this blog post and this paper, is a single AI model called NLLB-200, a state-of-the-art model capable of translating between 200 different languages, many of which are low-resource languages. Compared to the >25 African languages covered by front-end translation tools, NLLB-200 supports 55 African languages with “high-quality results.”

That has been Meta’s goal: to center previously marginalized low-resource languages by incorporating them into the general trend towards MT improvement. In the introductory video posted on Meta’s NLLB website, research engineer Vedanuj Goswami speaks of his grandmother, who used to write poetry in Assamese—a low-resource language with little online presence yet spoken by over 23 million speakers in Northeast India. “So my grandmother writes poems in Assamese, which is a low-resource language,” Goswami says. “and I envision a future in which where I can easily translate her poems into a high-resource language, so that the world can appreciate her poems.”

We have covered Meta’s frequent forays into advancing machine translation technology for quite some time, and we are surprised each time at how faithful Meta stays to its mission of uniting the world through the elimination of language barriers. More than anything, we are struck by how human-centered and intrinsically meaningful their work is; the researchers at Meta have a profound understanding of the historical and political reasons that have affected the currently biased state of machine translation and focus on democratizing and leveling the playing field for speakers of all languages. Meta’s research paper on NLLB-200 claims the following: “those who communicate in English, French, German, or Russian—languages which have long enjoyed institutional investments and data availability—stand to gain substantially more from the maturation of machine translation than those who speak Catalan, Assamese, Ligurian, or Kinyarwanda… Many of these languages escape researchers’ gaze for a confluence of reasons, including constraints conjured up by past investments (or lack thereof), research norms, organizational priorities, and Western-centrism to name a few. Without an effort to course correct, much of the internet could continue to be inaccessible to speakers of these languages.”


The challenges

While numerous technical difficulties plague MT development for low-resource languages (data scarcity, high-cost data procurement, for example), Meta is determined to overcome these challenges by utilizing multilingual systems. NLLB-200 is one big step in the journey to a universal machine translation model. However, it is important to identify the specific challenges that low-resource languages face, especially as they are integrated into a multilingual framework such as NLLB-200.

Modern MT systems rely heavily on large amounts of data; these data-hungry systems require “millions of sentences carefully matched between languages.” To account for the lack of such parallel datasets, current models mine data from the web, leading to poor data quality due to a difference in the source text for each language. Furthermore, the researchers note that such data is “often full of incorrect or inconsistent spellings and is missing accent marks and other diacritical marks.”


How it works

Structure-wise, the Meta researchers utilize a four-step approach in developing No Language Left Behind. First, the researchers analyze and identify problems in low-resource translation from the viewpoint of the speakers of those languages. Second, the researchers learn to create training data for the low-resource languages. Third, they use this data to develop cutting-edge translation models. Lastly, the researchers evaluate their efforts and results. The process is summed out neatly in the following diagram:

That is not all, however; NLLB-200 was not created out of a vacuum devoid of ideas, but rather, builds on years of research previously done. The researchers note the importance of FLORES-200, a “many-to-many multilingual dataset that allows [researchers] to measure translation quality through any of the 40,602 total translation directions.” According to the Meta blog post, FLORES-200 has the power to evaluate translation systems across various media, including “health pamphlets, films, books, and online content.” With FLORES-200, the model undergoes extensive scrutiny so that the language standards that evaluate the model are rigid and of high quality.

There’s also LASER3, which helps researchers “mine web data to create parallel datasets for low-resource languages.” LASER3 is an upgraded version of LASER, previously used for “zero-shot transfer in natural language processing (NLP)”; LASER3 utilizes a Transformer model “trained in a self-supervised manner with a masked language modeling objective.” In applying LASER3 to NLLB, the researchers were able to produce massive amounts of sentence pairs for all languages, including low-resource ones.

Developing and utilizing such evaluation methods and data creation techniques allowed the researchers to develop “mixture-of-experts networks” which allow for smoother integration of low-resource languages into the multilingual models. Such networks allowed the researchers to “iterate, scale, and experiment with a single multilingual model much more easily than with hundreds or even thousands of different bilingual models.”

At the same time, the researchers note that they were always wary of data toxicity and quality. Using LID-200 models, the researchers filtered out the data and eliminated noise from the data gleaned from the internet. Then, the researchers composed toxicity lists, which were used to “assess and filter potential and hallucinated toxicity.” All this serves to clean up and improve the data and its quality, “reducing the risk of what is known as hallucinated toxicity, where the system mistakenly introduces toxic content during the translation process.” Example “hallucinations” can be found below.

The researchers proudly announce that “in total, NLLB-200’s BLEU scores improve on the previous state of the art by an average of 44 percent across all 10k directions of the FLORES-101 benchmark.” What is more exciting, however, is that for some of the African and Indian languages covered by NLLB-200, the increase in score is more than 70 percent compared to modern translation systems. Shown below is the comparison between NLLB-200 and existing state of the art models.

Lastly, the researchers conducted human evaluations on the various languages covered by the models, assessing the effectiveness of the NLLB-200. Through a process of analyzing the results and evaluations of both human and machine, the researchers were able to “reflect on the creation process, analyzing the risks and benefits of our research from a societal standpoint.” To that end, the researchers have open-sourced all the data, models, and benchmarks relevant to the NLLB-200 to aid further research carried out by individuals and third parties. 

All this makes for an efficient multilingual system that effectively leverages new techniques of data generation and language integration to power low-resource language translation in ways never attempted before. The integrated diagram of the entire process is detailed below, taken from the paper. Below that is a timeline of all the previous developments that have come before NLLB-200.



Meta has partnered with the Wikimedia Foundation—renowned for hosting Wikipedia and other relevant sources of free information—to improve the status of low-resource languages in Wikipedia. The researchers note that there are large disparities in information that don’t reflect the true population of low-resource language speakers. For example, “there are around 3,260 Wikipedia articles in Lingala, a language spoken by 45 million people… contrast that with a language like Swedish, which as 10 million speakers in Sweden and Finland and more than 2.5 million articles.” 

This will be the first major application of NLLB-200, although Meta’s decision to open-source their entire project means other groups are welcome to utilize the model for better communication purposes. In fact, Meta AI is also providing “up to $200,000 of grants to nonprofit organizations for real world applications for NLLB-200,” in hopes that their “slew of research tools” will “enable other researchers to extend this work to more languages and build more inclusive technologies.”

Meta’s dream of a linguistically connected world doesn’t just stop at facilitating communication between speakers of different languages, nor does it only pertain to social media, commerce, and other conventionally important fields in today’s rather capitalistic societies. While written media today are dominated by a handful of languages, the researchers believe that “NLLB will help preserve language as it was intended to be shared rather than always requiring an intermediary language that often gets the sentiment/content wrong.”

The word “preserve” catches our attention. Not only will NLLB-200 translate low-resource languages into high-resource ones, but in the reverse process—of high-resource languages being translated into low-resource ones—written data in the low-resource languages will have been created. And in fostering the use of low-resource languages (as a result of mitigating conformity to high-resource languages), the world will manage to stay as linguistically diverse as possible, all the while giving speakers of low-resource languages the power to scribe their languages in writing (or perhaps, speech-format data), effectively “preserving” their linguistic existence in some shape or form.

The researchers also dream of the NLLB project helping to “advance other NLP tasks, beyond translation.” They give examples of “building assistants that work well in languages such as Javanese and Uzbek or creating systems to take Bollywood movies and add accurate subtitles in Swahili or Oromo.” For speakers of high-resource languages, it is hard to fathom a world in which a piece of writing or visual media has not yet been translated into their language. 

Many people are under the assumption that adopting a lingua franca (English, in this era) is the most effective, democratic way to communicate with others. After all, what is more democratic than a shared language? However, the Meta researchers’ efforts aim to deconstruct this notion of a lingua franca. With the technological and scientific developments of this age, there is no reason for people to succumb to the power of a more influential language. In this sense, the researchers announce their mission statement: “the ability to build technologies that work well in hundreds or even thousands of languages will truly help to democratize access to new, immersive experiences.” In 2022, democratization is not assimilation. It is accepting that the future is many, not one.




New Post-Editing Paradigm Ties for Best Paper at ACL 2022

The North American Chapter of the Association for Computational Linguistics will host its 2022 Annual Conference taking place this year in a hybrid format—both online and in-person in Seattle, Washington. The conference will be taking place from July 10 to 15 and will be bringing together researchers and scientists in the computational linguistics, natural language processing, and machine translation fields from all over North, Central, and South America. The conference features tutorial sessions, workshops, and a variety of talks and sessions regarding the numerous topics within the aforementioned fields of study. 

Hundreds of papers, long and short, were accepted by NAACL; of them, “Automatic Correction of Human Translations” written by Berkeley researchers and translators at Lilt Services tied for the best new task paper and won best new resource paper. Today, Sprok DTS would like to introduce some of the findings outlined in this paper, as we believe that the research carried out by these scholars and translators will greatly impact the course of machine translation as we know it.


Automatic Correction of Human Translations

Authors: Jessy Lin, Geza Kovacs, Aditya Shastry, Joern Wuebker, John DeNero

We are already familiar with the notion of post-editing, or automatic post-editing (APE), in which human translators are tasked with correcting machine-powered translations. While post-editing is believed to reduce the time spent translating and thus has been adopted as a genuine  translation service worldwide, the authors of the paper point out an important fact: “a tremendous amount of translated content in the world is still written by humans.” And while we believe human translation to be nearly perfect and fluent, that is often not the case as human translators also “do make errors, including spelling, grammar, and translation errors.”

Lin et al. in their paper “Automatic Correction of Human Translation” introduce the idea of “translation error correction” (TEC), which reverses the role of the human and machine in a traditional APE setting, instead assigning machines with the task of correcting a translation completed by a human translator. This flipped paradigm more accurately depicts the role of human translators as bearing the brunt of translation work. TEC not only poses a new paradigm of effective, efficient translation but also examines the current state of development in the field of machine translation, seeing how far we’ve come in enabling a machine to enhance and improve our writing.

In short, Lin et al.’s paper consists of four parts. First, the authors release ACED, “the first corpus for TEC containing three datasets of human translations and revisions generated naturally from a commercial translation workflow.” Then, the authors utilize ACED to analyze the types of errors that human translators make. Third, the authors train TEC, adjusting the corpus so that the types of errors shown are more human-like in nature. Lastly, the TEC model is evaluated in its usability via a human-in-the-loop user study. In doing so, the authors empirically prove that TEC is a novel, quasi-viable method of machine translation.


The ACED Corpus

One of the paper’s strengths is that it offers a whole new corpus; given its inquiry into human-made errors, the corpus is a real-world benchmark, “containing naturalistic data from a task humans perform, rather than manually annotated data.” The translations in the corpus were carried out by professional translations at Lilt, a localization service provider; all the translators involved in the paper have at least 5 years of experience under their belts. 

The authors note that the ACED corpus is diverse. The corpus consists of data from three accounts, each from a different domain and exhibiting different kinds of errors and difficulties. There is the ASICS dataset, which is primarily marketing material for ASICS, complete with product names and the like. The second is the EMERSON dataset, which includes industrial product names and vocabulary in the manufacturing domain. The third is the DO dataset, which are translations of software engineering tutorials. 


Error Classification

The authors classify the human errors/edits present in the dataset into three major categories: monolingual edits, bilingual edits, and preferential edits. Monolingual edits are “identifiable from only the target-side text,” and include typos, grammar, and fluency errors as subcategories. Bilingual edits deal with “mismatches between the source and target text,” such as mistranslations and omissions. Preferential edits are more nuanced edits, marked as “inconsistent with the preferences of the customer” and include terminology or stylistic preferences.

TEC is different from previous research, the authors claim, because concerns itself with error correction, not error detection and quality estimation. After annotating error labels for the three datasets (random samples except for ASICS, which was evaluated in full), the percentage of each error type was calculated. Compared to APE, in which 74% of sentences are marked as having a fluency error, only 22% of the ACED corpus are marked as such; instead, ACED features more monolingual grammar, bilingual, and preferential errors which are “notably underrepresented in APE.” It is true; APE is mainly geared toward improving the fluency of a text, as human translators are grappling with the rather machine-like translationese of MT. TEC is substantially different in this aspect, as human translations are more prone to humanlike errors—grammar, word usage, and omission.


Building a TEC Model

Now comes the technical part: the authors propose a TEC model and compare it to other  models that are “designed for related tasks,” so as to “determine whether they are also effective for TEC.” All the models are designed to use the Transformer neural architecture, and are “trained on 36M sentences from the WMT18 translation task.” The authors also use a “joint English-German vocabulary with 33k byte pair encoding subwords.”

Specifically, the authors employ 5 models total for this research: TEC (the premise of this research), MT, GEC, APE, and BERT-APE. Each model is configured slightly differently. For example, the English-German neural machine translation model is trained and fine-tuned on only the source and revised translation, omitting the original translation. The encoder-decoder (monolingual) GEC model utilizes a process that corrupts the revised translation, ignoring the source sentence. APE and BERT-APE are also built, modeled, and processed differently in their own ways, making ample use of the three parts of ACED’s data tuple: the source sentence, the original translation, and the revised translation.

Image credits: “Automatic Correction of Human Translations.” Here, s refers to the source text, t to the original translation, and t’ to the revised translation.


The Results

The authors utilize MaxMatch scores (M2) to score the performances of the models. M2 is a “standard metric for GEC that aligns” the original translation and the edited translation to “extract discrete “edits.”” When rated using M2, TEC exhibits the best overall score on all datasets. The MT model and the GEC model both underperform. Despite their structural similarities, TEC and APE show a large difference in performance. As for the BERT-APE model—which is “first fined-tuned to achieve state-of-the-art on APE before fine-tuning on ACED”—scores low, as it makes one too many edits, leading to low precision. In sum, the authors conclude that their “results emphasize that models that excel at correcting machine errors cannot be assumed to work well on human translations.”

Image credits: “Automatic Correction of Human Translations.” Note the high performances outlined in bold.

The authors carry out another analysis, this time of the fluency and “per-category errors” using the ASICS dataset. The analysis is designed to calculate “overall sentence-level accuracy” and the “accuracy per error type.” For these analyses, the TEC model also “achieves the best score overall on both alternative metrics over all sentences,” although “various models outperform on specific error types.”

Image credits: “Automatic Correction of Human Translations.” Note the high performance scores outlined in bold.


The User Study

In the last part of the research, Lin et al. profess their interest in the feasibility of TEC as a practical system. After all, all this research will have no use if it is not significantly helpful in aiding human translators with their jobs. The authors design an interface that integrates their TEC model into a more applicable setting. 

First, they had 9 professional translators familiarize themselves with the jargon of the ASICS data set, then were each assigned 74 sentences to look over. Half of the assigned sentences provided an editing suggestion by TEC, and the other half was left unsupervised to be translated directly by the human translator.

The authors then analyzed the following:

1. Whether the TEC suggestion, if shown, was accepted or declined

2. Total time spent reviewing each sentence

3. Number of edit operations (insertions and deletions) the user made

4. Levenshtein edit distance from the original text to the final text

The results showed that “79%” of the TEC suggestions were accepted. The authors found that, when suggestions are shown to the translators, the review time is “significantly less on sentences where the reviewers accepted the suggestion, compared to sentences where they declined.” This can perhaps be explained, the authors claim, by the possibility that incorrect revision suggestions by TEC distract and slow down translators. Furthermore, there is a “significant reduction in the number of insertions+deletions” when the suggestions are shown, which could possibly mean that the “TEC system suggestions help to significantly reduce the amount of manual typing that the user must perform.”

The 9 translators were also surveyed on their experience with and opinions of the TEC system. Five of the reviewers commented that “reliability is critical”; some of the TEC suggestions were incorrect, leading the translator to double-check for other errors, consequently wasting time. Three reviewers expressed their wish of developing the use case of TEC to include that of a “memory aid or substitute for researching client-specific requirements.” One translator noted that TEC could be used as an instructive tool. 


The Best Paper Committee expressed their high regard for this paper, especially impressed with the way the paper “introduces a new corpus, describes a new task, and proposes a new way of leveraging advances in NLP to support stakeholders in the (human) translation workforce.” The committee also appreciated the “in-depth usability study that showed how participatory design and evaluation can contribute to more wide-spread adoption of NLP-assisted technologies.”




Two ACL 2022 Papers that Combat Gender Bias in Machine Translation

Gender bias is a serious problem afflicting NLP and MT today. Given the inequalities of the world (not necessarily limited to gender in this sense, but also racial, class-wise, etc.) the machines we create take after our flaws, learning and internalizing much of the hatred we exhibit on a daily basis. We covered gender bias a few weeks back, introducing major articles that deal with gender bias. Today, we are delving into two papers that explicitly cover gender and other biases in machine translation, chosen from the wide pool of amazing research topics introduced at the ACL 2022.


Measuring and Mitigating Name Biases in Neural Machine Translation

Authors: Jun Wang, Benjamin I. P. Rubinstein, Trevor Cohn

Bias exists in machine translation in many forms: gender, racial, positional, etc. One previously uncovered area of bias is name bias, in which machines wrongly infer gender and other categorical information based on a person’s name. The authors note how “leading systems are particularly poor at [estimating the gender of names] and come up with a method that can help mitigate such biases in neural machine translation.

The reason for such gendered or name bias is the fact that a major portion of textual corpora used to train language models deal with men; sample sentences dealing with women are few. In the past three years, there has been research into how “NMT systems can still be biased even when there are explicit gender pronouns in the input sentences,” which goes to show how skewed NMTs can be in their identification of genders. All this leads to the fact—which the results of this paper prove—that “NMT systems often mistake female names for males, but the reverse is rarely seen.”

What are the consequences of such misgendering? The authors note that, “as an important category of named entity, person names are particularly sensitive to translation errors since they refer to real-world individuals, and systematic biases may cause serious distress to users, and reputational damage, libel or other legal consequences for vendors.” In a world that’s becoming more reliant on machine translation by the day (unrevised machine translation of news articles is now a common practice in many media outlets, for example), the skewing of genders to favor he/him pronouns instead of she/her makes for an inaccurate, harmful depiction of the world in which all major decision-makers and subjects are male, to the exclusion of women.

This problem is especially malignant in “languages with rich grammatical gender” such as German, which would translate the English sentence “[PER] is a Royal Designer” as “[PER] ist ein königlicher Designer” if the subject were male or “[PER] ist eine königliche Designerin” if the subject were female. 

Previous research has focused mostly on gender pronouns (he, she, they, etc.), which NMT still has a hard time processing. Such a problem is compounded when it comes to names, which are, at face value, devoid of immediate gender markers. For names, “gender is not explicitly marked, but is only implied, and the translation system must deduce the gender in order to correctly inflect the translation.”

In this paper, Wang et al. devise a template with which a translation system’s gender identification can be evaluated. The authors then test it on German and French:

With the template above, the authors replaced the pronoun with “a set of 200 full names and 200 first names,” then tested for accuracy using a wide array of off-the-shelf translation systems and other custom-trained models. 

The results are grim, but not surprising. The authors found that “the NMT system favours male names, with all results far better than for female names, even for the commercial translation systems.” Also of note is that “in general, the larger [the] corpus, the less the name bias is present.” The authors attribute this phenomenon to the fact that a larger amount of data leads to exposure to a wider variety of names, allowing the MT system to “better distinguish their gender.” This isn’t always a solution, however, especially for low-resource languages.

Aside from name biases, Wang et al. also coin the term “sentiment biases,” which have to do more with the kind and nuance of adjectival phrases MT systems resort to when translating. For example, in English, a foreign word meaning “famous” can be translated as “renowned” (a positive connotation) or “notorious” (a negative connotation). For words that can be translated in either a positive or negative nuance, the authors call these words “sentiment ambiguous words”: “a kind of homograph which has both commendatory and derogatory meanings.” Sentiment biases have to do with these kinds of words.

To test sentiment bias, the researchers used English to Mandarin Chinese, as “a cross-language family translation” would be more fruitful for the experiment. Similar to their previous gender bias evaluation of German and French, the authors devised a series of templates, which were fed to the system using a set of gendered names. The results were then evaluated for two evaluation metrics: “word-level positiveness” and “sentence-level positiveness.” On the side, the researchers also used the templates to test for racial and nationality biases, as well as “intersectional racial and gender bias.”

The results prove that there does exist some “evidence of gender bias, such as preferential treatment of actors over actresses, and female politicians and entrepreneurs over their male counterparts.” As for intersectional racial-gender evaluations, “Black man” and “Japanese man” were found to have the most negative connotations.

While not a panacea, the researchers offer a method through which these biases can be mitigated. The main reason for these inequalities, the researchers note, is due to unbalanced training data, which effectively skews the results toward favoring men over women. A simple way to “balance out gender biases is to add a number of female data to balance the ratio of male to female sentences.” Taking into account that “obtaining new data can be difficult… for low-resource languages,” the authors introduce SWITCHENTITY: a data augmentation method that switches names in the training corpus as a means to provide an MT system with more training material using feminine pronouns, articles, and names.

After applying SWITCHENTITY (SE) to three of the custom models mentioned before, the authors checked for translation accuracy. Results of SE application show that SE “has a substantial effect on gender inflection when both translating en->de and en->fr.” This improvement is witnessed across models and datasets, which leads us to conclude that “the SE method has a significant effect of mitigating biases for those models.”

This is not to say that SE is the answer to all our problems. The researchers note that “despite these improvements, the bias remains large.” Furthermore, the extent of this research only pertains to a binary gender system, which omits transgender and non-binary pronouns and names, making for incomplete research into the full breadth of the spectrum of human sexuality. Also of note is the small set of language pairs—it would be interesting to see how this research would play out if applied to low-resource languages.


Under the Morphosyntactic Lens: A Multifaceted Evaluation of Gender Bias in Speech Translation

Authors: Beatrice Savoldi, Marco Gaido, Luisa Bentivogli, Matteo Negri, Marco Turchi

Another long paper regarding gender bias in MT to be accepted by the ACL 2022 is Savoldi et al.’s “Under the Morphosyntactic Lens: A Multifaceted Evaluation of Gender Bias in Speech Translation.” The paper gives a thorough review of the state of MT today in terms of dealing with gender bias, then proposes the natural, gender-sensitive MuST-SHE corpus—designed to overcome the limitations of current evaluation protocols—alongside a three-direction (English-French/Italian/Spanish) evaluation method.

Referencing other researchers such as the likes of Matasović (2004) and Gygax et al. (2019), the writers Savoldi et al. note how “gendered features [of language] interact with the—sociocultural and political—perception and representation of individuals… by prompting discussions on the appropriate recognition of gender groups and their linguistic visibility.” In other words, language—and ensuing machine translation based on language thereof—are deeply tied to the political dimensions of language and the way language both permeates and reflects real-life representations of gender. In that sense, gender bias is a serious, critical problem in the realm of machine translation as MT comments directly on the unequal, gendered landscape of natural languages today. 

The writers point out inherent problems in present-day gender bias evaluation benchmarks, which: “i) do not allow us to inspect if and to what extent different word categories participate in gender bias, ii) overlook the underlying morphosyntactic nature of grammatical gender on agreement chains, which cannot be monitored on single isolated words (e.g. en: a strange friend; it: una/o strana/o amica/o).” Current benchmarks simply aren’t precise, nuanced, and robust enough, the authors claim, and fail to account for the grammatical features of gendered languages. A well-functioning evaluation would take into account both part of speech and agreement chains—something that the authors’ very own MuST-SHE corpus takes into account.

MuST-SHE is notable in that it differentiates between 6 parts-of-speech categories: “i) articles, ii) pronouns, iii) nouns, and iv) verbs,” with adjectives distinguished between “v) limiting adjectives with minor semantic import that determine e.g. possession, quantity, space (my, some, this); and vi) descriptive adjectives that convey attributes and qualities, e.g. glad, exhausted.” Upon utilizing MuST-SHE and evaluating performance based on the corpus, the researchers discovered that, “while all POS are subject to masculine skews, they are not impacted to the same extent.” 

Unlike many papers, the authors of “Under the Morphosyntactic Lens” finish the paper with an “impact statement,” in which they discuss the broader impact of the research. The authors rehash the dangers of gender bias in MT, citing that the behavior of Mts to “systematically and disproportionately [favor] masculine forms in translation” is “problematic inasmuch it leads to under-representational harms by reducing feminine visibility.” Next, the authors posit that this research paves the way for future NLP and MT research, as “Under the Morphosyntactic Lens” creates a space in which to ponder upon and renegotiate the depth and extent to which data and models should be sensitized to gender nuances. 

These two papers are two of few papers published on gender bias in MT, among 604 long papers that were accepted by the ACL 2022. While the number is small, these two papers raise awareness of the importance of combating gender bias—alongside other types of biases—in machine translation. 




Kinyarwanda Comes to the Forefront at ACL 2022

In our previous blog post, we introduced ACL 2022—the 60th Annual Meeting of the Association for Computational Linguistics—and two papers featured there: one on the disambiguation bias benchmark DiBiMT and another on low-resource speech synthesis on Indigenous languages in Canada. Today, we’re adding one more to the list of research that caught our attention. Antoine Nzeyimana of the University of Massachusetts Amherst and Andre Niyongabo Rubungo of the Polytechnic University of Catalonia won Best Linguistic Insight Paper with this research, titled “KinyaBERT: a Morphology-aware Kinyarwanda Language Model.”

As can be inferred by the title, the paper introduces a two-tier BERT architecture that, unlike previous pre-trained models, can take advantage of a morphological analyzer to translate low-resource, morphologically complex languages like Kinyarwanda. Past research in natural language processing has focused majorly on English, Chinese, and European languages spoken by economically prevalent powers, which has “exacerbated the language technology divide between the highly resourced languages and the underrepresented languages.” The authors note that this effectively hinders NLP research, as it is “less informed of the diversity of the linguistic phenomena.”

KinyaBERT, then, is all the more relevant. A “simple yet effective two-tier BERT architecture for representing morphologically rich languages,” KinyaBERT paves the way for morphologically rich, underresourced languages previously ignored by most NLP research. And the results are outstanding; whereas previous models are “sub-optimal at handling morphologically rich languages,” KinyaBERT outperforms baseline scores by 2 to 4.3%, proving that low-resource language translation is still viable on pre-trained language models.

The authors start off by noting that “recent advances in natural language processing (NLP) through deep learning have been largely enabled by vector representations (or embeddings) learned through language model pre-training.” BERT is one such example; it is pre-trained on vast amounts of text, then fine-tuned on downstream tasks, which greatly enhances performance on many NLP tasks. To this date, models pre-trained on monolingual data still outperform multilingual models. With all this said, KinyaBERT is an effort to utilize the advantages of a model pre-trained on monolingual data, except with a morphologically rich language such as Kinyarwanda.

There are a number of reasons why previous models don’t work as well for morphologically complex languages. For one, models like BERT use “statistical sub-word tokenization algorithms such as byte pair encoding (BPE),” which are “not optimal for morphologically rich languages” due to linguistic characteristics such as morphological alternations and non-concatenative morphology. The table below shows how both monolingual BPE models and multilingual BPEs fail, after being trained on 390 million tokens of Kinyarwanda text, to identify the correct morpheme:

Many factors complicate the Kinyarwanda language (like all other languages, Kinyarwanda is complex, although the reasons for its complexity are not as known as those of European languages). For example, Kinyarwanda has 16 noun classes, and “modifiers (demonstratives, possessives, adjectives, numerals) carry a class marking morpheme that agrees with the main noun class.” Furthermore, “the verbal morphology also includes subject and object markers that agree with the class of the subject or object”: markers that are used by “users of the language to approximately disambiguate referred entities based on their classes.” 

This agreement property is utilized by the authors’ unsupervised POS tagger. Before pre-training the model, KinyaBERT must be fitted with a morphological analyzer that takes into account all word types that can be inflected in the Kinyarwanda language. Afterward, the authors compose a complex speech tagging algorithm that measures syntactic agreement and estimates first order transition measures.

Once all the tagging and analysis are complete, the morphology encoder takes over. The encoder is “applied to each analyzed token separately in order to extract its morphological features,” which are then “concatenated with the token’s stem embedding to form the input vector fed to the sentence/document encoder.” Shown below is a diagram of the architecture of KinyaBERT:

KinyaBERT is then pre-trained using 2.4 GB of Kinyarwanda text comprised of 390 million tokens/words and 16 million sentences taken from 840,000 documents and articles.  Upon evaluating the pre-trained model using the GLUE benchmark, the KinyaBERT achieves a “4.3% better average score than the strongest baseline.” For the named entity recognition task, the KinyaBERT achieves a “3.2% better average F1 score than the strongest baseline.”

Overall, KinyaBERT is yet another step in the right direction for machine translation and natural language processing. Many research fails to take into account that the omission of low-resource, underrepresented languages can be detrimental to NLP research in that such omission skews the linguistic landscape. “There are very few works on monolingual PLMs [pre-trained language models] for African languages…” the authors write. “we aim to increase the inclusion of African languages in NLP community by introducing a PLM for Kinyarwanda.” This paper is significant not only for these linguistic reasons but also because it enriches BERT research and architecture, paving a path for other underresourced languages to follow suit.

Alongside Pine et al.’s “Requirements and Motivations of Low-Resource Speech Synthesis for Language Revitalization” covered in our last post, this paper on KinyaBERT makes up a small but critically essential part of NLP and MT research that sheds a sorely needed spotlight on underresourced languages. It’s important that NLP research heed and remain open to the possibilities that previously disregarded languages bring to the table.

Alongside papers, both long and short, the ACL 2022 hosted keynote speeches and panels on a wide variety of topics. Particularly relevant to the topic of underresourced language representation was the keynote panel on “supporting lingusitic diversity,” whose panelists included researchers and speakers of the following languages: Seneca (USA), Minangkabau (Indonesia), Creole languages, Irish, Wixaritari (Mexico), and Luo and Kiswahili (Kenya). 

The panel poses important questions: “How do the tools and techniques of computational linguistics serve the full diversity of the world’s languages? In particular, how do they serve the people who are still speaking thousands of local languages, often in highly multilingual, post-colonial situations?” Here are other questions covered by the panel, based on the following topics:

Diverse Contexts. What is the situation of local languages where panel members are working? Are there multiple languages with distinct functions and ideologies? What are the local aspirations for the future of these languages. How are people advocating for language technology on the ground? How did the work begin? What does success look like?

Understanding Risks. Do the people who provide language data fully understand the ways their data might be used in future, including ways that might not be in their interest? What benefit are local participants promised in return for their participation, and do they actually receive these benefits? Are there harms that come with language standardisation? What principles of doing no harm can we adopt?

New Challenges. How can we provide benefits of text technologies without assuming language standardisation, official orthography, and monolingual usage? When working with local communities, do we always require data in exchange for technologies, or is a non-extractive NLP possible? How do we decolonise speech and language technology? At the beginning of the International Decade of Indigenous Languages 2022–2032, we ask: how do we respond as a community, and how can our field be more accessible to indigenous participation?

Going forward from here, the inclusion of underresourced languages is essential to the integrity and future of NLP research, and it is important that researchers ask themselves these questions in regard to the extent of languages covered. Issues concerning which languages are chosen to be represented in certain academic research, or for what purposes these languages are utilized: these are questions that are closely intertwined with the history and politics of such languages. In that sense, discussion around language should be intentional and deliberate.

“KinyaBERT: a Morphology-aware Kinyarwanda Language Model” seeks to do exactly that: create spaces for intentional and deliberate discussion around the supremacy of European languages in the NLP discourse. KinyaBERT actively dismantles and combats the prevalency of European languages, shifting the discourse into a more equal, balanced dynamic in which underresourced languages can voice their presence and existence.




Innovating Low-Resource Speech Synthesis & Combating Ambiguity at the ACL 2022

From May 22 to 27, students and researchers from all over the world gathered in Dublin—or stayed in the comforts of their homes—to attend the 60th annual Meeting of the Association for Computation Linguistics. The ACL is the most renowned organization when it comes to natural language processing, and this year saw more than 2,000 authors contribute 604 long papers and 98 short papers to the conference, as well as a number of exciting events including keynote speeches and a panel on supporting linguistic diversity. 

Given the organization’s history as a machine translation-focused group (when it was founded in 1962, the organization was called the Association for Machine Translation and Computation Linguistics), a good portion of the accepted articles dealt with machine translation. And of those, many interesting titles and ideas caught our attention, such as Antoine Nzeyimana and Andre Niyongabo Rubungo’s “KinyaBERT: a Morphology-aware Kinyarwanda Language Model”—which was selected for the Best Linguistic Insight Paper—and “Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity” by Yao Lu et al.

ACL—alongside WMT (the Conference on Machine Translation)—yearly introduces some of the best, most innovative, and most creative research on MT and NLP the world has to offer. We find it very helpful to examine some of the highlighted papers; reading through them, we see the breadth of topics and values that are being pioneered by researchers in the realm of natural language processing. In other words, seeing what topics are covered here is a good gauge of what the future of natural language processing looks like.

Without further ado, here are some of the papers we would like to share with you.


“DiBiMT: A Novel Benchmark for Measuring Word Sense Diambiguation Biases in Machine Translation” (Best Resource Paper)

Authors: Niccolò Campolungo, Federico Martelli, Francesco Saina, Roberto Navigli

One of the most common errors we face when utilizing front-end machine translation is ambiguity. Machines have a much harder time distinguishing between the multiple meanings of a polysemous word. The authors give an example: “He poured a shot of whiskey.” The proper Italian translation, taking into account the fact that the word “shot” here means “a small quantity,” is “Versò uno goccio di whiskey.” However, it wouldn’t be strange to see a machine translation such as “Versò uno sparo di whiskey,” in which “sparo” refers to “gunshot.”

According to the authors, research has been carried out in the last few decades to investigate this issue, such as Gonzales et al.’s 2017 development of ContraWSD, a dataset of German-to-English and German-to-French instances of lexical ambiguity complete with contrastive examples of incorrect translations. Other studies such as Liu et al. in 2018 have improved on language models “via context-aware word embeddings,” culminating in Emelin et al.’s 2020 research that introduces a “statistical method for the identification of disambiguation errors in neural MT (NMT)” and demonstrates that “models capture data biases within the training corpora.”

However, the authors point out problems with previous research: “i) they are not based on entirely manually-curated benchmarks; ii) they rely heavily on automatically-generated resources to determine the correctness of a translation; and iii) they do not cover multiple language combinations.” To combat these shortcomings, the authors introduce “DiBiMT,” which, to their knowledge, is the “first fully manually-curated evaluation benchmark aimed at investing the impact of semantic biases in MT in five language combinations, covering both nouns and verbs.” In other words, the authors provide a framework for evaluating lexical ambiguity within a text, allowing other researchers to better understand the phenomenon of ambiguity and bias in machine translation and come up with ways to combat it.

To build DiBiMT, the authors needed “i) a set of unambiguous and grammatically-correct sentences containing a polysemous target word” and “ii) a set of correct and incorrect translations of each target word into the languages to be covered.” To select the sentence sets they were to use, the authors carried out a number of steps, starting with item structurization and notation, followed by dataset annotation. With a fully prepared DiBiMT, the authors evaluated 7 machine translation systems—from frontend ones such as Google Translate and DeepL to non-commercial ones such as MBart50 and M2M100—and reviewed the results.

And the results were interesting, to say the least. DeepL outperformed every single other model, but those other models performed particularly poorly, earning extremely low scores—between 20% and 33%. Most shocking of all: Google Translate performs the worst across all languages. Alongside general accuracy, the DiBiMT also measures semantic biases using four novel metrics, where DeepL also vastly outperforms its competitors. Another interesting result of the research is that DiBiMT’s conclusions support the existing literature’s stance regarding verbs, which are considered harder for MTs to translate than nouns.

Campolungo et al.’s research point out a very important flaw in the current state of MT: natural language is much too often ambiguous, and machines don’t have enough—or any—understanding of context to respond to this urgent need. The ambiguity can range anywhere from simple miscommunications (a gunshot of whiskey instead of a shot, for example) to straight-up offensive translations (“se taper le cul par terre”—to laugh out loud—translated as “banging ass on the floor.”) All jokes aside, ambiguity hinders MT greatly, and this research is one step in the right direction: eliminating bias and ambiguity to make for more accurate, context-aware translation models. After, Sun Tzu says that you have to know your enemy to defeat them.


“Requirements and Motivations for Low-Resource Speech Synthesis for Language Revitalization” (Best Special Theme Paper Award)

Authors: Aidan Pine, Dan Wells, Nathan Thanyehténhas Brinklow, Patrick Littell, Korin Richmond

This paper grabbed our attention, as it focuses on building speech synthesis systems for Kanien’kéha, Gitksan, and SENĆOŦEN, three Indigenous languages spoken in Canada, and the motivation and development behind such systems. The paper, while maintaining its technical integrity, reads almost like a sociology paper, given its in-depth insight into the sociological reasons why such speech synthesis systems are necessary.

The authors start out with an overview of the current state of Indigenous languages spoken in Canada. While there are about 70 Indigenous languages spoken in the country, most of these languages have less than 500 fluent speakers as a result of the “residential school system and other policies of cultural suppression.” This isn’t to lament the death of Indigenous languages; on the contrary, the authors note that such “‘doom and gloom’ rhetoric that often follows endangered languages over-represents vulnerability and under-represents the enduring strength of Indigenous communities who have refused to stop speaking their languages despite over a century of colonial policies against their use.”

In other words, Pine et al. are less concerned with the death of these Indigenous languages, as they have survived thus far through the will and perseverance of their speakers; furthermore, there has been growing interest in the revival of Indigenous languages. The authors also note that such language revitalization efforts “extend far beyond memorizing verb paradigms to broader goals of nationhood and self-determination,” and that programs endorsing language revitalization can have “immediate and important impacts on factors including community health and wellness.” Also taking into account the fact that the UN has declared 2022-2032 to be an International Decade of Indigenous Languages and the amazing news that the number of Indigenous language speakers in Canada has increased by 8% between 1996 and 2016, these concerns regarding language revitalization are more relevant than ever.

The authors argue for the necessity of speech synthesis for these Indigenous languages, as it would complement previously existing pedagogical grammar models (like the Kanien’kéha verb conjugator, Kawennón:nis) and “add supplementary audio to other text-based pedagogical tools.” However, speech synthesis itself isn’t exactly a much-developed field; for Indigenous languages in Canada, the problems are compounded by the fact that most Canadian Indigenous languages are low-resource, as they lack data and speakers. 

With these issues in mind, the research delves into a more technical aspect of language revitalization. How much data is exactly needed to configure such a system so that it meets certain pedagogical standards? What would the evaluation of such a system look like? And, “how is the resulting system best integrated into the classroom?”  To answer these questions, the paper takes a number of steps to gauge the current state of speech synthesis models and investigate the feasibility of speech synthesis given the amount of data necessary to support the authors’ goals. The paper also delineates the steps taken to build what is now “the first neural speech synthesis systems for Indigenous languages spoken in Canda.”

The authors then carried out listening test evaluations of the speech synthesis models, to which participants reacted positively. The evaluations “showed encouraging results for the naturalness and acceptability of voices for two languages, Kanien’kéha and Gitkan, despite limited training data availability.” In conclusion, the findings of the research “show great promise for future work in low-resource TTS for language revitalization, especially as they come from systems trained from scratch on such limited data, rather than pre-training on a high-resource language and subsequent fine-tuning on limited target language data.”




New Speech-to-Speech Translation (S2ST) Paradigms for Spoken Languages

Earlier in the month, Meta AI highlighted an area of its research dealing with speech-to-speech translation. In an article titled “Advancing direct speech-to-speech modeling with discrete units,” Meta AI researchers Ann Lee and Juan Pino alongside product manager Jeff Wang and product designer Ethan Ye briefly share Meta’s current research in the speech-to-speech translation field. Building on previous research, the current project is “the first direct S2ST system trained on real-world open sourced audio data instead of synthetic audio for multiple language pairs.”

Compared to text translations—which have been around for millennia—speech-to-speech translation is a relatively new field of study, made available by recent developments in automatic speech recognition (ASR), voice synthesis (TTS), and machine translation (MT). One of the first demonstrations of general speech translation was in 1999 when researchers from five different countries demonstrated a rudimentary version of speech-to-speech translation in French, English, Korean, Japanese, and German as part of the C-STAR II consortium. The same year, NEC succeeded in developing “Tabitsu,” an automatic speech-to-speech translation software for notebook PCs equipped with 50,000 Japanese and 25,000 English words related to travel and tourism.

What makes Meta AI’s foray into speech-to-speech translation different is that, unlike previous research, it aims to dramatically simplify the system process—hence, the title: direct speech-to-speech modeling. Previous models take a “cascaded approach,” meaning they first transcribe the spoken phrase into text, which is then machine-translated then output into speech again using text-to-speech (TTS) synthesis. A majority of research focuses on this cascaded system, with many researchers succeeding in streamlining the process or increasing the efficiency of each part of the setup. Nascent research in the past couple of years has given rise to more innovative approaches, such as the Translatotron—which “directly translates mel-spectrogram of the source speech into spectrogram features of the target speech”—and Kano et al.’s 2021 proposal to “build a single deep-learning framework by pre-training ASR, MT and TTS models separately and connecting them with Transcoder layers.”

Much like these new paradigms for S2ST, Meta AI’s research tries to avoid relying on intermediate text generation and thus cut out much of the cascading, intermediary steps. Meta instead opts for “discrete units,” which more efficiently handles the process, as it can “disentangle linguistic content from prosodic speech information.” Below is a diagram of how the model works with discrete units:

A diagram depicting the structure of Meta’s direct S2ST model using discrete units. Image credits:

Another difference from the previous models is that Meta incorporates newly available, real-world S2ST data in model training. S2ST doesn’t have a big enough parallel training dataset, meaning it relies on TTS to “generate synthetic target speech for model training.” Meta, in “Textless speech-to-speech translation on real data” was able to utilize real data from the VoxPopuli S2S data and the automatically mined S2S data.

So what do the results look like? Is Meta AI’s discrete unit-based any more efficient or accurate than previous models? Testing their model on the Fisher Spanish-English speech translation corpus (which consists of 139,000 sentences of Spanish conversations and corresponding Spanish/English text transcriptions), the researchers found that the framework reaches 6.7 BLEU compared to a previous model—an improvement. When the model is trained without text transcripts, it is comparable to models that are trained with text supervision but predict spectrograms.

Meta’s direct speech-to-speech model with discrete units provides not just better translation quality, but also heightened efficiency in terms of runtime, FLOPS, and max memory. With its bigger training data and new paradigm, Meta’s new model works even better for unwritten languages, which have previously been largely ignored due to the fact that it is near impossible to translate without text transcriptions. 

This research is part of Meta AI’s efforts to expand its reach so that more languages are covered by the translation services the company offers. Alongside this research are other projects aimed to democratize language prevalency in text translations—such as its M2M-100 model, the first multilingual machine translation model that translates between 100 languages without relying on English data—as well as its No Language Left Behind project and Universal Speech Translator initiative.




Reimagining Machine Translation with MIT’s Hallucinatory VALHALLA

As we write this, the world is reeling from DALL-E 2: an OpenAI artificial intelligence program that, in the simplest terms, draws what you say. The Eiffel Tower enjoying a skinny dip? You got it. Midsommar, but make it cyberpunk? You got it. On Twitter, TikTok, and Instagram, people are sharing their craziest inputs as DALL-E turns it into a believable, hilarious work of art, like these picture of Winnie the Pooh robbing a 7-Eleven store:

Or these pictures of a dumpster fire, but painted by Monet:

DALL-E uses a 12-billion parameter version of the GPT-3 Transformer model—a natural language processing model that gives DALL-E its trademark wealth of knowledge and robustness to handle the weirdest of commands. It also uses an autoregressive transformer—in which the model takes past values and applies them to create new values. That’s how it understands input such as “Mickey Mouse and Donald Duck in a Mexican standoff”—because it takes from its wide range of past values (“Mickey Mouse,” “Donald Duck,” “Mexican standoff”) and merges it into a new value. This is a grossly simplified explanation of what DALL-E actually does.

Last month, on May 31, researchers at US San Diego, IBM, and MIT have jointly published a research paper on VALHALLA, a “visual hallucination framework” and a close cousin of DALL-E. While the structure and framework of DALL-E make it suitable for creating artistic, creative pieces of visual media, VALHALLA utilizes visual modes to try and improve machine translation. While not as vivid or concrete as DALL-E’s manifestations, VALHALLA also predicts discrete “hallucinated visual representations”; combined with the input text, the picture is then used to figure out the target translation. To the best of the researchers’ knowledge, VALHALLA is the first work that “successfully leverages an autoregressive image transformer jointly with the translation transformer to hallucinate discrete visual representations.”

An overview of VALHALLA’s structure. Image credits:

In previous blog posts, we’ve spoken of machine translation and its numerous iterations and paradigms. From its rules-based paradigm, MT has developed into statistical MT models, from which it has now evolved into neural network-based models. VALHALLA continues the tradition of reforming paradigms; this time around, machine translation is multimodal—meaning it utilizes a variety of modes, such as text and pictures. 

After all, even the best of previous models—the neural network-based multilingual models come to mind—currently are mostly text-only; the ones that do utilize visual context usually “require manually annotated sentence-image pairs as the input during inference.” In regard to the former type, VALHALLA researchers cite that such systems “lack any explicit grounding to the real world.” As such, “there has been a growing interest in developing multimodal MT systems that can incorporate rich external information into the modeling process.”

An illustration of VALHALLA’s visual hallucination technique. Image credits:

As in the figure above, it is important to note that “the underlying visual perception [of a situation in the physical world] is shared among speakers of different languages.” The above sentence—“A snowboarder wearing a red coat is going down a snow-covered slope” shares the same visual perception as “Ein snowboarder in einem roten anzug fährt eine schneebedeckte piste herunter.” 

VALHALLA has exhibited great improvements over previous models. Using three public datasets—Multi30K, Wikipedia Image Text (WIT), and WMT2014—the researchers rate VALHALLA’s performance, and the results are quite astounding. When evaluated using Multi30K, VALHALLA “significantly outperforms the text-only baselines on all three test sets.” Furthermore, VALHALLA “outperforms all compared methods, achieving best BLEU and METEOR scores under both multimodal and text-only translation settings.” VALHALLA performs exceptionally across the board in all datasets. 

The ramifications of VALHALLA are both exciting and hopeful. According to its performance review when evaluated with the WMT dataset, VALHALLA “outperforms all the compared methods in both well- and under-resourced settings. The improvements over text-only baseline are more significant in under-resourced scenarios, which is of significant practical value.” The report also dedicates a good amount of analysis to how VALHALLA performs well even when translating under limited textual context. 

With the advent of natural language processing as it breaches the sphere of visual media, it’s time to welcome these changes to machine translation paradigms. Such is the beauty of cutting-edge scientific developments: research like this allows us to venture outside the comforts of what we once thought science should be and imagine new frameworks and concepts through which we can reform tired structures. All this leads to the eventual democratization of information and communication, we hope; even the most neglected of languages will, soon enough, have a chance to be heard alongside widely spoken lingua francas. 

At the same time, we are wary as to the feasibility of these new developments and techniques. When will they be readily available for application in practical functions? Or is this all part of a vague vaporwave trend—grandiose notions of technology bridging gaps when, in reality, we find ourselves just as incapable of communicating ourselves to one another as in the olden days? 

But the realm of machine translation proves to be, time after time, rife with positive, groundbreaking change that renegotiates our connection to one another and language. We hope that VALHALLA, like its preceding paradigms and developments, will revolutionize the way we think of translation and language. 




Meta’s Open-Source LLM, 24 New Languages for Google Translate, and Amazon NLP: Language Industry Updates for May 2022

We here at Sprok DTS hope everyone is having a beautiful spring. We are back with some exciting news in the language industry, which seems to forge ahead with brilliant new developments in the realm of machine translation despite all the hardships plaguing the world. From Amazon’s grandiose plans to increase the number of languages supported by Alexa to Google’s coverage of 24 new Asian and African languages, machine translation researchers work harder than ever to bridge the gap between nations and cultures, bringing us closer in these desperate times.


Developing Zero-Resource Machine Translation for Wider Coverage in Google Translate

Research scientists Isaac Caswell and Ankur Bapna recently announced that Google’s front-end translation service Google Translate will now be covering an additional 24 under-resourced languages. Included are Asian, South American, and African languages spoken by large populations but lack substantial data for proper model training, such as the Assamese (spoken by 25 million people in Northeast India), Quechua (spoken by 10 million people in Peru, Bolivia, Ecuador, etc.), and Luganda (spoken by 20 million people in Uganda and Rwanda). This brings the total number of covered languages to an impressive 133.

Before, Google Translate’s language offerings were predominantly European; for example, the service provided support for Frisian, Maltese, Icelandic, and Corsican, all of which have less than 1 million native speakers, but not for Bhojpuri (nearly 51 million speakers) or Oromo (nearly 24 million). This latest update is an effort to represent the world’s languages in a more proportionate manner and envision a more linguistically diverse world in which native speakers in South Asia, Africa, and South America can better enjoy the benefits of machine translation to aid with communication.

While the update was long overdue, it was no easy task for the researchers. The 24 languages added are some of what research scientists call “long-tail languages,” or languages with scarce data sets that require machine learning techniques that can “generalize beyond the languages for which ample training data is available” (Bapna et al., 4). If languages previously offered by Google Translate were trained—as language models usually are—using parallel text datasets, these 24 languages required the use of monolingual text.

We’ve covered this kind of “zero-shot,” monolingual model training before on this blog; bilingual language data training is becoming outdated due to its inefficacy in lesser-spoken languages, and these 24 languages are testament to how successfully zero-shot machine translation can function when given ample time and resources. In the case of Google Translate, researchers “train[ed] a single giant translation model on all available data for over 1000 languages” (Caswell and Bapna). 

A graph comparing the amount of parallel data and the monolingual data available for major languages. Image Credits:

There are other important, more scientific, parts to the process. Finding high-quality, effective data was difficult for these under-resourced languages, as “many publicly available datasets crawled from the web often contain more noise than usable data” (Caswell and Bapna). As a result, researchers instead “trained a Transformer-based, semi-supervised LangID model… to better generalize over noisy web data.” In doing so, the researchers were able to end up with high-quality content and data to train the model.

Caswell and Bapna note that communicating with native speakers was critical to developing machine translation services for these under-resourced languages. The researchers “collaborated with over 100 people at Google and other institutions who spoke these languages” (Caswell and Bapna), receiving help with tasks like developing filters to remove out-of-language content or transliterating between various scripts used by a language, among other tasks. 

There is still a long way to go for these languages—and multilingual, zero-shot language models. Caswell and Bapna stress that “the quality of translations produced by these models still lags far behind that of the higher-resource languages supported by Google Translate,” and advise users to practice caution when interpreting translation outputs.

A graph indicating the quality (RTTLangIDChrF) of translations by the number of monolingual sentences available (horizontal axis.) Image Credits:


Amazon’s MASSIVE Dataset for a Multilingual Alexa

While Google’s artificial intelligence research is admirable, it doesn’t offer its users a virtual assistant that’s as well known as Apple’s Siri or Amazon’s Alexa. Late last month, Amazon’s senior applied scientist Jack G. M. FitzGerald over at the Alexa AI’s Natural Understanding Group made an exciting announcement: Amazon is releasing its very own dataset called MASSIVE, which is “composed of one million labeled utterances spanning 51 languages, along with open-source code, which provides examples of how to perform massively multilingual NLU modeling.” 

In contrast to Google’s focus on natural language processing—after all, Google Translate is, above anything, a translating processor of language—Amazon focuses more on natural language understanding (NLU), given that users of Alexa are always in conversation with Alexa, who must first understand the intent and purpose of its users’ utterances before providing an apt answer. FitzGerald explains this in more simple terms: “given the utterance “What is the temperature in New York?”, an NLU model might classify the intent as “weather_query” and recognize relevant entities as “weather_descriptor: temperature” and “place_name: new york.”” (FitzGerald)

MASSIVE is Amazon’s answer to developing a more robust multilingual paradigm for Alexa. MASSIVE is short for Multilingual Amazon SLURP (SLU resource package) for Slot Filling, Intent Classification, and Virtual-Assistant Evaluation—what a mouthful—and contains “one million realistic, parallel, labeled virtual-assistant text utterances spanning 51 languages, 18 domains, 60 intents, and 55 slots” (FitzGerald). With this, Alexa would be able to carry out similar intent-identification and response-evaluation processes like the one above in 51 languages. Furthermore, the creation of such a dataset and paradigm means that Alexa—and other virtual assistants—can continue to build on preexisting datasets to cover more languages, including under-resourced ones that have previously been largely overlooked due to their lack of sample data and text.

Amazon has released and open-sourced the dataset and processes on GitHub, which means researchers and scholars can learn from, study, and make improvements to the dataset. Alongside the release of the dataset, Amazon is also hosting the Massively Multilingual NLU 2022 competition, inviting competitors to participate in two tasks. In the first task, competitors will be able to train and test a single model on all 51 languages based on the full MASSIVE dataset. In the second one, competitors will be able to fine-tune a pretrained model only with English-labeled data and test it on all 50 non-English languages, according to FitzGerald, who hopes that wide participation in these tasks can aid in developing and improving zero-shot learning for machine translation and natural language understanding.


Meta Democratizes Access to Large-Scale Language Models

Earlier this month, Meta announced that it will be open-sourcing its Open Pretrained Transformer (OPT-175B), which is a “language model with 175 billion parameters trained on publicly available data sets.” In a Meta AI blog, research engineer Susan Zhang, research scientist Mona Diab, and research director Luke Zettlemoyer explain that large language models have been fundamental to recent developments in NLP and AI research, allowing machines to “generate creative text, solve basic math problems, answer reading comprehension questions, and more.” However, these large language models are heavily-guarded secrets, accessible to only a few high-resource labs. This is a problem, claims Zhang et al., as secretive, isolationist policies on model access largely hinder developments. 

With Meta’s release of OPT-175B, researchers and scholars in the larger NLP and AI community can access and pry into the inner workings of a large language model—the largest to ever be publicly released—and examine “both the pretrained models and the code needed to train and use them” (Zhang et al.). Meta does note that the model will come with a noncommercial license, only to be used for academic, government, civil society, and industry research purposes around the world, so as to prevent misuse.

While OPT-175B signifies a new chapter in community engagement in NLP research, it is also a trial, examining how the AI community will manage and uphold the tenets of integrity and fair use as such important information is shared online. In the words of Zhang et al.: “We believe the entire AI community — academic researchers, civil society, policymakers, and industry — must work together to develop clear guidelines around responsible AI in general and responsible large language models in particular, given their centrality in many downstream language applications” (Zhang et al.). 

The concerns are not only political and moral in nature; AI research can be power-hungry, costly, and harmful to the environment. Meta had this in mind, and as such, OPT-175B has been designed so that it only uses a fraction of the carbon footprint as a previous model. Community engagement and usage of such large-size models is a new frontier, and companies and individual researchers are advised to be careful of their intentions and purposes as they utilize the powers of OPT-175B.





How Translation Will Enrich International Biological Research (and Hopefully, Save the World)

In early 2020—when COVID-19 was slowly making its way across the globe—Nintendo released Animal Crossing: New Horizons, a social simulation game that kept more than 37 million people around the world company. Ported specifically for the Nintendo Switch console, the game allows users to build and decorate their own island, befriend cute, quirky animal neighbors, expand their houses by taking out loans, and catch and collect local flora and fauna.

The lattermost activity—completing a collection of fish, insects, flowers, fossils, and subaquatic creatures—is one of the most exciting quests in the game; users donate their findings to an owl NPC in an enormous museum, which they can visit at any time to enjoy and learn more about their discoveries. For people who’ve grown up in the city and didn’t know much about wildlife, Animal Crossing: New Horizons was a testament to how exciting—and diverse—the natural world can be. Localized into the major Romance languages alongside Dutch, Russian, and the three major East Asian languages, the game has cultivated a following of avid international users who, despite their linguistic differences, are united by the version of nature they encounter in this simulation game.

Animal Crossing: New Horizons is also proof that every culture has a different relationship to nature, evidenced by the widely different names languages use to call the same species of fish or insect. The small, colorful fish known as bitterling in English (fishable in the game’s many rivers) is known as amarguillo in Spanish; in Japan, it’s known as tanago, and in Russia, it takes the name gorchak. For players who speak one of these 16 languages the game has been localized in, the game allows them to reconnect with nature—divorced from them due to the long-lasting pandemic—in their own language.


The reality

However, in the academic discourse of biology, the natural world isn’t as vibrant or diverse as the various names we call our wildlife. In a landmark research published in 2021 by more than 60 biological researchers around the world titled Tapping into non-English-language science for the conservation of global biodiversity, it is revealed that “to date, non-English-language studies have largely been ignored in evidence synthesis,” and that “the underuse of non-English-language science” is prevalent across disciplines. 

Academia has unknowingly adopted English as its primary—and most times, the sole—language of knowledge dissemination, rendering scientific works in other languages largely ignored. If we were to explain this in terms of Animal Crossing: New Horizons, any kind of research or documentation done about the bitterling in another language—any papers about amarguillo or tanago or gorchak—would be cast aside. The only research that matters are the ones about the bitterling, written in English (although the fish itself would be called by its genus, Rhodeus, perhaps).

This is a problem, because “non-English-language studies provide crucial evidence for informing global biodiversity conservation.” Of the 419,679 peer-reviewed papers screened by the researchers, there were 1,234 non-English-language studies that provide evidence on the effectiveness of biodiversity conservation interventions, as compared to 4,412 English-language studies within the same criteria. The researchers posit that “incorporating non-English-language studies can expand the geographical coverage… of  English-language evidence by 12% to 25%… and taxonomic coverage by 5% to 32%.” There is so much biodiversity being left out because non-English-language research is ignored in the English-dominated discourse of biology.

The researchers note how history is proof of the necessity of international, linguistically diverse research. For example, primary research into developing the Nobel Prize-winning antimalarial drug was first published in Mandarin.Eduardo H. Rapoport’s influential study of biodiversity, Areografía: estrategias geográficas de las especies, was published in Spanish. Even now, “many of the earliest papers on COVID-19 were written, again, in simplified Chinese.” But still, such non-English-language contributions to science and scientific communities—and hence, the broader society—are seldom quantified.




To examine the broader picture of linguistic variety in academia, the researchers carried out an intense “assessment of non-English-language studies’ contribution to evidence synthesis—the process of compiling an summarising scientific information from a range of sources.” More specifically, the researchers screened 419,679 peer-reviewed papers in 326 journals, published in 16 languages, and identified non-English studies “testing the effectiveness of interventions in biodiversity conservation.” 

The sixteen languages chosen were the top 15 non-English languages, ranked by their number of conservation-related publications. The list excludes Swedish and Dutch (as the researchers could not find any native speakers of these languages to collaborate with) and covers three additional languages (Arabic, Hungarian, and Ukrainian.) 38 total native speakers of the 16 languages were recruited, taking into account their education and area of study to make sure that they were capable of understanding the content of the studies covered. The searchers were then trained for a wide range of academic abilities necessary for the research.

The 326 journals were chosen by native speakers who were also academics in the field, rating each journal by its relevancy. Of most relevancy were journals in ecology, biodiversity conservation, and taxonomic studies, such as “ornithology, mammalogy, herpetology, plant sciences, etc.” Less relevant were journals in other related disciplines, such as “agricultural/forest sciences and general zoology.” The researchers aimed to include most of the journals that fell in the “most relevant” category and tried to include journals in the just “relevant” category as much as possible.

In the final step of the research, searchers scanned the “title and abstract of every peer-reviewed non-English-language paper published in the journal and by reading the main text of all papers” that met the eligibility criteria for the research. Their findings were then tested to see if they met the two major criteria they had for their contribution to biodiversity studies, documented for various categories of metadata, then compared for important information such as study designs, study locations, yearly changes, proportions of eligible studies by journal, and the species covered.




The research revealed that non-English-language studies “expanded the geographical coverage based on English-language studies” for amphibians by 12%, birds by 16%, and mammals by 12%. The 1,234 non-English-language studies together made up a total of “1,954 unique species recognised by the International Union for Conservation of Nature (IUCN),” including 40 amphibians (6 threatened), 564 birds (37 threatened), and 194 mammals (22 threatened). As for species not covered by English-language studies, the non-English-language studies provided scientific evidence for an additional 9 amphibians, 217 birds, and 64 mammals. Non-English-language studies also “increased the evidence coverage of threatened species by 23% for birds and 3% for mammals. “

Regarding the quality of the studies, the research found that “the proportion of eligible studies in each journal varied among languages.” Japanese had the highest proportion of eligible studies in a journal with 26.7%, followed by Hungarian with 15.3%, then French (12.9%), then German (9.1%). 

The number of eligible non-English-language studies increased significantly over the years since 2000 in 6 of the languages studied, with Portuguese and Russian showing a notable increase and traditional Chinese showing a “marginally significant increase.” This finding refutes the popular perception that “the number of non-English-language studies providing evidence is declining.” If anything, the number of relevant studies in languages other than English is growing, if slowly overall.

However, the results of the study do support the popular perception that “non-English-language studies tend to be based on less robust study designs.” More specifically, studies in 10 of the 16 languages covered were more likely to adopt less robust study designs compared to studies done in English. As such, there is also a correlation with the quality of the studies, which suffers when the study designs are less robust. 



In all, this research goes to prove that “synthesising non-English-language studies could be an effective avenue for reducing the exisiting, severe gaps in the geographical and taxonomic coverage of available scientific evidence for biodiversity conservation.” There is a divide between English and non-English studies, and taking the latter into account in the international discourse on biodiversity is very “critical to halting the ongoing biodiversity crisis.”

After all, the natural world doesn’t distinguish or discriminate based on language or culture. In an interconnected society like this, the demise of one natural habitat is detrimental to another—even if it’s miles and miles away. However, current gaps in evidence hinder conservation-related decision-making on a global level. The researchers proffer three ways in which non-English-language studies could be particularly important:

Over one-third of scientific documents on biodiversity conservation are published in languages other than English.

Gaps in globally compiled English-language evidence are often found in areas where English is not widely spoken.

Important evidence in biodiversity conservation is routinely generated by local practitioners, who often prefer publishing their work in their first language, which, for many, is not English.

Based on the research done, non-English-language studies were shown to largely disprove common misconceptions, instead proving the following:

A considerable amount of scientific evidence underpinning effective conservation is available in non-English languages.

The number of published studies providing such evidence has been increasing in many languages.

Non-English-language studies can provide evidence that is relevant to species (including threatened species) and locations (including highly biodiverse regions, such as Latin America) for which little or no English-language evidence is available.

As noted before, non-English-language studies tend to utilize less robust study designs and thus be of lesser quality than their English-language counterparts. It is important to note, as the researchers state, that “blindly discarding such lower-quality, yet relevant, studies—a common practice in conventional evidence syntheses—could unnecessarily delay, misinform, or hinder evidence-based decision-making, especially… for emergent issues, such as pandemics, where making the best use of available evidence is an urgent challenge.”

But the solution offered by this research isn’t meant to be perfect. Incorporating more non-English-language studies can help diversify and enrich—and hence drastically improve—international studies into biodiversity and conservation, but even this account of studies carried out in the world’s top 16 language isn’t enough to cover everything. The researchers confess that this study does not “fully address the large evidence gaps in some regions faced with the most pressing issues including biodiversity loss, such as Southeast Asia, tropical Africa, and Latin America.”

What are some feasible solutions, then, that can be taken to address these inherent issues in the academic foray into biodiversity? More local evidence—any kind, as long as they’re based on robust study designs—is imperative, and should be encouraged in any local language. This method should also be complemented by “the distribution of free teaching materials to facilitate the testing of conservation interventions.”


What does this mean for translators?

With an increase in global communication in academia comes a greater need for translation services that properly ensure that no part of the research is lost. There will be a rise in demand for translators equipped with the proper experiences and knowledge to accurately translate academic, scientific texts. While we are not yet sure how far the impact of this research will travel, it’s nice—at times like these—to know that translators aren’t just sitting idly by, typing away at their computers. They play an integral role in protecting the world’s ecosystems.

When we think about it, translators are often deemed to be vehicles of communication; they are never the creators of meaning, but rather, messengers that ensure a message is properly received. But without this crucial link between languages, there is so much that can be lost in translation. This is true not only for biodiversity, but also for healthcare, business, and practically every other major field of study and industry. 

We envision a world that, funny as it sounds, is akin to Animal Crossing: New Horizons, in which languages no longer serve as barriers, but rather multiply and enrich our lived experiences—whether that be with our quirky neighbors or our local flora and fauna. And this research—available here on PLOS—is just another proof that translators are here to stay.


Taking Pathways to a New Level with the Pathways Language Model (PaLM)

A few weeks ago, we shared some exciting news about Google’s latest foray into artificial intelligence. Their new, exciting development? An AI architecture called Pathways, which, unlike previous AI systems that are designed for specific purposes, can more efficiently and innovatively handle tasks with a flexibility never before seen in artificial intelligence.

Earlier this month, Google software engineers Sharan Narang and Aakanksha Chowdhery—with a number of other Google researchers—introduced the Pathways Language Model (PaLM): a scaled application of Pathways to natural language processing. PaLM is a 540-billion parameter, dense decoder-only Transformer model that has been trained with the Pathways system, utilizing 6,144 chips and lots and lots of data samples. According to Slator’s Anna Wyndham, this makes PaLM larger than Microsoft and NVIDIA’s Megatron-Turing NLG, DeepMind’s Gopher, and OpenAI’s GPT-3.

And the results are astounding; the researchers “evaluated PaLM on hundreds of language understanding and generation tasks, and found that it achieves state-of-the-art few-shot performance across most tasks, by significant margins in many cases.”


The history

Recent years have seen the rise of larger neural networks trained for comprehension and language generation. OpenAI’s GPT-3 was the first model to show that large language models (LLMs) can be utilized for few-shot learning (learning with limited data sets) to great effect without the need for “large-scale task-specific data collection or model parameter updating.” Microsoft/NVIDIA’s Megatron-Turing NLG and DeepMind’s Gopher have shown even more outstanding results, as they exhibit more efficient processes and more accurate results. But Narang and Chowdhery note that “much work remains in understanding the capabilities that emerge with few-shot learning as we push the limits of model scale.” In other words, what are large language models capable of if we push them to their very limits?


How they do it

PaLM is an answer to that question, to the unforeseen capabilities of large language models to set new standards in artificial intelligence and language processing. The researchers used data parallelism across two Cloud TPU v4 Pods, which is a significant rise in scale compared to previous LLMs, most of which were trained on a single TPU v3 Pod or multiple TPU v3 Pods. Thanks to this parallelism strategy, as well as a “reformulation of the Transformer block that allows for attention and feedforward layers to be computed,” PaLM boasts a training efficiency of 57.8%, which the researchers claim to be “the highest yet achieved for LLMs at this scale.”

As for data, the researchers trained PaLM using both English and multilingual datasets including high-quality web documents, books, Wikipedia, conversations, and GitHub code. Also of note is the addition of a “lossless” vocabulary, which “preserves all whitespace (especially important for code), splits out-of-vocabulary Unicode characters into bytes, and splits numbers into individual tokens, one for each digit.” In other words, PaLM is geared towards efficient data management and handling, so as to perform better over a wide range of areas.


The results

PaLM exhibits what the researchers call “breakthrough capabilities on numerous very difficult tasks.” The researchers evaluated PaLM on 29 frequently-used English natural language processing tasks, such as question-answering tasks, cloze and sentence-completion tasks, Winograd-style tasks, in-context reading comprehension tasks, common-sense reasoning tasks, SuperGLUE tasks, and natural language inference tasks. It’s also remarkable that PaLM has also performed well on multilingual NLP benchmarks, including translation, despite the fact that a mere 22% of the training corpus was in a language other than English. The following chart depicts PaLM’s performance improvement compared to previous state-of-the-art (SOTA) systems:

The researchers also charted the scaled improvements of PaLM using the Beyond the Imitation Game Benchmark (BIG-bench), a suite of more than 150 new language modeling tasks. Compared to Gopher and Chinchilla, PaLM showed outstanding results, outperforming the two with the sheer power of its parameters. Here is the chart that compares the three systems:

The researchers note, however, that “PaLM’s performance as a function of scale follows a log-linear behavior similar to prior models, suggesting that performance improvements from scale have not yet plateaued.” Furthermore, at 540 billion parameters, PaLM actually outperforms the average performance of humans. All this is to say that PaLM is wonderfully effective at what it does, and that it knows no bounds—the ceiling is nowhere to be seen for this one.


The tasks

The researchers showcase some of the feats of PaLM, starting with its “impressive natural language understanding and generation capabilities on several BIG-bench tasks.” They note how PaLM is able to differentiate between cause and effect, comprehend conceptual combinations in contexts, and—get this—even guess a movie based on a series of emojis.

PaLM is also capable at reasoning tasks, thanks to its chain-of-thought prompting. This means that PaLM excels at multi-step arithmetic and common-sense reasoning compared to previous LLMs. This following comparison shows how PaLM drastically outperforms its previous counterparts:

With 8-shot prompting, PaLM succeeds in solving 58% of the problems, 3% higher than the prior top score of 55%. It is also only 2% lower than the 60% average of problems solved by 9- to 12-year-olds, who are the supposed target audience for the questions solved by PaLM. 

But what is perhaps most impressive about PaLM’s results is its ability to “generate explicit explanations for scenarios that require a complex combination of multi-step logical inference, world knowledge, and deep language understanding.” When given a novel joke it has never seen, PaLM provides a high-quality explanation, like this example:

Lastly, despite the fact that only 5% of its pre-training dataset was code, PaLM exhibits strong performance in coding tasks, rivaling the prowess of the fine-tuned Codex 12B while using 50 times less Python code for training. The researchers attribute this phenomenon to earlier findings, which state that “larger models can be more sample efficient than smaller models because they transfer learning from both other programming languages and natural language data more effectively. Here are some examples of the coding problems PaLM solves with ease:


Ethical considerations

While all these new development and achievements are remarkable and exciting, researching artificial intelligence and language models should be examined and treated with care, given that their learning process relies heavily on taking in information available on the web. The developers of PaLM took great care to document their ethical considerations and processes, keeping a “datasheet, model card and Responsible AI benchmark results” and carrying out “analyses of the dataset and model outputs for biases and risks.” In the words of Narang and Chowdhery:

While the analysis helps outline some potential risks of the model, domain- and task-specific analysis is essential to truly calibrate, contextualize, and mitigate possible harms. Further understanding of risks and benefits of these models is a topic of ongoing research, together with developing scalable solutions that can put guardrails against malicious uses of language models.

PaLM showcases what LLMs are capable of once they’re pushed to their limit. With its more efficient, smarter processes, the conscience of language models is starting to resemble that of humans. Make sure to check out the Google blog post and their paper, in which they go into more detail about their research processes.