From Twilight to Warcraft: A Translator on Fantasy, Games, and AI

Author: Gan Zhenlong

Hey everyone, I’m Gan Zhenlong. I’d like to thank CRUX Language Services for featuring me and I’m incredibly honored to have the opportunity to introduce myself here.

I studied psychology at Simon Fraser University (Canada), and boast extensive experience translating bi-directionally between Chinese and English. The most popular work I’ve translated is likely Midnight Sun from the Twilight series.

Translating Fantasy: An Exploration into World of Warcraft

I have translated a range of fantasy works, including the World of Warcraft novels and the Throne of Glass series. The World of Warcraft novels, in particular, are a distinctive form of fantasy writing: as narrative extensions of a long-running game franchise, their lore, terminology, and characters are already firmly established within the player community.

As a long-time World of Warcraft player, I’ve occasionally found myself slipping into translator mode during gameplay—identifying and reporting mistranslations to in-game Game Masters that were later corrected by the studio. For example:

#1: Mistranslation from Misunderstanding of Idioms

In the tavern at Honor Hold in Hellfire Peninsula, an NPC bartender mocks an NPC warlock who has just entered. The warlock’s reply was rendered in Chinese as a direct, word-for-word translation meaning “don’t start with me.” While technically faithful to the surface structure of the English sentence, this phrasing is unnatural in Chinese and fails to convey the intended message. Based on the context, I deduced the original line as “Don’t you start with me,” a common English idiom used to mean “don’t provoke me” or “don’t pick a fight.” Because the idiomatic meaning was not carried over, the translated line read awkwardly and obscured the character’s intent. I suggested revisions such as “Don’t give me that attitude” or “Don’t start something with me,” which better reflect the warlock’s emotional response and fit the conversational context.

#2: Mistranslation Caused by Polysemy

In Crystalsong Forest in Northrend, a tree-like creature drops an item upon death that was translated into Chinese as “Broken Record.” However, clicking on the item makes it clear that the object is a piece of wood. The English source text is evidently “broken log.” In this context, log should be understood as “a piece of wood,” not “a record” or “logbook.” The mistranslation appears to stem from interpreting the word log through a technical or nautical sense, resulting in a semantic mismatch with the in-game object.

For me, identifying mistranslations through the Chinese text itself and successfully reconstructing the original intent is not merely part of the gameplay experience, but an ongoing exercise in linguistic sensitivity and translation awareness. A translator must do more than master two languages; they need an intuitive grasp of context, tone, character, and player expectations in order to avoid the trap of mistranslations that appear correct on the surface but miss the mark in use.

A Study in Translation: Novels vs. Games

Translating a novel is a bit like translating subtitles while watching a film: the length of your translation is entirely at your discretion.

Game translation, however, is more like translating subtitles without the movie screen, and sometimes without any audio at all. Game translators must render the dialogue accurately, but also keep it concise to ensure players aren’t forced to strain their eyes reading long blocks of text in game. 

One of the most immediate differences between the two lies in character limits. When translating a novel, the translator can freely adjust sentence length to match tone, pacing, and emotion. In games, by contrast, many types of text (e.g., menu items, button prompts, equipment names, and UI descriptions) come with strict length constraints. Exceeding those limits can cause layout issues, broken interfaces, or even game crashes. As such, game translators must preserve the core message while compressing the language, maintaining both semantic clarity and natural flow.

A second key difference is the lack of context. When translating a novel, the translator has full access to the story, character backgrounds, and narrative context, making it easier to judge tone and intent. Game text, however, is often highly fragmented. You might be given a single line such as “Don’t move!” without any indication of who is speaking, to whom, or under what circumstances. Unless explanatory notes are provided—and often they are not—the translator must infer these missing details through contextual reasoning. Failure to do so often lead to mismatched character tones or misaligned character personalities.

Finally, there is the challenge of tone management. One defining feature of many Western games is how natural and conversational the dialogue feels: grand speeches may be rare, but banter is everywhere. A character’s tone can also easily shift as the story or gameplay progresses. Within branching dialogue, combat barks, or quest interactions, a single character may move from calm to provocative to angry in the space of just a few lines. This requires the translator to assess tonal range based on character identity and situation, and to make nuanced adjustments so that three lines can sound distinct while still remaining true to the same character.

The Value and Potential of Novel/Game Translators in the Age of AI

This is a particularly fascinating subject for me and so I’d like to explore this more in depth. 

It brings to mind an often-discussed hypothetical question: Would you knowingly board a flight that is not manned by a human pilot?

Modern aircrafts are capable of automated taxiing, takeoff, cruising, and landing. In theory, an entire flight can be handled by computers alone. Yet I believe that most passengers still prefer knowing that an experienced human pilot is present in the cockpit. Setting aside other factors, one crucial difference is that human pilots possess emotion and an understanding of the consequences of failure—something AI does not.

The same applies to translation. In theory, AI is indeed capable of handling most sentence-level translations and conversion. In practice, however, today’s AI systems are largely LLMs (Large Language Models) rather than true AGI (Artificial General Intelligence). While LLMs can simulate human linguistic logic, they are still unable to genuinely experience the emotions, tone, and subtext conveyed through human language. As a result, translations generated through such simulation, whether in books or games, can have a noticeable impact on the audience’s experience.

As such, the rise of AI does not signal the end of human translators. On the contrary, it introduces a new model of human-machine collaboration. AI performs extremely well when dealing with clearly structured text and straightforward contexts, and its speed is astonishing. Humans, however, are inherently complex, and human language often carries contradictory emotions, subtle tonal shifts, cultural nuances, and layers of implication beneath the surface. This is precisely where AI tends to reveal its limitations. In the translation of novels and game narratives, these “human” qualities are often what determine the overall quality of the translation.

As I see it, current LLM-based AI has two incredibly powerful functions. The first is fact-checking. In both novel and game translation, translators frequently encounter obfuscate terms—such as the mechanisms of a machine, or highly specialized terminology. In the past, translators used Google search by attempting to guess the right keywords, sifting through results, and hoping one source contained the actual answer. Today, AI tools (including AI-enhanced Google Search) allow translators to ask questions in simple language and receive similarly phrased explanations in return. For me, fact-checking has always been critical: if a translator does not understand the term, mistranslation becomes almost inevitable. 

The second major strength of AI lies in text generation. While AI may not replace human translators, it functions as a tireless assistant—available 24/7, responding instantly, and endlessly patient. It can generate large amounts of customized reading material based on the languages I am studying, such as English or Japanese, and can even adopt the voices of characters from games or anime I enjoy. This makes the learning process both engaging and challenging. For a professional translator committed to lifelong learning and consistent improvement, AI is Batman’s Alfred: always on standby to offer technical support. But on the battlefield of translation, the one at the helm with the pen is ultimately the translator.

And so no, I am not concerned with AI replacing translators. The real question is whether translators will continue to learn and evolve. In this day and age, translators who can collaborate effectively with AI, make smart use of tools, and still retain their uniquely human sensitivity to language and judgment will continue to add depth and color to the bridges that connect people through language.

Translation often reflects the rise and fall of different industries. In this column series, Translation Across Industries, CRUX invites translators and professionals from a wide range of fields—both within and beyond the translation industry—to share practical perspectives on the role translation plays in their respective sectors, and to explore the insights and nuances that matter in real-world practice.

Gan Zhenlong is a graduate of the Department of Psychology at Simon Fraser University in Canada and has worked as a professional English translator for many years, covering a wide range of genres. Notable translations include Twilight: Midnight Sun, as well as novel series based on World of Warcraft, Star Wars, and Throne of Glass.