Localisation has always been one of the most expensive parts of shipping a game internationally. Translating a game with 50,000 words of dialogue into ten languages costs, conservatively, $50,000 to $100,000 through traditional localisation services. For a small Australian indie studio on a $200,000 total budget, that’s a non-starter.

AI translation is changing this equation, and the improvements in quality over the last 18 months have been significant enough that several Australian studios are now shipping games with AI-assisted localisation.

What’s changed

Earlier AI translation tools — basically Google Translate with a gaming dictionary — produced results that ranged from awkward to unintentionally hilarious. Cultural nuance was lost. Idioms were translated literally. Character voice was flattened into generic sentence structures.

The current generation of tools is materially better. LLM-based translation systems understand context, maintain character voice across conversations, and handle idiomatic language reasonably well. They’re not perfect, but they’ve crossed the threshold from “embarrassingly bad” to “good enough for most purposes.”

The specific improvements that matter for games:

Context awareness. Modern tools can process entire dialogue trees rather than individual strings. This means they understand that a character who speaks casually in one scene should speak casually in the next. Previous tools translated each line in isolation, producing jarring inconsistencies.

Cultural adaptation. Better tools don’t just translate words — they adapt references and idioms for the target culture. An Australian colloquialism that doesn’t translate directly can be replaced with an equivalent expression in the target language rather than a confused literal translation.

Terminology consistency. In-game terms — spell names, item names, location names — are handled consistently throughout the translation. This was a persistent problem with older tools that would translate the same term differently in different contexts.

How Australian studios are using it

I spoke with four Australian indie studios that have used AI-assisted localisation. The approach varies, but the general workflow is similar:

1. AI generates the initial translation for all target languages.
2. A native speaker reviews the AI output and corrects errors.
3. In-game testing verifies that translations fit UI elements and make sense in context.

The key insight is “AI-assisted,” not “AI-only.” No studio I spoke with ships AI translations without human review. But the human review step is dramatically faster and cheaper than translating from scratch.

One Melbourne studio estimated that AI-assisted localisation cost them roughly 30 percent of what traditional localisation would have cost. They translated their narrative adventure game into eight languages for under $15,000, a price that would have been impossible through traditional services.

Working with AI agency Sydney helped one studio build a custom translation pipeline that integrated directly with their game engine, automating the export/import process and reducing the manual overhead of managing translations across multiple languages.

The quality question

Is AI-assisted localisation as good as professional human localisation? No. A skilled human translator with gaming experience produces better results, full stop. The nuance, creativity, and cultural sensitivity of a professional translation is noticeably superior.

But the comparison isn’t between AI-assisted and professional human localisation. For most Australian indie studios, the comparison is between AI-assisted localisation and no localisation at all.

A game available only in English reaches roughly 25 to 30 percent of the global gaming market. Adding Japanese, Chinese (Simplified), Korean, Spanish, Portuguese, German, French, and Russian covers over 80 percent. For a small studio, the revenue difference between those two numbers can determine whether the studio survives.

Decent AI-assisted localisation that reaches 80 percent of the market is better than no localisation that reaches 30 percent.

Where AI localisation still struggles

Humour. Comedy is the hardest thing to translate well, and AI still struggles with it. Jokes that depend on wordplay, cultural references, or timing rarely survive AI translation intact. If your game relies heavily on humour, you need human translators for the funny parts at minimum.

Highly poetic or literary text. Games with dense, literary prose — think Disco Elysium or Planescape: Torment — need human translators. The subtlety and intentionality of literary writing is beyond what AI currently handles well.

Voice acting direction. Even if the text translation is good, translating voice acting scripts requires understanding of performance context — emphasis, emotion, timing. AI provides the words but not the performance direction that voice actors need.

Gendered languages. Languages with grammatical gender (Spanish, French, German) create challenges when the player character’s gender is variable. AI tools have improved at this but still make errors that require human correction.

The future

AI game localisation will continue to improve rapidly. Within two to three years, the quality gap between AI-assisted and professional human localisation will narrow further, particularly for straightforward dialogue and UI text.

The role of human translators will shift from full translation to review, cultural adaptation, and handling the creative elements that AI can’t. This means fewer total translation jobs but higher-value jobs that focus on the parts where human judgment matters most.

For Australian indie studios, this is unambiguously good news. More studios will be able to reach international audiences. More games will be playable in more languages. And the creative output of the Australian industry will find audiences that language barriers previously blocked.