I never thought I’d have strong feelings about supertitles. They’re those scrolling translations above the stage that help you follow what’s happening when the soprano is singing in Italian and you’re sitting there pretending you understand. They’re functional. They’re furniture.

And now they’re the subject of a surprisingly heated debate in opera circles, because several companies — including at least one in Australia — have started experimenting with AI-generated supertitles for live performances. As someone who spent years in the chorus watching the supertitle operator try to keep up with a conductor who’d suddenly decided to take the tempo at double speed, I have thoughts.

What’s Actually Happening

Traditionally, supertitles for opera are created by specialist translators who work with the libretto months before a production opens. They don’t do literal translation — they condense, they time each line to match the musical phrasing, and they make choices about tone and register that shape how the audience experiences the drama.

A good supertitle translator is part linguist, part poet, part dramaturg. The role is more skilled than people realise.

The AI alternative uses large language models trained on opera libretti and existing translations to generate supertitles in near real-time. Some systems use pre-generated AI translations that are then synced to the performance. More ambitiously, a few companies are testing systems that listen to the live audio feed, identify where in the score the performance is, and display the corresponding supertitle dynamically.

The Bayerische Staatsoper in Munich ran a trial of AI-synced supertitles last season, and several North American companies have quietly experimented with AI-generated translations for less commonly performed works where finding a specialist human translator is difficult.

The Case For AI Supertitles

Let me be fair to the technology, because there are legitimate arguments in its favour.

Cost. Commissioning supertitles for a new production costs $5,000-15,000 depending on the work’s length and language. For smaller companies operating on razor-thin margins, that’s real money. AI-generated translations could reduce that cost dramatically, potentially making it viable to supertitle works that would otherwise go untranslated.

Speed. When a company mounts a rarely performed work or a world premiere, the supertitle translator often works under crushing time pressure. AI can generate a working draft in hours rather than weeks, giving human editors more time to refine rather than starting from scratch.

Rare languages. Not many supertitle translators work in Czech, Russian, or Hungarian. AI systems trained on multilingual opera corpora can handle these languages more readily, expanding the repertoire that can be made accessible to non-native speakers.

Dynamic syncing. The live-sync technology is genuinely impressive. Traditional supertitles are operated by a person pressing a button at the right moment, which means they’re only as good as the operator’s reflexes and familiarity with the score. AI syncing can track the musical performance in real-time and adjust timing automatically. In theory, this could be more precise than a human operator.

The Case Against

Now let me tell you why I think full AI replacement would be a mistake.

Opera translation isn’t just translation. When Violetta sings “Amami, Alfredo” in La Traviata, a literal translation is “Love me, Alfredo.” But a good supertitle translator knows that the dramatic weight, the musical context, and the character’s desperation might be better served by “Alfredo — love me!” or simply “Love me.” These are artistic choices that reflect deep understanding of the drama, the music, and the performance tradition.

Current AI systems are competent at literal translation but poor at these kinds of dramaturgical judgements. They don’t understand that a particular phrase lands differently in Act 3 than it would in Act 1, because the character has been transformed by events the AI doesn’t truly comprehend.

Nuance and register. Opera libretti span centuries of literary tradition. The language of a Mozart libretto by Da Ponte requires a completely different English register than a Puccini libretto by Illica and Giacosa, which is different again from a contemporary work. A skilled translator captures these differences. AI tends to flatten everything into a generic “translated text” voice.

Cultural context. Many opera texts contain references, double meanings, and wordplay that require cultural knowledge to translate meaningfully. When Leporello reads the catalogue of Don Giovanni’s conquests, the humour depends on understanding 18th-century social hierarchies. AI might translate the words correctly while missing the joke entirely.

The timing issue. I mentioned AI syncing sounds great in theory. In practice, live opera is messy. Singers breathe in unexpected places, conductors adjust tempi, stage mishaps happen. A human operator who knows the production can anticipate and adapt. AI syncing systems, from what I’ve seen, handle routine performances well but struggle with the unexpected — which is, of course, what makes live performance live.

Where I Land on This

I think the smart approach — and what I suspect most companies will eventually adopt — is AI-assisted human translation rather than full AI replacement.

Use AI to generate a rough first draft. Use AI to handle the timing sync. But keep a human translator in the process to make the artistic judgements, capture the nuance, and ensure that the supertitles serve the drama rather than just providing information.

The Royal Opera House apparently takes this hybrid approach already, using AI tools to speed up the initial translation process while keeping their supertitle team for final editorial decisions. That seems sensible.

What worries me is the companies that will see AI as a cost-cutting measure and skip the human step entirely. For well-known works with existing translations in the training data, the AI output might be passable. For new works, rare repertoire, or productions with unconventional interpretive choices, it’ll fall flat.

Supertitles are the audience’s window into the story. They shape comprehension, emotional response, and engagement in ways we barely notice when they’re done well — and absolutely notice when they’re done badly. Anyone who’s sat through a performance with poorly timed or badly translated supertitles knows the experience is significantly diminished.

Opera already struggles to attract new audiences. The last thing we need is for someone’s first experience to be accompanied by translations that read like a Google Translate output from 2018.

Use the technology. Just don’t forget that art requires artists — even in the parts of the production nobody thinks about.