Skip to main content

When it comes to using AI for translation in government and community communications, there’s a lot of buzz, and a lot of misunderstanding.

We recently spoke with Dr Erika Gonzalez Garcia, an Associate Professor in Translating and Interpreting and a member of the government’s Culturally and Linguistically Diverse Communities Health Advisory Group. Erika combines academic expertise with hands-on experience as a translator, interpreter and national language policy advisor.

She shared what’s really happening in the world of language services, and what needs to change if we want to meaningfully reach culturally and linguistically diverse (CALD) audiences in Australia.

On AI and Translation: “It’s great. But you always need a human.”

“With AI, look, it’s true that AI has advanced in a significant manner in the last few years. AI basically learns from what’s been produced before, right? So for languages like Spanish or French, where there’s a huge amount of material, AI is fantastic. It works really well. But the issue for me, the greatest issue in Australia, is the lack of capacity in certain languages, especially many community languages.”

She explained that while AI performs well for high-resource languages with large datasets, it still falls short when it comes to minority languages.

“Swahili, Kurdish, these don’t have such large data sets. And whatever you get from a machine, it’s often very insufficient. It still requires a lot of human input. Sometimes you get stuff from the machine that’s so bad, you spend more time correcting or postediting it than if you’d just translated it from scratch.”

On CAT Tools or Translation Memories

CAT tools, or computer-assisted translation tools, are often confused with AI, but they serve a different purpose.

“CAT tools are different from AI. CAT tools are software that we use. When the software identifies that we have translated this before, it gives you the translation.

For example, legislation might be amended or updated, but most of the bank of the text remains the same. The translator starts the project and the software automatically identifies what’s been translated before.”

But in Australia, many community language translators have not had access to this kind of training or technology.

On Tools: Why She Uses DeepL (and Not Google Translate)

When it comes to machine translation tools, she’s clear about what she uses and why.

“Would you use Google Translate? No. No. I’ve got a paid subscription to DeepL. And I’ll tell you why. If you’ve got a paid subscription, that allows for privacy and security of your translations compared to free versions.

It gives you a layer of privacy that Google Translate doesn’t. I pay like $167 a year. I can translate whole documents, and there are higher level subscriptions for those who require higher volumes. I use it as a translator and as an interpreter on a continuous basis.”

On Post-Editing: “We need to train post-editors, not just translators.”

She supports using AI where it makes sense, but stresses the importance of human oversight.

“So to me, it’s like, use it, by all means. It’s a great advantage. But still, you always need a human, and you always need to allocate a portion of the budget.

You cannot just churn a  translation in AI and just put it like that. Because sometimes the machine doesn’t pick up intercultural nuances.”

“So let’s say, terms that are culturally or diversity-linked.’ In Australia we use certain terms. In other countries they might use other stuff. A local translator should go through it to identify what the issues are.”

On Strategy: “You don’t need to translate everything for everyone.”

She emphasised the importance of targeting translations based on need.

“The most important thing is to gauge the level of English within those communities. Maybe you don’t need to translate everything for everyone. For example, if you have to translate or do an animation about aged care, imagine translating it into, I don’t know how many languages. But some of the languages, for example, don’t have enough migrants in aged care.

It would be like, okay, maybe you go and choose Italian, Greek, Mandarin – obviously Chinese – and Arabic, but maybe not Thai, because there’s still not enough elderly Thai population in aged care.

You do a bit of demographic research.”

On Plain English: “It saves a lot of translation, of course.”

Plain English came up as one of the simplest and most effective tools for improving communication.

“Plain English is helping not just the translators, because then you’ve got a source text that’s a lot simpler. Usually fewer words, easier to translate.

But also for people who have more limited understanding of the language.”

She explained that plain English is often misunderstood as just shortening text, when in fact it’s a skill in itself.

“It’s not just summarising. It requires a skill, like choosing some verbs over others. There are courses in plain English. Some departments are training staff in plain English drafting. It’s not something you can always rely on putting through ChatGPT.”

On Community Review Panels: “That was the real learning curve.”

She spoke about a major shift in thinking that came out of the pandemic: the value of community input.

“I would never before think that we need to involve the community to revise the translation and provide us with their opinion. But then you realise, yes.

Because I might be translating it correctly, but the register might be too high. Maybe it’s too wordy. Maybe it doesn’t sound natural to them.”

That insight has now become standard practice in many large-scale projects.

“Now most of the translation agencies for big projects involve community review panels before translations are actually out there.”

On Where to Start: Protocols and Resources

For anyone looking to improve how they approach translation and multicultural communications, she recommended starting with national guidelines.

“If you go to AUSIT, the Australian Institute of Translators and Interpreters, we’ve got the National Protocols for the Translation of Community Communications.

The Victorian Governments Better Practice Guide for Multicultural Communications is also very useful.”

Final Word

She summed up why language access is not just a policy detail, but a national priority.

“This is not a tick-boxing exercise. In Australia, half of the population has been born overseas or has a parent who was born overseas. According to the latest census, approximately 850,000 people don’t speak English well.

It’s part of Australia’s national identity and tapestry. And unfortunately, there’s no way around it. Translation & interpreting services are an important public service, just as the Ambulance Service or the Police is.”