Rita El Khoury / Android Authority
Gemini is still far from perfect, but it’s slowly grown on me. Specifically, though, it’s the Live conversation mode that appeals to me the most because it’s everything I ever wanted from Google Assistant, and then some. I can talk to the AI, interrupt it, ask it to repeat, correct it, and ask for more details — all in a very natural and relaxed chat.
But if you’re someone like me and you’re used to speaking three languages at the same time, often in the same sentence, and your brain works like that by default, making it tough to maintain an entire conversation in one language, then you’ve probably been chomping at the bits, waiting for Gemini Live to support multiple languages. With the March Pixel Drop, the feature is now here, and Oh. My. Heavens. Is it much better than I expected, or what?
Have you tried speaking to Gemini in multiple languages?
4 votes
More intuitive and reliable than Google Assistant’s multi-language
When Google announced it, I thought multi-language support in Gemini Live meant that I could have a conversation in English and then another conversation in French without manually changing the language. This has been the case with Google Assistant for years, except I had to manually set exactly which languages I wanted to use on Assistant, and it never really worked as well as expected.
With Gemini Live, as you can see in the video above, that’s not the case:
- I didn’t have to pick the language each time; I just started a new chat, and it understood me.
- Out of the box, it works with all languages supported by Live. I don’t have to limit myself to just two like with Assistant.
- Although I had some awkward silences from Gemini and had to repeat a few sentences, the AI’s success rate for recognizing different languages has been over 90% in my testing, and that’s more than Assistant could ever dream of.
I speak three languages near-natively (English, French, Arabic) and can understand and speak (with a thick accent) some Spanish, Italian, and German. So, I put this to the test and tried different chats with Gemini Live in all of these. It got me and all of my native and thick accents every time.
The only one I had trouble with is, oddly enough, my Arabic mother tongue. I could speak in formal written Arabic, but that just doesn’t come naturally to me. Instead, when I speak, it’s in the informal Lebanese dialect. Gemini, however, seems to speak a mix between a non-descript informal Levantine dialect and formal written Arabic. I blame this on the millions of regional dialects and how complicated and widely different they are, but even then, the success rate was higher than I ever anticipated or ever experienced with Assistant in Arabic.
All of this was already a win, but then I decided to push further. And that’s where Gemini Live figuratively blew my socks off.
Gemini Live’s multi-language skills work mid-chat and mid-sentence!
Rita El Khoury / Android Authority
Since I had such a positive experience with different chats in different languages, I wanted to see if Gemini could handle me switching languages mid-chat. So I started a simple discussion in English, then switched to French, Arabic, Spanish, Italian, German, and it followed me across all six, never breaking a sweat. You can see it in the video below.
Looking back at the transcript, I could see that it really understood every word I said in every language and switched its responses accordingly.
But I couldn’t stop there, could I? Now, I was curious to see if it could handle switching mid-sentence. So I started a sentence in English, finished it in French, and waited with bated breath for the answer. And it got it! I tried the other way. Success!
Honestly, at this point, I was internally screaming, “Sorcery!” After living with Google Assistant for 10 years and seeing it struggle to know the difference between “bonjour” and “bone joke,” I’d lost all hope in AIs and voice recognition algorithms. But Gemini Live restored that faith. Check it out in action:
I started mixing in Arabic and Spanish and kept switching mid-sentence, and it got all of those. It often answered in the first language I started my sentence with, but its answer was proof that it understood the entire question, not just the first part. It even opened my wound about Randal Kolo Muani’s last-minute miss in the last FIFA World Cup and teased me about Emiliano Martinez’s excellent save. Oh, well.
Rita El Khoury / Android Authority
Past that, I wanted to try destabilizing Gemini Live even more and push it to its limit. So, I started speaking like I normally do with my family and friends, mixing English, French, and Arabic in the same sentence — the true Lebanese way of speaking, if you will. To my absolute gobsmacking surprise, it got our famous, “Hi, kifak, ca va?” And it followed along just fine (aside from the awkward Arabic accent limitation that I mentioned earlier).
One word in a different language in the middle of an entire English sentence? No problem
Finally, I just went for the most extreme example I could think of: speaking an entire sentence in one language but putting one word in another. To be fair, this is how I speak with my husband 90% of the time. If we’re using English, a few words will elude us, and in the middle of our flow, we just use the French or Arabic word instead. Or if we’re speaking Arabic or French, we’ll intersperse some words in the other languages without giving it much thought. It’s how our brains work normally, and it’s why I’m never too comfortable talking with voice assistants because I have to force myself to just use one language. But Gemini Live got it.
I asked, “There’s a plant called habaq in Arabic, what’s that in English?” It told me that’s basil. I asked, “What’s the word for ciboulette in English?” It said chives. And when I said, “I have some small roquettes,” while rolling my r, it understood that I was talking about rocket/arugula leaves. And finally, when I asked what “jozt el tib” was in English, it correctly said it’s nutmeg. (Yes, I was in my kitchen and trying to get ideas for Gemini testing.)
Rita El Khoury / Android Authority
Looking back at the chat log during all of these, the transcript isn’t 100% accurate or in the right language. There’s “haba” and “rocket” and “rose to tibe,” while “ciboulette” isn’t even written in any way. But the answer proves that Gemini Live got the right word in the right language every single time.
And these aren’t just extreme cases. They’re all questions I’ve actually asked myself or used Google Translate for at one point in my life. You can’t believe how many times I want to search for recipes with zucchini and all my brain wants to type is “courgette recipes.” So I used to go translate it first, remember that’s zucchini, then come back to do my search. This multi-language support means I no longer have to do that silly first translation step. I can finally follow mom’s sfouf (curcuma cake) recipe with 3e2de safra (curcuma) and look for places to buy grains for my dad’s recommended ba2le (purslane) plant without breaking a sweat.
I went back to my testing and tried the same questions with ChatGPT’s voice chat mode. While it did get the French roquettes and ciboulette, it failed with the Arabic habaq and jozt el tib, telling me they’re fenugreek and cumin. Oof. I wouldn’t want fenugreek in my pesto.
After all of this testing, I can’t but tip my hat to the Gemini team for nailing multi-language support and making it work so impressively well from the get-go. Every time I pushed it further, I was surprised to see that it still kept up with me. This is the first AI agent that understands me the way I naturally speak, so I no longer have to remember the exact word in English if I want to carry on a conversation with it. I still have to morph my Arabic accent a bit to get it to understand me, but that’s a small price to pay for such a versatile AI voice agent. Once it understands the Lebanese dialect as is, though, it’ll be absolute perfection.
