For example, I canceled my Grammarly Pro subscription last year, letting Gemini — which I already pay for — handle all my proofreading needs.
Gemini has also eliminated the need to consider apps for transcribing meetings and taking notes, analyzing documents, or planning meals.
However, when I tried to use Gemini for mastering a new language, the results were a disappointingly mixed bag of sublime wins and disappointing fails.
For me, the challenge I faced was Tagalog. I married a Filipina immigrant two years ago, and while her English is very good, I exist in a circle of family and friends that all default to Tagalog.
Sadly, traditional apps have remarkably underserved that language. All the usual suspects, like Duolingo and Rosetta, treat the Filipino language as a second-class citizen, with rigid and unnatural drills.
Gemini promised to do better. In some ways, it did. But the reality was a bit more complicated.
The win
Gemini gets the nuances of the language
For text-based translation, Gemini nails it. It beats a traditional Tagalog-English dictionary hands down, and it picks up on things that Google Translate missed by being too literal.
One of the hardest things (for me, anyway) about learning a new language is realizing that not everything has a one-to-one mapping to English.
Traditional translation tools try to build that mapping, and language-learning apps work on rote-based repetitive drills.
Sometimes I need things to be explained, and Gemini does a stellar job as a tutor.
For example, I can ask a question like “What’s the difference between kumakain and nakikikain?” — both words related to eating a meal — and get an answer that makes sense.
Better yet, Gemini can help me avoid vernacular bear traps. When I was courting my wife, a couple of years before consumer AI tools existed, I was practicing Tagalog with her in a chat.
When she asked what I had for lunch, I replied with “Tapos lang ako kumain itlog ko.” Literally, that would translate to “I just finished eating my egg.”
What I didn’t know, and what Google Translate and Duolingo had failed to tell me, is that “itlog ko” means something different to a native Tagalog speaker. Thankfully, I learned that day that my wife has a sense of humor.
Punching that same phrase into Gemini immediately warns me that it “carries an unintended double entendre you will definitely want to avoid.”
The wall
The heteronyms will do you in
While Gemini nails written Tagalog by understanding nuance and context, trying to master spoken Tagalog using Gemini Live becomes an exercise in frustration.
Many languages are full of heteronyms — words that mean different things depending on how they’re said. If you’ve ever given someone a present, you know what I mean.
Tagalog is no exception. It relies heavily on syllable stress and glottal stops (which I still find challenging).
I’ve had a few funny moments at the dinner table trying to tell my sister-in-law that I’m full or asking someone to pass the vinegar.
In these cases, the word “puno” means both “full” and “tree,” depending on how it’s said. The same is true with “suka,” which can mean “vinegar” or “vomit.” Another word, “kaibigan,” which can mean “friend” or “lover,” was an absolute minefield when I was courting my wife.
It’s not hard to see the problem with that one — and Gemini is no help at all here. It just silently forgives my absolutely terrible accent.
Traditionally, Gemini Live has used a speech-to-text engine that transcribed what I said and then analyzed it as text, rather than putting it through a real-time audio analyzer.
It gets what I’m trying to say from the context, knowing that a conversation about a meal is more likely to end with me saying I’m full than declaring myself to be a conifer. But it doesn’t tell me that. Instead, it effectively validates those mistakes.
Until recently, even if I was self-aware enough to realize I’d pronounced something wrong, Gemini couldn’t help at all.
For example, let’s say I was attempting to say “Gusto ko mag-aral ng Tagalog para sa pamilya ko” (“I want to learn Tagalog for my family”), but realized that I stumbled over “mag-aral.”
If I asked Gemini Live, “Did I hit the stress right on mag-aral?” it would respond that it can’t hear “raw audio pitch,” and it would try to explain how I’m supposed to say it, rather than assessing how I was actually saying it.
Thankfully, recent updates to Gemini Live’s native audio processing have improved this. Now, if I ask about my pronunciation, Gemini can analyze my spoken audio and tell me if I’ve hit it correctly.
That’s an improvement, but it still requires manual intervention. Instead of carrying on a conversation naturally, I have to break character and ask, “Did I say that right?”
It interrupts the immersive conversational flow of learning the language, forcing me to fall back to more basic AI text chat. It won’t catch a mistake on its own.
It’s not only Gemini
Nobody’s nailed the sweet spot yet
To be fair, this is a problem that’s common to all AI tools. I use Gemini almost exclusively as I pay for it in my Google Workspace plan, but I’ve tried ChatGPT and Claude with similarly disappointing results.
Meanwhile, dedicated apps like Duolingo and Pimsleur offer better spoken audio features, but they focus too much on mainstream languages and lack the contextual intelligence Gemini excels at.
But for conversational AI models, it’s an unfortunate trade-off between creating natural, human-like conversational partners — which must be forgiving of vocal slip-ups and accents — and training an AI to detect vocal inflections.
I believe if anyone can solve this problem, it’s Google. It has world-class translation models, some of the best speech recognition, and incredible audio hardware on the Google Pixel.
Of all the AI assistants I’ve tried, Gemini also has the best “brain” for recognizing cultural nuance and grammar.
From a consumer perspective, the missing piece is no longer the audio processing. What Gemini Live needs now is a more focused language learning mode.
You can ask Gemini to respond with a pronunciation analysis after each sentence, but it doesn’t always get it right, and when it does, it’s not as seamless as it should be.
For one thing, it’s hard to get it to only call out mistakes. If you’re speaking properly, you still get an assessment after every phrase, which becomes too much.
This also turns it into a straight diction coach rather than a conversationalist. Gemini Live then starts fixating entirely on pronunciation while missing context and failing to analyze inflections.
You can prompt it to give feedback only when a correction is needed, but then it’s more likely to miss errors, especially if the conversation continues for long enough that it forgets the earlier prompt.
What Gemini Live needs is a dedicated “Language Tutor” mode that toggles it into being a proactive phonetic coach.
Without the ability to consistently interrupt bad accents in real time without otherwise breaking the flow, Gemini remains confined to the role of an excellent linguistic consultant that can help you with your text but remains cumbersome at assessing how you sound.
