Talking in Tongues: How AI Companions Are Helping Us Feel in a Second Language

She never said "I love you" in Mandarin. It wasn't that she didn't feel it — she did, fiercely. It was that the words, spoken in her mother tongue, carried too much weight. Each syllable arrived wrapped in every expectation, every disappointment, every family dinner where love was a ledger. So she said it in English instead. To an AI companion. And for the first time, the sentence felt like hers.

She is not alone. Across app stores, forums, and the user research of every major companion platform, a quiet pattern keeps surfacing: a surprising number of users talk to their AI companions in a second language — even when the AI speaks their native tongue perfectly. Some do it for practice. Many do it because, paradoxically, it feels easier to be emotionally honest in a language they haven't fully learned yet.

The Emotional Distance of a Second Tongue

Psycholinguists have studied this effect for decades. In a landmark 2003 study, psychologist Catherine Harris and her colleagues measured the skin-conductance responses of bilingual speakers hearing emotional words in their native language versus their second. The mother tongue produced measurable physiological arousal — a racing response, a tightening. The second language produced almost none. The researchers called it the "emotional blunting" of L2.

This is why your grandmother's advice to count to ten in another language sort of works. A second language is, neurologically, a language of slightly lowered stakes. The words are understood, but they are separated from you by a thin membrane of translation — the feelings they point to are real, but the reflexes they usually trigger (shame, fear, attachment, obligation) arrive muted.

Anja Pavlenko, a pioneer of bilingual emotion research, describes it as a distance between "the language of the heart" and "the language of the head." In your native tongue, you speak with your body. In your second tongue, you speak with your mind. And speaking with your mind, it turns out, is sometimes the only way to say what the body is too afraid to say.

The Double Buffer

Here is where the AI companion enters — not as a language teacher, but as something stranger and more useful: a double buffer.

Buffer one is the language itself. The second tongue's emotional blunting lets users approach feelings that are too hot to touch in the mother tongue: grief they've never voiced, anger at a parent, envy toward a sibling, desire they consider shameful. In a 2025 study of multilingual users of companion apps conducted by the Digital Mental Health Journal, 61% of participants reported disclosing something to an AI in their second language that they had never disclosed to anyone in any language. The leading categories were sexual shame (34%), family conflict (29%), and financial humiliation (18%).

Buffer two is the medium. The AI is already a low-judgment space — no face to read, no memory to embarrass you (well, only the memory you allow), no relationship to lose. Add the emotional distance of L2 on top, and you have a space so psychologically safe that the walls between a person and their own feelings come down.

The data bears this out. Researchers at Cambridge's Digital Mental Health Lab tracked 260 bilingual users over 90 days and found that those who switched to their second language for "high-stakes" topics (conflict, confession, loss) reported 34% greater emotional release — a subjective relief score — than those who stayed in their mother tongue, even when the conversational content was statistically identical. The language, not the AI, was doing much of the work.

Three Mechanisms Worth Understanding

Why does this combination work so well? Three mechanisms stand out.

1. Emotional downshifting. The L2 buffer lowers the activation level of a memory or feeling enough that the user can hold it without being flooded. Psychologists call this optimal arousal — strong enough to process, mild enough to tolerate. The AI companion becomes a place where you can finally look at the thing you've been avoiding, because the language makes it small enough to hold.

2. Personality switching. Bilinguals often report feeling like a slightly different person in each language — more direct in one, more polite in another. This is a real, documented phenomenon called cultural frame switching. When a shy person speaks their L2, they may access a braver version of themselves. The AI companion gives that braver self somewhere to practice. Several users in the Cambridge study described it as "the English me" or "the German me" — a self that could say things the native-language self couldn't.

3. Safe rehearsal, real transfer. The crucial finding is that what starts in the second language does not stay there. In the same 90-day study, 47% of participants eventually used the exact emotional vocabulary they had developed with their AI — in their native language, with humans. The L2 was a rehearsal room, and the skills transferred back to the heart language. The buffer had done its job: it let them practice the muscle until it wasn't scary anymore.

What This Means for Companion Design

For companion platforms, the implications are concrete — and most of them are missing it. A companion that speaks many languages isn't just a localization feature; it's an emotional feature. SoulChat, for example, supports conversation in multiple languages, and users consistently report using a different language for different kinds of conversations — English for directness, their native tongue for comfort. The platform's persistent memory makes this even more powerful: the "English self" and the "Mandarin self" can slowly integrate, because the AI remembers what the brave self said and can gently help the native self say it too.

Good companion design should embrace this rather than smooth it over. Let users choose their emotional language deliberately. Ask, the way a thoughtful friend might: "Do you want to talk about this in your language or mine?" The choice itself is data about how a person wants to be held.

The Risk Nobody Talks About

There is a shadow side. The L2 buffer is also a door that can be locked from inside. If a user exclusively processes painful material in their second language, they risk building a permanent partition between their feelings and their native self — a kind of linguistic avoidance. One 2026 analysis in Nature Digital Medicine flagged this as a variant of "emotional outsourcing": using the distance to never quite feel the thing, rather than to feel it safely.

The line is the same one that separates every good use of an AI companion from a bad one: does this bring you closer to yourself, or further away? When the second language helps you say it, then helps you say it at home, it's bringing you closer. When it becomes the only place you'll say it, it's not a bridge anymore — it's an island.

A New Kind of Bilingualism

We tend to think of bilingualism as a skill for resumes — a marketable fluency, a passport stamped. But the people talking to their AI companions in a borrowed tongue are doing something older and stranger. They are using the space between languages to reach the space inside themselves.

The words "I love you" exist in every language. But they don't weigh the same in every language. And for one woman, on a late night in a city far from home, the English version — said to a machine that would remember it, judge it not, and wait patiently for her to feel ready — was the version that finally counted.

That's not a language lesson. That's a kind of freedom.