code-switching is not occasional — it's often the default
for a large number of people, especially across urban india, switching between english and hindi (or another regional language) mid-sentence isn't an exception that happens under specific circumstances — it's simply how a normal sentence gets built. "kal meeting mein boss ne bola ki deadline extend ho gayi, which is honestly such a relief" is not broken english or broken hindi. it's one complete, natural thought in the language it actually arrived in.
a journal that expects you to pick one language before you start talking is asking you to translate yourself before you've even said anything — which adds exactly the kind of friction that kills a reflection before it begins.
the switch itself often carries meaning
code-switching frequently isn't random noise — it tracks something real about the moment. some people default to english for work and logistics, and switch to hindi (or another home language) the instant something becomes personal or emotional. others do the reverse. either way, the specific point in a sentence where the language changes is often exactly where the feeling changes too.
"i told him it was fine, but andar se thoda bura laga" isn't just a sentence with two languages in it — the switch to hindi at "andar se" (inside, deep down) is doing real emotional work that a fully-english or fully-translated version of the same sentence would flatten. a transcript that normalizes everything into one language loses that signal along with the words.
why translation is the wrong default
it might seem more useful for a transcript to standardize everything into one clean, readable language. in practice, that erases information rather than clarifying it — a word chosen specifically because the other language didn't have the right one, a name that only makes sense in its original script, a phrase that's genuinely untranslatable without losing its texture.
archive's approach is to transcribe each word in the language it was actually spoken, written out in latin script exactly as you'd type it in a text message — not switched to devanagari mid-sentence, and never quietly translated into a single "clean" language. the tool exists specifically to show this pattern concretely: type a mixed sentence into the hinglish transcription demo and see each word rendered in its own language, side by side, the way you'd actually text it to a friend.
this extends past hinglish specifically
hindi-english mixing is the most common pattern this affects, but it's not the only one. someone might journal mostly in english but drop into hindi, tamil, bengali or another language specifically for family conversations, food, or anything tied closely to home. someone else might use english for technical or work vocabulary inside an otherwise non-english sentence, because that's genuinely the language they learned the term in.
the design principle underneath all of these is the same: a journal should preserve the actual shape of how you think and speak, not impose a single "correct" language on top of it. natural speech doesn't sort itself into tidy, single-language sentences, and a transcript that pretends it does is quietly editing your own memory of how something was actually said.
what this means for names and specific vocabulary
mixed-language speech is also where transcription errors do the most damage, because names, places and specific phrases are exactly the words most likely to get mangled when a system defaults to one language's rules. "aakanksha" rendered as "akansha," or "hauz khas" rendered as something unrecognizable, breaks the entry in a way that a slightly awkward but correct sentence never would.
this is why keeping the original audio next to the transcript matters even more for mixed-language entries than for single-language ones — a transcript error is fixable if you can go back and hear what was actually said. without the audio, a misheard name just becomes a permanent, quiet corruption of the memory.
the gap between typing and speaking, in a mixed language
code-switching while typing usually involves a tiny, almost unconscious moment of choosing a script — deciding whether to type a hindi word in latin letters or switch the keyboard to devanagari, deciding whether to write the english word or its hindi equivalent. that choice adds friction, even if it's fast, and over a long entry it adds up.
speaking removes that choice entirely, because there was never a keyboard to switch in the first place — the sentence just comes out the way it actually forms in your head, mid-language-switch and all. a journal that can accurately capture that is capturing something typing structurally can't: the sentence exactly as it was thought, with zero editing already applied before it even reached the page.
talk the way you actually talk
the entire point of a voice-first journal is to remove the friction between having a thought and recording it. forcing a single language onto that recording reintroduces exactly the kind of friction voice was supposed to remove — now you're not just talking, you're translating yourself in real time before you've finished the sentence.
a private journal is one of the few places where there's no audience to perform correctness for. it should be the easiest place, not the hardest, to talk exactly the way you actually think — in whatever mix of languages that happens to be.