Show HN: I built this to talk Danish to my girlfriend – works with any language

Original Article ↗ Hacker News Discussion ↗

Overall concept and reception

  • Tool helps users write messages in a target language by mixing unknown words in braces; AI fills in correct phrasing.
  • Many commenters like the idea, name, and simple UX; see it as a motivating supplement to language learning, especially for relationships/couples.
  • Several report building similar tools (web, WhatsApp, SRS, couple-oriented apps), reinforcing this as a popular LLM use case.

Functionality, language support & examples

  • Creator says languages are auto‑detected and should “work with all languages,” limited by the underlying GPT model.
  • Dialect handling is not explicit; users suggest providing dialect as context.
  • Multiple native speakers criticize the Danish examples as unidiomatic or incorrect, including a spelling mistake and “shopping” → “handle,” which changes meaning.
  • Some are confused whether the typos are intentional (to showcase correction) or just mistakes.

Reliability and technical issues

  • Many users across languages (Danish, Dutch, Chinese, Korean, Spanish, Hindi, Armenian, Indonesian, French, English) see “Unable to process correction” or “invalid text.”
  • Author states GPT/API credits were exhausted due to traffic and TTS isn’t enabled yet.
  • Some interpret failures as lack of multilingual support; others attribute it to the “HN hug of death.”
  • UI issues reported on iOS/macOS dark mode where text is nearly invisible.

Feature requests and improvements

  • Add:
    • Language selector to improve accuracy and reduce ambiguity.
    • Better, correct and realistic examples (with caching so they work even if backend is down).
    • Integration with established translators (e.g., DeepL) as a second check.
    • Rate limiting and/or client‑side use of users’ own API keys to avoid credit exhaustion and reduce data exposure.
    • Integration with messaging platforms like WhatsApp.

Comparison to using ChatGPT directly

  • Multiple commenters note the same behavior is trivially replicable in ChatGPT via a short prompt and question why a separate app is needed.
  • Counterpoint: dedicated tools can provide tailored UX (auto‑detection, inline corrections, chat‑specific workflows) even though they’re thin wrappers over LLMs.

Language, learning, and terminology tangents

  • Long subthread on “shopping” vs groceries and how loanwords spread across languages.
  • Discussion of difficulty learning Danish, especially pronunciation.
  • Debate over the term “expat” and its connotations versus “immigrant,” including class and integration implications.