Brood War Korean Translations

Original Article ↗ Hacker News Discussion ↗

AI-assisted translation is opening up decades of Korean StarCraft: Brood War strategy content to non-Korean speakers, but the project highlights how hard it is to handle game-specific slang, build-order notation, and cultural nuance. Commenters compare traditional tools like Google Translate with modern LLMs and ASR systems such as Whisper, noting that with the right prompting and domain dictionaries, LLMs can correct transcripts and produce far more idiomatic, jargon-aware translations. The thread also draws parallels to translating technical language in fields like software development and Go, and expresses a broader desire to preserve and share Korean esports knowledge that has long been inaccessible to foreign fans.

AI Translation vs Traditional Tools

  • Several commenters note that LLMs now often outperform older machine-translation systems like Google Translate, especially when properly prompted.
  • There is curiosity about whether Google Translate is already using newer LLM tech or still relies on older MT models.
  • Some highlight that LLM translations can be very good but often sound overly formal or robotic unless carefully prompted.

Domain-Specific Language & Jargon

  • The article is praised as a deep dive into how Brood War commentary has evolved into its own domain-specific language, beyond just Korean vs. English.
  • Similar issues are reported in other domains (Go, software engineering) where generic translators mis-handle specialized terms or choose literal meanings.
  • Several people equate “domain-specific language” here with “jargon,” noting that outsiders will find it opaque.

StarCraft Build Orders and Korean Slang

  • A large subthread explains that numbers like “12 hatch / 12 pool” refer to supply (effectively workers in early game) when structures are built, not quantity of buildings.
  • Korean terms like “front yard” (앞마당) for “natural expansion” spark debate over whether to translate literally (“courtyard/front yard”) or to match English esports jargon (“natural”).
  • Commenters share Korean BW slang differences from English (“multi/double” for expand, various terms for “cheese/all-in”) and note heavy use of English loanwords in Korean.

Speech Recognition & Contextual Biasing

  • A speech-recognition researcher points out that the original Korean transcripts (e.g., Whisper) contain domain errors that LLMs later “fix.”
  • They suggest contextual biasing with a slang dictionary to improve ASR output, citing tools in existing libraries; another commenter mentions similar work in faster-whisper.

Strategy, Cheese, and Game Theory

  • Long discussion about “cheese” builds: high-risk, often all-in strategies that can feel unfair but are essential for keeping opponents honest.
  • Comparisons are drawn to bluffing in poker and game-theoretic optimal frequencies of risky plays.

Broadcasts, Commentary, and Archives

  • Multiple people express desire for more Korean VODs and translated historical broadcasts.
  • There is frustration about copyright takedowns and concern that important esports history may be lost without institutional support.

Language, Borrowing, and Miscellany

  • Commenters discuss how technical communities import English terms into other languages (Spanish, German, Polish) versus coining native equivalents.
  • Minor nitpicks include misuse of “signal-to-noise ratio,” a typo of “Najdorf” in chess, and some terminology choices in the article.