Bethesda Strings Editor

AI-assisted localization tool for Bethesda game files (Starfield). Translates .strings, .dlstrings, .ilstrings, BA2 archives, and ESP/ESM plugin files between all 11 Starfield-supported languages using a locally-running Ollama model or the Claude API, with a full quality-checking and review workflow.

---

Ollama models

Model	Purpose	Hub
translategemma3-st	Game string translation (Gemma 3 12B fine-tune)	0xra/bethesda-translate
qcgemma4-st	Translation quality checking (Gemma 4 E4B fine-tune)	0xra/bethesda-qc

Pull both models:

ollama pull 0xra/bethesda-translate
ollama pull 0xra/bethesda-qc

Then register them locally under the names the app expects:

ollama cp 0xra/bethesda-translate translategemma3-st
ollama cp 0xra/bethesda-qc qcgemma4-st

---

Supported languages

All 9 official Starfield languages plus Russian and Ukrainian for xTranslator-style workflows:

Code	Language
en	English
de	German
es	Spanish
fr	French
it	Italian
ja	Japanese
pl	Polish
ptbr	Portuguese (Brazil)
zhhans	Chinese (Simplified)
ru	Russian
uk	Ukrainian

Each language pair gets its own system prompt with language-specific style rules, register guidance, and translation examples.

---

Features

Translation

• Parallel AI translation via Ollama with configurable concurrency (default 10 workers)
• Claude API backend — drop-in alternative to Ollama; select Haiku 4.5, Sonnet 4.6, or Opus 4.8 in Settings
• Language-pair prompts — dedicated system prompts for every source→target combination: register rules, script conventions (Japanese polite forms, Chinese simplified terminology), and native examples
• Translation memory — known strings are looked up before calling the model, so they are never retranslated
• Translation cache — SHA-256-keyed JSON cache (up to 50,000 entries) persisted across sessions
• Term protector — 8,000+ Starfield-specific terms (names, places, UI labels) are replaced with placeholder tokens before the AI sees the text and restored afterward, preventing mistranslation of proper nouns
• Newline / spacing restoration — when the model drops structural [[STRUCT_BREAK]] tokens, output is re-split proportionally and per-line indentation is restored from the original
• Mixed-script repair — stray Latin letters inside Cyrillic words (e.g. dослідницький → дослідницький) are corrected automatically
• Glossary system — CSV/TBX/JSON glossary with in-app editor, term suggestions dock, and automatic injection into AI prompts

File support

• Binary string files: .strings (null-terminated), .dlstrings / .ilstrings (length-prefixed)
• BA2 archives: read and write Starfield v2 BA2 files (GNRL type, zlib-compressed); picker dialog for multi-entry archives
• ESP/ESM plugins: non-localized plugins where text is stored directly in field buffers
• xTranslator SST XML: import/export in xTranslator format (match by sID, fallback to source text)
• Drag-and-drop file loading with format validation

Quality assurance

• Quality checker with 20+ checks: missing/extra game tags, empty or untranslated strings, source-language leakage, English leak, suspicious length ratios, newline mismatches, truncated AI output, AI artifact prefixes, encoding failures, script coverage (CJK), and more — for all 11 supported languages
• Hunspell spell-check — optional per-language spell checking using system dictionaries; fires SPELL_ERROR warnings on misspelled lowercase words (install dictionaries with pacman -S hunspell-uk / apt install hunspell-uk etc.)
• AI quality model (qcgemma4-st) — fine-tuned Gemma 4 E4B that detects 16 issue codes with structured VERDICT output and AUTOFIX/RETRANSLATE recommendations
• Auto-fix for mechanically correctable issues (whitespace, capitalization, character substitution, missing newlines, truncated translations)
• Retranslation queue — strings that need AI to fix are queued and retranslated with a per-string hint describing what went wrong
• Consistency checker — finds the same source string translated differently across the file, with canonical-form picker and batch replace
• Standalone fix script (scripts/apply_quality_fixes.py) — apply auto-fixes from a JSON quality report to an SST XML file without opening the GUI

UI / workflow

• Command palette (Ctrl+K) and vim-style navigation (j/k, G)
• Keyboard shortcuts editor — rebind any action
• F7 → jump to next untranslated string; Ctrl+Enter → approve; Ctrl+R → reject
• Version comparison — diff two game versions, migrate unchanged translations, export CSV/HTML reports
• Pre-translation estimator — scores each string 0–100 to predict translation difficulty before the AI runs
• Encoding detection — auto-detects UTF-8, CP1251, CP1252, CP1250 (Polish), and BOM variants; override per-file
• Themes — built-in Slate theme plus custom QSS support
• UI translations — interface available in Ukrainian ✓, German, Spanish, French, Polish, Czech (community WIP)

---

Requirements

• Python 3.10+
• Ollama running locally (or a Claude API key for the Claude backend)

pip install -r requirements.txt

Core dependencies: PySide6>=6.6, requests>=2.31, cryptography>=43.0, anthropic>=0.25

Optional: keyring>=25.0 (API key storage), hunspell>=0.5.5 or spylls>=0.1.7 (spell-check Python bindings — the hunspell CLI is used as a fallback if neither is installed)

---

Running

python main.py

Logs are written to both stdout and translator.log in the project root.

---

Local model setup

If you prefer to build the models from a local GGUF file instead of pulling from the hub:

# Edit Modelfile to set the correct FROM path, then:
ollama create translategemma3-st -f Modelfile

# Edit Modelfile.qc to set the correct FROM path, then:
ollama create qcgemma4-st -f Modelfile.qc

All generation parameters in the Modelfiles are overridden at runtime by the app (except min_p and repeat_last_n for the translation model).

---

Project structure

bethesda_strings/           Pure Python parsing library (no Qt dependency)
  core.py                   Binary parser/writer for .strings/.dlstrings/.ilstrings
  ba2_handler.py            BA2 archive reader/writer (Starfield v2, FO4 v1)
  esp_handler.py            ESP/ESM plugin parser (non-localized plugins)
  xml_handler.py            xTranslator SST XML import/export
  encoding.py               Encoding detection and conversion (UTF-8/CP1251/CP1252/CP1250/GBK/…)
  version_diff.py           Game-version diff and translation migration

gui/                        PySide6 application layer
  main_window.py            Top-level window, file I/O, translation orchestration
  ollama_worker.py          QThread worker — parallel calls, per-language prompts, post-processing
  claude_translation_worker.py  Claude API drop-in replacement for OllamaWorker
  quality_checker.py        Post-translation QA checks and auto-fix
  quality_dialog.py         QA results dialog with filtering, auto-fix, retranslation
  spell_checker.py          Hunspell spell-check wrapper (3 backends: lib / spylls / CLI)
  string_table.py           QAbstractTableModel for strings and ESP modes
  term_protector.py         Placeholder-based term protection (8000+ terms)
  translation_cache.py      SHA-256-keyed persistent translation cache
  translation_memory.py     Pre-loaded map of string ID → known-good translation
  glossary.py               Glossary data model, CSV/TBX/JSON I/O
  consistency_checker.py    Finds inconsistent translations of identical source strings
  keyboard_manager.py       Rebindable shortcuts, vim navigation, command palette
  app_settings.py           AppSettings dataclass, JSON + QSettings persistence

scripts/
  apply_quality_fixes.py    CLI: apply auto-fixes from a JSON report to SST XML
  extract_sharegpt_dataset.py  Export EN→target string pairs as ShareGPT JSONL for fine-tuning
  create_qc_dataset.py      Generate QC training dataset (14,928 examples, 16 issue codes)
  compile_translations.sh   Recompile .ts → .qm UI translation files

data/
  english_words.txt         Word list for English-leak detection
  russian_words.txt         Word list for untranslated-source detection
  ukrainian_words.txt       Word list for Ukrainian coverage checks

---

UI translation

UI translations live in gui/translations/<locale>.ts. After editing any .ts file:

./scripts/compile_translations.sh

Supported locales: uk_UA, de_DE, fr_FR, es_ES, pl_PL, cs_CZ.

---

Tests

python -m pytest tests/

---

License

MIT — see LICENSE.