Gxpcode Terminology-Aware Translator

Problem It Solves

Technical document translation often fails on terminology consistency, PDF layout information, and natural Chinese phrasing. Asking an LLM to translate a long PDF directly can cause term drift, lost page context, and literal patterns such as “be done → processed.” Gxpcode-translator splits translation into executable steps: it constrains the LLM with a terminology dictionary, parses PDFs into page-level assets, translates pages, and merges results with validation.

How It Works

• Configuration and routing: first run guides setup into config.json; text and PDF use different step sequences.
• PDF parsing: paddleocr/ produces page Markdown and element JSON, followed by truncation repair.
• Terminology scanning: term_matcher.py matches CSV dictionary terms using an AC automaton and word boundaries.
• Translation and export: the LLM translates with term mappings; PDF output is merged and validated, then exported as bilingual HTML and Markdown with logs.

Boundaries

It is best for office, product, or technical documents with a clear term list. Text mode returns translated text directly, while PDF mode emphasizes file outputs; missing dictionaries or complex PDF structures still require checking intermediate artifacts and validation results.