Academic Paper Translator

What problem it solves

Academic translation usually fails in two ways: terminology drift and “tidy-up” edits to formulas, citations, or numbers. For papers and technical docs, rendering Transformer inconsistently can distort a concept; altering \cite{...}, x_i, section numbers, or p-values can break compilation, layout, or the evidence trail. This skill targets that constrained task: it outputs a structurally matching translation without extra explanation, expansion, or editorializing.

How it works

It turns translation into a checkable process:
- Terminology consistency: scan the full text, extract terms, abbreviations, and proper names, inject the glossary into each batch, and emit a consistency checklist.
- Morphology freeze: preserve $...$, x_i, units, Fig.1, [1], algorithm numbers, and other notation; translate only natural language.
- Data fidelity: check numbers, percentages, confidence intervals, and statistics verbatim; do not “correct” source errors.
- Chunked processing: translate 20–30 paragraphs at a time with ±5 surrounding paragraphs; isolate and annotate abnormal segments instead of re-running the whole file.

Boundaries

It defaults to Simplified Chinese and fits academic papers, technical reports, LaTeX, or Markdown translation. It does not write code, call external APIs, or process non-academic content. Unpublished, sensitive, or high-stakes material still requires human review of terms and values.