""" Robert Zoller - Medieval Astrology Foundation Course Translation Pipeline (improved) Key improvements over the previous Jyotish-Prasana pipeline: 1. Per-page translation (natural page boundaries, ~2000 chars/page) 2. Medieval/Classical astrology terminology system prompt 3. Page header/footer cleanup before API call 4. Footnote detection and inline placement 5. Progress saved every 5 pages (JSON resume) Usage: python tools/translate_zoller.py """ import fitz import os, sys, json, time, re from pathlib import Path import requests # ---- Config ---- BASE_DIR = Path(r"D:/projects/micrograd_from_scratch") PDF_PATH = BASE_DIR / "古典占星著作" / "Robert Zoller - Medieval Astrology Foundation Course.pdf" OUTPUT_DIR = BASE_DIR / "output" PROGRESS_DIR = BASE_DIR / "progress" OUTPUT_NAME = "Zoller-Medieval-Astrology" DEEPSEEK_API_KEY = os.environ.get("DEEPSEEK_API_KEY", "") DEEPSEEK_API_URL = "https://api.deepseek.com/v1/chat/completions" OUTPUT_DIR.mkdir(parents=True, exist_ok=True) PROGRESS_DIR.mkdir(parents=True, exist_ok=True) # ---- System Prompt ---- SYSTEM_PROMPT = """你是一位中世纪/古典占星学（Medieval/Traditional Astrology）文献的专业翻译。你的任务是： 1. **翻译为中文**: 将英文授课文字稿翻译成流畅、准确的中文。注意： - 占星学术语保持专业和一致性： "sign" → "星座"、"house" → "宫位"、"planet" → "行星" "dignity" → "尊贵"、"debility" → "陷落"、"ruler" → "主宰星" "exaltation" → "擢升"、"triplicity" → "三分性"、"term" → "界"、"face/decan" → "面/十分度" "natal chart/figure" → "本命盘"、"horoscope" → "星盘" "delineation" → "解盘/解读"、"prediction" → "预测" " Arabic Parts/Lots" → "阿拉伯点/希腊点" "Hermetic" → "赫尔墨斯"、"Medieval" → "中世纪" - 保持作者 Robert Zoller 的口语授课风格（这是录音文字稿） - 列表、表格保持结构 2. **格式处理**: - 保留原文中的 [N] 引用标记 - 用 --- 分隔明显的话题转换 - 原文中的拉丁语/希腊语术语保留原文并附中文翻译 3. **页脚清理**: 忽略版权声明、URL、重复标题等页脚内容。 4. **只输出中文翻译，不加额外解释。**""" def clean_page(text: str) -> str: """Remove page headers, footers, copyright notices from extracted text.""" lines = text.split("\n") cleaned = [] # Patterns to remove skip_patterns = [ re.compile(r"^© Copyright.*All Rights Reserved$", re.IGNORECASE), re.compile(r"^http://", re.IGNORECASE), re.compile(r"^New Library Limited$", re.IGNORECASE), re.compile(r"^New Library Publication$", re.IGNORECASE), re.compile(r"^London WC1N 3XX$"), re.compile(r"^England$"), re.compile(r"^contact@new-library\.com$", re.IGNORECASE), re.compile(r"^All Rights Reserved\.?$", re.IGNORECASE), re.compile(r"^MEDIEVAL ASTROLOGY FOUNDATION COURSE$"), re.compile(r"^ORIENTATION$"), re.compile(r"^Robert Zoller\.?$"), re.compile(r"^\d+$"), # standalone page numbers (but keep [N] references) ] for line in lines: stripped = line.strip() if not stripped: cleaned.append("") continue skip = False for pat in skip_patterns: if pat.match(stripped): skip = True break if not skip: cleaned.append(stripped) # Remove consecutive blank lines (keep at most 1) result = [] prev_blank = False for line in cleaned: if line == "": if not prev_blank: result.append(line) prev_blank = True else: result.append(line) prev_blank = False return "\n".join(result).strip() def detect_footnotes(text: str) -> tuple[str, list[str]]: """Detect footnote markers and extract footnote content from end of page. Returns (body_text, footnote_list).""" lines = text.split("\n") # Look for footnote markers like [1], [2] etc. in body footnote_markers = re.findall(r"\[(\d+)\]", text) if not footnote_markers: return text, [] # Check if last few lines look like footnotes (start with number) footnotes = [] body_end = len(lines) for i in range(len(lines) - 1, max(0, len(lines) - 10), -1): line = lines[i].strip() if re.match(r"^\d+[\.\s]", line): footnotes.insert(0, line) body_end = i else: break body = "\n".join(lines[:body_end]).strip() return body, footnotes def translate_page(text: str, page_num: int, prev_summary: str = "") -> str: """Translate cleaned page text via DeepSeek API.""" body, footnotes = detect_footnotes(text) user_content = f"第 {page_num + 1} 页：\n\n{body}" if footnotes: user_content += f"\n\n本页脚注：\n" + "\n".join(footnotes) if prev_summary: user_content = f"上文摘要（第 {page_num} 页前的内容）：{prev_summary}\n\n{user_content}" headers = { "Authorization": f"Bearer {DEEPSEEK_API_KEY}", "Content-Type": "application/json", } payload = { "model": "deepseek-chat", "messages": [ {"role": "system", "content": SYSTEM_PROMPT}, {"role": "user", "content": user_content}, ], "temperature": 0.3, "max_tokens": 4096, } for attempt in range(3): try: resp = requests.post(DEEPSEEK_API_URL, headers=headers, json=payload, timeout=180) if resp.status_code == 200: return resp.json()["choices"][0]["message"]["content"] else: print(f" API error (attempt {attempt+1}): {resp.status_code} - {resp.text[:200]}") if resp.status_code == 429: time.sleep(10 * (attempt + 1)) except Exception as e: print(f" API exception (attempt {attempt+1}): {e}") time.sleep(5 * (attempt + 1)) return f"[翻译失败 - 第{page_num+1}页]\n\n{text[:1000]}" def main(): if not DEEPSEEK_API_KEY: print("ERROR: Set DEEPSEEK_API_KEY environment variable.") print(" CMD: set DEEPSEEK_API_KEY=sk-...") sys.exit(1) progress_file = PROGRESS_DIR / f"{OUTPUT_NAME}_progress.json" output_file = OUTPUT_DIR / f"{OUTPUT_NAME}.txt" # Load or init progress if progress_file.exists(): progress = json.loads(progress_file.read_text(encoding="utf-8")) print(f"Resuming: {len(progress)} pages already translated") else: progress = {} doc = fitz.open(str(PDF_PATH)) total = doc.page_count print(f"PDF: {PDF_PATH.name}") print(f"Pages: {total}") print(f"Output: {output_file}") print() # Open output for appending output_lines = [] for pg in range(total): page_key = str(pg) if page_key in progress: continue page = doc[pg] raw_text = page.get_text("text") cleaned = clean_page(raw_text) if len(cleaned) < 30: # Nearly empty page (diagram or blank) progress[page_key] = "[本页为图表或空白页]" print(f" Page {pg+1}/{total}: skipped (empty/diagram)") if pg % 5 == 0: progress_file.write_text(json.dumps(progress, ensure_ascii=False, indent=2), encoding="utf-8") continue # Get previous page summary for context prev_summary = "" if pg > 0 and str(pg - 1) in progress: prev_text = progress[str(pg - 1)] if len(prev_text) > 200: prev_summary = prev_text[:200] print(f" Page {pg+1}/{total}: translating ({len(cleaned)} chars)...", end=" ", flush=True) translated = translate_page(cleaned, pg, prev_summary) progress[page_key] = translated print(f"→ {len(translated)} chars") # Save progress every 5 pages if pg % 5 == 0 or pg == total - 1: progress_file.write_text(json.dumps(progress, ensure_ascii=False, indent=2), encoding="utf-8") time.sleep(1) # brief pause to avoid rate limiting # Rate limit time.sleep(0.5) doc.close() # Build final output print(f"\nBuilding output file...") with open(output_file, "w", encoding="utf-8") as f: f.write(f"============================================================\n") f.write(f"Robert Zoller - Medieval Astrology Foundation Course\n") f.write(f"中世纪占星学基础教程\n") f.write(f"翻译时间: {time.strftime('%Y-%m-%d %H:%M:%S')}\n") f.write(f"总页数: {total}\n") f.write(f"============================================================\n\n") for pg in range(total): page_key = str(pg) if page_key in progress: f.write(f"\n--- 第 {pg + 1} 页 ---\n\n") f.write(progress[page_key]) f.write("\n") print(f"Saved: {output_file}") print(f"Progress saved: {progress_file}") if __name__ == "__main__": main()