You are an expert subtitle creator specializing in semantic segmentation and bilingual subtitles.

## YOUR TASK:
Transform the following word-level timing data into properly segmented SRT subtitles with these requirements:

1. **Semantic Segmentation**: 
   - Group words into meaningful phrases and sentences
   - Each subtitle line should be a complete thought or phrase
   - **LENGTH RULES**: 
     * Try to use as much of the {max_chars_per_line} character limit as possible
     * ONLY break into new subtitle when exceeding {max_chars_per_line} characters
     * DO NOT break short sentences that fit within the limit
     * Example: If limit is 30, "这就构成了我们今天真正要去解开的核心谜题" (20 chars) should be ONE line, not broken
   - When breaking is necessary, break at natural pause points (commas, conjunctions, etc.)
   - Never break in the middle of a phrase or compound word like "一百八十度"

2. **Bilingual Format**:
   - If the original text is Chinese/Japanese/Korean/etc., create bilingual subtitles
   - Format: Original language on first line, English translation on second line
   - **CRITICAL**: If original is English, DO NOT provide translation field (set to null or empty string)
   - **IMPORTANT**: When translation is provided, it must ALWAYS be on ONE SINGLE LINE, never break it

3. **CRITICAL PUNCTUATION AND SPACING RULES**:
   - **ABSOLUTELY MUST REMOVE ALL QUOTATION MARKS**: Remove "" '' "" '' 「」 『』 and replace with spaces
   - REMOVE ALL Chinese punctuation marks: ，。！？；：（）【】〈〉《》
   - REMOVE ALL English punctuation at beginning and end of lines
   - NO punctuation at the start or end of any subtitle line
   - **IMPORTANT**: Quotation marks MUST be removed, not kept!
   - **Mixed Chinese-English**: Add spaces around English words in Chinese text
   - Examples of quote removal:
     * "混合政体" → 混合政体
     * "中道" → 中道  
     * 最好的政体是"混合政体" → 最好的政体是 混合政体
     * 追求"中道"，避免极端 → 追求 中道 避免极端
     * "认识世界" → 认识世界
     * "改造世界" → 改造世界
   - Other examples:
     * "因此，铭记历史，" → "因此 铭记历史"
     * "Hello, world!" → "Hello world"
     * "今天学习Python编程" → "今天学习 Python 编程"
     * "使用API接口" → "使用 API 接口"

4. **Timing Rules**:
   - Use the first word's start time as subtitle start
   - Use the last word's end time as subtitle end
   - Each subtitle should be 1-4 seconds long ideally
   - Never exceed 7 seconds for a single subtitle

5. **Output Format**:
   Return a JSON array with this structure:
   ```json
   [
     {
       "index": 1,
       "start": 0.123,
       "end": 2.456,
       "original": "原文内容没有标点",
       "translation": "Complete English translation on one line without breaks"
     },
     ...
   ]
   ```
   **IMPORTANT**: For English content, set "translation": "" (empty string) or "translation": null

## WORD-LEVEL DATA:
{words_json}

## IMPORTANT:
- Focus on natural reading flow and comprehension
- **MINIMIZE LINE BREAKS**: Only break when text exceeds {max_chars_per_line} characters
- **BAD EXAMPLE** (too many breaks):
  "这就构成了" (6 chars) → Break → "我们" (2 chars) ❌ WRONG - should be one line
- **GOOD EXAMPLE**:
  "这就构成了我们今天真正要去解开的核心谜题" (20 chars) → One subtitle ✅ CORRECT
- Never split compound words or numbers like "一百八十度", "API接口", etc.
- **VALIDATION**: Check your output - it should NOT contain any quotes "" '' "" '' 「」 『』
- MUST remove ALL punctuation marks as specified above, especially quotation marks
- **ENGLISH CONTENT RULE**: When original is English, set translation to "" or null - DO NOT duplicate English text
- When translation is provided for non-English content, it MUST NEVER be broken into multiple lines
- Return ONLY the JSON array, no explanations