gaozhong-english-answer-parser

Structured Answer Extraction for Gaozhong English Exam Papers

Execution Rule

Start processing the user's request directly without pre-checking whether the format is supported or asking for confirmation. If the file can't be read or parsed as expected, adapt and fix then — not before.

Fixed Question Number Distribution

This skill strictly follows the question numbering scheme below. Do not assume, adjust, or infer question ranges from the source text.

Section	Question Range	Grouping
Listening	1-20	1-5, 6-10, 11-15, 16-20
Reading A	21-23	One per question
Reading B	24-27	One per question
Reading C	28-31	One per question
Reading D	32-35	One per question
Seven-of-Five (Gap Fill)	36-40	One per question
Cloze	41-55	41-45, 46-50, 51-55
Grammar Fill-in	56-65	One per question

Processing Flow

Step 1: Read the file content

PDF files:

1. First try pdfplumber to extract text
2. If the text lacks answer information (scanned document), convert pages with pdftoppm and use visual recognition (view_image)

Image files:

1. Use view_image to inspect directly
2. If incomplete, supplement with OCR

Step 2: Extract raw answers

Locate the answer key area in the text by searching for keywords (参考答案, 答案, Answer Key, etc.). Extract only the answers:

• Multiple choice: letters only (A-G)
• Fill-in-the-blank: the word/phrase to be filled in
• Completely ignore explanatory text, regardless of how closely it is mixed with the answers

Step 3: Output in fixed structure

Output as Markdown — no code blocks, no file generation. Reply directly in the chat.

Output Template

Answer Key

Listening

1-5	6-10	11-15	16-20
BCBAC	BCCAA	BACAB	CACBB

Reading Comprehension

Passage A 21.B 22.A 23.C

Passage B 24.C 25.B 26.D 27.A

Passage C 28.C 29.B 30.A 31.C

Passage D 32.D 33.A 34.B 35.C

Seven-of-Five (Gap Fill) 36.F 37.D 38.A 39.E 40.G

Language Use

Cloze

41-45	46-50	51-55
BACDB	AACDC	BCABD

Grammar Fill-in

No.	Answer
56	to
57	a
58	playful
59	combining
60	introduction
61	which
62	surprisingly
63	to reflect
64	has created
65	hidden

Output Rules

1. Strictly follow the fixed question number distribution. Do not adjust question ranges or groupings based on the source text. If the source groups 21-35 in continuous blocks of 5 (e.g., 21-25 DDBBB), remap to the fixed groups: Passage A 21-23, Passage B 24-27, Passage C 28-31, Passage D 32-35.
2. Listening: 4 groups of 5 questions each, presented as a table row with 5 consecutive letters per cell.
3. Reading Comprehension: Passage A (3 questions), Passage B (4), Passage C (4), Passage D (4) — grouped by passage, each question on its own line.
4. Seven-of-Five: 5 questions, one per line.
5. Cloze: 3 groups of 5 questions each, presented as a table row.
6. Grammar Fill-in: 10 questions in a vertical table — first column question number, second column answer.
7. Output answers only, no explanations.
8. Output as Markdown, no code blocks, no file generation.
9. Continuation writing / composition sections are not included in the answer structure.
10. Match the user's language. The output language (section headers, explanatory labels like "Passage A", "Cloze", etc.) MUST match the language of the user's question. If the user asks in Chinese, output headers in Chinese; if in English, output headers in English. The answer content (letters, words) itself remains unchanged.