JLPT N5 Listening Point Comprehension Creator (Mondai 2 / selectText)

Source Data Location

backend/listening/data/selectText/
  001/                         <- Finished example (reference format)
    question_data.json
    audio.mp3
  clip_XXXXX/                  <- Raw source questions (5-6 per clip)
    data.json                  <- Contains questions[] array
    audio.mp3                  <- IGNORE

Output Location

backend/listening/data/selectText/
  {next_id}/                   <- Zero-padded 3 digits: 002, 003, ...
    question_data.json         <- Your output

Logic Patterns (Trap Strategies)

Based on analysis of 30 real JLPT N5 Point Comprehension questions, these are the 8 trap patterns used. Every question MUST use at least one.

Pattern 1: Fikir Degistirme (Change of Mind) — Most common (~30%)

A plan is proposed, then rejected or modified. The correct answer is always the final decision.

• Dialogue flow: Teklif A -> Itiraz/Ret -> Alternatif B -> Kabul
• Key markers: じゃなくて, でも, そうしましょう, それにします, それがいいですね
• Distractors: The initially proposed option, the rejected alternative, an unrelated option from dialogue

Pattern 2: Dikkat Dagitici Bilgi (Distractor Info) — ~20%

Multiple people's plans/situations are mentioned. The question asks about only ONE specific person.

• Dialogue flow: A says their plan, B says their plan, C says their plan
• Distractors: Other people's plans/situations

Pattern 3: Duzeltme (Correction) — ~10%

Speaker says something, then immediately corrects it.

• Dialogue flow: "X var... hayir, X yok, Y var"
• Distractors: The uncorrected (wrong) information

Pattern 4: Hesaplama (Calculation) — ~15%

Numbers from the dialogue must be combined (add/subtract/multiply) to get the answer.

• Question types: いくら, なんにん, なんまい, なんじ
• Distractors: Partial calculations, raw numbers from dialogue

Pattern 5: Zaman Cikarimi (Time Inference) — ~10%

A date/time is not stated directly but must be inferred from relative expressions.

• Key words: つぎのひ, ～まえ, おととい, ～ご, ～あと
• Distractors: The reference time, wrong calculations

Pattern 6: Olumsuz Eleme (Negative Elimination) — ~10%

Items/things that are NOT available or NOT allowed are listed. What remains is the answer.

• Key markers: ありません, だめ, ないでください, てはいけません
• Distractors: The eliminated/forbidden items

Pattern 7: Siparis/Secim (Order/Selection) — ~5%

Customer asks for options, rejects some, selects one.

• Dialogue flow: "X var mi?" -> "Yok" -> "Y olsun"
• Distractors: Rejected or unavailable items

Pattern 8: Neden/Sebep (Reason) — ~5%

Question asks why something happened. Multiple possible reasons are mentioned.

• Key word: どうして
• Distractors: Suggested but incorrect reasons

Question Data Schema — Fat JSON (question.json)

Output format follows the multi-language "Fat JSON" architecture used across the JLPT Bites ecosystem.

{
  "metadata": {
    "level": "N5",
    "topic": {
      "ja": "Japanese topic in hiragana",
      "tr": "Turkish", "en": "English", "de": "German", "fr": "French", "es": "Spanish", "ko": "Korean"
    }
  },
  "audio_url": null,
  "options": [
    "Japanese option 1",
    "Japanese option 2",
    "Japanese option 3",
    "Japanese option 4"
  ],
  "correct_option": 0,
  "transcriptions": {
    "ja": {
      "intro": "Setting + question sentence in hiragana",
      "dialogue": [
        { "speaker": "おとこのひと", "text": "Japanese dialogue line" }
      ],
      "question": "Question sentence repeated"
    },
    "tr": {
      "intro": "Turkish intro",
      "dialogue": [
        { "speaker": "Erkek", "text": "Turkish dialogue line" }
      ],
      "question": "Turkish question"
    },
    "en": { "intro": "...", "dialogue": [...], "question": "..." },
    "de": { "intro": "...", "dialogue": [...], "question": "..." },
    "fr": { "intro": "...", "dialogue": [...], "question": "..." },
    "es": { "intro": "...", "dialogue": [...], "question": "..." },
    "ko": { "intro": "...", "dialogue": [...], "question": "..." }
  },
  "analysis": {
    "vocabulary": [
      { "word": "Japanese word", "reading": "hiragana reading", "meanings": { "tr": "...", "en": "...", "de": "...", "fr": "...", "es": "...", "ko": "..." } }
    ],
    "grammar": [
      { "point": "Grammar pattern", "meanings": { "tr": "...", "en": "...", "de": "...", "fr": "...", "es": "...", "ko": "..." } }
    ]
  },
  "logic": {
    "tr": "Turkish explanation",
    "en": "English explanation",
    "de": "German explanation",
    "fr": "French explanation",
    "es": "Spanish explanation",
    "ko": "Korean explanation"
  }
}

Speaker Labels Per Language

Schema Rules

Distractor Design Rules

1. Every distractor must come from the dialogue. No random options. Each wrong answer should be a word/phrase that was actually mentioned or implied in the conversation.
2. Distractor sources by pattern:

3. Vary correct_option position. Do not always put the correct answer in the same slot. Distribute across 0, 1, 2, 3.

Workflow

Step 1 — SELECT SOURCE

• If the user specifies a clip folder + question number, use that as the source.
• If not specified, list available clips in backend/listening/data/selectText/clip_*/ and let the user choose.
• Read the source question from the clip's data.json.

Step 2 — ANALYZE SOURCE

• Identify the logic pattern (which of the 8 patterns above).
• Identify the critical entities being tested (objects, people, times, places, numbers).
• Note the dialogue structure (number of turns, speaker roles).

Step 3 — ENTITY SWAP

• Select new N5-level entities for the swap. Keep the same domain/category:
  • Food -> different food
  • Place -> different place
  • Time -> different time
  • Person/role -> same role structure
• Ensure the trap logic still works with new entities.
• The variation must be meaningfully different from the source (not just one word changed).

Step 4 — WRITE DIALOGUE

• Write a new Japanese dialogue using only N5 grammar and vocabulary.
• Preserve the same logic pattern and trap structure.
• Keep dialogue length at 4-7 turns.
• Use appropriate speaker labels based on the setting.

Step 5 — GENERATE OPTIONS

• Write exactly 4 Japanese text options.
• One option is the correct answer.
• Three options are distractors following the distractor design rules above.
• Options should be short phrases (3-8 words), parallel in structure.
• Randomize the position of the correct answer (vary correct_option).

Step 6 — WRITE ANALYSIS, LOGIC & TRANSLATIONS

• Select 3-7 key vocabulary words from the dialogue. Provide reading (hiragana) and meanings in all 6 languages (tr, en, de, fr, es, ko).
• Identify 1-3 grammar points used. Provide meanings in all 6 languages.
• Write a logic explanation in all 6 languages that:
  • States why the correct answer is correct
  • Explains why each distractor is wrong
  • References specific dialogue lines
• Translate the topic, intro, dialogue, and question into all 6 languages.
• Use the Speaker Labels Per Language table for dialogue speaker names.

Step 7 — DETERMINE OUTPUT ID

• List existing folders in backend/listening/data/selectText/ that match the pattern [0-9][0-9][0-9].
• Find the highest number.
• The new folder is {max + 1}, zero-padded to 3 digits.

Step 8 — DISPLAY & WAIT FOR APPROVAL

• Display the complete question_data.json to the user.
• Show a summary:
  • Source: clip folder + question number
  • Pattern: which logic pattern was used
  • Correct answer: option text + index
  • Distractors: brief explanation of each
• WAIT for user approval. Do NOT save until the user says "Onayla", "OK", "Tamam", "Approved", etc.

Step 9 — SAVE

• Create the output folder: backend/listening/data/selectText/{id}/
• Write question.json (Fat JSON format) to the folder.
• Confirm the file path to the user.

Settings & Mekan (Scene) Library

Use these settings for the intro line:

Question Word Library

Resources