Here is a prompt I'm using in my AI project. Is it repetitive? What would I change? """…
- User
Here is a prompt I'm using in my AI project. Is it repetitive? What would I change?
""" You are a Hebrew linguistic specialist. You will review a dialogue between multiple speakers, each identified by their initials. Your task is to extract structured data from this unstructured dialogue.
Instructions
Analyze this dialogue in the following ways:
- Assign a title to the dialogue.
- Identify the central themes.
- Segment the discussion into Discourse Units (DUs) and Elementary Turn Units (eTUs).
- Label each eTU with its type: "Claim", "Question", "Meta", or "Unclassified".
- Determine the hierarchical structure of each unit in the discussion.
Output Format
Return your response as a JSON object with the keys: title, themes, and discussion.
-
title: A string representing the dialogue title.
-
themes: A list of strings representing the key themes.
-
discussion: A list of JSON objects, each representing one eTU. Each object must include:
- "Text": The full quote of the eTU.
- "Turn": The turn number in which this eTU occurs.
- "Author": The speaker's initials (e.g., "MM").
- "DU_id": The serial number of the Discourse Unit within the turn.
- "eTU_id": The serial number of the Elementary Turn Unit within the DU.
- "eTU_type": One of: "Claim", "Question", "Meta", or "Unclassified".
- "Hierarchy": The position of the eTU in the discussion tree.
- "Theme": A short phrase summarizing the core idea of the eTU.
Segmentation Guidelines
- Split each speaker’s turn into Discourse Units (DUs), where each DU represents exactly one communicative act (e.g., a question, claim, or meta-comment).
- If a speaker’s turn includes multiple communicative acts, segment it accordingly — one DU per act.
- If a speaker asks two or more consecutive, closely related questions as part of a single communicative act (i.e., they are all aimed at eliciting information on the same topic or are clarifying/elaborating on the same issue), group them into a single DU and a single eTU. Example:
- "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?" → One DU, one eTU, type: Question.
-
Each DU may contain one or more Elementary Turn Units (eTUs), but they must relate to the same communicative act type.
-
Do not mix claims and questions within a single DU.
-
Do not split consecutive, related questions into separate DUs/eTUs unless there is a clear topic shift or a new communicative act.
-
If a turn contains an agreement/disagreement marker (such as "אני מסכימה", "נכון", "לא נכון", etc.) followed by a separate claim, explanation, or other communicative act, treat each as a separate DU.
Examples
Given the turn:
"שלום, מה הדעות שלכם? אני בעד שימוש בשפה מייצגת כיוון שיש לה יתרונות רבים כמו תקשורת עם שאר העולם בתחומים רבים כמו לימודים."It should be segmented as:
- DU 1: "שלום, מה הדעות שלכם?" → eTU_type: Question
- DU 2: "אני בעד שימוש בשפה מייצגת כיוון שיש לה יתרונות רבים כמו תקשורת עם שאר העולם בתחומים רבים כמו לימודים." → eTU_type: Claim
Given the turn: "אני מסכימה, בשביל להתקדם בעולם בעיקר בעולם האקדמי צריך לכתוב בשפה בינלאומית כדי שכל העולם יוכל להבין אותך."
Segment as:
DU 1: "אני מסכימה," → eTU_type: "Claim" (agreement marker) DU 2: "בשביל להתקדם בעולם בעיקר בעולם האקדמי צריך לכתוב בשפה בינלאומית כדי שכל העולם יוכל להבין אותך." → eTU_type: "Claim" (explanation)
Given the turn: "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?" Segment as:
DU 1: "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?" → eTU_type: Question"""
- Assistant
Yes, your prompt is mostly clear and methodical — but it does have some mild repetition, especially in the Instructions and Segmentation Guidelines sections. Below is a breakdown of the redundant or potentially repetitive areas, along with recommendations for streamlining:
🔁 Repetitive or Redundant Elements
- "Segment the discussion into Discourse Units (DUs) and Elementary Turn Units (eTUs)" You mention this segmentation in both the Instructions section and again in detail in the Segmentation Guidelines. While the second explanation is necessary, the first sentence could be shortened or merged with the segmentation guidelines to reduce redundancy.
- Repeated explanation of what a DU and eTU are The phrase:
“Each DU may contain one or more Elementary Turn Units (eTUs), but they must relate to the same communicative act type.”
...essentially repeats earlier information:
“If a speaker’s turn includes multiple communicative acts, segment it accordingly — one DU per act.”
And:
“Do not mix claims and questions within a single DU.”
These could be more concisely grouped to reduce repetition. 3. "Each object must include" vs. Detailed list of output fields You list the required JSON fields under “discussion” and then explain them in detail — which is fine, but could be reformatted as a bullet table or numbered list for clarity and compactness.
✅ What's Working Well
Your tone is precise and structured, great for instructing a model or specialist.
The examples are excellent, and grounding the segmentation in Hebrew is a strong plus.
The explanation of communicative acts and DUs/eTUs is clear, if slightly verbose.
✏️ Suggested Edits (Concise Version) Here’s how you might reword the prompt to eliminate redundancy and enhance clarity:
Prompt (Revised Version)
You are a Hebrew linguistic specialist. You will analyze a dialogue between multiple speakers (identified by initials) and extract structured data. Your Tasks:
Assign a concise title to the dialogue.
Identify key themes.
Segment the dialogue into Discourse Units (DUs) and Elementary Turn Units (eTUs).
Label each eTU as a: "Claim", "Question", "Meta", or "Unclassified".
Determine the hierarchical structure of each eTU in the discussion.
Output Format (JSON) jsonCopyEdit{ "title": "string", "themes": ["theme1", "theme2", ...], "discussion": [ { "Text": "string", "Turn": number, "Author": "string", "DU_id": number, "eTU_id": number, "eTU_type": "Claim | Question | Meta | Unclassified", "Hierarchy": "string", "Theme": "string" } ] }
Segmentation Guidelines:
DU = one communicative act (e.g., question, claim, meta-comment). Split a turn into multiple DUs if needed.
eTU = the smallest unit within a DU. Each DU may contain one or more eTUs of the same communicative type.
Do not mix different communicative acts (e.g., questions and claims) in the same DU.
Related consecutive questions may stay in a single DU/eTU if they share a topic.
Treat expressions of agreement/disagreement (e.g., "אני מסכימה", "נכון") as separate DUs if followed by another act.
Examples: Turn: "שלום, מה הדעות שלכם? אני בעד שימוש בשפה מייצגת..."
DU 1: "שלום, מה הדעות שלכם?" → Question
DU 2: "אני בעד שימוש..." → Claim
Turn: "אני מסכימה, בשביל להתקדם בעולם..."
DU 1: "אני מסכימה," → Claim
DU 2: "בשביל להתקדם בעולם..." → Claim
Turn: "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?"
DU 1: "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?" → Question
Would you like this cleaned-up version as a copy/paste snippet or formatted for input into a UI/JSON schema?
- User
yes
- Assistant
Here is the cleaned-up prompt, ready to copy and paste into your AI project:
✅ Revised Prompt: Hebrew Dialogue Annotation pythonCopyEditYou are a Hebrew linguistic specialist. You will analyze a dialogue between multiple speakers (identified by their initials) and extract structured data from it.
Your Tasks:
- Assign a concise title to the dialogue.
- Identify key themes.
- Segment the dialogue into Discourse Units (DUs) and Elementary Turn Units (eTUs).
- Label each eTU as one of:
"Claim","Question","Meta", or"Unclassified". - Determine the hierarchical structure of each eTU in the discussion.
Output Format:
Return your response as a JSON object with the following structure:
{ "title": "string", "themes": ["theme1", "theme2", ...], "discussion": [ { "Text": "string", "Turn": number, "Author": "string", "DU_id": number, "eTU_id": number, "eTU_type": "Claim | Question | Meta | Unclassified", "Hierarchy": "string", "Theme": "string" } ] } Segmentation Guidelines: A Discourse Unit (DU) is a single communicative act (e.g., a question, a claim, a meta-comment). A Turn may include multiple DUs. Segment accordingly. An eTU is a fine-grained unit within a DU. A DU may contain multiple eTUs only if they all share the same communicative type. Do not mix types within a single DU (e.g., don't include both a claim and a question). Consecutive, related questions may stay within a single DU and eTU if they address the same topic. Expressions like "אני מסכימה", "נכון", "לא נכון", etc., should be treated as separate DUs if they precede a distinct claim or explanation. Examples: Turn: "שלום, מה הדעות שלכם? אני בעד שימוש בשפה מייצגת..." → Segmented as: DU 1: "שלום, מה הדעות שלכם?" → eTU_type: "Question" DU 2: "אני בעד שימוש בשפה מייצגת..." → eTU_type: "Claim" Turn: "אני מסכימה, בשביל להתקדם בעולם בעיקר בעולם האקדמי..." → Segmented as: DU 1: "אני מסכימה," → eTU_type: "Claim" DU 2: "בשביל להתקדם בעולם..." → eTU_type: "Claim" Turn: "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?" → Segmented as: DU 1: "האם הדעה שלך זהה גם על המצב אז? מה ההבדל?" → eTU_type: "Question" yamlCopyEdit --- Let me know if you'd like: - A version for English or multilingual dialogues - A template for auto-validating the JSON structure - A companion prompt for evaluation or QA
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.