Kseniia

contest/README.md

@@ -25,7 +25,7 @@ pip install -r requirements.txt
 
 Run extraction:
 ```bash
-python extract_claims.py
+python extract.py
 ```
 
 Extraction enriches each object with a list of claims. Each claim has the following structure:

src/reclaim/extract_claims.py

@@ -152,11 +152,6 @@ def claims_from_text(self, text: str, tokens: List[int], tokenizer) -> List[Clai
             List of `Claim` objects with token-level provenance.
         """
         start_time = time.time()
-        stupid_claims = [
-            "Mary is a five-year old girl.",
-            "Mary likes playing piano.",
-            "Mary doesn't like cookies.",
-        ]
         # Guardrail cleanup: drop known garbage claims sometimes returned by the prompt.
 
         sent_list = []
@@ -165,10 +160,6 @@ def claims_from_text(self, text: str, tokens: List[int], tokenizer) -> List[Clai
         uniq_sentences: List[str] = []
         claim_list = doc2sentences(doc=text, mode="claims", schema=ClaimModel).claims
 
-        for s in stupid_claims:
-            if s in claim_list:
-                claim_list.remove(s)
-
         print("claim_list:", claim_list)
 
         final_claims: List[Claim] = []

src/reclaim/prompts.py

@@ -37,37 +37,89 @@
 
 
 SENTENCES_TO_CLAIMS_PROMPT = """
-Your task is to decompose the text into atomic claims.
-Claims should be a context-independent, fully atomic, representing one fact. Atomic claims are simple, indivisible facts that do not bundle multiple pieces of information together.
+Decompose the following text into atomic claims.
 
-### Guidelines for Decomposition:
-1. **Atomicity**: Break down each statement into the smallest possible unit of factual information. Avoid grouping multiple facts in one claim. For example:
-   - Instead of: "Photosynthesis in plants converts sunlight, carbon dioxide, and water into glucose and oxygen."
-   - Output: ["Photosynthesis in plants converts sunlight into glucose.", "Photosynthesis in plants converts carbon dioxide into glucose.", "Photosynthesis in plants converts water into glucose.", "Photosynthesis in plants produces oxygen."]
+## What is an Atomic Claim?
+An atomic claim is a single, verifiable statement of fact that:
+- Contains exactly ONE piece of information that can be true or false
+- Is self-contained (understandable without the original text)
+- Preserves the original meaning accurately
 
-   - Instead of: "The heart pumps blood through the body and regulates oxygen supply to tissues."
-   - Output: ["The heart pumps blood through the body.", "The heart regulates oxygen supply to tissues."]
+## Decomposition Rules
 
-   - Instead of: "Gravity causes objects to fall to the ground and keeps planets in orbit around the sun."
-   - Output: ["Gravity causes objects to fall to the ground.", "Gravity keeps planets in orbit around the sun."]
+### 1. Split Conjunctions
+Separate statements joined by "and", "but", "or", commas, or semicolons.
 
-2. **Context-Independent**: Each claim must be understandable and verifiable on its own without requiring additional context or references to other claims. Avoid vague claims like "This process is important for life."
+Input: "The heart pumps blood and regulates oxygen supply."
+Output: ["The heart pumps blood.", "The heart regulates oxygen supply."]
 
-3. **Precise and Unambiguous**: Ensure the claims are specific and avoid combining related ideas that can stand independently.
+### 2. Resolve Pronouns and References
+Replace pronouns (he, she, it, this, that) with their referents.
 
-4. **No Formatting**: The response must be a Python list of strings without any extra formatting, code blocks, or labels like "python".
+Input: "Marie Curie was a physicist. She won two Nobel Prizes."
+Output: ["Marie Curie was a physicist.", "Marie Curie won two Nobel Prizes."]
 
-### Example:
-If the input text is:
-"Mary is a five-year-old girl. She likes playing piano and doesn't like cookies."
-Extracted claims should be:
-"Mary is a five-year-old girl.", "Mary likes playing piano.", "Mary doesn't like cookies."
+### 3. Preserve Relationships, Don't Invent New Ones
+Keep the original semantic relationships. Don't split processes that are inherently unified.
+
+Input: "Photosynthesis converts carbon dioxide and water into glucose using sunlight."
+Output: [
+    "Photosynthesis uses sunlight.",
+    "Photosynthesis uses carbon dioxide.",
+    "Photosynthesis uses water.",
+    "Photosynthesis produces glucose."
+]
+NOT: "Photosynthesis converts sunlight into glucose." (factually wrong)
+
+### 4. Handle Lists Properly
+Each item in a list becomes a separate claim with the full context.
+
+Input: "Python supports integers, floats, and strings."
+Output: [
+    "Python supports integers.",
+    "Python supports floats.",
+    "Python supports strings."
+]
+
+### 5. Keep Comparisons Intact
+Don't split comparative statements.
+
+Input: "Jupiter is larger than Earth."
+Output: ["Jupiter is larger than Earth."]
+NOT: ["Jupiter is large.", "Earth is small."] (loses the comparison)
+
+### 6. Preserve Negations Accurately
+Keep negations with the correct scope.
+
+Input: "The test did not detect any errors."
+Output: ["The test did not detect any errors."]
+NOT: ["The test detected errors."] (inverted meaning)
 
-### Now, decompose the following text into atomic claims:
+### 7. Handle Conditionals as Single Claims
+Keep "if-then" relationships together.
+
+Input: "If the temperature exceeds 100°C, water boils."
+Output: ["Water boils if the temperature exceeds 100°C."]
+
+### 8. Skip Meta-Commentary
+Exclude statements about the text itself or the writing process.
+
+Input: "As mentioned earlier, the algorithm runs in O(n) time."
+Output: ["The algorithm runs in O(n) time."]
+(Skip "As mentioned earlier")
+
+## Output Format
+Return ONLY a valid Python list of strings. No explanations, markdown, or code blocks.
+
+Example:
+["First claim.", "Second claim.", "Third claim."]
+
+## Text to Decompose
 {doc}
 """
 
 
+
 DOC_TO_INDEPEDENT_SENTENCES_PROMPT = """
 Your task is to perform sentence segmentation and de-contextualization. 
 Let's define a function named process(input:str).