Rework prompt

apps/learnbefore-bff/src/get-words.ts

@@ -47,47 +47,60 @@ export async function* getWords(
 }
 
 function getPrompt(translationLanguage: string) {
-  return `Please process the input text as follows:
+  return `Please analyze the input text to extract valuable vocabulary, prioritizing words in three tiers:
 
-1. First detect the source language of the text and remove:
-   - Most frequently used words in that language (approximately top 5000)
-   - Basic vocabulary (A1/A2 level) including:
-     * Common verbs (equivalents of be, do, go, etc.)
-     * Basic adjectives (equivalents of good, bad, big, small)
-     * Time expressions
-     * Basic numbers and quantities
-     * Family terms
-     * Elementary nouns
-     * Question words
-     * Pronouns
-     * Articles (if language has them)
-     * Prepositions
-     * Conjunctions
-     * Basic adverbs
-     * Auxiliary/modal verbs
-     * Common greetings
-     * Basic location words
-     * Everyday action words
+1. High-complexity words:
+   - Academic vocabulary (B1-C2 level)
+   - Technical and specialized terms
+   - Domain-specific terminology
+   - Scientific and professional jargon
+   - Abstract concepts
+   - Literary or archaic terms
 
-2. For each remaining word:
-   - Convert to dictionary form
-   - Remove duplicates
-   - Keep order
-   - Create entry with:
-     * Original word
-     * Definition
-     * Russian translation
-     * Detected language code (ISO 639-1)
+2. Medium-complexity words (B1-B2 level):
+   - Less common everyday verbs (e.g., blaze, scatter, dodge)
+   - Descriptive vocabulary (e.g., graceful, peculiar, vivid)
+   - Phrasal verbs beyond basics
+   - Nature and environment terms
+   - Emotion and behavior words
+   - Specific actions and processes
+   - Words with multiple meanings
+   - Common metaphorical usage
+   - Words that native speakers use but learners often don't know
 
-3. Format each entry as JSON:
+3. Contextually valuable words:
+   - Words crucial for understanding the text's meaning
+   - Topic-specific vocabulary
+   - Words with cultural significance
+   - Terms that often appear in media/news
+   - Words with tricky usage patterns
+   - Terms that often cause confusion for learners
+
+Processing rules:
+- Maintain original order of appearance
+- Convert to dictionary form
+- Remove duplicates while preserving first occurrence
+- Consider word frequency in general usage (roughly 3000-15000 range for medium complexity)
+- Include words that might seem simple to native speakers but are often unknown to learners
+
+Format output as JSON:
 {
   "words": [
     {
       "word": "[Original word]",
-      "meaning": "[Definition in source language]",
+      "meaning": "[Definition in text language]",
       "translation": "[${translationLanguage} translation]",
       "languageCode": "[ISO 639-1 code]"
     }
   ]
-}`
+}
+
+Additional guidelines:
+- Include words that appear in upper-intermediate textbooks
+- Keep terms that might not be extremely complex but are still challenging for learners
+- Consider including words that:
+  * Have subtle usage differences from their synonyms
+  * Are common in native speech but rare in learner vocabulary
+  * Represent concepts that might be familiar but whose specific term might not be
+  * Are frequently used in certain contexts but not necessarily in basic conversation`
 }