feat(VocabPracticeDialog): enhance vocabulary item expansion logic

- Added methods to split phrase alternatives and expand pool item alternatives, allowing for better handling of multiple valid translations and maintaining alignment between learning and reference phrases.
- Updated the normalization process to incorporate these new methods, improving the accuracy and flexibility of vocabulary practice sessions.

frontend/src/dialogues/socialnetwork/VocabPracticeDialog.vue

@@ -600,23 +600,51 @@ export default {
 
       return (leftLooksShortFragment && rightLooksSentence) || (rightLooksShortFragment && leftLooksSentence);
     },
+    splitPhraseAlternatives(value) {
+      const text = String(value || '').trim();
+      if (!text) return [];
+      const parts = text
+        .split(/\s+\/\s+/)
+        .map((part) => String(part || '').trim())
+        .filter(Boolean);
+      return parts.length >= 2 ? parts : [text];
+    },
+    expandPoolItemAlternatives(item) {
+      const learning = String(item?.learning || '').trim();
+      const reference = String(item?.reference || '').trim();
+      const learningParts = this.splitPhraseAlternatives(learning);
+      const referenceParts = this.splitPhraseAlternatives(reference);
+
+      // Common case: one prompt with multiple valid translations.
+      // Split into separate vocab cards so each answer is trainable.
+      if (learningParts.length === 1 && referenceParts.length > 1) {
+        return referenceParts.map((ref) => ({ ...item, learning, reference: ref }));
+      }
+      if (referenceParts.length === 1 && learningParts.length > 1) {
+        return learningParts.map((lrn) => ({ ...item, learning: lrn, reference }));
+      }
+
+      // If both sides contain parallel alternatives of equal length, keep pairs aligned.
+      if (learningParts.length > 1 && learningParts.length === referenceParts.length) {
+        return learningParts.map((lrn, idx) => ({ ...item, learning: lrn, reference: referenceParts[idx] }));
+      }
+
+      return [{ ...item, learning, reference }];
+    },
     normalizePool(items = []) {
       const seen = new Set();
       return (Array.isArray(items) ? items : [])
-        .map((item, index) => {
-          const learning = String(item?.learning || '').trim();
-          const reference = String(item?.reference || '').trim();
-          if (!this.isTrainablePair(learning, reference)) {
-            return null;
-          }
+        .flatMap((item, index) => this.expandPoolItemAlternatives(item).map((candidate, altIndex) => ({ candidate, index, altIndex })))
+        .map(({ candidate, index, altIndex }) => {
+          const learning = String(candidate?.learning || '').trim();
+          const reference = String(candidate?.reference || '').trim();
+          if (!this.isTrainablePair(learning, reference)) return null;
           const key = `${this.normalize(learning)}|${this.normalize(reference)}`;
-          if (seen.has(key)) {
-            return null;
-          }
+          if (seen.has(key)) return null;
           seen.add(key);
           return {
-            ...item,
-            id: item?.id || item?.itemKey || item?.key || `${key}|${index}`,
+            ...candidate,
+            id: candidate?.id || candidate?.itemKey || candidate?.key || `${key}|${index}|${altIndex}`,
             learning,
             reference
           };
@@ -766,10 +794,7 @@ export default {
 
       // Handle full-answer alternatives like "A / B" as separate valid answers.
       // Word-level slash expansion alone does not cover this reliably.
-      const phraseAlternatives = base
-        .split(/\s+\/\s+/)
-        .map((part) => String(part || '').trim())
-        .filter(Boolean);
+      const phraseAlternatives = this.splitPhraseAlternatives(base);
 
       if (phraseAlternatives.length >= 2) {
         const expanded = phraseAlternatives.flatMap((part) => this.expandSingleAnswerVariants(part));