Add name translation fix for OPUS-MT - correct 'sand geese' to 'Zhang Shaian'

- Add _fix_name_translations() method to post-process OPUS-MT translations
- Fix incorrect translation of 章沙雁 (Zhang Shaian) as 'sand geese'
- Handle 'elders of the sand geese' -> 'Zhang Shaian Elder'
- Preserve 'Elder' title when original text contains 长老
- Apply fix automatically after OPUS-MT translation

document_processing_agent.py

@@ -548,6 +548,69 @@ class DocumentProcessingAgent:
         
         return self._qwen_model, self._qwen_tokenizer, self._qwen_device
     
+    def _fix_name_translations(self, translation: str, original_text: str) -> str:
+        """Fix known name translation errors in OPUS-MT output.
+        
+        OPUS-MT sometimes incorrectly translates Chinese names. This function
+        checks for known incorrect translations and replaces them with correct ones.
+        """
+        if not translation:
+            return translation
+        
+        # Check if original text contains the Chinese name 章沙雁
+        if "章沙雁" not in original_text:
+            return translation  # No need to fix if name not in original
+        
+        import re
+        
+        # Fix "章沙雁" (Zhang Shaian) mis-translations
+        # OPUS-MT translates 章沙雁 as "sand geese" (沙雁 = sand geese)
+        corrected = translation
+        
+        # Pattern 1: "elders of the sand geese" -> "Zhang Shaian Elder"
+        # This handles: "We have the ceremonial ceremony of the elders of the sand geese here"
+        if "长老" in original_text:
+            # Replace "elders of the sand geese" with "Zhang Shaian Elder"
+            corrected = re.sub(
+                r'\b(?:the\s+)?elders\s+of\s+the\s+sand\s+geese\b',
+                'Zhang Shaian Elder',
+                corrected,
+                flags=re.IGNORECASE
+            )
+            # Replace "sand geese elder" with "Zhang Shaian Elder"
+            corrected = re.sub(
+                r'\b(?:the\s+)?sand\s+geese\s+elder\b',
+                'Zhang Shaian Elder',
+                corrected,
+                flags=re.IGNORECASE
+            )
+            # Replace remaining "sand geese" with "Zhang Shaian" (if 长老 is present, add Elder)
+            corrected = re.sub(
+                r'\bsand\s+geese\b',
+                'Zhang Shaian',
+                corrected,
+                flags=re.IGNORECASE
+            )
+            # If we have "Zhang Shaian" but original had 长老, make sure we have "Zhang Shaian Elder"
+            if "Zhang Shaian" in corrected and "Zhang Shaian Elder" not in corrected:
+                # Only add Elder if it's in a context where it makes sense (not in the middle of a sentence)
+                corrected = re.sub(
+                    r'\bZhang\s+Shaian\b(?!\s+Elder)',
+                    'Zhang Shaian Elder',
+                    corrected,
+                    count=1  # Only replace first occurrence to avoid over-correction
+                )
+        else:
+            # If no 长老, just replace "sand geese" with "Zhang Shaian"
+            corrected = re.sub(
+                r'\bsand\s+geese\b',
+                'Zhang Shaian',
+                corrected,
+                flags=re.IGNORECASE
+            )
+        
+        return corrected
+    
     def _validate_translation_quality(self, translation: str, original: str) -> bool:
         """Validate translation quality. Returns True if translation is acceptable."""
         if not translation or len(translation.strip()) < 2:
@@ -998,6 +1061,9 @@ English:"""
                 # Decode result
                 opus_result = tokenizer.decode(translated[0], skip_special_tokens=True)
                 
+                # Fix known name translation errors in OPUS-MT output
+                opus_result = self._fix_name_translations(opus_result, text)
+                
                 # HYBRID DECISION: Choose best result
                 # Prefer Qwen if available and valid, otherwise use OPUS-MT
                 if qwen_result and self._validate_translation_quality(qwen_result, text):