translator.py

# translator.py
from gradio_client import Client
import re
import time
import config
import logging

# 設定日誌
logger = logging.getLogger(__name__)

try:
    # 💡 真正的 NLLB 運算大本營在這裡
    client = Client("https://ai-labs.ilrdf.org.tw/kari-seejiq-tnpusu-ai-hmjil/")
except Exception as e:
    logger.error(f"翻譯引擎初始化失敗: {e}")

def get_clean_value(res):
    if isinstance(res, dict) and 'value' in res: 
        return res['value']
    if isinstance(res, list) and len(res) > 0: 
        return res[0]
    return res

def is_gibberish(text):
    """防亂碼濾網：檢查是否違反基本拼寫原則"""
    text_lower = text.lower()
    if re.search(r'[fvz]', text_lower):
        return True
    if re.search(r'[^aeiouwy\s]{6,}', text_lower):
        return True
    return False

def calculate_confidence(text, is_chinese, tribe_name):
    """智慧信心值判定核心"""
    if not is_chinese and tribe_name == "太魯閣" and is_gibberish(text):
        return "low", "低 (疑為亂碼或不符拼寫規範)"
    
    if is_chinese and (len(set(text)) < 3 and len(text) > 6):
        return "low", "低 (語意不清或無效輸入)"

    complex_keywords = ["法律", "規定", "科技", "系統", "網路", "醫療", "政策", "分析", "資料", "會議", "機制", "科學", "專業", "名詞"]
    if is_chinese:
        if len(text) > 25 or any(kw in text for kw in complex_keywords):
            return "medium", "中 (含專業術語或複雜長句)"
    else:
        if len(text.split()) > 10:
            return "medium", "中 (複雜句結構)"

    return "high", "高 (一般生活簡單句)"

def basic_format(text, direction):
    """💡 使用超快速的 Python 原生語法進行基本格式化"""
    text = str(text).strip()
    if direction == "zh_to_native" and text:
        # 確保輸出的族語第一個字母大寫
        text = text[0].upper() + text[1:]
    return text

def translate(text: str, tribe_name: str):
    # 💡 確保傳入資料庫的名稱是正確的，避免 RAG 抓空
    db_tribe_name = "太魯閣語" if tribe_name == "太魯閣" else tribe_name

    if tribe_name not in config.TRIBE_MAP:
        return {"error": f"暫時不支援 {tribe_name}"}

    try:
        is_chinese = bool(re.search(r'[\u4e00-\u9fa5]', text))
        
        # 計算信心值
        conf_level, conf_desc = calculate_confidence(text, is_chinese, tribe_name)

        if conf_level == "low":
            return {
                "source": text,
                "target": "⚠️ 無法精準翻譯，請確認拼字或語意是否完整。",
                "direction": "unknown",
                "tribe": tribe_name,
                "db_tribe": db_tribe_name,
                "confidence_level": conf_level,
                "confidence_desc": conf_desc
            }

        # 💡 測速計時起點
        api_start_time = time.time()

        # 正常翻譯流程 (純 NLLB，無 Gemini 負擔)
        if is_chinese:
            back_code = get_clean_value(client.predict(ethnicity=tribe_name, api_name="/lambda_1"))
            raw_result = get_clean_value(client.predict(
                text=text, src_lang="zho_Hant", tgt_lang=back_code, api_name="/translate_1"
            ))
            direction = "zh_to_native"
            final_result = basic_format(raw_result, direction)
        else:
            go_code = get_clean_value(client.predict(ethnicity=tribe_name, api_name="/lambda"))
            raw_result = get_clean_value(client.predict(
                text=text, src_lang=go_code, tgt_lang="zho_Hant", api_name="/translate"
            ))
            direction = "native_to_zh"
            final_result = basic_format(raw_result, direction)

        # 💡 測速計時終點
        api_duration = time.time() - api_start_time
        logger.info(f"📡 [遠端 API 呼叫耗時] 字數: {len(text)} | 等待 AI Labs 伺服器回應耗時: {api_duration:.2f} 秒")

        return {
            "source": text,
            "target": final_result,
            "direction": direction,
            "tribe": tribe_name,
            "db_tribe": db_tribe_name, # 💡 提供給後續 deep_analyzer 進行 RAG 檢索使用
            "confidence_level": conf_level,
            "confidence_desc": conf_desc
        }
    except Exception as e:
        logger.error(f"翻譯過程出錯: {e}")
        return {"error": f"翻譯過程出錯: {str(e)}"}