Translation và Multilingual

1from langchain_openai import ChatOpenAI
2from langchain_core.prompts import ChatPromptTemplate
3from langchain_core.output_parsers import StrOutputParser
4
5llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.3)
6
7translate_chain = (
8    ChatPromptTemplate.from_messages([
9        ("system", """Ban la dich gia chuyen nghiep.
10        Dich tu {source_lang} sang {target_lang}.
11        Giu nguyen y nghia, tone, va format.
12        Khong giai thich, chi tra ve ban dich."""),
13        ("human", "{text}")
14    ])
15    | llm
16    | StrOutputParser()
17)
18
19result = translate_chain.invoke({
20    "source_lang": "Vietnamese",
21    "target_lang": "English",
22    "text": "Hoc AI khong kho neu ban co lo trinh phu hop va kien tri."
23})
24print(result)

1from pydantic import BaseModel, Field
2from typing import List
3
4class TranslationOutput(BaseModel):
5    translation: str
6    alternatives: List[str] = Field(description="2-3 cach dich khac")
7    notes: List[str] = Field(description="Ghi chu ve context, idiomatic expressions")
8
9context_translate = (
10    ChatPromptTemplate.from_messages([
11        ("system", """Dich voi context awareness.
12        Domain: {domain}
13        Tone: {tone}
14        Cung cap: ban dich chinh, alternatives, va notes."""),
15        ("human", "Dich tu {source} sang {target}:\n{text}")
16    ])
17    | llm.with_structured_output(TranslationOutput)
18)
19
20result = context_translate.invoke({
21    "domain": "technology",
22    "tone": "professional",
23    "source": "Vietnamese",
24    "target": "English",
25    "text": "Chay mo hinh AI tren may cuc bo giup giam chi phi cloud."
26})
27
28print(f"Translation: {result.translation}")
29for alt in result.alternatives:
30    print(f"  Alt: {alt}")

1import asyncio
2
3async def translate_batch(texts, source, target):
4    tasks = [{"text": t, "source_lang": source, "target_lang": target} for t in texts]
5    results = await translate_chain.abatch(tasks, config={"max_concurrency": 10})
6    return results
7
8texts_vi = [
9    "Xin chao, toi la AI assistant.",
10    "Hom nay thoi tiet dep qua!",
11    "Ban da an com chua?",
12]
13
14translated = asyncio.run(translate_batch(texts_vi, "Vietnamese", "English"))
15for original, translated_text in zip(texts_vi, translated):
16    print(f"VI: {original}")
17    print(f"EN: {translated_text}\n")

1class LanguageInfo(BaseModel):
2    language: str
3    confidence: float
4    script: str
5    is_formal: bool
6
7detect_chain = (
8    ChatPromptTemplate.from_messages([
9        ("system", "Detect language of the text. Return language name, confidence, script type, formality."),
10        ("human", "{text}")
11    ])
12    | llm.with_structured_output(LanguageInfo)
13)
14
15# Auto-detect and translate
16def auto_translate(text, target="English"):
17    lang = detect_chain.invoke({"text": text})
18    if lang.language.lower() == target.lower():
19        return text
20    return translate_chain.invoke({
21        "text": text,
22        "source_lang": lang.language,
23        "target_lang": target
24    })

1class LocalizedContent(BaseModel):
2    text: str
3    cultural_notes: List[str]
4    adapted_references: List[str]
5
6localize_chain = (
7    ChatPromptTemplate.from_messages([
8        ("system", """Localize content cho thi truong {market}.
9        Khong chi dich ma con:
10        - Adapt cultural references
11        - Adjust tone cho local audience
12        - Convert units, formats
13        - Note van hoa dac biet"""),
14        ("human", "{content}")
15    ])
16    | llm.with_structured_output(LocalizedContent)
17)
18
19# Localize marketing content
20result = localize_chain.invoke({
21    "market": "Vietnam",
22    "content": "Our Black Friday sale starts at $99! Save big this Thanksgiving weekend."
23})

1from langchain_core.runnables import RunnableParallel
2
3# Generate content in multiple languages simultaneously
4multi_lang = RunnableParallel(
5    english=ChatPromptTemplate.from_messages([
6        ("system", "Write in English."),
7        ("human", "Write product description: {product}")
8    ]) | llm | StrOutputParser(),
9    
10    vietnamese=ChatPromptTemplate.from_messages([
11        ("system", "Viet bang tieng Viet."),
12        ("human", "Viet mo ta san pham: {product}")
13    ]) | llm | StrOutputParser(),
14    
15    japanese=ChatPromptTemplate.from_messages([
16        ("system", "Write in Japanese."),
17        ("human", "Write product description: {product}")
18    ]) | llm | StrOutputParser()
19)
20
21results = multi_lang.invoke({"product": "AI-powered smart home assistant"})

1class QualityScore(BaseModel):
2    accuracy: int = Field(ge=1, le=10)
3    fluency: int = Field(ge=1, le=10)
4    terminology: int = Field(ge=1, le=10)
5    overall: int = Field(ge=1, le=10)
6    issues: List[str]
7
8quality_chain = (
9    ChatPromptTemplate.from_messages([
10        ("system", """Danh gia chat luong ban dich.
11        So sanh original va translation.
12        Score 1-10 cho: accuracy, fluency, terminology, overall."""),
13        ("human", "Original: {original}\nTranslation: {translation}")
14    ])
15    | llm.with_structured_output(QualityScore)
16)