教程：将翻译器与 Azure AI 服务配合使用

翻译器是一项 Azure AI 服务，可用于执行语言翻译和其他与语言相关的操作。 本教程介绍如何使用翻译器在 Azure Synapse Analytics 上构建智能的多语言解决方案。

本教程将演示如何结合使用翻译器和 MMLSpark 来实现以下目的：

如果没有 Azure 订阅，请在开始前创建一个试用帐户。

• Azure Synapse Analytics 工作区，其中 Azure Data Lake Storage Gen2 存储帐户配置为默认存储。 你需要成为所使用的 Data Lake Storage Gen2 文件系统的存储 Blob 数据参与者。
• Azure Synapse Analytics 工作区中的 Spark 池。 有关详细信息，请参阅在 Azure Synapse 中创建 Spark 池。
• 在 Azure Synapse 中配置 Azure AI 服务教程中所述的预配置步骤。

打开 Synapse Studio 并创建新笔记本。 若要开始，请导入 MMLSpark。

import mmlspark
from mmlspark.cognitive import *
from notebookutils import mssparkutils
from pyspark.sql.functions import col, flatten

使用在配置前的步骤中配置的链接翻译器。

ai_service_name = "<Your linked service for translator>"

Translator 服务的核心操作是翻译文本。

df = spark.createDataFrame([
  (["Hello, what is your name?", "Bye"],)
], ["text",])

translate = (Translate()
    .setLinkedService(ai_service_name)
    .setTextCol("text")
    .setToLanguage(["zh-Hans", "fr"])
    .setOutputCol("translation")
    .setConcurrency(5))

display(translate
      .transform(df)
      .withColumn("translation", flatten(col("translation.translations")))
      .withColumn("translation", col("translation.text"))
      .select("translation"))

["你好，你叫什么名字？","Bonjour, quel est votre nom?","再见","Au revoir"]

音译是指基于拼音相似性将脚本（表音符号系统）中的单词或短语从一种语言转换为另一种语言的过程。

transliterateDf =  spark.createDataFrame([
  (["こんにちは", "さようなら"],)
], ["text",])

transliterate = (Transliterate()
    .setLinkedService(ai_service_name)
    .setLanguage("ja")
    .setFromScript("Jpan")
    .setToScript("Latn")
    .setTextCol("text")
    .setOutputCol("result"))

display(transliterate
    .transform(transliterateDf)
    .withColumn("text", col("result.text"))
    .withColumn("script", col("result.script"))
    .select("text", "script"))

如果你知道你需要翻译，但不知道将发送到 Translator 服务的文本的语言，则可以使用语言检测操作。

detectDf =  spark.createDataFrame([
  (["Hello, what is your name?"],)
], ["text",])

detect = (Detect()
    .setLinkedService(ai_service_name)
    .setTextCol("text")
    .setOutputCol("result"))

display(detect
    .transform(detectDf)
    .withColumn("language", col("result.language"))
    .select("language"))

"["en"]"

标识文本段中的句子边界的位置。

bsDf =  spark.createDataFrame([
  (["Hello, what is your name?"],)
], ["text",])

breakSentence = (BreakSentence()
    .setLinkedService(ai_service_name)
    .setTextCol("text")
    .setOutputCol("result"))

display(breakSentence
    .transform(bsDf)
    .withColumn("sentLen", flatten(col("result.sentLen")))
    .select("sentLen"))

"[25]"

使用终结点，可以获取字词或短语的替代翻译。

dictDf = spark.createDataFrame([
  (["fly"],)
], ["text",])

dictionaryLookup = (DictionaryLookup()
    .setLinkedService(ai_service_name)
    .setFromLanguage("en")
    .setToLanguage("es")
    .setTextCol("text")
    .setOutputCol("result"))

display(dictionaryLookup
    .transform(dictDf)
    .withColumn("translations", flatten(col("result.translations")))
    .withColumn("normalizedTarget", col("translations.normalizedTarget"))
    .select("normalizedTarget"))

执行字典查找后，可以将源文本和翻译传递到 dictionary/examples 终结点，以获取在句子或短语的上下文中显示这两个字词的示例列表。

dictDf = spark.createDataFrame([
  ([("fly", "volar")],)
], ["textAndTranslation",])

dictionaryExamples = (DictionaryExamples()
    .setLinkedService(ai_service_name)
    .setFromLanguage("en")
    .setToLanguage("es")
    .setTextAndTranslationCol("textAndTranslation")
    .setOutputCol("result"))

display(dictionaryExamples
    .transform(dictDf)
    .withColumn("examples", flatten(col("result.examples")))
    .select("examples"))

[{"sourcePrefix":"I mean, for a guy who could ","sourceSuffix":".","targetPrefix":"Quiero decir, para un tipo que podía ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"Now it's time to make you ","sourceSuffix":".","targetPrefix":"Ahora es hora de que te haga ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"One happy thought will make you ","sourceSuffix":".","targetPrefix":"Uno solo te hará ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"They need machines to ","sourceSuffix":".","targetPrefix":"Necesitan máquinas para ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"That should really ","sourceSuffix":".","targetPrefix":"Eso realmente debe ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"It sure takes longer when you can't ","sourceSuffix":".","targetPrefix":"Lleva más tiempo cuando no puedes ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"I have to ","sourceSuffix":" home in the morning.","targetPrefix":"Tengo que ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":" a casa por la mañana."},{"sourcePrefix":"You taught me to ","sourceSuffix":".","targetPrefix":"Me enseñaste a ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"I think you should ","sourceSuffix":" with the window closed.","targetPrefix":"Creo que debemos ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":" con la ventana cerrada."},{"sourcePrefix":"They look like they could ","sourceSuffix":".","targetPrefix":"Parece que pudieran ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"But you can ","sourceSuffix":", for instance?","targetPrefix":"Que puedes ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":", por ejemplo."},{"sourcePrefix":"At least until her kids can be able to ","sourceSuffix":".","targetPrefix":"Al menos hasta que sus hijos sean capaces de ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"I thought you could ","sourceSuffix":".","targetPrefix":"Pensé que podías ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"I was wondering what it would be like to ","sourceSuffix":".","targetPrefix":"Me preguntaba cómo sería ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."},{"sourcePrefix":"But nobody else can ","sourceSuffix":".","targetPrefix":"Pero nadie puede ","targetTerm":"volar","sourceTerm":"fly","targetSuffix":"."}]

为了确保关闭 Spark 实例，请结束任何已连接的会话（笔记本）。 达到 Apache Spark 池中指定的空闲时间时，池将会关闭。 也可以从笔记本右上角的状态栏中选择“停止会话”。

• 查看 Synapse 示例笔记本
• MMLSpark GitHub 存储库