Analyse & Translate Text

NLP terms — defined before code

What the Language service can do

Calling the Language service from Python

Install: pip install azure-ai-textanalytics azure-core
The endpoint format for Azure AI Language is: https://your-name.cognitiveservices.azure.com/

import os
from azure.ai.textanalytics import TextAnalyticsClient
from azure.core.credentials import AzureKeyCredential

# ── Create the client ─────────────────────────────────────────────────
# TextAnalyticsClient is the main client for Azure AI Language
# endpoint: your Language resource endpoint (cognitiveservices.azure.com)
# credential: your API key wrapped in AzureKeyCredential
client = TextAnalyticsClient(
    endpoint   = os.environ.get("LANGUAGE_ENDPOINT"),
    credential = AzureKeyCredential(os.environ.get("LANGUAGE_KEY"))
)

# ── Documents to analyse ──────────────────────────────────────────────
# The API accepts a LIST of documents — batch processing in one call
# Each document is a string. IDs are assigned automatically.
documents = [
    "The new laptop is absolutely fantastic — best purchase I've made all year!",
    "Delivery was painfully slow and the packaging was completely destroyed.",
    "The product itself is fine but the customer service is outstanding."
]


# ════════════════════════════════════════════════════════════════════
# 1. SENTIMENT ANALYSIS
# ════════════════════════════════════════════════════════════════════
print("--- Sentiment Analysis ---")

# analyze_sentiment() sends all documents at once and returns results in the same order
# show_opinion_mining=True enables aspect-based sentiment (Opinion Mining)
sentiment_results = client.analyze_sentiment(
    documents,
    show_opinion_mining=True
)

# Iterate through results — one result per document
for i, result in enumerate(sentiment_results):
    if result.is_error:   # Always check for errors
        print(f"Doc {i}: Error — {result.error}")
        continue

    # result.sentiment: "positive", "negative", "neutral", or "mixed"
    print(f"Doc {i}: {result.sentiment} "
          f"(pos={result.confidence_scores.positive:.2f}, "
          f"neg={result.confidence_scores.negative:.2f})")

    # result.sentences: per-sentence sentiment
    for sentence in result.sentences:
        print(f"  Sentence: '{sentence.text[:40]}' → {sentence.sentiment}")


# ════════════════════════════════════════════════════════════════════
# 2. NAMED ENTITY RECOGNITION
# ════════════════════════════════════════════════════════════════════
print("\n--- Named Entity Recognition ---")

ner_docs = ["Sarah Johnson from Microsoft called about the meeting on March 15th in Seattle."]

# recognize_entities() identifies named entities in text
ner_results = client.recognize_entities(ner_docs)

for result in ner_results:
    if not result.is_error:
        for entity in result.entities:
            # entity.text: the actual text found ("Sarah Johnson")
            # entity.category: the type ("Person", "Location", "DateTime", "Organization")
            # entity.confidence_score: how confident (0-1)
            print(f"  Entity: '{entity.text}' | Category: {entity.category} | "
                  f"Confidence: {entity.confidence_score:.2f}")


# ════════════════════════════════════════════════════════════════════
# 3. KEY PHRASE EXTRACTION
# ════════════════════════════════════════════════════════════════════
print("\n--- Key Phrase Extraction ---")

# extract_key_phrases() returns the main topics of each document
kp_results = client.extract_key_phrases(documents)

for i, result in enumerate(kp_results):
    if not result.is_error:
        # result.key_phrases: a list of strings
        print(f"Doc {i}: {', '.join(result.key_phrases)}")


# ════════════════════════════════════════════════════════════════════
# 4. LANGUAGE DETECTION
# ════════════════════════════════════════════════════════════════════
print("\n--- Language Detection ---")

mixed_docs = ["Hello, how are you?", "Bonjour, comment allez-vous?", "Hola, ¿cómo estás?"]

# detect_language() identifies the language of each document
lang_results = client.detect_language(mixed_docs)

for i, result in enumerate(lang_results):
    if not result.is_error:
        # result.primary_language.name: "English", "French", "Spanish"
        # result.primary_language.iso6391_name: "en", "fr", "es"
        # result.primary_language.confidence_score: 0-1
        print(f"Doc {i}: {result.primary_language.name} "
              f"({result.primary_language.iso6391_name}) — "
              f"confidence {result.primary_language.confidence_score:.2f}")


# ════════════════════════════════════════════════════════════════════
# 5. PII DETECTION & REDACTION
# ════════════════════════════════════════════════════════════════════
print("\n--- PII Detection ---")

pii_docs = ["Please contact John Smith at john.smith@email.com or call 416-555-0123."]

# recognize_pii_entities() detects and redacts PII
pii_results = client.recognize_pii_entities(pii_docs)

for result in pii_results:
    if not result.is_error:
        # result.redacted_text: original text with PII replaced by asterisks
        print(f"Redacted: {result.redacted_text}")
        for entity in result.entities:
            print(f"  PII found: '{entity.text}' | Category: {entity.category}")

Translating text between languages

Azure AI Translator is a separate service from Azure AI Language — it does translation only, but does it very well across 100+ languages. The API uses a different endpoint format and authentication pattern.

Translator capabilities

Python — Text Translation

Install: pip install requests (Translator uses REST directly — no dedicated SDK is required for simple calls)

import os, requests, json

# Azure AI Translator uses a GLOBAL endpoint — not resource-specific
# You authenticate with a subscription key in the request header
TRANSLATOR_ENDPOINT = "https://api.cognitive.microsofttranslator.com"
TRANSLATOR_KEY      = os.environ.get("TRANSLATOR_KEY")
TRANSLATOR_REGION   = os.environ.get("TRANSLATOR_REGION")  # e.g. "eastus"


# ── Text Translation ──────────────────────────────────────────────────
# Translate text to one or more target languages in a single request

# URL: base endpoint + /translate + API version
url = f"{TRANSLATOR_ENDPOINT}/translate?api-version=3.0"

# Headers contain authentication — different from other Azure AI services
# Ocp-Apim-Subscription-Key: your Translator resource key
# Ocp-Apim-Subscription-Region: the region your Translator resource is in
headers = {
    "Ocp-Apim-Subscription-Key"   : TRANSLATOR_KEY,
    "Ocp-Apim-Subscription-Region": TRANSLATOR_REGION,
    "Content-Type"                : "application/json"
}

# Body: a list of text objects to translate
# You can translate multiple strings and to multiple languages in one call
body = [
    {"text": "Hello, how are you today?"},
    {"text": "Thank you for your business."}
]

# Params: specify target languages with "to" — can pass multiple "to" values
# from_param is optional — if omitted, language is auto-detected
params = {
    "to": ["fr", "es", "de"]   # Translate to French, Spanish, and German simultaneously
}

# Make the POST request
response = requests.post(url, headers=headers, json=body, params=params)

# Parse the JSON response
translations = response.json()

# Process results — one result per input text
for i, result in enumerate(translations):
    print(f"\nOriginal ({i+1}): {body[i]['text']}")
    # result["translations"] is a list — one entry per target language
    for t in result["translations"]:
        print(f"  {t['to'].upper()}: {t['text']}")


# ── Language Detection (via Translator) ───────────────────────────────
# Use when you need to detect language AND are using the Translator service
detect_url = f"{TRANSLATOR_ENDPOINT}/detect?api-version=3.0"

detect_response = requests.post(
    detect_url,
    headers = headers,
    json    = [{"text": "Guten Morgen, wie geht es Ihnen?"}]
)
detect_result = detect_response.json()

# result[0]["language"] → ISO code e.g. "de"
# result[0]["score"] → confidence 0-1
print(f"\nDetected: {detect_result[0]['language']} "
      f"(confidence: {detect_result[0]['score']:.2f})")

Document Translation

For translating entire documents (PDFs, Word files, PowerPoint), use the Document Translation feature — an async batch operation using Azure Blob Storage.

Analyse & Translate — what the exam tests

Test your understanding