Translation Guide

The Nepal Entity Service provides powerful translation capabilities between English and Nepali, supporting both Devanagari script and Romanized Nepali text.

Overview

The translation service uses AWS Bedrock (Claude Sonnet 4.5) to provide:

• English ↔ Nepali Translation: High-quality translation between languages
• Devanagari Support: Native Nepali script translation
• Romanized Nepali: Handles transliterated Nepali text (e.g., "Ma bhat khanchu")
• Auto-Detection: Automatically detects source language
• Transliteration: Provides romanized versions of Nepali text

Quick Start

Basic Translation

Translate English to Nepali:

nes translate --to ne "Ram Chandra Poudel"

Output:

Detected language: English

Translation: राम चन्द्र पौडेल
Transliteration: Ram Chandra Paudel

Translate Nepali to English:

nes translate --to en "राम चन्द्र पौडेल"

Output:

Detected language: Nepali

Translation: Ram Chandra Poudel

Romanized Nepali

The translator handles Romanized Nepali text:

nes translate --to en "Ma bhat khanchu."

Output:

Translation: I eat rice.

Configuration

Environment Variables

Configure the translation service using environment variables:

AWS Credentials

The translation service requires AWS credentials with Bedrock access. Configure credentials using one of these methods:

1. Environment Variables:
   

   export AWS_ACCESS_KEY_ID=your_access_key
   export AWS_SECRET_ACCESS_KEY=your_secret_key
   export AWS_REGION=us-east-1
   

2. AWS Credentials File (~/.aws/credentials):
   

   [default]
   aws_access_key_id = your_access_key
   aws_secret_access_key = your_secret_key
   

3. AWS Profile:
   

   export AWS_PROFILE=my-profile
   nes translate --to ne "Hello"
   

4. IAM Role (when running on AWS infrastructure)

CLI Reference

Command Syntax

nes translate [OPTIONS] [TEXT]

Options

Examples

Basic translation:

nes translate --to ne "Ram Chandra Poudel"

Specify source language:

nes translate --from en --to ne "Ram Chandra Poudel"

Use different region:

nes translate --region us-west-2 --to ne "Hello"

Read from stdin:

echo "Ram Chandra Poudel" | nes translate --to ne

Translate file content:

cat names.txt | nes translate --to ne

Use Cases

Translating Names

Translate Nepali names to English:

nes translate --to en "राम चन्द्र पौडेल"
# Output: Ram Chandra Poudel

nes translate --to en "पुष्प कमल दाहाल"
# Output: Pushpa Kamal Dahal

Translate English names to Nepali:

nes translate --to ne "Ram Chandra Poudel"
# Output: राम चन्द्र पौडेल

Translating Political Terms

nes translate --to en "राष्ट्रपति"
# Output: President

nes translate --to en "प्रधानमन्त्री"
# Output: Prime Minister

nes translate --to en "संसद"
# Output: Parliament

Batch Translation

Translate multiple items using a shell script:

#!/bin/bash
# translate_names.sh

names=(
    "राम चन्द्र पौडेल"
    "पुष्प कमल दाहाल"
    "केपी शर्मा ओली"
)

for name in "${names[@]}"; do
    echo -n "$name -> "
    nes translate --to en "$name" | grep "Translation:" | cut -d: -f2
done

Mixed Language Text

Handle text with both English and Romanized Nepali:

nes translate --to en "Ram Chandra Paudel Nepal ko rashtrapati hun."
# Output: Ram Chandra Poudel is the President of Nepal.

Language Detection

The translator automatically detects the source language based on script:

• Devanagari characters → Detected as Nepali (ne)
• Latin characters → Detected as English (en)

When auto-detection is used, the output shows the detected language:

nes translate --to en "राम चन्द्र पौडेल"

Output:

Detected language: Nepali

Translation: Ram Chandra Poudel

Transliteration

When translating from Nepali to English, the service provides transliteration (romanized version):

nes translate --to en "राम चन्द्र पौडेल"

Output:

Detected language: Nepali

Translation: Ram Chandra Poudel
Transliteration: Ram Chandra Paudel

Best Practices

1. Use Explicit Source Language for Consistency

While auto-detection works well, specify the source language for consistent results:

nes translate --from ne --to en "राम चन्द्र पौडेल"

2. Set Environment Variables for Configuration

Instead of passing options every time, set environment variables:

export AWS_REGION=us-west-2
export AWS_BEDROCK_MODEL_ID=global.anthropic.claude-sonnet-4-5-20250929-v1:0

nes translate --to ne "Hello"

3. Use Pipes for Batch Processing

Process multiple lines efficiently:

cat names.txt | while read line; do
    nes translate --to en "$line"
done

4. Handle Errors Gracefully

Check exit codes in scripts:

if nes translate --to ne "Hello" > /dev/null 2>&1; then
    echo "Translation successful"
else
    echo "Translation failed"
fi

Troubleshooting

Error: Failed to initialize translation service

Cause: AWS credentials not configured or insufficient permissions.

Solution:
1. Verify AWS credentials are set up correctly
2. Ensure the IAM user/role has Bedrock access
3. Check the AWS region supports Bedrock

Error: Unsupported model

Cause: The specified model ID is not supported.

Solution: Use the default model or check available models in your AWS region.

Empty or incorrect translations

Cause: Input text may be ambiguous or the model may need more context.

Solution:
1. Specify the source language explicitly with --from
2. Ensure the input text is clear and well-formed
3. Try breaking complex sentences into smaller parts

Slow performance

Cause: Network latency or AWS Bedrock API response time.

Solution:
1. Use a region closer to your location
2. Process translations in batches when possible
3. Consider caching frequently translated terms

Programmatic Usage

For programmatic access to translation features, see the Scraping Service Guide or the Jupyter notebook at notebooks/05_translation_service.ipynb.

Related Documentation

• Contributor Guide
• Scraping Service Guide
• Usage Examples