[GH-ISSUE #82] Critical RTL/Persian Support Issue - 80M+ Users Affected #61

Open
opened 2026-10-26 21:02:16 +00:00 by bsakaguchi · 0 comments
Owner

Originally created by @sanginabadim on GitHub (Dec 23, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-LLM/issues/82

Technical Issue Report: Persian RTL Text Rendering

  1. PROBLEM SUMMARY
    DeepSeek has severe Right-to-Left (RTL) text display issues for Persian/Farsi language, making the product nearly unusable for Persian-speaking users.

  2. TECHNICAL DETAILS

Core Issue:

Text displays left-aligned (LTR) instead of right-aligned (RTL)

Incorrect character ordering in mixed Persian-English text

Broken hierarchical structures (lists, tables, code blocks)

Misplaced emojis and punctuation

Test Cases Showing Failure:

"این یک متن فارسی است" → Displays left-aligned ❌

"Hello دنیا" → Shows as "Hello دنیا" instead of "دنیا Hello" ❌

Numbered lists show incorrect alignment ❌

Emojis appear at wrong positions ❌

  1. IMPACT ASSESSMENT

User Base Affected:

80+ million Persian speakers (Iran, Afghanistan, Tajikistan, diaspora)

Major Middle East market opportunity at risk

Business Consequences:

Users migrating to competitors (ChatGPT has proper RTL support)

Negative reviews spreading in Persian tech communities

Lost market share in high-growth region

  1. REPRODUCTION STEPS

Ask DeepSeek any question in Persian

Observe text alignment - starts from left instead of right

Test mixed Persian-English content

Check list and table formatting

  1. EXPECTED BEHAVIOR

Persian text should display right-aligned (RTL direction)

Mixed content should handle proper bidirectional text

Structures should maintain correct alignment

Emojis should position based on context

  1. REQUESTED ACTIONS

Immediate (1-2 weeks):

Implement direction: rtl and text-align: right for Persian text

Fix basic tokenizer for language detection

Apply CSS fixes for text alignment

Short-term (1 month):

Redesign Persian tokenizer

Retrain model on proper RTL data

Implement hierarchical structure support

Long-term (3 months):

Architectural changes for multilingual RTL support

Dedicated Persian language team

Comprehensive testing framework

  1. USER FEEDBACK EXCERPTS
    "We love DeepSeek but cannot use it with broken Persian display"
    "ChatGPT shows Persian correctly - why can't DeepSeek?"
    "80 million Persian speakers waiting for a fix"

  2. CONTACT FOR FOLLOW-UP
    [Your email/contact info here]
    Available for testing and feedback on fixes

  3. ADDITIONAL NOTES

This is a business-critical issue affecting user retention

Persian market represents significant growth opportunity

Quick resolution will prevent further user migration

Attachments:

Screenshots of RTL issues available upon request

Detailed technical analysis available

This issue requires immediate attention from your engineering team.
The Persian-speaking community is actively discussing this problem, and users are migrating to competitors daily.

Please provide:

Acknowledgement of this issue

Timeline for fixes

Point of contact for updates

Thank you for your urgent attention to this critical matter.

sanginabadim
Concerned Persian-speaking User

Originally created by @sanginabadim on GitHub (Dec 23, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-LLM/issues/82 Technical Issue Report: Persian RTL Text Rendering 1. PROBLEM SUMMARY DeepSeek has severe Right-to-Left (RTL) text display issues for Persian/Farsi language, making the product nearly unusable for Persian-speaking users. 2. TECHNICAL DETAILS Core Issue: Text displays left-aligned (LTR) instead of right-aligned (RTL) Incorrect character ordering in mixed Persian-English text Broken hierarchical structures (lists, tables, code blocks) Misplaced emojis and punctuation Test Cases Showing Failure: "این یک متن فارسی است" → Displays left-aligned ❌ "Hello دنیا" → Shows as "Hello دنیا" instead of "دنیا Hello" ❌ Numbered lists show incorrect alignment ❌ Emojis appear at wrong positions ❌ 3. IMPACT ASSESSMENT User Base Affected: 80+ million Persian speakers (Iran, Afghanistan, Tajikistan, diaspora) Major Middle East market opportunity at risk Business Consequences: Users migrating to competitors (ChatGPT has proper RTL support) Negative reviews spreading in Persian tech communities Lost market share in high-growth region 4. REPRODUCTION STEPS Ask DeepSeek any question in Persian Observe text alignment - starts from left instead of right Test mixed Persian-English content Check list and table formatting 5. EXPECTED BEHAVIOR Persian text should display right-aligned (RTL direction) Mixed content should handle proper bidirectional text Structures should maintain correct alignment Emojis should position based on context 6. REQUESTED ACTIONS Immediate (1-2 weeks): Implement direction: rtl and text-align: right for Persian text Fix basic tokenizer for language detection Apply CSS fixes for text alignment Short-term (1 month): Redesign Persian tokenizer Retrain model on proper RTL data Implement hierarchical structure support Long-term (3 months): Architectural changes for multilingual RTL support Dedicated Persian language team Comprehensive testing framework 7. USER FEEDBACK EXCERPTS "We love DeepSeek but cannot use it with broken Persian display" "ChatGPT shows Persian correctly - why can't DeepSeek?" "80 million Persian speakers waiting for a fix" 8. CONTACT FOR FOLLOW-UP [Your email/contact info here] Available for testing and feedback on fixes 9. ADDITIONAL NOTES This is a business-critical issue affecting user retention Persian market represents significant growth opportunity Quick resolution will prevent further user migration Attachments: Screenshots of RTL issues available upon request Detailed technical analysis available This issue requires immediate attention from your engineering team. The Persian-speaking community is actively discussing this problem, and users are migrating to competitors daily. Please provide: Acknowledgement of this issue Timeline for fixes Point of contact for updates Thank you for your urgent attention to this critical matter. sanginabadim Concerned Persian-speaking User
Sign in to join this conversation.
No labels
bug
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepSeek-LLM#61
No description provided.