Challenges in Tamil Content Moderation

Summary of Content Moderation Issues for Tamil Speakers on Social Media

• Content Moderation Challenges:

  • Content moderation refers to the enforcement of platform policies, local laws, and user safety on social media.
  • It is more complex for languages other than English, leading to inequitable outcomes.

• Survey Findings:

  • Over 100 Tamil-speaking users reported inconsistent moderation, particularly chronic over-moderation.
  • Users experienced unexplained takedowns of political speech while encountering high levels of hate speech and harassment.
  • Individuals from marginalized groups (women, LGBTQ+, caste-oppressed) reported greater penalties in moderation.

• User Experiences:

  • Users voiced feelings of trust erosion towards social media platforms.
  • Tamil-speaking women reported differing treatment compared to English-speaking counterparts, highlighting cultural nuances.

• Technical Limitations:

  • Companies struggle with moderation in non-English languages due to limited digitized data for training moderation systems.
  • Indic languages, especially Dravidian languages like Tamil, are notably under-resourced.
  • Automated moderation tools are predominantly trained on English data, leading to inadequacies in Tamil moderation capabilities.
  • Tamil users often communicate in "Tamlish" (a mix of Tamil and English), complicating the moderation process.

• Company Practices:

  • Social media companies often employ a “coverage model,” prioritizing minimal resource allocation unless crises arise.
  • There is a lack of hiring for subject matter, linguistic, or regional expertise in Tamil and South Asia.

• Trust Issues:

  • Many users feel distrust due to non-disclosure of reasons for post removals, especially amid governmental pressures to delete content.

• Recommendations for Improvement:

  • Companies should develop tools that empower users to control their online experience and manage information.
  • A shift towards community-driven moderation rather than centralized, top-down approaches is recommended.
  • Establishing robust engagement with local language experts and researchers can enhance moderation effectiveness.

• Support Initiatives:

  • Groups such as Karya and Tattle aim to create high-quality datasets and user-controlled filters to combat harassment.
  • Fostering research and natural language processing capabilities through organizations like Vaani NLP and the Center for Tamil Natural Language Processing Research (CTNLPR) can be beneficial.

• Broader Implications:

  • Chronic underinvestment in language capabilities has led to users facing disproportionate content moderation experiences.
  • Addressing these issues is vital for restoring trust and fairness in online platforms, especially for non-English-speaking communities.

This overview highlights the systemic issues faced by Tamil-speaking users in content moderation on social media platforms, emphasizing the need for tailored solutions and investment in local linguistic capabilities.