Multilingual Corpus for Online Safety and Moderation in Nepal
MuLCAM is Nepal's first open, context-aware linguistic corpus focused on Technology-Facilitated Gender-Based Violence (TFGBV). It documents how harm and abuse appear in Nepali digital spaces across languages, scripts, and cultural contexts, and makes this data openly available for research, moderation, and public-interest technology.
Online gender-based violence is rising in Nepal, and complaints have increased exponentially in recent years, with women constituting a significant proportion of survivors. According to the Cyber Bureau of Nepal Police, in fiscal year (FY) 2023/24 alone 19,730 cybercrime complaints were filed, of which 8,745 cases were related to violence against women, with cases ranging from harassment and impersonation to blackmail and non-consensual sharing of intimate images. 382 complaints involved girls and 767 involved individuals from gender and sexual minority groups.
Nepali digital communication is multilingual, contextual, and often informal. Contents on the digital spaces are on romanized Nepali, slang, coded language, grawlix, and obfuscation. Existing moderation systems which have been largely trained on global or English-language datasets, fail to recognize these patterns.
We have categorized these causes into the following key problem areas:
Many TFGBV incidents go unnoticed and unreported across languages and communities. Without context-aware, multilingual moderation systems, instances of online harm remain undetected and unaddressed. This invisibility perpetuates harm and prevents accountability.
No multilingual Nepal-based dataset of slurs and harassment language exists. Platforms cite multilingual gaps as a barrier to effective moderation. Most systems are trained on English or global datasets and fail to recognize patterns in Nepali digital communication, including Romanized Nepali, slang, dialects, and coded language.
Tech developers often lack gendered and intersectional understanding when building moderation systems. Without diverse perspectives and lived experiences informing technical design, tools fail to recognize harm in its actual context and may perpetuate bias.
Evidence unavailability is a major reason for inaction and slow investigation and justice. Without comprehensive, context-aware datasets documenting how harm appears in Nepali digital spaces, it becomes difficult to hold perpetrators accountable, demonstrate patterns to platform moderators, and build evidence for policy advocacy.
MuLCAM addresses this gap by building a locally grounded, multilingual corpus that reflects how harm actually appears in Nepali digital spaces.
We approach these problems through corpus development, tools development, and research.
An evolving, feedback-driven dataset of slurs, derogatory terms, and contextual harassment across languages.
Tools designed to detect, monitor, and flag gender-based harmful language—without stripping users of agency.
MuLCAM translates feminist values into technical and governance choices.
Safety across languages and contexts
Ensuring all communities can access and benefit from safe digital spaces across languages and cultural contexts.
Governance, accountability, and justice by design
Promoting governance, accountability, and justice by design in safety systems and moderation tools.
Open knowledge over platform extraction
Promoting open knowledge and public-interest technology over proprietary solutions and platform extraction.
Lived experiences at the center
Centering lived experiences in the design and development of safety tools and moderation systems.
Intersectional, consent-first systems
Building intersectional, consent-first systems that respect user agency and diverse identities.
MuLCAM welcomes collaboration with feminist and human rights organizations, academic institutions, digital safety initiatives, and open data and civic tech communities.
Partner with UsSupport includes using and citing the dataset, sharing the project, providing financial or institutional support, and collaborating on research and tool partnerships.
Learn MoreContributions are accepted through structured submission and review processes. All contributions are quality-checked before inclusion.
Start Contributing