Soumedhik/AjamiMorph

Official implementation of AjamiMorph: A multi-method consensus framework for unsupervised morphological discovery in Hausa Ajami. Includes the Hausa Ajami morpheme inventory and consensus-based segmentation pipeline. # AjamiMorph: Zero-Annotation Morphological Discovery for Hausa Ajami **AjamiMorph** is a zero-annotation framework for discovering morphemes in Hausa Ajami (Hausa written in Arabic script) through multi-method consensus. This repository contains the complete implementation of the methods presented in our EACL 2026 AbjadNLP workshop paper. ## 📄 Paper > **AjamiMorph: Zero-Annotation Morphological Discovery for Hausa Ajami via Multi-Method Consensus** > Soumedhik Bharati, Shibam Mandal, Prithwish Ghosh, Swarup Kr Ghosh, Sayani Mondal > *EACL 2026 - AbjadNLP Workshop (Oral Presentation)* > Sister Nivedita University ### Abstract Hausa Ajami (Hausa written in Arabic script) remains severely under-resourced for computational morphology. We present **AjamiMorph**, a zero-annotation framework that discovers morphemes through consensus among three unsupervised methods: Byte Pair Encoding (BPE), transition-based boundary detection using Pointwise Mutual Information (PMI), and distributional affix mining with Hausa-specific phonotactic filtering. On a Hausa Ajami Bible corpus (637,414 tokens), AjamiMorph identifies **1,611 high-confidence morphemes** achieving **99.9% coverage** with a linguistically realistic distribution (66.0% stems, 22.6% suffixes, 11.4% prefixes) and recovers 77.8% of known Hausa affixes. --- ## 🌟 Key Features - **Zero-Annotation**: No manually labeled training data required - **Multi-Method Consensus**: Combines BPE, transition PMI, and affix mining - **Linguistically Grounded**: Incorporates Hausa phonotactic constraints - **Statistically Validated**: Permutation tests and chi-square significance testing - **High Coverage**: 99.9% token coverage with compact morpheme inventory - **Extensible**: Modular architecture for adaptation to other languages --- ## 📊 Results Summary | Metric | Value | |--------|-------| | **Morpheme Inventory Size** | 1,611 morphemes | | **Token Coverage** | 99.9% | | **Type Coverage** | 99.9% | | **Known Affix R …