Yared37/amharic-pos-tagging-bilstm

# Amharic Part-of-Speech Tagging Using BiLSTM This project implements a **Part-of-Speech (POS) tagging model for the Amharic language** using a **Bidirectional Long Short-Term Memory (BiLSTM)** neural network. It aims to advance **Natural Language Processing (NLP)** research for low-resource languages, particularly Amharic — one of the most widely spoken languages in Africa. --- ## Overview Part-of-Speech tagging is a fundamental task in NLP that assigns grammatical labels (e.g., noun, verb, adjective) to words in a sentence. While English and other major languages benefit from extensive datasets and pre-trained models, **Amharic lacks robust open-source POS taggers**. This project bridges that gap by training a deep learning model specifically designed for Amharic text. --- ## Model Architecture The model uses a **BiLSTM** network to capture both past and future context of words in a sentence. This bidirectional nature allows the model to understand how surrounding words influence a token’s grammatical role. **Model Pipeline:** 1. Text preprocessing and tokenization 2. Word embedding representation 3. BiLSTM layers for context understanding 4. Dense output layer with softmax activation for tag prediction **Framework:** TensorFlow **Language:** Python --- ## Dataset - **Source:** Amharic tagged corpus (mention your dataset if public, e.g. “the Amharic POS dataset from the Ethiopian NLP research group”). - **Format:** Each sentence is tokenized with corresponding POS labels. - **Preprocessing:** Text normalization, token encoding, and sequence padding. --- ## Evaluation The model’s performance is measured using: - **Accuracy** - **Precision** - **Recall** - **F1-score** Results demonstrate strong performance in capturing linguistic patterns unique to Amharic morphology and syntax. --- ## Example Output **Input Sentence:** > አባቴ ቤት ሄደ። **Predicted POS Tags:** > አባቴ → Noun > ቤት → Noun > ሄደ። → Verb