fenan-yosef/Amharic-Medical-Chat

# ADAMA SCIENCE AND TECHNOLOGY UNIVERSITY ## SCHOOL OF ELECTRICAL ENGINEERING AND COMPUTING ### DEPARTMENT OF SOFTWARE ENGINEERING --- # An Explainable Amharic NLP Chatbot for Symptom Understanding **Course:** Natural Language Processing **Project Type:** NLP System Design & Implementation (Explainable Prototype) **Team Members:** * Bereket Melese * Fenan Yosef --- ## 1. Introduction Natural Language Processing systems are often evaluated not only by their performance, but also by how well their internal decisions can be explained. This is especially important in low‑resource languages such as Amharic, where large annotated datasets and pretrained domain‑specific models are limited. This project presents an **explainable hybrid NLP chatbot prototype** that understands Amharic health‑related queries at a fundamental linguistic level. The goal is not to build a full medical diagnosis system, but to demonstrate **how raw Amharic text is transformed step‑by‑step into structured meaning** using well‑established NLP principles. The system is intentionally designed to expose all intermediate representations — tokens, features, vectors, similarity scores, and final decisions — allowing every design choice to be justified and defended academically. --- ## 2. Problem Definition Amharic speakers frequently use **colloquial and metaphorical expressions** to describe physical discomfort. For example: > "ራሴን ሰንጥቆ ያመኛል" While humans easily interpret this as a *severe headache*, most NLP systems fail because: * The language is metaphorical rather than literal * Amharic is a low‑resource language * Direct translation loses clinical nuance The core NLP challenge addressed in this project is: > **How can we transform colloquial Amharic sentences into structured, machine‑understandable representations while preserving semantic and intensity information?** --- ## 3. Project Objectives ### Primary NLP Objectives * Demonstrate text normalization, feature extraction …