faniyonm/Machine-Translation-for-PSAs-from-English--Swahili--Somali

A natural language processing project that evaluates translation quality of Public Service Announcements using the NLLB model, back translation, BLEU score evaluation, and text analytics. # Evaluating Machine Translation for Public Service Announcements (PSAs) ## Project Overview Public Service Announcements (PSAs) play an important role in informing communities about safety, security, and public awareness issues. In multilingual regions, these messages must often be translated into multiple languages to ensure accessibility. This project evaluates the performance of neural machine translation models when translating PSAs into Somali. The study investigates whether multi step translation affects translation accuracy compared with direct translation. The project uses the **No Language Left Behind (NLLB) model** developed by Meta and evaluates translation quality using the **BLEU score**, a standard metric for machine translation evaluation. --- ## Objectives This project focuses on the following tasks: ### 1. Translation from Swahili to Somali Public Service Announcements are translated from Swahili into Somali using a neural machine translation model. ### 2. Back Translation The Somali translations are translated back into English to evaluate whether the original meaning of the messages is preserved. ### 3. BLEU Score Evaluation The BLEU score is calculated to quantitatively measure the quality of the round trip translation. ### 4. Direct English to Somali Translation Evaluation The original English PSAs are translated directly into Somali and a BLEU score is calculated. This allows comparison between direct translation and multi step translation to determine whether multiple translations reduce accuracy. --- ## Translation Model This project uses the **NLLB model** available through the Hugging Face Transformers library. Model used: ``` facebook/nllb-200-distilled-600M ``` This model supports translation across many languages including English, Swahili, and Somali. --- ## Project Workflow The translation pipeline follows these steps: 1. Load the PSA dataset 2. Translate the text into Somali 3. Perform back translation into Engli …