Portfolio Project

Description

Considering words as a sequence of letters, an encoder-decoder architecture is used in this project to convert Hebrew words to their English translations. After mapping strings to token ids and vice versa, the encoder reads words character by character and outputs code vector, and the decoder takes that code vector and produces translations character by character. Here minimal Levenshtein distance is used. It measures how many characters we need to add/remove/replace from model translation to make it perfect. Reinforcement learning is employed in this project to translate the words with self-critical sequence training.

Technology