CS5720 - Week 11
Slide 212 of 220

Machine Translation with Seq2Seq

Machine Translation

Machine Translation (MT) is the automatic conversion of text from one language to another while preserving meaning, style, and context using computational methods.
Evolution of MT:

Rule-based (1950s-1980s): Grammar rules + dictionaries
Statistical (1990s-2000s): Learn from parallel texts
Neural (2014+): End-to-end deep learning
Transformer-based (2017+): Current state-of-the-art
🌍 Impact:
Modern neural MT has made cross-language communication accessible to billions, breaking down language barriers in real-time!

Seq2Seq Architecture

Sequence-to-Sequence Model
Encoder
Input Processing
Context Vector
Fixed Representation
Decoder
Output Generation
Key Components:

Encoder: Processes source sentence into fixed representation
Context Vector: Compressed encoding of input meaning
Decoder: Generates target sentence word by word
Attention: Focuses on relevant input parts (later improvement)

Interactive Translation Demo

Source Language
Target Language
Translation Examples
BLEU Score
Measures n-gram overlap with reference translations
METEOR
Considers synonyms and word order
Human Evaluation
Fluency, adequacy, and overall quality
Prepared by Dr. Gorkem Kar