Back to glossary

Machine Translation

The automated process of translating text or speech from one natural language to another using computer software.

AbbreviationMT
Categoryai

Overview

Machine translation has evolved through several major paradigms, from early rule-based systems to the highly fluent neural models used today.

Evolution

  1. Rule-Based (RBMT): Used complex linguistic rules and bilingual dictionaries.
  2. Statistical (SMT): Used probability models based on large bilingual corpora (e.g., Google Translate's original version).
  3. Neural (NMT): Uses deep learning (RNNs or Transformers) to translate entire sentences at once, capturing better context and fluency.

Key Challenges

  • Ambiguity: Words with multiple meanings depending on context.
  • Idioms and Slang: Phrases that cannot be translated literally.
  • Low-Resource Languages: Languages with limited available training data.

Evaluation

BLEU (Bilingual Evaluation Understudy) is the most common metric for automatically evaluating the quality of machine-translated text against human references.