Machine Translation

Cross translation of human languages

Example could be translating English to French or Vice Versa

Statistical Machine Translation (SMT)

Machine translation using hand picked mappings

• Used throughout in 1990s - 2010s

• Pre-mapping of words and sentences are done by humans

  These mappings are used to translate a given word or sentence in one language to another

• The Rosetta Stone was such a medium for translation

  • The Rosetta Stone is a large stone tablet (ca. 196 BCE) that was discovered in Egypt in 1799.

    It is important because it allowed researchers to translate ancient Egyptian hieroglyphic writing for the first time.

• Challenges

  • Multiple mappings exists for different words such as:

    • One to One
    • One to Many
    • Many to One
    • Many to Many

  • Lots of Feature Engineering needed to be done in-order to produce the mappings

Neural Machine Translation (NMT)

Machine Translation using Neural Networks

• The Neural Network architecture is called sequence-to-sequence (seq2seq)
• The Encoder is given the source language and Decoder generates the translated or required language