AI · Theory

Machine Translation

Theory

62/215

Machine Translation

Machine translation maps text from one language to another. Old: phrase tables. Now: encoder–decoder / transformers. Issues: word order, gender, names, low-resource languages. Always evaluate with humans, not only BLEU.

Example — Hindi “kal” can be yesterday or tomorrow from context. A model that ignores context looks fluent and wrong.

Output — idea / do / check for Machine Translation. Fill those three in the viva.

Exam tip

Neural MT + one linguistic trap.

Example

# Machine Translation
print("idea :", "Machine Translation")
print("do   : one tiny example on paper")
print("check: one failure case")

Output: idea / do / check for Machine Translation. Fill those three in the viva.

Short notes

DefSource → target language.
RuleContext / morphology traps.

AI · Theory

Machine Translation

← All stacks

Theory

62/215

Machine Translation

Example — Hindi “kal” can be yesterday or tomorrow from context. A model that ignores context looks fluent and wrong.

Output — idea / do / check for Machine Translation. Fill those three in the viva.

Exam tip

Neural MT + one linguistic trap.

Example

# Machine Translation
print("idea :", "Machine Translation")
print("do   : one tiny example on paper")
print("check: one failure case")

Output: idea / do / check for Machine Translation. Fill those three in the viva.

Short notes

DefSource → target language.
RuleContext / morphology traps.

Machine Translation

Example

Short notes

Questions

Machine Translation

Example

Short notes

Questions