5 - machine translation and where are we

Definition of Machine Translation

MT is the use of computers to automate some or all of the process of translating from one language to another

What are some MT challenges

One word in one language can be translated in many words in other language, and sometimes there is no obvious/exact translation; There are syntactic divergences; Collocations are especially tricky; Bias in different languages

Example of a colocation

“I missed the flight” vs. “perdi o avião” (something like “I lost the plane”)

Example of bias

What is SMT

Statistical Machine Translation (SMT) is the name for a class of approaches that build probabilistic models of faithfulness (or fidelity or adequacy) and fluency, and then combine these models to choose the most probable translation

What is DEEP LEARNING-BASED MACHINE TRANSLATION

1. Multilingual language models are trained on data from multiple languages simultaneously, allowing them to learn representations that capture commonalities and differences across languages

2. Multilingual models alleviate the dependency on large parallel corpora by using monolingual data and leveraging the transfer of linguistic structures learned from well-resourced languages