Machine Translation
PV061
Pavel Rychlý
NLP Centre, FI MU
20 Sep 2023
Pavel Rychlý ·Machine Translation ·20 Sep 2023 1 / 6
Optional Assigments
You can earn extra points for the exam.
Pavel Rychlý ·Machine Translation ·20 Sep 2023 2 / 6
Errors in translations
(1-2 points for a group)
find erros in machine translaton
choose any system (Google, DeepL, ...)
translate a document
report and describe errors
try to find shortest sentence containing that error
Pavel Rychlý ·Machine Translation ·20 Sep 2023 3 / 6
Errors in traning data
(1-2 points for a group)
look at some training data for translation
find errors
wrong alignments
wrong language
non-fluent sentences
Pavel Rychlý ·Machine Translation ·20 Sep 2023 4 / 6
Compare small optimizations
(max 15 points)
find a NMT system
train with original setup
train with a small optimizaton
different tokenizer (BPE/SentencePiece/HFT)
shared/separate embeddings
compare results
Pavel Rychlý ·Machine Translation ·20 Sep 2023 5 / 6
Replicate SotA results on own data
(max 15 points)
find a NMT implementation
run on own data (Cz-En ficet)
compare results
Pavel Rychlý ·Machine Translation ·20 Sep 2023 6 / 6