Explore data-driven strategies for Neural Machine Translation in this 41-minute lecture by Graham Neubig. Delve into various data augmentation techniques, including back translation, meta-back translation, and pivoting for high-resource languages. Learn about machine translation evaluation methods, such as BLEU scores, BERT Score, and COMET. Examine the challenges of high and low-resource languages, and discover approaches like transfer learning and dictionary-based augmentation. Gain insights into word alignment, word-by-word data augmentation, and reordering techniques to enhance translation quality.
Overview
Syllabus
Introduction
Machine Translation Evaluation
Manual Evaluation
Human Evaluation Shared Tasks
Blue Scores
Shortness Penalty
Bert Score
BlueRT
Comet
Bart Score
Meta Evaluation
Database Strategies
High and Low Resource Languages
Data Augmentation
Back Translation
Training Schedule
Generating Translations
In iterative back translation
Metaback translation
Metaback translation issues
High resource languages augmentation
High resource languages pivoting
Monolingual data copying
Transfer learning
Dictionarybased augmentation
Word alignment
Word by word data augmentation
Reordering
Assignment
Taught by
Graham Neubig
Related Courses
-
CMU Multilingual NLP 2020 - Data Augmentation for Machine Translation
-
CMU Multilingual NLP 2022 - Unsupervised Machine Translation
-
CMU Multilingual NLP 2022 - Translation and Translation Data
-
CMU Multilingual NLP - Unsupervised Translation
-
CMU Multilingual NLP 2020 - Multilingual Training and Cross-Lingual Transfer
-
CMU Advanced NLP: Bias and Fairness
