Overview
Explore conditional generation in advanced natural language processing through this comprehensive lecture from CMU's CS 11-711 course. Delve into encoder-decoder models, conditioned generation techniques, and search algorithms. Learn about ensembling methods and evaluation metrics for NLP tasks. Discover various types of data used for conditioning in language models. Gain insights into practical applications such as machine translation, text summarization, and dialogue response generation. Examine both supervised and unsupervised evaluation approaches, including BLEU and ROUGE scores. Master the concepts of ancestral sampling, beam search, and model ensembling to enhance generation quality.
Syllabus
Intro
Conditional Language Models
Formulation
Transformation
Sampling
ancestral sampling
arg max
beam search
model ensembling
different architectures
Translation
Summarization
Dialogue Response Generation
Evaluation
Blue and Rouge
Evaluation Metrics
Unsupervised Metrics
Taught by
Graham Neubig