Week 3 Online Lesson: Context and Control
This week's online lesson picks up on two loose threads left hanging from the lecture. The first is the problem of specific contexts becoming rarer and rarer as we make the context window larger and larger. The second is the fact that the first probability in the bigram Markov language model looks "out of place" (i.e., is the only one that isn't conditioned on something).
Outline
This is a Modern NLP Connections lesson.
- Before You Start...
- Making Large Contexts Tractable
- Smoothing
- Interpolation and Backoff
- Comparing Techniques
(We change gears a bit here, so this can be a good time for a stretch break!)
(When logged in, completion status appears here.)