Lecture 15: Computational Lexical Semantics

Today's topic is word sense disambiguation.  This is a very "heated" topic and we'll discuss in class (not on the public blog!) why this might be!

Here are some examples of the verb "drive"... which are different senses?

1. Can you drive this four-wheel truck?
2. We drove to the university every morning.
3. She drove me to school every day.
4. He drives me mad.
5. She is driven by her passion.
6. They drove back the invadors.
7. She finally drove him to change jobs.
8. He drove a nail into the wall.
9. The player drove the ball far out into the field.
10. She is driving away at her doctoral thesis.
11. What are you driving at?
12. My new truck drives well.
13. He drives for the taxi company in DC.
14. The car drove around the corner.
15. The farmers drove the cows into the barn.
16. We drive the beltway to work.
17. He drove a golf ball across the street.
18. I hit the ball with a bat and drove it into left field.
19. The workers drove a tunnel through the hill.
20. These engines are driven by steam.
21. The hunter drove the forest looking for deer.
22. After finding deer, he drove it into open ground.

We'll talk about this example after you've clustered these examples into senses.

Two of the most important concepts in WSD are:

1. One sense per discourse (which may or may not be true...)
2. One sense per collocation (which also may or may not be true...)

The Yarowsky algorithm is an "unsupervised" algorithm (i.e., doesn't really require people to label data) for doing WSD that has inspired a ton of other work in WSD as well as other areas (including some of my own!!!).  Wikipedia has a good description of the algorithm if you don't like the one in the book; we'll talk about how and why it works in class.