hw4: tagging and introduction to parsing

about this assignment

We started this assignment with the in-class exercise on October 9.

part 1: finish your tagger!

Make the tagger that we started in the in-class exercise work! All the changes you have to make are in hmm_trellis.py, so that's the only file you have to turn in. Unless you make some other changes. (this part is due for the homework ping on Thursday 18 October!)

part 2: naive chart parser

Take the following grammar:

S -> NP VP           NP -> people
VP -> V NP           NP -> fish
VP -> VP PP          NP -> tanks
NP -> NP PP          NP -> rods
NP -> NP NP          V -> people
PP -> P NP           V -> fish
                     V -> tanks
                     P -> with

Implement a CKY parser that can return all parses for "people fish tanks", "people fish with rods", and "people fish tanks with rods". CKY parsing is described at the beginning of Chapter 13 in Jurafsky and Martin! There's nice pseudocode that you can use on page 440. We'll build this live in class on Thursday 18 October.

HARD MODE

More things to try!

• any HARD MODE option from the page about the tagging exercise: Add better smoothing. Or do a thorough job of evaluating the tagger. Or see how well your tagger does against somebody else's!
• Use a different programming language: Show us your lambdas. Or is this not Enterprise enough for you? Java it up!
• Build a broader-coverage grammar: Get some sentences from the wild (just a few!) and write grammar rules that allow you to parse them. How many possible parses are there, for your real-world sentence, with your grammar? (recommended.)

turning it in

Turn in your hmm_trellis.py, whatever code you write for your CKY parser, and a text file hw4.txt, describing how to run your parser, and any thoughts, questions, difficulties, or HARD MODE extensions you'd like to discuss. Use Oncourse! Due on Thursday October 25 at 1:30pm.