NLP Homework 4: Parsing

We assume that you’ve made a local copy of http://www.cs.jhu.edu/~jason/465/hw-parse/ (for example, by downloading and unpacking the zipfile there) and that you are currently in that directory.

QUESTION 3.

First try out the recognizer that we provided.

./recognize.py papa.gr papa.sen
./recognize.py -v papa.gr papa.sen    # verbose output

Copy the program recognize.py to parse.py. Then edit parse.py to transform our unweighted recognizer into your probabilistic parser. You should be able to call it as

./parse.py papa.gr papa.sen | ./prettyprint

Notice that the docstring for the Agenda class includes some sample calls and their intended output. The bottom of the script uses doctest to check that those calls behave as intended. You may want to use doctest (or unittest) to document and test your own classes and methods, too.

You are free to ignore part or all of recognize.py and just write your own solution from scratch, if you prefer. If your solution is not in Python, you should still submit a parse.py, which can just simply run your program, for example like this:

#!/usr/bin/env python3
import sys
import subprocess
subprocess.run(["java","-jar","myparser.jar"]  # invoke my parser
                + sys.argv[1:])                # on the same args

QUESTION 4.

Notice that even recognize is very slow on a large grammar, and of course parse will be at least as slow:

./recognize.py --progress wallstreet.gr wallstreet.sen
./parse.py --progress wallstreet.gr wallstreet.sen

So copy parse.py to parse2.py and make that version faster.

(The --progress option displays the fraction of columns that have been processed so far. In addition, the -v option reports the number of PREDICT, SCAN, and ATTACH actions for each sentence. You could enhance it to report more detailed statistics if you like.)