Assignment 11: Incremental finite-state parsing

For this assignment you need to run the xfsp application, Salah Aït-Mokhtar's tool for developing incremental finite-state parsers. Please read the documentation in

http://www.cis.upenn.edu/~cis639/docs/xfsp/

Start by creating a directory that contains the following five files:

You first need to compile the xfst parser scripts, NP.xfst and PP.xfst, to create the corresponding network files NP.fst and PP.fst. These files will be created by the scripts. Note that the transducers are inverted before they are saved. The reason is that xfsp applies the transducers in the upward direction. Next, launch xfsp with the command

    /pkg/cis639/bin/xfsp english.xfsp

and try it on the following text

Although the drivers should have been checking their speedometers to verify their speed, most of them were too afraid to take their eyes off the road out of fear that something might emerge from the fog.

You should see the following result:

###########################
PP.fst applied:
###########################

[S Although/Conj-Sub NP[ the/Det-Def drivers/Nn-Pl]NP should/Aux
have/V-Pres been/V-PaPart-be checking/V-Prog NP[ their/Det-Poss
speedometers/Nn-Pl]NP to/Part-Inf verify/V-Pres NP[ their/Det-Poss
speed/Nn-Sg]NP ,/Punct-Comma most/Adj-Sup of/Prep-of them/Pron
were/V-Past-Pl-be too/Adv afraid/Adj to/Part-Inf take/V-Pres NP[
their/Det-Poss eyes/Nn-Pl]NP PP[ off/Prep NP[ the/Det-Def
road/Nn-Sg]NP]PP out/Prep of/Prep-of fear/Nn-Sg that/Conj-Sub
something/Pron might/Aux emerge/V-Pres
 PP[ from/Prep NP[ the/Det-Def fog/Nn-Sg]NP]PP
./Punct-Sent ]S

Although, xfsp always applies all the transducers that are marked for application, it only reports the name of the last transducer in the chain. In this case, both the NP and PP transducers were applied.

The Task

Start by improving the NP and PP parsers. Note that some noun phrases and prepositional phrases in the above example are not recognized correctly. Then add a recognizer for VPs. Finally, add a fourth transducer that eliminates all the part of speech tags leaving just the original words and the structure you have introduced with your parse rules.