Bi-directional definite-clause grammar for converting problems specified in Controlled Natural Language to Answer Set Programs, and verbalizing them back into language
This project is based on the paper "Specifying and Verbalising Answer Set Programs in Controlled Natural Language" by Rolf Schwitter. I recreated what was done in the paper but on a much smaller scale, focusing on creating a bi-directional grammar suitable for both processing and generation of program specifications in CNL.
The program big.pl (stands for Bi-directional Grammar) should be run using SWI-Prolog. To give it a test run, simply load the file using command: [big]. And then query: test.
Having done that, you will see the output of the program for a pre-defined example.
If you want to run the program on YOUR OWN INPUT SENTENCES, you have to modify the file big.pl:
- Change the variable Sentences in test (line 206).
Note that each sentence must be a list of words, where all words must be separated by a comma, and a sentence must end with the special symbol '.'. Each sentence must start with the system word 'every'/'no' or have an agent in it, e.g. like in [bob,dreams,'.'], where bob is an agent. System words are: 'every', 'no', 'is', 'is not', 'does not', '.'.
See samples.txt for samples of input sentences that you can try running.
- (optional) Broaden the lexicon.
In case you'd like to introduce new words and/or agents not included in the program, you have to add them yourself in big.pl. Current lexicon is defined in lines 175-198. For each of your custom words, add a line following the format of:
lexicon(cat:noun, wform:[student], arg:X, term:student(X)).
where cat argument is the category of the word (noun, verb, det, etc.); wform stands for word-form, i.e. the exact word that appears in a sentence; arg specifies argument of the term that the word refers to; term denotes the actual term that will be used in the internal format and in the ASP. For each of your custom agents, add a line of the format agent(name)., where name is the name of your agent, e.g. bob.
Note that currently the program only supports the following categories: noun (cat:noun), verb (cat:iv), determinant ('every', 'no'), adjective (cat:jj). All sentences must have a simple structure, no articles ("a", "an", "the"), no "and"/"or"-linked components.
Program big.pl gives out four outputs at once:
- Input sentences
- Internal representation
- ASP representation
- Generated (back) sentences
Please refer to Report.pdf for more detail about the representations.
Note that 3. ASP representation is a valid answer-set program, which you can solve using e.g. clingo, which I encourage you to try.
Created by Yevheniia Soroka, MS student, Stony Brook University, NY (CSE505: Computing with Logic, Fall 2018)
Special thanks to Rolf Schwitter for guidance on how to get started and sharing useful resources.