Author: Chen Zhang, with NYU Courant Institute of Mathematical Science
Email: [email protected]

Brief Description

This is the solution to assignment 4 of CSCI_GA.2590 Natural Language Processing Spring 2017, taught by Prof. Grishman. The code carries out Part-of-Speech tagging using HMM model.
This very submission got the BEST performance among all my peers.   很惭愧，只做了一点微小的贡献。

Disclaimer

If you want to use this software for academic purposes, aka assignments, please refer to the ACADEMIC INTEGRITY.

Features

• 1st-Order Markov Model and 2nd-Order Markov model are hybrided in the solution.
• Unknown words handled using
  • Hapax Legomena with open class to approximate the basic POS distribution over unknown word
  • morphological features
    • 16 catagories, with regard to Capitalizations, hyphens, digits, and symbols.
    • each word counts only once
  • suffix features
    • 2-letter suffices with adequate number of times occurance
    • each word conuts only once
• Vectorized and Matricized computation for better performance

Requiremnts

• numpy

Instrucions

Place the tagPOS_hmm.py in the parent directory of WSJ_POS_CORPUS_FOR_STUDENTS

There are 3 ways to run the tagger:

python tagPOS_hmm.py

This will take WSJ_POS_CORPUS_FOR_STUDENTS/WSJ_02-21.pos as the training file,
and let the user enter space divided tokens with sentence structure as input.

python tagPOS_hmm.py testfile

This will take *WSJ_POS_CORPUS_FOR_STUDENTS/WSJ_02-21.pos* as the training file,  
and use  *WSJ_POS_CORPUS_FOR_STUDENTS/testfile.words* as test input.  
*./testfile.pos* will be generated as output.   

python tagPOS_hmm.py testfile training-file1.pos training-file2.pos ...

This will take 

- *WSJ_POS_CORPUS_FOR_STUDENTS/training-file1.pos*, 
- *WSJ_POS_CORPUS_FOR_STUDENTS/training-file2.pos*, 
- ... 

as the training files,  
and use  

- *WSJ_POS_CORPUS_FOR_STUDENTS/testfile.words* 

as test input.  

- *./testfile.pos* 

will be generated as output.  

Reference

1. Thorsten Brants. 2000. TnT: a statistical part-of-speech tagger. In Proceedings of the sixth conference on Applied natural language processing (ANLC '00). Association for Computational Linguistics, Stroudsburg, PA, USA, 224-231. DOI=http://dx.doi.org/10.3115/974147.974178
2. Ralph Weischedel, Richard Schwartz, Jeff Palmucci, Marie Meteer, and Lance Ramshaw. 1993. Coping with ambiguity and unknown words through probabilistic models. Comput. Linguist. 19, 2 (June 1993), 361-382.
3. Ratnaparkhi, A., 1996. A maximum entropy model for part-of-speech tagging. In Proceedings of the conference on empirical methods in natural language processing (Vol. 1, pp. 133-142).