shimaokasonse / VerySimpleChunker

A neural network based chunker written in Torch7.

VerySimpleChunker

About

VerySimpleChunker is an efficient, off-the-shelf shallow parser. Trained on the CONLL-2000 shared-task, it achieved 92% accuracy. It has it’s strength in speed, thanks to high performance machine learning package Torch7 .

Requirements

• lua ( > 5.2 )
• Torch7

Downloading

$git clone https://github.com/shimaokasonse/VerySimpleChunker

$cd VerySimpleChunker

Parsing

An example input (sample.txt) would look like as follows:

This
is
a
sample
sentence
.

I
think
torch
7
is
one
of
the
best
deep
learning
frameworks
.

To parse this document, just run the following command:

$th parse.lua sample.txt > output.txt

The parsed document (output.txt) becomes:

This	B-NP	
is	B-VP	
a	B-NP	
sample	I-NP	
sentence	I-NP	
.	O	

	
I	B-NP	
think	B-VP	
torch	B-NP	
7	I-NP	
is	B-VP	
one	B-NP	
of	B-PP	
the	B-NP	
best	I-NP	
deep	I-NP	
learning	I-NP	
frameworks	I-NP	
.	O	

	

That's it !!