python-zpar is a python wrapper around the ZPar parser. ZPar was written by Yue Zhang while he was at Oxford University. According to its home page: ZPar is a statistical natural language parser, which performs syntactic analysis tasks including word segmentation, part-of-speech tagging and parsing. ZPar supports multiple languages and multiple grammar formalisms. ZPar has been most heavily developed for Chinese and English, while it provides generic support for other languages. ZPar is fast, processing above 50 sentences per second using the standard Penn Teebank (Wall Street Journal) data.
I wrote python-zpar since I needed a fast and efficient parser for my NLP work which is primarily done in Python and not C++. I wanted to be able to use this parser directly from Python without having to create a bunch of files and running them through subprocesses. python-zpar not only provides a simply python wrapper but also provides an XML-RPC ZPar server to make batch-processing of large files easier.
python-zpar uses ctypes, a very cool foreign function library bundled with Python that allows calling functions in C DLLs or shared libraries directly.
IMPORTANT: As of now, python-zpar only works with the English zpar models since the interface to the Chinese models is different than the English ones. Pull requests are welcome!
Currently, python-zpar only works on 64-bit linux and OS X systems. Those are the two platforms I use everyday. I am happy to try to get python-zpar working on other platforms over time. Pull requests are welcome!
In order for python-zpar to work, it requires C functions that can be called directly. Since the only user-exposed entry point in ZPar is the command line client, I needed to write a shared library that would have functions built on top of the ZPar functionality but expose them in a way that ctypes could understand.
Therefore, in order to build python-zpar from scratch, we need to download the ZPar source, patch it with new functionality and compile the shared library. All of this happens automatically when you install with pip:
pip install python-zpar
IMPORTANT: On OS X, the installation will only work with
installed using either macports or
homebrew. The zpar source cannot be compiled with
clang. If you are having trouble compiling the code after cloning
the repository or installing the package using pip, you can try
to explicitly override the C++ compiler:
CXX=<path to c++ compiler> make -e
CXX=<path to c++ compiler> pip install python-zpar
If you are curious about what the C functions in the shared library
module look like, see
To use python-zpar, you need the English models for ZPar. They can be
downloaded from the ZPar release page here.
There are three models: a part-of-speech tagger, a constituency parser, and a
dependency parser. For the purpose of the examples below, the models are
english-models directory in the current directory.
Here's a small example of how to use python-zpar:
from six import print_ from zpar import ZPar # use the zpar wrapper as a context manager with ZPar('english-models') as z: # get the parser and the dependency parser models tagger = z.get_tagger() depparser = z.get_depparser() # tag a sentence tagged_sent = tagger.tag_sentence("I am going to the market.") print_(tagged_sent) # tag an already tokenized sentence tagged_sent = tagger.tag_sentence("Do n't you want to come with me to the market ?", tokenize=False) print_(tagged_sent) # get the dependency parse of an already tagged sentence dep_parsed_sent = depparser.dep_parse_tagged_sentence("I/PRP am/VBP going/VBG to/TO the/DT market/NN ./.") print_(dep_parsed_sent) # get the dependency parse of an already tokenized sentence dep_parsed_sent = depparser.dep_parse_sentence("Do n't you want to come with me to the market ?", tokenize=False) print_(dep_parsed_sent) # get the dependency parse of an already tokenized sentence # and include lemma information (assuming you have NLTK as well # as its WordNet corpus installed) dep_parsed_sent = depparser.dep_parse_sentence("Do n't you want to come with me to the market ?", tokenize=False, with_lemmas=True) print_(dep_parsed_sent)
The above code sample produces the following output:
I/PRP am/VBP going/VBG to/TO the/DT market/NN ./. Do/VBP n't/RB you/PRP want/VBP to/TO come/VB with/IN me/PRP to/TO the/DT market/NN ?/. I PRP 1 SUB am VBP -1 ROOT going VBG 1 VC to TO 2 VMOD the DT 5 NMOD market NN 3 PMOD . . 1 P Do VBP -1 ROOT n't RB 0 VMOD you PRP 0 SUB want VBP 0 VMOD to TO 5 VMOD come VB 3 VMOD with IN 5 VMOD me PRP 6 PMOD to TO 5 VMOD the DT 10 NMOD market NN 8 PMOD ? . 0 P Do VBP -1 ROOT do n't RB 0 VMOD n't you PRP 0 SUB you want VBP 0 VMOD want to TO 5 VMOD to come VB 3 VMOD come with IN 5 VMOD with me PRP 6 PMOD me to TO 5 VMOD to the DT 10 NMOD the market NN 8 PMOD market ? . 0 P ?
Detailed usage with comments is shown in the included file
python zpar_example.py -h to see a
list of all available options.
The package also provides an python XML-RPC implementation of a ZPar
server that makes it easier to process multiple sentences and files by
loading the models just once (via the ctypes interface) and allowing
clients to connect and request analyses. The implementation is in the
zpar_server that is installed when you install the
package. The server is quite flexible and allows loading only the
models that you need. Here's an example of how to start the server
with only the tagger and the dependency parser models loaded:
$> zpar_server --modeldir english-models --models tagger parser depparser INFO:Initializing server ... Loading tagger from english-models/tagger Loading model... done. Loading constituency parser from english-models/conparser Loading scores... done. (65.9334s) Loading dependency parser from english-models/depparser Loading scores... done. (14.9623s) INFO:Registering introspection ... INFO:Starting server on port 8859...
zpar_server -h to see a list of all options.
Once the server is running, you can connect to it using a client. An
example client is included in the file
can be run as follows (note that if you specified a custom host and port
when running the server, you'd need to specify the same here):
$> cd examples $> python zpar_client.py INFO:Attempting connection to http://localhost:8859 INFO:Tagging "Don't you want to come with me to the market?" INFO:Output: Do/VBP n't/RB you/PRP want/VBP to/TO come/VB with/IN me/PRP to/TO the/DT market/NN ?/. INFO:Tagging "Do n't you want to come to the market with me ?" INFO:Output: Do/VBP n't/RB you/PRP want/VBP to/TO come/VB to/TO the/DT market/NN with/IN me/PRP ?/. INFO:Parsing "Don't you want to come with me to the market?" INFO:Output: (SQ (VBP Do) (RB n't) (NP (PRP you)) (VP (VBP want) (S (VP (TO to) (VP (VB come) (PP (IN with) (NP (PRP me))) (PP (TO to) (NP (DT the) (NN market))))))) (. ?)) INFO:Dep Parsing "Do n't you want to come to the market with me ?" INFO:Output: Do VBP -1 ROOT n't RB 0 VMOD you PRP 0 SUB want VBP 0 VMOD to TO 5 VMOD come VB 3 VMOD to TO 5 VMOD the DT 8 NMOD market NN 6 PMOD with IN 5 VMOD me PRP 9 PMOD ? . 0 P INFO:Tagging file /Users/nmadnani/work/python-zpar/examples/test.txt into test.tag INFO:Parsing file /Users/nmadnani/work/python-zpar/examples/test_tokenized.txt into test.parse
Note that python-zpar and all of the example scripts should work with both Python 2.7 and Python 3.4. I have tested python-zpar on both Linux and Mac but not on Windows.
If you want to use ZPar in your node.js app, check out my other project node-zpar.
Although python-zpar is licensed under the MIT license - which means that you can do whatever you want with the wrapper code - ZPar itself is licensed under GPL v3.
- Improve error handling on both the python and C side.
- Expose more functionality, e.g., Chinese word segmentation, parsing etc.
- May be look into using CFFI instead of ctypes.