Ckylark - A latent-annotated probabilistic context-free grammar (LAPCFG) parser.

This software generates the phrase structure of given input sentence using latent annotated probabilistic context-free grammar (LAPCFG) model proposed by [Petrov et al., 2006] [Petrov & Klein, 2007a] [Petrov & Klein, 2007b].

Since original LAPCFG Parser sometimes makes failed parses in parse-time, Ckylark avoids this problem using below approaches:

• Using probabilities of unknown words for parse-time smoothing.
• Rollbacking coarse grammar if parsing failed.

If you want to read more, or cite Ckylark when you use it in a paper, please reference the following paper:

• Yusuke Oda, Graham Neubig, Sakriani Sakti, Tomoki Toda, Satoshi Nakamura. Ckylark: A More Robust PCFG-LA Parser. Proceedings of NAACL: Demo Track. 2015.

Ckylark is a portmanteau of both "CKY" and "skylark."

You need following tools to build Ckylark.

• GCC 4.7 or later
• Boost 1.49 or later
• autotools

You simply run below:

cd /path/to/Ckylark
autoreconf -i
./configure
make
(sudo) make install

For simply use, you can type below command to parse your sentences:

src/bin/ckylark --model (model prefix) < (your word-segmented corpus)

--model requires the prefix of model file like data/wsj in this repository. (data/wsj is English model. if you need to parse Japanese sentences, use data/jdc instead)

For example,

$ echo "This is a pen ." | ckylark --model data/wsj
( (S (NP (DT This)) (VP (VBZ is) (NP (DT a) (NN pen))) (. .)) )

Ckylark uses the text dump files of original Berkeley Parser models.

You can also use your original models made by GrammarTrainer and WriteGrammarToTextFiles of Berkeley Parser.

And you can also use the pre-trained models listed below:

Ckylark Models (site language; Japanese)

If you want to see all options, please type below:

src/bin/ckylark --help

• Yusuke Oda (@odashi) - Most coding
• Koichi Akabe (@vbkaisetsu)
• Graham Neubig (@neubig)

We are counting more contributions from you.

Ckylark | Yusuke Oda

If you find an issue, please contact Y.Oda

• yus.takara (at) gmail.com
• @odashi_t on Twitter