CCG Supertagging: 2014

This is the software used for the following publication:

Weakly-Supervised Bayesian Learning of a CCG Supertagger
Dan Garrette, Chris Dyer, Jason Baldridge, and Noah A. Smith
In Proceedings of CoNLL 2014

Data setup

Put the English, Chinese, and Italian data into the following directories:

 data/ccgbank
 data/ccgbank-chinese
 data/ccgbank-italian

The files should be arranged as follows:

$ ls data/ccgbank/AUTO
00	02	04	06	08	10	12	14	16	18	20	22	24
01	03	05	07	09	11	13	15	17	19	21	23
$ ls data/ccgbank-chinese/AUTO
00	02	04	06	08	10	20	22	24	26	28	30
01	03	05	07	09	11	21	23	25	27	29	31
$ ls data/ccgbank-italian/pro
civil_law.pro.txt	jrc_acquis.pro.txt	newspaper.pro.txt    

Running the experiments

First, compile the code and generate the run script:

$ ./compile

• $tagset was tested with the following values: ccgfeat, ctbfeat, tutfeat
• $tdcut was tested with the following values: 0.1, 0.01, 0.001, 0.0

1. Uniform

    target/start dhg.ccg.run.Conll2014Run em   $tagset  --iterations 50             --tdcut $tdcut  --trinit un  --eminit un   --ccgtrinit-comb false --ccgtrinit-catprior false
   

2. B08

    target/start dhg.ccg.run.Conll2014Run em   $tagset  --iterations 50             --tdcut $tdcut  --trinit un  --eminit un   --ccgtrinit-comb 0.95  --ccgtrinit-catprior cplx,w0.5
   

3. FFBS B08

    target/start dhg.ccg.run.Conll2014Run ffbs $tagset  --samples 200 --burnin 100  --tdcut $tdcut  --trinit un  --eminit un   --ccgtrinit-comb 0.95  --ccgtrinit-catprior cplx,w0.5
   

4. FFBS B08 + catgram

    target/start dhg.ccg.run.Conll2014Run ffbs $tagset  --samples 200 --burnin 100  --tdcut $tdcut  --trinit un  --eminit un   --ccgtrinit-comb 0.95  --ccgtrinit-catprior tdipcfg,t0.6,f0.5,m0.8,w0.5,l1000
   

5. FFBS B08 + catgram + trinit

    target/start dhg.ccg.run.Conll2014Run ffbs $tagset  --samples 200 --burnin 100  --tdcut $tdcut  --trinit tde --eminit un   --ccgtrinit-comb 0.95  --ccgtrinit-catprior tdipcfg,t0.6,f0.5,m0.8,w0.5,l1000
   

6. FFBS B08 + catgram + trinit + eminit

    target/start dhg.ccg.run.Conll2014Run ffbs $tagset  --samples 200 --burnin 100  --tdcut $tdcut  --trinit tde --eminit tde  --ccgtrinit-comb 0.95  --ccgtrinit-catprior tdipcfg,t0.6,f0.5,m0.8,w0.5,l1000