-
Notifications
You must be signed in to change notification settings - Fork 1
/
readme
51 lines (30 loc) · 3.39 KB
/
readme
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
Syllabic Bayesian word segmenter
Lawrence Phillips
10/10/2014
#############################
Included are all of the files necessary to run the Bayesian word segmenter from Goldwater, Griffiths, Johnson (2010) over a syllabified corpus as documented in Phillips & Pearl (2014).
All code was created with a UNIX environment in mind. It is highly recommended that all code be run in Linux/Mac or similar environment.
#############################
CREATING DPSEG FILE
#############################
Before the model can be run, you'll first need to compile the dpseg executable.
First, acquire the Boost program_options library. Directions can be found at http://www.boost.org/doc/libs/1_56_0/more/getting_started/unix-variants.html
If you're having trouble installing the boost library, another set of useful directions can be found at http://ubuntuforums.org/showthread.php?t=1180792
You'll need to download the boost library and extract it from the archive. From the command line, move to the extracted directory and install using './bootstrap.sh --exec-prefix=/usr/local' (you can limit the number of libraries installed by using the --with-libraries flag). Then build the program_options library using the built-in function './b2 install'. This should throw the header files in /usr/local/include and the libraries in /usr/local/lib.
Then, in the 'dpseg_files' folder, compile the program 'dpseg' by using the 'make' function. If the program does not compile, check the Makefile. In particular, make sure the variable CFLAGS is pointing to the folder containing the boost header files and that the variable LIBS is pointing to the folder containing the boost libraries (.so files).
Finally, make dpseg executable using 'chmod +x dpseg' and copy the file to the parent folder.
#############################
INSTRUCTIONS
#############################
Run the file 'syllable-conversion/run-syllabification.pl' from that directory to create the appropriate syllabified corpus files
Run 'corpora_clean/CreateTrainTestSets.pl' from that directory in order to create train/test sets. Then run 'RemoveLineFinalSpaces.pl' to clean up these files somewhat.
Run the files 'run_dpseg/RunUnigram-Eng.pl' and 'run_dpseg/RunBigram-Eng.pl' to create the model output. Note that these files can take some time. In particular, the DMCMC (OnlineMem) learner was not thoroughly optimized, and in some cases may take a week in order to complete.
The output files are difficult to compare. To consolidate the data run 'analysis/ReadInData.pl' which will create a 'results.txt' file located in the folder 'results/English/'
#############################
ADDITIONAL SCRIPTS
#############################
The 'analysis' folder contains a number of scripts to accomplish different tasks:
Run_score_output.pl - Take the output of the learners and adjust F-scores to account for 'reasonable' errors. These are printed to 'results/English/adjusted_results.txt'. A list of reasonable errors found is printed in 'okayerrors/eng/'
read_okayerrors.pl - Reads the reasonable error files and prints a summary in the 'okayerrors' folder. The summary shows what each error should have been.
read_contexterrors.pl - Similar to 'read_okayerrors.pl', except prints the output of the learner rather than showing what the error should have been.
'frequency.pl' - Calculates a frequency metric to determine for each learner how frequent (on average), the words they correctly identified were.