About the Chinese Word Demo

Chinese is written without spaces between the words. This demo illustrates how to break Chinese into words.

Annotation Standard

Because the notion of word is not a natural one in Chinese, different "standards" have been developed for determining wordhood. This demo is based on the Academia Sinica corpus as distributed as part of SIGHAN 2005.

LingPipe's Word Segmentation

LingPipe's word segmentation is performed using a noisy channel model, where the source is a character language model trained on segmented text and the channel simply allows spaces to be freely inserted.

Model Quality

The demo uses a 4-gram character language model for the source of the noisy channel. This model is roughly 30MB in compiled form. 5-grams perform significantly better (about 25% fewer errors in the SIGHAN evaluation), but the model is 80MB.

Sentences

The training data provided a single sentence per line, so this segmenter assumes that sentences are already marked. LingPipe does not yet have a Chinese sentence detector, so this is left to the user. Performance should be OK even without sentences marked.

Chinese Word XML Markup

This demo only provides first-best output, with words wrapped in token elements.

Chinese Word Demo on the Web

The demo is hosted on the web at the following URL:

Academia Sinica Corpus

http://lingpipe-demos.com:8080/lingpipe-demos/word_zh_as/textInput.html

For detailed information about using web demos, including web form, file upload and web service instructions, see the web demo instructions

Chinese Word Demo via GUI

To launch the demo in a GUI, first change directories to the command directory and then invoke the demo batch script. Note: Parameters are set in the GUI, not as arguments to the launch script.

Windows Operating System

Academia Sinica Corpus

> cd %LINGPIPE_HOME%\demos\generic\bin
> gui_word_zh_as.bat 

Unix-like Operating Systems

Academia Sinica Corpus

> cd %LINGPIPE_HOME%\demos\generic\bin
> sh gui_word_zh_as.sh

For detailed information about running demos in a GUI, see the GUI demo instructions

Chinese Word Demo via Shell Command

Shell commands may be run over single files, all of the files in a directory, or using standard input/output.

Running over a Directory

Academia Sinica Corpus

> cd $LINGPIPE/demos/generic/bin
> cmd_word_zh_as.bat -inDir=../../data/testdir -outDir=/testout

Running a Single File

Academia Sinica Corpus

> cd $LINGPIPE/demos/generic/bin
> cmd_word_zh_as.bat -inFile=../../data/testdir/foo.txt -outFile=foo.out.xml

The other genres are handled the same way, with different suffixes in place of general_brown.

Running through a Pipe (Standard input/output)

Academia Sinica Corpus

> cd demos/generic/bin
> echo See Spot. See Spot run. | cmd_word_zh_as.bat

Running in Unix-like Operating Systems

For unix-like operating systems such as Unix, Solaris, Linux, or Macintosh OS X:

For detailed information about running demos from the command line, see the command line demo instructions

Chinese Word Demo Scripts

The following scripts are available in $LINGIPE/demos/generic/bin for running the demo. Note that each script comes in four flavors, distinguishing command line from GUI, and the Windows DOS shell from the Unix shell.

Genre Corpus Mode Windows DOS Unix/Linux/Mac sh
General Academia Sinica Command cmd_word_zh_as.bat cmd_word_zh_as.sh
GUI gui_word_zh_as.bat gui_word_zh_as.sh

Chinese Word Demo Parameters

The following is a complete list of parameters for the demo.

General Demo Parameters

These parameters apply to every version (web/GUI/command) of every demo.

Parameter Description Usage Constraints
inCharset Input character set Optional. Defaults to platform default.
outCharset Output character set
contentType Input content type May be one of:
  • text/plain
  • text/html
  • text/xml
Defaults to text/plain.
removeElts Element tags to remove Optional. May only be used with contentType=text/html or contentType=text/xml. Each value may be comma-separated list. If neither of these are specified, all text content is processed.
includeElts Elements to annotate

Command-Line Only Parameters

These parameters apply to every command-line demo, but are not relevant for the GUI or web versions of the demos.

Parameter Description Usage Constraints
inFile Readable input file May not be used with inDir. If either is not specified, defaults to standard input or output.
outFile Writeable output file
inDir Readable input directory May not be used with inFile or outFile. If used, inDir and outDir must both be specified.
outDir Writeable output directory