Skip to content

Latest commit

 

History

History
executable file
·
195 lines (160 loc) · 8.24 KB

POS.md

File metadata and controls

executable file
·
195 lines (160 loc) · 8.24 KB
Raw

Vitk -- Part-of-Speech Tagging

The part-of-speech tagger of Vitk implements a conditional Markov model (CMM), a common probabilistic model for sequence labelling. In essence, CMM is a discriminative model which models the conditional probability distribution P(tag sequence|word sequence). This probability is decomposed into a chain of local probability distributions P(tag|word) by using the Markov property. Each local probability distribution is a log-linear model (also called a maximum entropy model).

Modes

The tagger has three modes: tag, eval, and train. In each run, it executes in one of these modes.

  1. In the tag mode, the tagger loads a pre-trained CMM, reads and tags an input text file and writes the result to the console or to an output file if specified.
  • A pre-trained CMM is located in a data folder of the tagger. Vitk comes with a pre-trained CMM for Vietnamese, which is located in the directory dat/tag/vi/cmm. This CMM is trained by using 10,165 tagged sentences of a Vietnamese Treebank.
  • The input text file is a plain text in UTF-8 encoding which has been word segmented, and it will be tagged line by line.
  1. In the eval mode, the tagger loads a pre-trained CMM, reads and evaluates the accuracy of the tagger on an input text file, and reports the performance of the tagger to the console.
  • A pre-trained CMM is similar to that in the tag mode described above.
  • However, the input text file now contains manually tagged sentences, each sentence on a line, in a simple format as follows: Đất/N nghèo/A trở_mình/V ./. That is, each word is paired with its correct tag, separated by the slash character.
  • The result of an evaluation is the accuracy of the tagging. The tagger reports token accuracy, that is the percentage of tokens which have been correctly tagged. If the toolkit is run with verbose option, the tagging speed is also reported. All the results are printed to the console window.
  1. In the train mode, the tagger reads manually tagged sentences from an input text file, trains a CMM and saves the model to an output file. There are some options for used in training which will be described in detail in the next section.
  • The input text file contains correct tagged sentences in the simple format as described in the eval mode above. The result of training is a CMM, which is then saved to an output file.
  • Pre-trained tagging models provided by Vitk are also created by running the tagger in this mode.

Arguments

The tagger of Vitk has following arguments:

  • -a <mode>: the mode, or action of the tagger, which is either tag, eval, or train. If this argument is not specified, the default mode of the tagger is tag.
  • -i <input-file>: the name of an input file to be used. This should be a text file in UTF-8 encoding. If the tagger is in the tag mode, it will read and tag every lines of this file. If it is in the eval or train mode, it will preprocess the file to get pairs of word and tag sequences for use in evaluating or training.
  • -o <output-file>: the name of an output file containing the tagging results (in the tag mode). Since by default, Vitk uses Hadoop file system to save results, the output file is actually a directory. It contains text files in JSON format, and will be created by Vitk. Note that this directory must not already exist, otherwise an error will be thrown because Vitk cannot overwrite an existing directory. If this parameter is not specified, the result is printed to the console window.
  • -cmm <cmm-file>: the name of a CMM file containing a tagging model. In the tag mode, this is a pre-trained CMM, while in the train mode, this is the resulting CMM of the training. If this argument is not specified, the tagger will use the default file named dat/tag/vi/cmm. This contains binary files in the parquet format of Apache Spark.
  • -dim <dimension>: this argument is only required in the train mode to specify the number of features (or the domain dimension) of the resulting CMM. The dimension is a positive integer and depends on the size of the data. Normally, the larger the training data is, the greater the dimension that should be used. Vitk implements the feature hashing trick, which is a fast and space-efficient way of vectorizing features. As an example, we set this argument as 160,000 when training a CMM on about 10,000 tagged sentences of the Vietnamese treebank.
  • -reg <lambda>: this argument is only used in the train mode to specify the L2-regularization term of the objective function to be minimized. This is a non-negative real value. The default value of lambda is zero, that is there is no regularization. Using a small lambda often helps avoid overfitting and speed up the training. As an example, we set lambda=1e-6 when training a CMM on about 10,000 tagged sentences of the Vietnamese treebank.

Running

Suppose that Apache Spark has been installed in ~/spark, Vitk has been installed in ~/vitk, data files required by this tool have been copied to /export/dat/tag. To launch Vitk, open a console, enter the folder ~/spark and invoke an appropriate command. For example:

  • To tag an input file and write the result to an output file, using the default pre-trained CMM:

./bin/spark-submit ~/vitk/target/vn.vitk-3.0.jar -t tag -a tag -i <input-file> -o <output-file>

Because the default mode of the tagger is tag, we can drop the argument -a tagin the command above.

There is not any -m argument in the command above, therefore, Vitk runs in the stand-alone cluster mode which uses a single local machine and all CPU cores.

  • To tag an input file, write the result to an output file, using a specified pre-trained CMM:

./bin/spark-submit ~/vitk/target/vn.vitk-3.0.jar -t tag -a tag -i <input-file> -o <output-file> -cmm <cmm-file>

  • To evaluate the accuracy on a gold corpus, using the default pre-trained CMM:

./bin/spark-submit ~/vitk/target/vn.vitk-3.0.jar -t tag -a eval -i <input-file>

  • To train a tagger on a gold corpus:

./bin/spark-submit ~/vitk/target/vn.vitk-3.0.jar -t tag -a train -dim 10000 -reg 1e-6 -i <input-file>

The resulting CMM has 10 thousand dimensions and is saved to the default directory /export/dat/tag/vi/cmm. If you want to save the result to a specific directory, append the -cmm argument, for example:

./bin/spark-submit ~/vitk/target/vn.vitk-3.0.jar -t tag -a train -dim 10000 -reg 1e-6 -i <input-file> -cmm /tmp/myCMM

If you wish to run Vitk on a Spark cluster, all you need to do is to specify the master URL of the cluster, such as:

./bin/spark-submit ~/vitk/target/vn.vitk-3.0.jar -t tag -a tag -i <input-file> -o <output-file> -m <master-url>

For your convenience, Vitk includes a sample file containing 1,000 tagged sentences which are extracted from the Vietnamese treebank.

Experimental Results

On the Vietnamese treebank:

  • 9,000 tagged sentences for training, 1,165 tagged sentences for testing.
  • Training parameters: dimension=160,000; lambda=1e-6.
  • Training accuracy: 99.65%; test accuracy: 95.17%.
  • Tagging speed on a single machine: tag 221,112 tokens in 216 milliseconds, that is, about 1,105,000 tokens per second.

Vietnamese Part-Of-Speech Tagset

The tagset used by Vitk is that of the Vietnamese treebank. There are 18 different tags:

  1. Np - Proper noun
  2. Nc - Classifier
  3. Nu - Unit noun
  4. N - Common noun
  5. V - Verb
  6. A - Adjective
  7. P - Pronoun
  8. R - Adverb
  9. L - Determiner
  10. M - Numeral
  11. E - Preposition
  12. C - Subordinating conjunction
  13. CC - Coordinating conjunction
  14. I - Interjection
  15. T - Auxiliary, modal words
  16. Y - Abbreviation
  17. Z - Bound morphemes
  18. X - Unknown

References

Back to Vitk