Train-files: Comma-separated list of training files Test-files: Comma-separated list of training files Train-dir: Directory containing training files

Test-dir: Directory containing test files Save-model: Whether to save the trained model Model: Filename for the model (saving a trained model or reading a saved model) using the command line parameters listed below: You can change the default values, as well as specify entire directories of training data, etc. The above will learn the parameters of the model using regularized AdaGrad, log loss and some default hyperparameters. For example, to train on the Ontonotes corpus requires about 16GB. If you are training on a lot of data, you may need to increase the amount of memory allocated to the JVM. This will train a model on the given training data, testing accuracy on the given test data, and saving the trained model to a file called “ForwardPosTagger.factorie”. Java -classpath -Xmx $memory cc.ForwardPosTrainer -owpl $trainfile $testfile $save $model Modelname = 'ForwardPosTagger.factorie' trainfile = '-train-file=/path/to/training/data' testfile = '-test-file=/path/to/test/data' save = '-save-model=true' model = '-model= $modelname ' To use our tagger as a dependency in your Maven project, you will need to add the IESL repository: The easiest way to get started is by using our tagger through the FACTORIE NLP command line tool. Our tagger is both fast and accurate, processing more than 20K tokens/second and achieving 97.22% accuracy on WSJ sections 22-24. The default tagger in Factorie is the model trained on Ontonotes, a cross-domain dataset guaranteeing relatively high inter-annotator accuracy, called OntonotesForwardPosTagger. We provide two pre-trained models, one trained on the Ontonotes English corpus and one trained on Penn Treebank WSJ sections 0-18. It is trained using regularized AdaGrad, with l1, l2, learning rate, delta, feature count cutoff, and number of training iterations hyperparameters tuned via grid search using a development set. Unlike the above, our tagger does not perform dynamic model selection. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (ACL12), pp. Fast and Robust Part-of-Speech Tagging Using Dynamic Model Selection. Our tagger uses a fast and accurate forward model based heavily on the one described in the following paper: We achieve this by training a simple probabilistic model that, given a token of text, predicts its part-of-speech given its form and context in the sentence. A part-of-speech tagger (POS tagger) is a program that takes human-language text as input and attempts to automatically determine the grammatical POS tags (noun, verb, etc.) of each token in the text.

Speech tagger