BaselineChipaSystem

This will end up as a significant part of the tasks/evaluation chapter.

Experiments that we'll do repeatedly...

• disambiguate es-gn for the top 100 words in es, using the Bible
• also es-qu, using the Bible
• also es-en and en-es using the Bible
• also es-en and en-es using Europarl

baseline system looks like this

• take the one-to-many alignments from cdec as ground truth
• for each source-language word, train a classifier
• do 10-fold cross validation with that classifier

classifiers that we'll use are...

• scikit-learn maxent classifiers
• can try other classifiers too
• try some different feature selection and regularization approaches

feature set that we'll use...

• try a few different context window sizes: definitely at least 2,3,4,5
• also bag of words over the whole sentence
• try some dependency features: we can run a parser for both es and en as source languages

some relevant links...

• http://scikit-learn.org/stable/modules/feature_selection.html
• http://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
• http://scikit-learn.org/stable/modules/feature_selection.html#l1-feature-selection
• http://scikit-learn.org/stable/modules/generated/sklearn.svm.LinearSVC.html#sklearn.svm.LinearSVC