The task is to predict email authors based on the unigram or bigram language model. You can download the training, validation and testing sets here or use the ones provided in CMS (they are the same). The data set is of the following format:
author1 word1 word2 ...
author2 word1 word2 ...
...
"author" represents the unique id of the email author and is followed by the text of the email.
You should only be using the training data and the validation data for developing your author prediction model. The test data should only be used at the end for producing the evaluation results.