TMI: Type-Mention Inference

Files have about the same format as FACC1 annotations, plus some extra columns whose formats should be self-explanatory. The last column is editorial judgment on the correctness of various algorithms. Normally, you would only use rows that have a 1 in the last column.

• The byte spans in the files below were detected to have various discrepancies wrt ClueWeb files, partly as a result of charset discrepancies. We have recovered a large fraction of the gold annotations with standardized documents as byte sequences here.
• tmi_correct_only.tsv contains some annotations which are correctly predicted by TMI.
• editorial_judgement.tsv contains all the annotations on which editorial judgement was recorded.

Thanks to project staff Shashank Gupta, interns Prakhar Awasthi and Suraj Prajapati, and to Yahoo! editors Bhavana Srinath, Akshi Ghosh, Alisha Chatterjee, Anupriya S, Anvita Arora, Ashwini Sagar, Jenil Princia Dhason, Shahbaz Ahmed Khan, Sneha Johari, Sudipa Banerjee, Mona Parvaresh, Dipthi Diwakar, Amit Singh, Jinu Sabastian, Marshall Andrews, Tojo Thomas, Vijay Christy, and Yogesh Kumar, for generous help with preparing the data.