Speciteller is a tool for predicting sentence specificity using lightweight features. Lightweight means Speciteller does not need you to perform any processing to your sentences other than word tokenization. Speciteller tells you how much detail is present in a sentence by giving it a rating ranging from 0 (most general) to 1 (most detailed).

For example, for the two sentences below,

  1. "Evidence of widespread cheating has surfaced in several states in the last year or so ."
  2. Speciteller's specificity rating: 0.047500

  3. "California ’s education department suspects adult responsibility for erasures at 40 schools that changed wrong answers to right ones on a statewide test ."
  4. Speciteller's specificity rating: 0.897984


If you use Speciteller please cite the following paper:

Junyi Jessy Li and Ani Nenkova. 2015. Fast and Accurate Prediction of Sentence Specificity. Twenty-Ninth Conference on Artificial Intelligence (AAAI). [bib]


Speciteller is implemented using Python 2.7. It also needs numpy and liblinear (not included in package).

To use/download the tool please head to the github page.

The following lexicon are used for feature extraction:

Data and resources: speciteller_data.tar.gz

* The full package is distributed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.

Please send comments and feedback to J. Jessy Li.