Advanced users

Please note that if you are not already familiar with this system, then you are definitely at the wrong place! If you need a quick user manual, please read the Quick (15-minutes) manual. If you need a detailed user manual please read the detailed manual.

---

How to Run the System from another program

@TODO: What if one wants to call/use an entailment engine from one's own Java program?

Configuration Files

Overview of the common configuration (first draft)

@TODO

Configuration file format and syntax (first draft)

@TODO

Sample configurations? How to customize or develop users' own config?

@TODO

Entailment Decision Algorithms (EDAs)

What is an EDA?

An Entailment Decision Algorithm is a Component which takes a Text-Hypothesis pair in input and returns one of a small set of answers. A complete entailment recognition system is trivially an EDA. However, in the interest of re-usability, generic parts of the system should be made into individual Components. Entailment Decision Algorithms communicate with Components through generic specified interfaces.

Which EDAs are already implemented in the EOP?

ClassificationEDA the current implementation of the classifier is like kMeans. The training calculates the average scores for all the features, and store them in the model. The testing calculates the distance between the input instance and the model scores and select the closest label.

EditDistanceEDA implements edit distance algorithms. In the training phase it calculates a distance threshold that is applied during the test phase so that pairs resulting in a distance below the threshold are classified as ENTAILMENT, while pairs above the threshold are classified as NONENTAILMENT.

MaxEntClassificationEDA is an EDA based on a prototype system called TIE (Textual Inference Engine). It uses the OpenNLP MaxEnt package to train a GIS Model in order to classify Entailment T-H pairs from Non-Entailment ones. Currently, it works for both English and German. The compatible components are: 1) distance components 2) scoring components 3) lexical knowledge components.

How to modify or extend EDAs?

@TODO