NLP Notes

- T and H are seen as two syntactic graphs to reduce the textual entailment

- In principle, textual entailment is a transitive oriented relation holding in one these:

• T semantically subsumes H (e.g. H: The cat eats the mouse, T: the devours the mouse) eat generalizes devour
• T syntactically subsumes H (e.g. H: The cat eats the mouse, T: the cat eats the mouse in the garden)
• T directly implies H (e.g. H: The cat killed the mouse, T: the cat devours the mouse)

Assumptions:
- H is supposed to be a sentence describing completely a fact in an assertive or negative way
- H should be a simple subject-verb-object sentence

Measure:
- E(XDG(T), XDG(H)), where XDG(T) and XDG(H) are syntactic representation of T and H as eXtended Dependancy Graph.
- E(XDG(T), XDG(H)) has to satisfy:

• E(XDG(T), XDG(H)) not equals E(XDG(H), XDG(T))

• the overlap between XDG(T) and XDG(H) has to describe if a subgraph of XDG(T) implies the graph XDG(H)

[more reading must be done]

The Paper: http://www.cs.biu.ac.il/~glikmao/rte05/zanzotto_et_al.pdf