Section: New Results
One important characteristic of natural language is the huge flexibility on the ways in which we can express the same information. Many NLP applications like Question/Answering, information retrieval, and so on, need to deal with this diversity efficiently and accurately. Recognizing textual entailment, i.e., the question of whether the information contained in a given text T1 can be inferred from the information provided by another text T2 , is the core inference task for such systems.
Textual entailment recognition is a very difficult task, and systems presented at conferences like the RTE Challenge (http://www.pascal-network.org/Challenges/RTE2 ) still perform very poorly.
LED has recently started investigating this issue, and Paul Bedaride's Master Thesis  investigates how lexical information encoded as a description logic ontology can be used in this task.