Linguistic Inference and Textual Entailment
Hauptseminar: Computerlinguistik, 2. Studienabschnitt
Leitung: PinkalOrt: Geb. C7 2, Konferenzraum 2.11
Zeit: Di 16 - 18 Uhr
Beginn: 24.04.2007
Geeignet für: B.Sc./M.Sc.
Semantic interpretation as assignment of literal meaning representations to sentences is useful for a number of natural language processing tasks, in itself. To arrive at a more complete picture of human semantic processing, and to enhance the performance of many NLP applications, we need to model inference mechanisms that derive the required information (say, the query submitted to a Q&A system) from one or several pieces of directly given linguistic meaning information (e.g., in the semantic document markup). Traditionally, inference has been modelled through logical deduction, and brought to work via different types of theorem provers. A variety of proposals have been made to tailor formal inference systems to the special needs of linguistic processing, to improve efficiency or extend coverage. More recently, "textual entailment" or textual inference has been proposed as a more natural and useful concept of inference, based on humans' intuitive judgments as to whether a sentence is a plausible consequence of another sentence or piece of text. Textual entailment is modelled through a combination of knowledge-lean information (like word overlap) with deeper semantic information processing and machine learning techniques.
In the seminar, we will inspect and discuss logic-based approaches to linguistic inference (like Hobbs et al. 1990), as well as the textual entailment concept (a seminal paper is Monz/de Rijke 2001), and systems developed for the shared RTE task ("Recognising Textual Entailment").
Formalities
B.Sc. Regelstudienzeit 6. SemesterM.Sc. Standard time slot 2nd semester, CL, LT