Fecha: 28-29 Febrero 2012
Ponente: Ekaterina Ovchinnikova
Lugar de celebración: Sala 6.02, ETSI Informática, UNED (mapa)
Talk 1: Introduction into inference-based natural language understanding (28/02/2012, a las 12h00)
In order to understand a natural language expression it is usually not enough to know the literal ("dictionary") meaning of the words used in this expression and compositional rules of the corresponding language. Much more knowledge is involved in discourse processing; knowledge, which may have nothing to do with the linguistic competence but is rather related to our general conception of the world and reasoning abilities.
This talk discusses how knowledge-intensive natural language understanding (NLU) can be modeled in a computational framework, with a main focus on inference-based NLU. Two important components of each inference-based NLU system are a knowledge base and an inference machine. In this talk, I will mainly focus on these two components and discuss existing sources of machine-readable knowledge and reasoning procedures applicable to NLU. In the second part of the talk, I will present and discuss experiments on recognizing textual entailment, semantic role labeling, and paraphrasing of noun-noun constructions, which are knowledge-intensive NLU tasks.
Talk 2: ILP-based weighted abduction for natural language understanding (29/02/2012, a las 12h00)
This talk focuses on discourse processing based on a mode of inference called weighted abduction. This framework is appealing because it is a realization of the observation that we understand new material by linking it with what we already know. It instantiates in natural language understanding the more general principle that we understand our environment by coming up with the best explanation for the observables in the environment. Hobbs et al. (1993) show that the lowest-cost abductive proof provides the solution to a whole range of natural language pragmatics problems, such as word sense disambiguation, anaphora and metonymy resolution, interpretation of noun compounds and prepositional phrases, detection of discourse relations, etc.
Abduction-based discourse processing was a thriving area of research in the 1980s and 1990s. But in spite of successful theoretical progress and small-scale systems, work on large-scale, ``real life'' systems foundered on a complexity problem: Reasoning procedures were not efficient enough. In this talk, I will present a recently developed implementation of weigthed abduction, which is based on Integer Linear Programming. In this approach, the abductive reasoning problem is formulated as an ILP optimization problem. The ILP-based system achieves a significant processing speed-up and is applicable on a large scale.
This talk will be quite technical, therefore I would like to make it possibly informal and interactive. For those, who is interested in reading related literature in advance, I recommend the following two papers: