Michael White is visiting from Ohio State on 10/6 Friday. He works in natural language processing. His talk abstract and bio are below. His seminar is 1:30-2:30 in White 411.
In this talk, I will discuss two scenarios where it is important to avoid generating sentences with structural ambiguities. In the first
part of the talk, I will investigate whether statistical parsers can be used for self-monitoring in natural language generation in order to avoid so-called “vicious” ambiguities, namely those where the intended interpretation fails to be considerably more likely than alternative ones. In this part of the talk, I will demonstrate that although using statistical parsers for this purpose is more difficult than one might expect—since automatic parsers too often make errors that human readers would be unlikely to make—by training a ranking model using features from the generator together with multiple parsers, successful self-monitoring can be achieved. In the second part of the talk, I will investigate whether natural language generation can be used to automatically construct disambiguating paraphrases for structurally ambiguous sentences—that is, paraphrases that clarify the competing interpretations of a structurally ambiguous sentence. Here I will present an experiment which suggests that by simply asking naive annotators to clarify which paraphrase is closer in meaning to the original sentence, the resulting paraphrases can potentially enable meaning judgments for parser training and domain adaptation to be crowd-sourced on a massive scale.