Generating Disambiguating Paraphrases for Use in Crowdsourced Judgments of Meaning

Adapting statistical parsers to new domains requires annotated data, which is expensive and time consuming to collect. Using crowdsourced annotation data as a “silver standard” is a step towards a more viable solution and so in order to facilitate the collection of this data, we have developed a system for creating semantic disambiguation tasks for use in crowdsourced judgments of meaning. In our system here described, these tasks are generated automatically using surface realizations of structurally ambiguous parse trees, along with minimal use of forced parse structure changes.NSF grant IIS-1319318No embargoAcademic Major: Computer and Information Scienc