Evocation
We are working on a system to add a new relationship to WordNet that connects all parts of speech. Our goal is to bootstrap human ratings of evocation to the entirety of WordNet via machine learning.
People
|
Jordan Boyd-Graber (*), Christiane Fellbaum, Dan Osherson, Rob Schapire
|
|
This work has been supported by the National Science Foundation
|
ImageNet
Building ImageNet
ImageNet is a proposed extension of WordNet using labeled images to
illustrate the synsets' underlying concepts besides the original
dictionary definitions. Currently, the image database we use to build
the ImageNet comes from ESP Game. Each image has a list
of captions words about what can be perceived in the image which is
agreed by many people.
We designed a study to evaluate the synset-image assignments of ImageNet
against human decisions of image assignments, and are investigating the effectiveness of using images for communication.
People
|
Xiaojuan Ma(*), Jordan Boyd-Graber, Sonya Nikolova, Christiane Fellbaum, Dan Osherson, Perry Cook, Rob Schapire, Moses Charikar, Chandra Barnett
|
Sense Disambiguation
In order to apply WordNet to untagged corpora, techniques must be
developed to perform word sense disambiguation (i.e. determine which
WordNet synset corresponds to a particular word in the text). We are
applying machine learning techniques to effect accurate word sense
disambiguation, thereby allowing a variety of NLP techniques which
leverage WordNet to be applied to a wide body of corpora.
People
|
Dave Blei(*), Miroslav Dudik, Jonathan Chang, Jordan Boyd-Graber, Dan Osherson, Rob Schapire
|
From WordNet to a Knowledge Base for Question Answering
People
|
Christiane Fellbaum(*), Peter Clark (Boeing), Jerry Hobbs (ISI/USC)
|
Robust Extraction of Meaning
People
|
Christiane Fellbaum(*), Chris Manning (Stanford), Andrew Ng (Stanford), Dan Jurafsky (Stanford)
|
(*) Contact person
|
