The FrameNet project has been building a lexical database containing highly detailed information about the syntax and semantics of the different senses of English words, with applications such as question answering and information extraction. The database is based on Charles Fillmore's theory of Frame Semantics, relating word meanings to conceptualizations of situations and their participants, ranging from simple actions like Placing to complex events such as a Criminal process, with subevents like Arraignment and Trial. The generalizations are derived from annotated example sentences from large text corpora.
In this planning grant, input will be gathered from researchers in the NLP field regarding how the FrameNet data is currently being used, how it can be made more useful to a wider range of users, and how it can be enhanced so as to enable new directions of research. As part of this, a series of conferences will be held, both face-to-face and by videoconference, to plan future FrameNet development and cooperation and alignment of data with other lexical resources and annotation projects, such as WordNet, PropBank, OntoNotes, etc., with the ultimate goal of improving results in sense disambiguation, semantic role recognition, and natural language understanding in general.
The FrameNet project is creating a lexicon of English, designed to be both human- and machine-readable, which contains very detailed information about the grammatical structures in which words appear in sentences. The FrameNet database has been downloaded by thousands of researchers around the world, and its detailed lexical information is proving useful in many Natural Language Processing (NLP) tasks, including question answering, information extraction and machine translation. The grant "FrameNet 3: Upgrading FrameNet for the NLP Community" was requested in order to collect input from the NLP research community as to what the next steps for the FrameNet project should be. The major activity planned for this grant was to hold a meeting of leaders in NLP research to discuss the current state of the FrameNet database, its relation to other NLP resources, and its future development. The meeting was held May 1, 2010; in order to reduce travel (and thereby, both the expense and the environmental effects of flying), it was organized as a video tele-conference, with 18 participants at two meeting sites, UC Berkeley and Princeton University, another video link to the University of the Basque Country, and a phone link to NSF. The main conclusions of the meeting included the following: (1) we need more regular meetings of the NLP community concerning lexical resources, (2) we need better alignment between existing lexical resources, both by adding lexical units to smaller resources and by creating formal links between resources, and (3) while resources naturally differ in their approach, purpose and technical terminology, these differences between resources must be explained more clearly to the users. A wiki was set up to post the conclusions from the meeting and allow participants to edit them. The public website for FrameNet (http://framenet.icsi.berkeley.edu) has also been updated with help from this grant, adding a forum section to encourage continuing discussion among NLP researchers using FrameNet, a glossary to help clarify the terminology used in FrameNet (https://framenet.icsi.berkeley.edu/fndrupal/glossary), and a demo of the kind of annotation on which FrameNet is based (https://framenet.icsi.berkeley.edu/fndrupal/annotation_tool).
