Learning to Align across Languages: Toward Multilingual FrameNet

TitleLearning to Align across Languages: Toward Multilingual FrameNet
Publication TypeConference Paper
Year of Publication2018
AuthorsGilardi, L., & Baker C. F.
Keywordscross-lingual resources, Frame semantics, lexical resources, Semantic roles
Abstract

The FrameNet (FN) project, developed at ICSI since 1997, was the first lexical resource based on the theory of Frame Semantics, and documents contemporary English. It has inspired related projects in roughly a dozen other languages, which, while based on frame semantics, have evolved somewhat independently. Multilingual FrameNet (MLFN) is an attempt to find alignments between them all. The degree to which these projects have adhered to Berkeley FrameNet frames and the data release on which they are based varies, complicating the alignment problem. To minimize the resources needed to produce the alignments, we will rely on machine learning whenever that’s possible and appropriate. We briefly describe the various FrameNets and their history, and our ongoing work employing tools from the fields of machine translation and document classification to introduce a new relation of similarity between frames, combining structural and distributional similarity, and how this will contribute to the coordination of the FrameNet projects, while allowing them to continue to evolve independently.

Acknowledgment

This material is based in part upon work supported by the U.S. National Science Foundation under Grant No (1629989). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

URLhttp://www.icsi.berkeley.edu/pubs/ai/alignacrosslanguages18.pdf