Frame Blends Nomination System
Overview
This site is the project entry for 2020 GSoC project built by Wenyue Xi (Suzie) and mentored by Professor Mark Turner, Profesor Tiago Torrent, Professor Cristóbal Pagán Cánovas and Fred Belcavello. For this project, Suzie developed multiple approaches that can nominate some kinds of Frame Blends, built a preliminary system of Frame Embedding, and made an interactive prototype of the human-in-the-loop Frame Blends Nomination System. This site is developed by Suzie to present this project as a public source for the Red Hen Lab community, and also supports developers or users who may participate in this project in the future.
Useful Resources
Suzie's daily progress blog of 2020 Google Summer of Code(GSoC) Frame Blends project, including all the meeting slides and related papers
Suzie's Github repository for 2020 Google Summer of Code(GSoC) Frame Blends project, including coding files and data
This page is a place to show some personal reflection from Suzie Xi for the 2020 GSoC journey with Red Hen Lab
The offical project page on GSoC website
Human-in-the-loop Frame Blends Nomination System
The flowchart and presentation slides below shows the overall workflow of Human-in-the-loop Frame Blends Nomination System. Two human roles are invoved in the primary process: admin and analyst. This section also includes user guides for Admin and analysts, and an additional entry for potential nomination algorithm contributers.
Tutorial Video for Running HPC Pipeline
Video Transcript
Rapid Annotator Documentation
Human-in-the-loop System Design Documentation
Guide for Administrator
Guide for Analysts
Guide for Algorithm Contributor
Flowchart
Presentation Slides
Frame Embedding
The method of Frame Embedding is inspired by Word Embedding from Natural Language Processing(NLP). There are currently very limited attempt of this method(by 2020 Aug), but some previous papers mentioned some potential approaches of building or using Frame Embedding in their tasks.
These are two existing examples:
Using Embeddings to Compare FrameNet Frames Across Languages
Semantic Frame Embeddings for Detecting Relations between Software Requirements
Both papers above didn't design a relatively complete system of Frame Embedding. Instead, they mention "Frame Embedding" as a new concept and suggest that it may be useful for some specific NLP tasks. And the way Frame Embedding has been constructed is straightforward. As mentioned in Semantic Frame Embeddings for Detecting Relations between Software Requirements by Alhoshan, Batista-Navarro, and Zhao, they present a tentative approach to build the Frame Embedding:
" The word embedding vectors resulting from the previous step were then used to form an embedding- based representation of semantic frames, i.e., frame embeddings. That is, for any given semantic frame F, we collected the vectors corresponding to the LUs that evoke it. The average of these LU vectors is then computed and taken as the frame embedding for F. For instance, as 11 LUs are associated with the Creating frame in FrameNet, a vector containing the average over the 11 word embedding vectors corresponding to these LUs was obtained as part of this step."
The Frame Embedding in this project follows the suggested straightforward approach that calculates the average vector of all LU word embedding vectors as the Frame Embedding vector. The Frame Embedding method is applied to all the frame data from FrameNet 1.7, and have been used in various nomination algorithms.
Frame Embedding Doc
Attention Mechanism
Word Embedding and Word2Vec
Word2Vec is a more recent model that embeds words in a lower-dimensional vector space using a shallow neural network. Word2Vec model is presented by Google researchers with the following paper:
Efficient Estimation of Word Representations in Vector Space
Distributed Representations of Words and Phrases and their Compositionality
Gensim is the Python implementation of Word2Vec, and a tutorial of Gensim shows below:
FrameNet
FrameNet maps meaning to form in contemporary English through the theory of Frame Semantics. FrameNet is the important foundation that Frame Embedding has been based on, and this project mostly using the data from FrameNet 1.7. For more information, please check the offical website of FrameNet: Berkeley FrameNet.
This is a full guide published by FrameNet, and you can also find it from FrameNet website FrameNet II: Extended Theory and Practice.