Paper Title: Exploring Content Selection in Summarization of Novel Chapters
Paper Author(s): Faisal Ladhak, Bryan Li, Yaser Al-Onaizan, Kathleen McKeown
What was the most important point (in your opinion) of this paper?
While there are far more efficient summarization tools for news articles, books summarization with natural language processing techniques is a novel approach.
List four (4) facts that you learned from this paper.
- A literature survey of novel summarization gives an unsupervised Textrank approach to Cliffsnotes and GradeSaver dataset of novel/summary pairs.
- The paper details the scholar’s approach to novel summarization by matching for best alignment of referential summary statements. They are extracted at different levels of granularity.
- Each summary is produced with new alignment approaches which human evaluation and automated metrics both rank better.
- Further studies will indicate if the abstractive summarization process is applicable to much larger pre-trained language models.
Write a summary of the paper. Please use complete sentences and take us from the start of the paper through the end. What were the main themes in the paper, how were they integrated together, and how do they apply to what you have previously learned (in this class or another class)?
The paper discusses techniques for generating novel summaries using chapter-wise summarization pairs of online study guides. A matching algorithm is used to align the summary and chapter sentences. The datasets are of five study guides and Project Gutenberg. At different levels of granularity, that is, levels of complexity, extracting the sentence summary for each of the chapters. The paper claims that this is an improved form of summarization of longer text with automatic metrics and a crowd-sourced pyramid analysis.
The themes explored in the paper range from shortcomings of news summarization, improvements over sentence-level evaluation novel summarizations, and stable alignments of the summary texts. Future work in the field is about sentences that are more fluid in a contextual summary.
In this class we learnt about text summarization with n-gram analysis and by term frequency.
Give an example from natural language (found in a news site online) that illustrates one of the problems described in the paper. Your examples should not be taken from a paper or textbook, but should rather be ones that you discover on your own. For your example, show the steps that a human would take to solve it.
Discuss any background knowledge of the world that a human must have to solve this problem as well as whatever language related knowledge they must have. Clearly illustrate how the problem would be solved, and show the steps that lead a human to a solution.
Background knowledge of the world required for a human to summarize chapters in a novel are the following:
- Know or learn the language that the novel is written in.
- Reading the chapter in the novel or study guide in question.
- Understand the essence of the chapter.
- Shorten text.
X. Yuan et al., “Interactive Language Learning by Question Answering,” arXiv:1908.10909 [cs], Aug. 2019, Accessed: Oct. 28, 2020. [Online]. Available: http://arxiv.org/abs/1908.10909.
1 Comment