I feel that's a bit different. When skimming or reading a book, even though we don't remember all the information typically a lot of the information is dependen

I feel that's a bit different. When skimming or reading a book, even though we don't remember all the information typically a lot of the information is dependent on prior information, so we form a sparse knowledge graph from the entire text. If I'm understanding the article, this approach would not use the skipped words to influence the output, so I thinks ita necessarily different.