On Syntactical Graphs-of-Words

A graph-of-words is a graph representation of natural language text based on proximity in the linear text reading order: the vertices are the words, and edges are induced by k left and right neighbours of the words. Vertices representing same or similar words are then contracted. We propose graphs-of-words where edges are instead induced on paths in the syntax trees (we investigate both dependency and constituency trees). We discuss some properties, advantages, and disadvantages of classic and new graphs-of-words on texts extracted from literature, as well as from a technical Q &A database.