Learning Shortest Paths in Word Graphs∗

In this paper we briefly sketch our work on text summarisation using compression graphs. The task is described as follows: Given a set of related sentences describing the same event, we aim at generating a single sentence that is simply structured, easily understandable, and minimal in terms of the number of words/tokens. Traditionally, sentence compression deals with finding the shortest path in word graphs in an unsupervised setting. The major drawback of this approach is the use of manually crafted heuristics for edge weights. By contrast, we cast sentence compression as a structured prediction problem. Edges of the compression graph are represented by features drawn from adjacent nodes so that corresponding weights are learned by a generalised linear model. Decoding is performed in polynomial time by a generalised shortest path algorithm using loss augmented inference. We report on preliminary results on artificial and real world data.