Narrative similarity as common summary: Evaluation of behavioral and computational aspects


One of the main efforts of recent computational linguistics is to formalize the process of identifying and evaluating similarity between narratives, which is argued to be a key concept for all human behavior. Analyses of the data of 52 adults that participated in our empirical study offered evidence that supports the position that narrative similarity can be equated to the existence of a common summary between the narratives involved. As a further step for applying the above hypothesis, we introduced our own methods for measuring similarity of narratives through the notion of summary and compared them with some of the existing lexical-matching similarity measures. Comparisons of each computational measure with the actual similarity judgments of human participants revealed that methods that merely measure the number of shared words between two stories were unable to capture human similarity judgments, especially for stories of moderate similarity levels. However, the summary-based methods of our approach managed to reproduce human ratings for story pairs across all the range of similarity (from high similarity to low similarity).

In Literary and Linguistic Computing (LLC).

Understanding the mechanism of identifying similarities among stories is the key for developing efficient computational methods that are able to adequately evaluate story similarity. Existing lexical matching algorithms were found to be inferior when compared with our summary-based approach, as they failed to anticipate actual human judgments of similarity between stories. However, the present study is only a first effort towards the direction for putting into practice the hypothesis that narrative similarity can be defined through the existence of a common summary.

If we wish to put the above ideas into work, there is still a lot more that needs to be done. In order to apply our approach into real settings, we first need to suggest some automatic mechanisms which are necessary for running our methods. First of all, in order to compare two stories using our methods, we need an adequate common summary of the stories. Methods for automated summarization of texts are already available (Hovy & Lin, 1999; Dalianis, 2000; Hassel, 2003; Orasan et al., 2004). However, for the specific needs of our approach we need to develop methods able to automatically summarize two documents simultaneously, and produce their common summary. Devising a method that would automatically produce candidate summaries that would be the less abstract among the common summaries of both stories, would offer the first big step towards the applicability of our approach. As a second step, we also need to develop tools that would automatically evaluate the produced summary, with regards to each of the stories under consideration.