Classification of Literary Style that takes Order into Consideration

Abstract

The statistical analysis of the heterogeneity of the style of a text often leads to the analysis of contingency tables of ordered rows. When multiple authorship is suspected, one can explore that heterogeneity through either a change-point analysis of these rows, consistent with sudden changes of author, or a cluster analysis of them, consistent with authors contributing exchangeably, without taking order into consideration. Here an analysis is proposed that strikes a compromise between change-point and cluster analysis by incorporating the fact that parts close together are more likely to belong to the same author than parts far apart.The approach is illustrated by revisiting the authorship attribution of Tirant lo Blanc.

Outlier detection for multivariate categorical data

“…The second extension can be modeled by letting the weights in the mixture distribution for the random effects of the i ‐th row, ν i = ( ν i 2 , …, ν i J ), vary from row to row, where ω = ( ω 1 ,…, ω n ). This idea was proposed by Fernandez and Green for Poisson mixtures for spatial data, and it was used by Puig et al for multinomial cluster analysis. As a consequence of , the probability that the i th row is allocated to the outlier group, ω i , will change from row to row, and the set of allocation variables, ζ = ( ζ 1 , …, ζ n ), will not be identically distributed because π ( ζ i = 1| ω ) = ω i .…”

“…where ω = (ω 1 ,…,ω n ). This idea was proposed by Fernandez and Green 13 for Poisson mixtures for spatial data, and it was used by Puig et al 14 for multinomial cluster analysis. As a consequence of (2.5), the probability that the ith row is allocated to the outlier group, ω i , will change from row to row, and the set of allocation variables, ζ = (ζ 1 , …, ζ n ), will not be identically distributed because π(ζ i = 1|ω) = ω i .…”

“…The hierarchical structure through which the δ i link the probabilities of being outlier, ω i , for rows that are close will depend on the particular example. One could, for example, model that dependency through a conditional autoregressive structure mimicking the one used in disease mapping to obtain spatially smoothed estimates of Poisson means (Besag et al and Mollie), and the one used in stylometric analysis to take the order of texts into consideration (Puig et al). Instead, in the example on the plays by Shakespeare in Section 4, that will be done through a repeated measurement type of structure that takes into account the fact that acts belonging to the same play are more likely to belong to the same cluster than acts from different plays.…”

Bayesian Analysis of the Heterogeneity of Literary Style

“…Puig, Font & Ginebra (2015) propose an alternative analysis that treats the two stylometric variables separately, but incorporates the fact that chapters close together are more likely to belong to the same author than chapters that are far apart. In this case, the results of the analysis are similar.…”

Bayesian analysis of textual data

“…Finally, note that different from the previous case study, in this example texts (chapters) are ordered sequentially, and that order is not taken into consideration in the cluster analysis model used here. Puig, Font and Ginebra (2014) proposes an alternative analysis that treats the two stylometric variables separately, but incorporates the fact that chapters close together are more likely to belong to the same author than chapters that are far apart. In that way, one strikes a compromise between change-point analysis, as-suming all neighboring chapters to belong to the same cluster except the boundary ones, and the kind of cluster analysis considered here, that treat all chapters exchangeably, as if order did not matter whatsoever.…”