A Statistical Foray into Contextual Aspects of Intertextuality

TY - CHAP

T1 - A Statistical Foray into Contextual Aspects of Intertextuality

AU - Manjavacas, Enrique

AU - Karsdorp, F.B.

AU - Kestemont, Mike

PY - 2020/10

Y1 - 2020/10

N2 - Intertextuality is a highly productive concept in literary theory. The pervasiveness of intertextuality in literary texts has lead simultaneously to a proliferation of applications with often divergent interpretations of the concept of intertextuality, as well as a recurrent interest in studying it from a computational point of view. Despite the potential of data-driven, bottom-up approaches, most computational research into intertextuality has focused on the matter of text reuse detection, exploiting surface-level properties to improve the performance of retrieval systems. In the present study, we utilize the Patrologia Latina -- a substantial collection of religious texts spanning over a millennium of Latin writing (3rd to 13th centuries) -- to provide a large-scale systematic study of biblical intertexts. On the basis of multi-level statistical models, we investigate two axes of intertexts: the degree of lexical similarity, and the degree to which intertexts are thematically embedded in the context. Furthermore, we investigate the extent to which the following contextual sources of variation help explain the distribution of intertexts along the aforementioned axes: first, we analyze the effect of authorship: do authors differ in the way they compose their intertexts? Secondly, we inspect factors related to the source collection (i.e. the Bible) to elucidate whether the authority and tradition of particular books exert an influence on the observed intertexts: do certain books trigger a more allusive or quotational intertext type? Finally, we take into account the dominant topic surrounding the intertext location and examine associations between the distribution of dominant topics and intertext types. On the one hand, our analysis indicates that both axes (lexical similarity and thematic embedding) play partially complementary roles in our computational account of intertextual types. On the other hand, we find that biblical books and, more strongly, dominant topics constitute important factors of variation, while the authorial signal remains comparatively weak.

AB - Intertextuality is a highly productive concept in literary theory. The pervasiveness of intertextuality in literary texts has lead simultaneously to a proliferation of applications with often divergent interpretations of the concept of intertextuality, as well as a recurrent interest in studying it from a computational point of view. Despite the potential of data-driven, bottom-up approaches, most computational research into intertextuality has focused on the matter of text reuse detection, exploiting surface-level properties to improve the performance of retrieval systems. In the present study, we utilize the Patrologia Latina -- a substantial collection of religious texts spanning over a millennium of Latin writing (3rd to 13th centuries) -- to provide a large-scale systematic study of biblical intertexts. On the basis of multi-level statistical models, we investigate two axes of intertexts: the degree of lexical similarity, and the degree to which intertexts are thematically embedded in the context. Furthermore, we investigate the extent to which the following contextual sources of variation help explain the distribution of intertexts along the aforementioned axes: first, we analyze the effect of authorship: do authors differ in the way they compose their intertexts? Secondly, we inspect factors related to the source collection (i.e. the Bible) to elucidate whether the authority and tradition of particular books exert an influence on the observed intertexts: do certain books trigger a more allusive or quotational intertext type? Finally, we take into account the dominant topic surrounding the intertext location and examine associations between the distribution of dominant topics and intertext types. On the one hand, our analysis indicates that both axes (lexical similarity and thematic embedding) play partially complementary roles in our computational account of intertextual types. On the other hand, we find that biblical books and, more strongly, dominant topics constitute important factors of variation, while the authorial signal remains comparatively weak.

KW - intertextuality

KW - text re-use

KW - multi-level analysis

M3 - Contribution to conference proceedings

VL - 2723

SP - 77

EP - 96

BT - Proceedings of the Workshop on Computational Humanities Research (CHR 2020)

A2 - Karsdorp, Folgert

A2 - McGillivray, Barbara

A2 - Nerghes, Adina

A2 - Wevers, Melvin

PB - CEUR Workshop Proceedings

ER -