Quotations, Relevance and Time Depth: Medieval Arabic Literature in Grids and Networks

Abstract

This contribution deals with the use of quotations (repeated n-grams) in the works of medieval Arabic literature. The analysis is based on a 420 millions of words historical corpus of Arabic. Based on repeated quotations from work to work, a network is constructed and used for interpretation of various aspects of Arabic literature. Two short case studies are presented, concentrating on the centrality and relevance of individual works, and the analysis of a time depth and resulting impact of a given work in various periods. 1 Quotations and Their Definition The relevance of individual works in a given literature and the time depth of such relevance are of interest for many reasons. There are many methods that can reveal such relevance. The current contribution is based on quotation extraction. Quotations, both covert and overt, both from written and oral sources, belong to constitutive features of medieval Arabic literature. There are genres which heavily depend on establishing credible links among sources, especially the oral ones, where a trusty chain of tradents is crucial for the claims that such chains accompany. Other links may point to the importance of a given work (or its part) and may uncover previously unseen relations within a given literature or a given genre/register, or reveal connections among genres/registers within a given literature. As such, the results are interesting in a wide research range, from linguists or literature theorists to authors interested in the interactions of various subsets of a given literature. The research on quotations, their extraction and detection is rich in the NLP, but the algorihms used are based mainly on the quotation-marker recognition, e.g. Pareti et al. (2013), Pouliquen et al. (2007) and Fernandes et al. (2011), or on the metadata procesing (e.g. Shi et al. 2010), to name just a few examples. It can be said that most of the contributions focus on issues different from the one described in this contribution and choose a different approach. Our understanding of quotations in this project is limited to similar strings of words, i.e. the quotations are very close to borrowings or repetition of verbatim or almost verbatim passages. Technically, it can be viewed as an n-gram that is being repeated in at least two works. These repeated n-grams create links that exhibit some hierarchy, e.g. on the chronological line. The only approach known to us that can be paralleled to ours is the one described in Kolak and Schilit (2008) for quotation mining within the Google Books corpus with algorithm searching for verbatim quotations only. In a different context and without direct inspiration we developed an algorithm that is tolerant to a certain degree of lexical and morphological variation and word order variability. The reason for this tolerance is both the type of the Arabic language (flective morphology and free word order), but also the fact that the quotations in medieval Arabic literature tend not to be very strict. Despite of the fact that the matching is not so rigorous, we assume that the length of n-grams we use drastically decreases possibilities of random matches. The frequency of such n-gram repetition in various literary works can point to several aspects, however, in this contribution we will limit ourselves to interpreting such links in a rather cautious and not too far-reaching manner, mainly as pointers to the fact that the writer of the book where the quotations appear was also a reader of the book from which the quotations stem and that he was to a certain degree influenced by it. This does not necessarily mean that the lineage of quotations is complete in our picture, for we