James Joyce is one of the most admired, emulated and mythologised masters of twentieth-century prose. His Ulysses (1922) is among the cardinal texts of literary modernism produced between the world wars. The text creates its numinous effect through the repetition of short phrases over the course of a quarter of a million words. As early as 1929, critics invoked the musical device of leitmotif to explain this form of literary repetition (Curtius, 1929). Leitmotif describes a signature phrase or cue that accompanies and signals the presence of a character, locale or theme in a work. The device achieved a new importance in the nineteenth century through the opera of Richard Wagner, and it quickly migrated from music to literature in the writings of Édouard Dujardin, Thomas Mann and Marcel Proust. In Joyce’s hands, the pervasive use of the device amounts to “a sort of linguistic magic-realism” (O’Callaghan, 2011). But whereas musical leitmotif is conveyed through auditory recall – listeners recognise a brief musical phrase as an instance of leitmotif – written language cannot always offer this immediacy. The problem is a singular one: how do readers and how do computational tools recognise and unify the discrete instances of leitmotif that are distributed across a text?
The proposed paper will report on a series of experiments that combine methods of corpus and computational linguistics with close reading to gauge a fuller extent of the repetition in the novel and to assess its worth as leitmotif. From the perspective of text analysis, leitmotif is the purposeful repetition of textual fragments (or n-grams) in a text or over a collection of textual documents. Our computer assisted retrieval of leitmotifs is based on the following set of assumptions: First, for a sequence of words to be a leitmotif, it must capture a (human) reader’s attention through distinctive word choice or unsual collocation. This distinctiveness ensures a reader can recall earlier occurrences. Second, a given sequence of words functions as a leitmotif only if it occurs in a limited number of chapters or episodes of Ulysses. The sequence must occur in more than one episode but not in all eighteen. Key to distinguishing a given sequence is a limitation in the number of its occurences. Finally, previous scholarship has observed that Joyce not only repeated leitmotifs verbatim but also paraphrased, transposed, abbreviated or otherwise altered their individual elements (see Büchler et al., 2014 in this context). A computational approach needs to take stock of this state of affairs.
At the current stage of our research, we focus on bigrams. By treating each episode of the novel as a single document, we have constructed a document collection to measure the inverse document frequency (idf) of bigrams in the collection. The use of idf metrics enables us to retrieve those bigrams occur in a given number of episodes. Initially we set a threshold of occurrence at two, three and four episodes; in the light of the results produced we are continually revising this threshold. As a second step, the uniqueness or distinctiveness of bigrams was investigated by measuring the probability of their occurrence in a contemporary corpus of English texts put together from documents in Project Gutenberg and Archive.org. This helped to filter out very common constructions that enjoy a high idf. Named entities were also eliminated from the list of bigrams. We have also examined bigrams for “fuzziness” – gauging whether the constituent elements reoccur but with words inserted between them or in a reversed or otherwise transposed order. We have produced a list of around two hundred bigrams that are candidates for consideration as leitmotifs. Finally, a striking feature of Joyce’s style in Ulysses is the frequent use of compound coinages. Not only did he insert such neologisms into the text, but he also split them, reusing the constituent units of compounds in close proximity. This stylistic device also contributes to the pervasive sense of repetition in Ulysses, functioning as an alternative to and special case of leitmotif. To identify this type of repetition, we have retrieved and split all compound coinages in the novel, and examined whether their constituents reoccur within a given word distance.
The main obstacle to identification is the “protean nature” of leitmotif itself (Bribitzer-Stull, 2015). Whereas a highly distinctive phrase like Joyce’s “Agenbite of inwit,” which occurs in Ulysses seven times (Joyce, 1922), can readily be identified as a leitmotif, the associative potential inherent in even very short units of language means it is not always clear how one is to distinguish leitmotif from mere linguistic flukes or from instances of more general intertextuality (Kristeva, 1967). Examples of the latter include lexical priming (Hoey, 2005) and natural language collocation – for example, if an author writes “he opened the,” a reader of English would reasonably expect the next word to be “door.” At the moment, our list of candidate leitmotifs is checked by close reading bigrams within the context of the sentences in which they occur in the novel. Those bigrams that are accepted or judged as valid instances of leitmotif are added to a purpose-built database.
In addition to extending the inventory of leitmotifs identified in Ulysses (see, for example, Schutte, 1982), the experiments and the subsequent analysis have another key goal. They are meant to test the hypothesis that leitmotif not only offers a way to flag the presence of a character, locale or theme in the novel, but also creates a series of “primitive hyperlinks” or “analogue hyperlinks” (Cope and Phillips, 2006) threaded throughout Ulysses. Critics have often invoked hypertext to explain the myriad connections established within Joyce’s work (e.g. James, 1999), but our research will lead to an online hypertext edition of Ulysses that allows the reader to explore the non-linear reading paths created by leitmotif.