Introduction
Fiction texts are in the majority of cases made up of stretches of narrative and sequences of dialogue between characters appearing in the novel. In the narrative parts of the novel, the author describes the characters, e.g. how they look and feel, their actions, and also the surroundings/ space/ time the novel is set in. Between these, often longer, stretches of narrative, we have sorter sequences of dialogue between characters figuring in the novel, as e.g. (from Angela Thirkell's "O, These Men, These Men"):
| Within three hours he was at Beechwood where the Danvers family lived. Before he could get out of his car, Anna Danvers, James’s younger sister, came running out to meet him, pleased at the surprise of seeing Caroline’s delightful cousin Francis at such an unexpected hour. |
| "I have brought Caroline with me," said Francis. "Anna, be an angel and ask no questions, but take her up to your room. Where are Mr. and Mrs. Danvers?" |
| "In the library, I expect. Oh, Francis, what is the matter? Is it James?" |
| "It is." |
| A very quiet, polite Caroline got out of the car. |
The first and last row contain narratives, the ones in between dialogue or what is sometimes called direct speech (DS). Note that DS is made up of the reported or quoted speech within quotation marks and a so-called reporting clause, said Francis. (second row). In this case the reporting clause consists of the reporting verb, 'said' and the speaker 'Francis' only. A reporting clause my well contain mini-narratives, usually of how something is said reflecting the speaker's attitude, appearance or action, e.g. "Is he dead?" asked Mrs. Danvers, suddenly looking old. (ibid.)1
Preprocessing steps
The preprocessing steps are the same as before, only that the input this time was words occurring within quotation marks only.
The input was also limited in that the sequence of quoted speech had to occur within an orthographic sentence, i.e., a stretch of text starting with a capital letter and ending in full stop, question mark, exclamation mark or a dash. However, the quoted/reported speech did not need to start the sentence. My procedure for extracting DS has a recall of approx. 70% and a precision of more that 90%. The major drawback is that it does not extract quoted speech spanning several sentences. I tested BookNLP, which does in fact extract DSs spanning several sentences, but it seems to rely on carefully coded texts in the sense that if it could not match a pair of quotations marks, " or ', it lost track of the start and end of quoted text and produced the wrong output.
Results
After having created my model of quoted text, I output all pairs with a cosine similarity score of >= 0.6. This resulted in a list of 158 pairs, all of which, apart from 11 texts, were text pairs by the same author. This seems to indicate, strongly in some cases (CS score > 0.7), that certain authors let their characters talk or formulate themselves in a similar fashion in several of their novels. If the books have the same characters, e.g. the same detective(s) or police officer(s) as the main protagonists, this can perhaps be expected.
The few pairs of texts with a score of >= 0.6, where the authors were different were, as expected, the authors writing under various pen names (see CBF and cosine similarity: Experiment no. 2) and also these:
- Bryher (historical) <> Wallace Breem (historical)
- Douglas Reeman (historical) <> C. S. Forester (war)
- Hilary Mantel (historical) <> Margaret Campbell Barnes (historical)
- Margaret Campbell Barnes (historical) <> Hilary Mantel (historical)
- Niall Griffiths (general) <> Irvine Welsh (general)
- Patrick O'Brian (historical) <> C. Northcote Parkinson (historical)
- Richard Llewellyn (general) <> Alexander Cordell (general)
- Robert Graves (historical) <> Allan Massie (historical)
- Susan Hill (general) <> Daphne du Maurier (general)
Note the overlap in genre between these pairs/authors.
I also performed the same comparison lowering the SC score threshold to >= 0.5. This resulted in 143 novels where the DS in this novel had a medium cosine similarity with one or more other novels in the corpus. The number of unique authors was 93. These are not large numbers when the corpus contains 1,746 novels and 669 different authors, but it may still prompt further research into why and to what extent these particular 143 novels and 93 authors share features (words) in their dialogue that yields such relatively high CS scores. Percentagewise, the genre 'historical' is by far the genre with the most pairwise SC score >= 0.5 with 59 out of the 143. (The other genres/numbers are 'adventure' with 5, 'crime' 24, 'general' 38, 'romance' 13 and 'war' 4.)
Wallace Breem's Eagle in the Snow compares favourably, in the sense of reaching our CS comparison threshold, with nine other novels. These are three novels by Bryher, two by Mary Renault and one by Alfred Duggan, Allan Massie, R. F. Tapsell and Simon Scarrow respectively. All have been tagged as 'historical'.
1 Think it. Say it. Reported. A century of reporting verbs in British fiction (1900-2019).