Names in Novels: an Experiment in Computational Stylistics

Names in Novels: an Experiment in Computational Stylistics
van Dalen-Oskam, Karina, Huygens Institute for the History of the Netherlands - KNAW, The Hague, karina.van.dalen@huygensinstituut.knaw.nl
________________________________________
Introduction
The growing amount of digital texts and tools is slowly but definitely changing the way literary scholars design their research. This paper will describe the effects for one research topic: ‘literary onomastics’, the study of the usage and functions of names in literary texts. Research in literary onomastics usually is qualitative in nature and focuses on 'significant' names in literary texts. No quantitative or comparative studies have been published yet. Several researchers, however, have pointed out that names can only be called significant if they are studied in the context of all the names - the so-called 'onymic landscape' - in a text, oeuvre, genre etc. (e.g. Sobanski 1998). This question is comparative by nature and implies the wish for a more quantitative, computer-assisted approach.
Research questions
As to name usage, first we have to find out what is normal. We want to know how many name forms usually occur in a novel, and how many of these are personal names, place names, and other names. As to name functions, a useful disctinction is between plot internal and plot external names. Plot internal names refer to characters, places or other entitities which only ´exist´ in the fiction of the story. Plot external names refer to persons, places or other entities which are known to exist or to have existed in the real world. Most place names are plot external, referring to real countries, cities, streets, etc. and thus have a reality enhancing function. In fantasy novels, however, place names are usually fictional and thus plot internal, enhancing the unreality, the fantasy of the story. Plot external personal names seem to function as characterizations of the fictional characters, describing e.g. their political or cultural preferences.
It is expected that different authors, genres, time periods or even languages apply these different types and functions in different ways, showing different trends which we want to discover in what we like to call comparative literary onomastics (Van Dalen-Oskam 2005, 2006). The ultimate aim of the research is to develop a method to compare the name usage and functions on as large a scale as possible, explicitly also across languages. So what is needed to do that?
Method and corpus
The amount of names in a text can be expressed in the percentage of the total amount of tokens in the text. For that, we need digital texts of fair to good quality. To tag all the names, we need named entity recognition and classification (NERC) tools. Different forms of the same name (e.g. Patrick and Patrick's) need to be grouped by a lemma (PATRICK in this case). Different name forms for the same person or place need to get the same identification (e.g. the name Alfred and the name Issendorf both belong to the character identified as ALFRED ISSENDORF). To find out whether we can compare the resulting percentages across languages, we focused on a corpus of modern Dutch and English novels and their translations into the other language. We found we have to include two other levels of aggregation: mentions and name tokens. Mentions are occurrences of a name irrespective of the number of tokens used. So several name tokens can be used in one mention. This distinction is necessary because different languages have different tokenization rules. The Dutch personal name Gerrit-Jan, e.g., with a hyphen and therefore counted as one token, is translated in English as Gerrit Jan, resulting in two tokens. Our corpus consists of 22 Dutch and 22 English novels, added with the translation into the other language of ten in each group.
Theory and practice
Comparative research can only be done when many scholars collaborate. We will have to make sure that all those scholars encode their texts in the same way, considering the same tokens as names. This may sound easy, but it is not. Even name theorists have different definitions of what a name is (Van Langendonck 2007). Guidelines had to be set. We decided to limit the tagging to the 'prototypical' names, so those names that are considered names by readers in general. Something is a name if it refers to a unique person, place, or object. So we excluded currencies, days of the week, months, etc. For cases leading to discussion we defined additional rules.
We found that not many modern novels are digitally available yet. Furthermore, NERC tools proved to be not good enough yet and are using partly different categories than our research needs (Sekine and Ranchod 2009). This meant much more manual work than we expected. The tagging and the statistics are now done by means of several perl scripts written by André van Dalen and an ingenious Excelsheet with macros designed by David Hoover. These tools can be seen as a prototype for the webservices we need for this type of comparative stylistic research.
Results (selection)
In Figure 1, the percentage of name tokens in a set of novels is shown. The selection of novels in this graph consist of sixteen novels of which fifteen are Dutch and one is (American-)English (Of the Farm), added with the English translation of one of the Dutch novels (The Twin, translation of Boven is het stil).
Fig. 1

Full Size Image
The graph gives as a first impression that we can expect around 2-3 percent of the tokens in a novel to be names. For this small selection, it is noteworthy that the four novels with the highest percentages (around 5 percent) are all books for children or young adults. Furthermore, the English novels do not show up in extreme positions compared to the other ones.
Figure 2 gives more insight in the role of plot internal versus plot external names in the same set of novels. The novels are presented in the same order as in Fig. 1. We can see that the four books for children/young adults are still exeptional in the percentage of plot internal names. We also find that three of them have a rather small amount of plot external name tokens.
Fig. 2

Full Size Image
The paper will present the case of names in Gerbrand Bakker’s novel Boven is het stil / The Twin. Dutch readers have the impression that the novel does not contain many names, while readers of the English translation have the opposite impression and point out that especially geographical names abound. The comparative computational approach shows that both reader groups are correct and wrong at the same time. The novel has around 2% of name tokens, which is at the lower end of what seems to be normal. But in the amount of different names (lemmas), geographical names (place names) do have a special role when we look at the ratio between personal names and place names (Figure 3). The trend here is that a novel contains more different personal names than place names. In only two cases (Boven is het stil and its translation The Twin, and Fenrir) a novel has more different place names. This suggests that place names have an extra stylistic function in these two novels. It will be shown in which ways place names function in the Bakker novel and how this analysis enriches the understanding of the novel and of the stylistic possibilities of place names for an author.
Fig. 3

Full Size Image
We could only show a small part of all interesting observations to be made about the usage of names in a corpus of modern Dutch and English novels. But these first results make us anxious for more, in the expectation that this approach may lead to an acceptable method for a.o. across language comparison of stylistic elements.
Conclusion
We conclude that the preliminary results are sufficiently interesting to go into the stylistic analysis of name usage and functions in novels more deeply. Names also seem promising stylistic elements for a comparison across languages. The currently available tools which could be expected to be helpful for this type of research, proved to be insufficient. We therefore plan to develop a set of interrelated webservices which will assist the scholar in the recognition, categorization, further tagging, and statistical analysis of names in novels.
References:
Sekine, S. and Ranchod, E. 2009 Named Entities. Recognition, classification and use, John Benjamins Amsterdam/Philadelphia
Sobanski, I. 1998 “The Onymic Landscape of G.K. Chesterton’s Detective Stories, ” Proceedings of the XIXth International Congress of Onomastic Sciences, Aberdeen , 1996 373-378
Van Dalen-Oskam, K. 2005 “Vergleichende literarische Onomastik, ” Namenforschung morgen: Ideen, Perspektiven, Visionen, Brendler, A. und S. Brendler Baar Hamburg 183-191 (link)
Van Dalen-Oskam, K. 2006 “Mapping the Onymic Landscape, ” Il nome nel testo. Rivista internazionale di onomastica letteraria VIII; Atti del XXII Congresso Internatzionale di Scienze Onomastiche, Pisa, 28 agosto – 4 settembre 2005 93-103
Van Langendonck, W. 2007 Theory and typology of proper names, Mouton de Gruyter Berlin / New York