Crunching numbers in the 'Night Watches of Bonaventura': A computer-aided contribution to author's lexicography

“Crunching numbers in the 'Night Watches of Bonaventura': A computer-aided contribution to author's lexicography”

Barbara Arnold University of Exeter b.arnold@exeter.ac.uk

Analysing the language of a literary text is a complex undertaking. In order to obtain a complete survey of the literary idiolect of a particular writer, as defined by his use of individual word and syntactic style, author dictionaries have proved to be a reliable and informative source of data. Such data, in the form of a corpus, lends itself to analytical interpretation by providing a basis for literary and linguistic exegesis. This presentation will introduce a project, namely the construction of a dictionary based on a German novel of the early 19th century, 'Nachtwachen von Bonaventura' (Night Watches of Bonaventura'). Various aspects of computer-based research in author lexicography will be discussed, focussing on statistical methods that can be used as a basis for extracting characteristic features of style and lexis.

1. Preliminary notes

The creation of a dictionary based on the 'Night Watches of Bonaventura' presented itself as a rewarding and challenging task for a number of reasons. Firstly, it would make a contribution to the rather neglected area of author lexicography in general (Mattausch 1990, 1558). Secondly, because a dictionary of this particular text has not previously been compiled, although the novel is widely considered to be a valuable source for literary, linguistic and computational research. A dictionary of the 'Night Watches' appeared able to serve all of these purposes simultaneously. Being a systematically ordered collection of language material and with its XML-encoded structure, the lexicon is intended to be a prototype example of the capabilities of text-based computing in the humanities, offering easy access to semantic definitions as well as to a set of sense relations. Following decades of speculation concerning authorship of the novel, which was published under the pseudonym 'Bonaventura' in 1804, it can now be taken as certain that August Klingemann, who had formerly been regarded as a rather marginal figure in German Romanticism, indeed wrote the 'Night Watches'. The techniques employed to unlock the puzzle of who wrote as 'Bonaventura' ranged from frequency counts (Fleig 1985) to sophisticated computer-aided approaches (Wickmann 1974). Irrefutable evidence of Klingemann's authorship emerged when an autograph originating from him was found in 1987, citing the 'Night Watches' among his other works (Haag 1987). Knowing who 'Bonaventura' really was marked the beginning of a new epoch in research on the novel. Now it is the text itself that attracts scholars, as one of the most recent publications indicates (Kaminski 2001).

2. Crunching numbers in the ‘Night Watches of Bonaventura’

Compiling a dictionary and processing items of vocabulary lexicographically both require one simple question to be resolved at the very outset: "What is a word?" Defining the term 'word' as a primitive carrier of meaning that exists independently of the syntactical structure (Duden-Grammatik 1998, 870), is certainly an attractive assumption, but it provides no help in carried out computer-aided text analysis for a language like German, which is extraordinarily rich in displaying phenomena such as separable prefixes that appear as "words" in the data file but are not independent of other elements of the sentence. Neither is it clear whether abbreviations, numbers or symbols such as Greek letters, which are employed quite frequently throughout the novel, ought to be treated as other lemmata or extracted from the word list and presented in an appendix. Equally troublesome is the problem of proper names. One wonders whether proper names are proper words at all or whether their status within the system of language is somewhat different as they refer to individuals rather than general ideas (Lerner/Zimmermann 1991, 349). Reviewing the statistical aspects of the corpus material leads to some fascinating conclusions about the essential ingredients of a literary text. "Crunching numbers" in Bonaventura's 16 'Night Watches', the 16 chapters in which the novel is organized, works on several levels. Counting all the separate "word" units within each sentence produces a total of 38,785 tokens to be identified by a word-processing programme. Using Tustep to generate a Kwic-Index of these parts of speech produces a list of 8,522 types. This number shrinks again after having reduced the inflected word forms to their canonical form. This means that we can reduce the 'Night Watches' to a list of 5739 lemmata in terms of our dictionary. Not all words appear with equal frequency in the novel. Not surprisingly are there more contextual occurrences of articles, conjunctions and auxiliary verbs than of nouns and other mere grammatical words. With a total number of 6,692 occurrences for all the nouns they only form 17.25% of the whole novel, while this number is easily topped just by counting all the occurrences for the definite articles der, die, das (3,576 occurrences), the conjunction und (1,689 occurrences) and the pronouns er, sie, es(1,575 occurrences). Although these functional words clearly dominate the text, their semantic meaning remains mainly on the surface. Focussing on the most frequent nouns on the other hand is much more rewarding in terms of meaning and interpretational value. A survey such as that given in figure 1 produces a puzzling effect: Mensch ('human being'; 83 occurrences), Teufel ('devil'; 56 occurrences), Nacht ('night'; 53 occurrences), Kopf ('head'; 51 occurrences), Hand ('hand'; 50 occurrences), Zeit ('time'; 50 occurrences), Gott ('god'; 49 occurrences), Himmel ('heaven/sky'; 49 occurrences), Liebe ('love'; 47 occurrences) and Tod ('death'; 43 occurrences) are only ten different types, but they seem to be enough to express the content of the 'Night Watches', disregarding the rest of the novel.

Figure 1. Figure 1: frequency count of nouns

Another surprising figure is the percentage of those words that appear only once in the whole of the text: 54.8% of all lemmata are employed only once throughout the novel, leaving 3,146 entries in the dictionary with single occurrences. Some words show a significantly high number of occurrences concentrated in certain 'Night Watches' only. The verb lieben ('to love') for instance occurs 30 times throughout the whole of the novel, with an average number of zero to four occurrences per chapter. Although the semantics of 'love' are implicitly dealt with in almost each 'Night Watch', there is only one statistical peak in chapter XIV with 12 references close together. The antonymic hassen ('to hate') on the other hand appears only 12 times showing no significant peak at all (figure 2). Retrieving this sort of information might most conveniently be supported by a computational programme, e.g. Tom Horton's WordCluster ( http://www.cs.virginia.edu/~horton/cs494/examples/wordcluster/wdc/readme.txt) as the method applied so far, despite using Tustep, still requires several repetitive procedural actions and quite a considerable amount of tedious manual interferences.

Figure 2. figure 2: distributional relations of 'lieben / Liebe - hassen / Hass '

Expressing lexical material in terms of such precise statistics contributes to a deeper understanding of what poetic texts consist of. For the 'Night Watches' surveys like these become even more relevant, especially as doubts are still expressed as to the work's authorship. Assuming that Klingemann was involved in authorship, one recent publication nevertheless suggests that some parts might originate from another writer, Clemens Brentano (Braeuer-Ewers 1995, 13). Computer-generated comparative studies of a variety of narrative prose showing distributional relations like the above may certainly help to develop a further insight into the concept of an 'idiolectal fingerprint' (Cluett 1976).

3. Outlook

Analysing frequency counts and distributional relations are certainly not the only way in which a dictionary of Klingemann's 'Night Watches of Bonaventura' can be used in a linguistic context. As an illustrative article (figure 3) indicates, the dictionary can be used for different purposes.

Figure 3. figure 3: example of a dictionary article

Besides lemma, word class, frequency, definition and contextual reference there is a sophisticated system of cross-references at the bottom of each entry. Here hints on word-formation are given and sense-related pointers create a network structure of meanings. Each article is generated by consulting a number of lexicographical reference works. Comparing the semantic commentaries of dictionaries published during the 18th and 19th century (Adelung 1793-1801; Campe 1807-11; Grimm 1854-1961) with a reliable reference work of contemporary German (Duden 1999) leads to insights into the history of words, including aspects of lexicalisation and changes in meaning and usage. The dictionary on the ‘Night Watches of Bonaventura’ is just one part of a larger project dealing with August Klingemann. As mentioned above linguistic research has to a great extent concentrated on isolating typical idiolectal ways of expression within the novel for the purpose of proving authorship. Now that we know its author it makes sense to compare the vocabulary employed in the ‘Night Watches’ to that of other items of written work left by Klingemann. The application of IT facilities may provide interesting insights into these, too, pointing the way to further potential of computer-aided research in the field of author lexicography.

Johann Christoph Adelung. Grammatisch-kritisches Wörterbuch der hochdeutschen Mundart. Leipzeig: Breitkopf und Härtel, 1793. 5 vols.

Ina Braeuer-Ewers. Züge des Grotesken in den 'Nachtwachen von Bonaventura'. Paderborn: Schöningh, 1995.

Joachim Heinrich Campe. Wörterbuch der Deutschen Sprache. Braunschweig: Schulbuchhandlung, 1807. 5 vols.

Robert Cluett. Prose Style and Critical Reading. New York: Teachers College Press, 1976.

Duden. Das große Wörterbuch der deutschen Sprache. Ed. Dudenredaktion. Mannheim / Leipzig / Wien: Bibliographisches Institut, 1999. 10 vols.

Duden. Grammatik der deutschen Gegenwartssprache. Ed. Dudenredaktion. Mannheim / Leipzig / Wien: Bibliographisches Institut, 1998.

Horst Fleig. Literarischer Vampirismus. Klingemanns 'Nachtwachen von Bonaventura'. Tübingen: Niemeyer, 1985.

Jacob Grimm Wilhelm Grimm. Deutsches Wörterbuch. Leipzig: Hirzel, 1854. 32 vols.

Ruth Haag. “Noch einmal: Der Verfasser der Nachtwachen von Bonaventura.” Euphorion. 1987. 81: 286-297.

Nicola Kaminski. Kreuz-Gänge. Romanexperimente der deutschen Romantik. Paderborn: Schöningh, 2001.

August Klingemann. Nachtwachen von Bonaventura. Penig: Dienemann, 1804.

Jean-Yves Lerner Thomas E. Zimmermann. “Eigennamen.” Semantik. Ein internationales Handbuch der zeitgenössischen Forschung. Ed. Arnim von Stechow Dieter Wunderlich. Berlin: Walter de Gruyter, 1991. 349-370.

Josef Mattausch. “Das Autoren-Bedeutungswörterbuch.” Wörterbücher. Ein internationales Handbuch zur Lexikographie. Ed. Franz Josef Hausmann Oskar Reichmann Herbert Ernst Wiegand. Berlin: Walter de Gruyter, 1990. vol 2: 1549-1562.

Dieter, Wickmann. “Zum Bonaventura-Problem: Eine mathematisch-statistische Überprüfung der Klingemann-Hypothese.” Zeitschrift für Literaturwissenschaft und Linguistik. 1974. 4: 13-29.