This is a working paper at an early stage. All feedback is welcome, to alaric@cantab.net. The URLs of files linked to from this page may not be stable, so don't rely on them. The URL of this page will, however, contain or point to this article long-term and can, therefore, be cited.

References will be found at http://www.alarichall.org.uk/teaching/alaricsnotes.php. Where possible, references link to online versions of the texts cited.

A stemma of Nikulás saga leikara

Alaric Hall and Sheryl McDonald Werronen, University of Leeds

This is a nascent working paper, gradually establishing a stemma of the Icelandic romance saga Nikulás saga leikara. The saga survives in no medieval manuscripts (and so was omitted from Kalinke and Mitchell's 1985 bibliography of the genre), but does seem to have been part of the now fragmentary fifteenth-century manuscript Stockholm, Royal Library, Perg. fol. nr 7 (Sanders 2000, 17, 21). The saga does survive in over sixty post-medieval manuscripts, however, in two main recensions, and was also twice printed in popular editions: in Winnipeg by the Heimskringlu Prentstofa (1889) and in Reykjavík by Helgi Árnason (1912). This makes it an interesting example of Canadian-Icelandic literature, and an unusually late example of Icelandic readerships for printed romance sagas. The saga was edited by Wick (1996), on whose work our own depends; we also benefit from the fact that it happens to be particularly well represented at handrit.is. Despite the dearth of other commentary on the text in past scholarship, Nikulás saga leikara is both interesting in its own right and for its possible relationship with Nítíða saga fræga (McDonald Werronen 2013, 102--16). The project of completing this stemma complements our ongoing work on the manuscript and print transmission of a wider range Icelandic romance-sagas from the Middle Ages down into the early twentieth century, and what it can tell us about Icelandic scribes, readers, social networks, literature and culture (see Hall and Parsons forthcoming on Konráðs saga keisarasonar; McDonald 2009 and 2012, and McDonald Werronen 2013, on Nítíða saga; Hall working paper on Sigurgarðs saga).

We are establishing the stemma mostly using the methods developed by Hall and Parsons (forthcoming). The ongoing collection of samples from the saga is available here. A hint of what an unrooted stemma generated using the phylogenetic program Pars will look like en route to a fully worked out stemma is afforded by this interim example:

The main methodological advance afforded by this paper arises from our use of Wick's edition: Wick gave complete variants from six early manuscripts of the longer of the two recensions of the saga. We have rendered these variants as a spreadsheet. This is probably the first stemma of an Icelandic saga to be based on all variants from the oldest manuscripts. (NB we did not attempt to clean up the sometimes faulty optical character recognition of Wick's thesis in making this, so the text in the spreadsheet sometimes looks a bit grisly; but the analysis was based on consultation of her actual, and correct, text.) The saga (counting in the A-text) is about 5,000 words long (5,104 according to the word-count, but the OCR errors mean this won't be completely accurate). From this we generated this Pars infile, recording 2039 sites of variation ('parsimony-informative characters') in these six manuscripts (thus one parsimony informative character per 2.5 words), with an average of 2.3 variants per character. Compared with later manuscripts of the same saga, this is a fairly low rate of variability: we are dealing in this section of the Nikulás saga tradition with very conservatively copied manuscripts, or with a high surviving proportion of the manuscripts, or both. We will be using this dataset to develop Hall's past work on sampling in stemma-making (Hall and Parsons forthcoming). A few quick and dirty experiments suggest that te result is stable down to 150 characters, but then the software starts to produce multiple possibilities.

The stemma for the top of the saga produced by Pars, rendered using Drawgram, using the factory settings and the full dataset is:

But we'll be subjecting this to some human analysis too!

Using weights function in Pars, I sampled every Nth character of the six-manuscript dataset. This, then, took an even sample throughout the dataset. NB that I need to add finite numbers of characters into this table, as the proportion of the text sampled is not in itself very important, I reckon (Hall and Parsons 2013, §XXXXX). It would be interesting to run this again adding one 0 to the beginning of the weights file (and taking one character off the end obviously) so as to examine a completely different dataset for each proportion; I've done this for a few, labelling the weights file starting with '10...' as 'a', the file starting with '01...' as 'b', the file starting with '001...' (where possible) as 'c', etc.

proportion of characters # stemmas drawgram concensus tree structure same as tree for 100% sample?
1/1 1 Y
1/2a 1 Y
1/2b 1 Y
1/3a 1 Y
1/3b 1 N
1/3c 1 Y
1/4a 1 Y
1/4b 1 Y
1/5a 1 Y
1/5b 1 Y
1/6 1 Y
1/7 1 Y
1/8 1 Y
1/9 1 Y
1/10 1 Y
1/11 1 Y
1/12a 3 Y
1/12b 3 Y
1/13 1 Y
1/14 1 Y
1/15 1 Y
1/16a 2 Y
1/16b 3 N
1/17a 1 N
1/17b 1 N
1/18a 1 Y
1/18b 1 N
1/19 2 N
1/20 2 N
1/21 1 Y
1/22 1 N
1/23 2 N
1/24 1 N
1/25 1 N
1/26 2 N
1/27 1 Y
1/28 1 N
1/29 1 N
1/30 2 Y
1/31 2 N
1/32 2 N
1/33 2 N
1/34a 15 N
1/34b 2 Y
1/35a 1 Y
1/35b 1 N
1/36 1 N
1/37 2 N
1/38 2 N
1/39 4 N
1/40a 17 N
1/40b 1 N
Even with a third of the dataset, it is possible to come to a different conclusion from the dominant one. Conversely, results consistent with the full dataset occur right down to 1/35 (58 characters). But it is clear that the reliability generally starts to falter around 1/15 (136 characters). Almost invariably, the positions of A and D are clear, and it is clear that B, C, E, and F have a lost common ancestor, but the precise relationships between these four are problematic. So then I took contiguous 150-character chunks, and these were the results (NB the last 89 characters of the saga aren't represented here because they didn't make up a whole chunk. I need to run it again but counting from the end):
150-character chunk # stemmas drawgram concensus tree structure same as tree for 100% sample?
1 1 N
2 1 N
3 1 N
4 1 N
5 3 N
6 1 Y
7 2 Y
8 3 N
9 1 N
10 2 Y
11 1 N
12 1 Y
13 1 N
Again, A and D are generally fine, but there's a lot of variability in dealing with the other four.
300-character chunk # stemmas drawgram concensus tree structure same as tree for 100% sample?
1 4 N
2 1 N
3 2 Y
4 1 N
5 2 Y
6 (only 238 chars) 1 N


Acknowledgements

This project is indebted to the kindness and hospitality of the National Library of Iceland, the Stofnun Árnamagnússonar in Reykjavík and the Arnamagnæanske Hꜹndskriftsamling in Copenhagen. Sheryl's involvement was funded by a Leeds University Teaching Fellowship held by Alaric.