Algorithmic Empathy: Nick Montfort’s ‘Megawatt’

  • Hannes Bajohr
  • ·
  • 2015-01-01

On January 10, Frohmann Verlag will publish my novel Durchschnitt (“Average” – I will make an English version at some point), my conceptual take on the canon. In the context of 0x0a, conceptual digital literature has mostly been discussed as either leaning towards poetry or lacking any genre – in any case, removed from the realism that tends to reign in the novel. Yet it is worth exploring what digital literature might contribute to the novel itself (or the other way around).

 


Every year since 1999, November has been America’s National Novel Writing Month or NaNoWriMo. As a collective enterprise in self-discipline, it is supposed to help amateur authors “dust off their literary ambitions and finally write that novel they’ve always dreamed of.” The condition: To produce a novel of 50,000 words or more in a month, regardless of coherence, relevance and overall quality. On NaNoWriMo.org, one can find incentives in the shape of boyscoutish “badges” for word milestones already achieved.

What is interesting about it is that the novel writing month, acting prophylactically against procrastination, uses two types of constraint: the time limit and the word target. These are reminiscent of the self-limitations typical of conceptual literature. However, these constraints don’t touch upon the formal aspects of the novels, which often end up as fan fiction things or – more augustly – as exactly that type of realism the ‘great novel’ is meant to achieve, even if in reality it is nothing but a simulation of the gestures of seriousness that the literary field considers exemplary of ‘high’ fiction. In the US, this includes the boot-campishness of what’s called creative writing, which conceives challenges like the NaNoWriMo’s as an obstacle courses one fights through to arrive at realism (No Plot? No Problem! is the title of the literary self-help book by NaNoWriMo founder Chris Baty – which doesn’t intend to keep the writer happily plotless, but to encourage him to just sit down, write, and let it emerge).

Of course, this form of literature is conceptual in potential only. One could simply add on formal limitations, so as to encourage a flurry of conceptual novels, a yet-to-be founded NaCoNoWriMo. Or one could limit the means of production itself, as did code artist Darius Kazemi. His version is called NaNoGenMo – National Novel Generation Month. Transferring the “creative” from the novel to a novel making code, it lets the machine do the writing. Already for the second time, hundreds of code savvy writers pledged to design scripts and programs that would generate just that 50,000 word novel instead of writing it themselves. (‘Novel’ is understood pragmatically, paratextually: If it’s called a novel, it is one.) The results have been surprisingly varied, and some of them quite extraordinary. In the coming weeks, I would like to look at some of them, hoping to find out something about digital literature, what it is, how it works – and, above all, what it can do for the novel.

I’ll start with a captivating work by Nick Montfort (recently mentioned here as a practical-minded code historian of, among others, Brion Gysin). He generated the novel Megawatt. Both reconstruction and intensification of Samuel Beckett’s highly artificial novel Watt, Montfort’s novel takes passages from the original that exhibit systematic mannerisms, and simulates them with a Python script. But instead of simply generating automatically what Beckett had written manually (which would already be – Pierre Menard style – quite extraordinary), he took Beckett beyond Beckett and extended these passages according to the immanently derived rules of the urtext. There is, for example, a short passage at the beginning of Beckett’s book, in which the protagonist, Watt, is unable to follow what’s being said to him because he hears voices in his head:

Now these voices, sometimes they sang only, and sometimes they cried only, and sometimes they stated only, and sometimes they murmured only, and sometimes they sang and cried, and sometimes they sang and stated, and sometimes they sang and murmured, and sometimes they cried and stated, and sometimes they cried and murmured, and sometimes they stated and murmured, and sometimes they sang and cried and stated, and sometimes they sang and cried and murmured, and sometimes they cried and stated and murmured, and sometimes they sang and cried and stated and murmured, all together, at the same time, as now, to mention only these four kinds of voices, for there were others. And sometimes Watt understood all, and sometimes he understood much, and sometimes he understood little, and sometimes he understood nothing, as now.

After a quick glance, you will notice that Beckett employs one of the simplest rules of generating text: the permutation of combinatory possibilities from a finite set of elements (just as Gysin had done in his permutation poems). The voices have four possible states – they “sang,” “cried,” “stated,” and “murmured” – which they can adopt alone or together with other states, and Beckett runs through their combinations. Additionally, Watt understands “all,” much,” “little,” and “nothing” of what they say.

Thus, this paragraph resembles two functions – one for the voices and one for the understanding – which could, fully automated, be produced by a script with the same result. This is exactly what Montfort does in the first chapter of Megawatt, “The Voices.” But since Beckett concedes that more voices are possible (“for there were others”), and because Montfort knows that in a series of permutations the number of possibilities increases exponentially with each additional element, he adds four more verbs to Beckett’s: “babbled,” “chatted,” “ranted,” “whispered.” Likewise, Watt now can understand also “most,” “half,” “less,” and “bits.” As a result, Montfort’s version is, at seven pages, about twenty times longer than the original paragraph. This is just the first page, for which I put in black text the retained original sentences – you can see how the distances between them increase, because more elements result in longer series of combinations:

Watt heard voices. Now these voices, sometimes they sang only, and sometimes they cried only, and sometimes they stated only, and sometimes they murmured only, and sometimes they babbled only, and sometimes they chattered only, and sometimes they ranted only, and sometimes they whispered only, and sometimes they sang and cried, and sometimes they sang and stated, and sometimes they sang and murmured, and sometimes they sang and babbled, and sometimes they sang and chattered, and sometimes they sang and ranted, and sometimes they sang and whispered, and sometimes they cried and stated, and sometimes they cried and murmured, and sometimes they cried and babbled, and sometimes they cried and chattered, and sometimes they cried and ranted, and sometimes they cried and whispered, and sometimes they stated and murmured, and sometimes they stated and babbled, and sometimes they stated and chattered, and sometimes they stated and ranted, and sometimes they stated and whispered, and sometimes they murmured and babbled, and sometimes they murmured and chattered, and sometimes they murmured and ranted, and sometimes they murmured and whispered, and sometimes they babbled and chattered, and sometimes they babbled and ranted, and sometimes they babbled and whispered, and sometimes they chattered and ranted, and sometimes they chattered and whispered, and sometimes they ranted and whispered, and sometimes they sang and cried and stated, and sometimes they sang and cried and murmured, and sometimes they sang and cried and babbled, and sometimes they sang and cried and chattered, and sometimes they sang and cried and ranted, and sometimes they sang and cried and whispered, and sometimes they sang and stated and murmured, and sometimes they sang and stated and babbled, and sometimes they sang and stated and chattered, and sometimes they sang and stated and ranted, and sometimes they sang and stated and whispered, and sometimes they sang and murmured and babbled, and sometimes they sang and murmured and chattered, and sometimes they sang and murmured and ranted, and sometimes they sang and murmured and whispered, and sometimes they sang and babbled and chattered, and sometimes they sang and babbled and ranted, and sometimes they sang and babbled and whispered, and sometimes they sang and chattered and ranted, and sometimes they sang and chattered and whispered, and sometimes they sang and ranted and whispered, and sometimes they cried and stated and murmured, and sometimes they cried and stated and babbled, and sometimes they cried and stated and chattered, and sometimes they cried and stated and ranted, and sometimes they cried and stated and whispered, and sometimes they cried and murmured and babbled, and sometimes they cried and murmured and chattered, and sometimes they cried and murmured and ranted, and sometimes they cried and murmured and whispered, and sometimes they cried and babbled and chattered, and sometimes they cried and babbled and ranted, and sometimes they cried and babbled and whispered, and sometimes they cried and chattered and ranted, and sometimes they cried and chattered and whispered,…

Montfort’s own creative contribution consists of the first three words, the merely expositional first sentence. The rest – Beckett’s text as well as the extensions – have been generated through code alone. Its output is, first, what Beckett had written (the text in black) – but then not only what he could have written, but also what he must have according to his own rules (the gray text). Megawatt is thus a form of algorithmic empathy, which is not a copy but a reconstructive comprehension which can claim that it was done in the spirit of Beckett with more legitimacy than any epigonal text, any parody or pastiche ever could. Megawatt, for Watt, comes close to what Borges’s/Menard’s Don Quixote is for Cervantes’s Don Quixote. It is also – because it comprehends not only reconstructively but also productively – what Ulysses is for the Odyssey – a beyond that exceeds its model.

On the one hand, Montfort confirms Jessica Pressman’s thesis that digital literature is returning to the constructive mechanisms of the historical avant-gardes while, as “digital modernism,” executing them with more appropriate means and much more consistently; on the other, he also shows how even the strictest re-enactment of rules already contains the seed of productive growth, which might precisely result from the stubbornly narrow-minded algorithm that is uncreative in appearance only.

For literary criticism, Megawatt should belong to the paradigmatic text-ontological border and example cases that come up in seminars on “What is Literature?” It is the best possible embodiment of how digital and analog literature differ. For this difference lies in a twofold concept of the text: While Watt literally consists of only one text – any intertextuality notwithstanding – Megawatt consists of two: the code and the output. So far, I have only talked about the output, the ‘novel.’ But at the end of his book, Montfort shows us the code that created it – and which you could take and put through a Python interpreter yourself, re-enacting the performativity of the twofold text. What this show is that in the Digital, as is 0x0a’s basic assumption, text is action and thought.

Thus, in Megawatt’s appendix, you will find the mere 27 lines of code that generate the seven pages of “The Voices.” After defining the ‘combine function, which puts together the final text (lines 3-12), Montfort first shows how to conceive of Beckett’s own text as a set of elements in a Python array (lines 14-15; non-executed lines are commented out with a “#”); after that, he introduces his own, extended set of elements (lines 16-18). The rest of the script puts these elements together (lines 21-27) and adds the connecting phrases Beckett had used as well (e.g. “and sometimes they,” line 20.)

 	1  #### THE VOICES
	2  text.append('\n# I\n\n')
	3  def combine(num, words):
	4     final = []
	5     if num > 0 and len(words) >= num:
	6         if num == 1:
	7             final = final + [[words[0]]]
	8         else:
	9             final = final + [[words[0]] + 
	10            c for c in combine(num – 1, words[1:])]
	11        final = final + combine(num, words[1:])
	12    return final
	13  
	14 ## In Watt the voices = ['sang', 'cried', 'stated', 'murmured']
	15 ## And Watt understood = ['all', 'much', 'little', 'nothing']
	16 ## Here the voices did eight things and there are eight levels:
	17 voices = ['sang', 'cried', 'stated', 'murmured', 'babbled', 'chattered', 'ranted', 'whispered']
	18 understood = ['all', 'most', 'much', 'half', 'little', 'less', 'bits', 'nothing']
	19 para = ''
	20 preface = ', and sometimes they '
	21 for num in range(len(voices)):
	22    for word_list in combine(num + 1, voices):
	23        para = para + preface + ' and '.join(word_list)
	24        if len(word_list) == 1:
	25            para = para + ' only'
	26 para = ('Watt heard voices. Now these voices,' + para[5:] +
	   ', all together, at the same time, as now, to mention ' + 
	   'only these ' + spelled_out[len(voices)] + ' kinds of voices, for ' + 
	   'there were others. And sometimes Watt understood ' + 
	   ', and sometimes he understood '.join(understood) + ', as now.')
	27 text.append(para) 

This deeper text, the code behind the output, also illustrates the inherent proximity of code poetry and conceptualism – the relation of idea and execution is analogous to that of code and output. It is, however, only an analogy, not an identity, for the output is determined finitely (as long as there are no random variables). In conceptualism there is still the erratic (or even: expressive) moment in which the idea is executed, which can yield different results from the same concept (contrary to what Sol Lewitt claimed, the result is not “perfunctory” after all).

Of course, this is simply the first chapter of the book, which deserves a thorough analysis. Its nine chapters, except for the last, are all forms of permutation, which are employed in very different and effective ways – from a mega-biblical genealogy in the fourth chapter (“my father’s mother’s mother’s mother’s and my mother’s father’s father’s father’s and my mother’s father’s father’s mother’s [earth]”) to the eighth, an absolute mega-chapter that lists Mr. Knott’s ever changing appearance. It runs from page 25 to page 239 and must be among the longest character descriptions in literature, “to mention only the figure, stature, skin, hair, eyes, body type, facial hair, and posture.”