Hyperrhiz 24

Biometric Poetics: The Case of Eververse

Justin Tonra
National University of Ireland Galway

David Kelly
National University of Ireland Galway

Citation: Tonra, Justin and David Kelly. “Biometric Poetics: The Case of Eververse.” Hyperrhiz: New Media Cultures, no. 24, 2021. doi:10.20415/hyp/024.e04

Abstract: Eververse was a yearlong conceptual poetry project which used a poet’s biometric data as the basis for generating verse. This article describes the project’s conceptual contributions to the field of electronic literature and its technical development. Eververse operated by collecting biometric data from the poet with a commercial fitness tracking device; this data was sent to a custom-built poetry generator which deployed a number of processes from the domains of Natural Language Generation and Sentiment Analysis to generate poetry; the form and content of this poetry was designed to vary according to specific changes in the biometric data, resulting in a poetry that conspicuously correlated with the poet’s daily activities; this poetry was published in real-time on the project website and the full poem and associated data have now been archived. In addition to providing details on the technical implementation of Eververse, this article includes discussion that situates the work within the tradition of electronic literature and analyses its unique inscription of biometric data. The article examines that feature in the contemporary context of the quantified self, but also in its engagement with historic poetic theories of composition, creativity, and the textualisation of the body.

Keywords: biometrics, poetry, poetics, electronic literature, quantified self, text generation

Introduction

Eververse was a yearlong project which synthesised perspectives from the humanities and sciences to develop critical and creative explorations of poetry and poetic identity in the digital age. Deploying tools and methods from poetic theory, data processing and analysis, and Natural Language Generation (NLG), Eververse used data from a quantified self device to automatically generate and publish poetry which correlated with the poet’s varying physical states. The poet is a creative vessel or conduit, according to a common theory of poetry, absorbing the sensory input of the world and producing poetic output in turn. Responding in literal fashion to such theories of poetic embodiment, Eververse used biometric data from a commercial fitness tracking device (Fitbit) worn by the poet in a custom-built poetry generator. This generator utilised a number of different NLG techniques to output poetic text whose form and content varied in response to different physical sensations and experiences in the poet’s waking and sleeping life, as manifested in the biometric data. The resulting poetry was published in real time on the project website and the yearlong poem and its accompanying data are now archived there.

The formal changes employed by Eververse were influenced by theories of poetry which linked proprioception and expression. For example, following Charles Olson’s injunction that “the line comes ... from the breath, from the breathing of the man who writes” (291), Eververse’s poetic lines decreased in length as the poet’s heart rate increased and breath contracted. Similarly, the response to the randomness of the dream sleep (REM) state was an increased fragmentation and irregularity in the poetic form. Content, too, reflected these variations, as heightened-sentiment vocabulary was generated to reflect the emotional intensification implied by an increased heart rate, while REM sleep generated the surreal vocabulary and images characteristic of dreams. In a broader figurative sense, Eververse’s conceptual resolve was to explore the role of the body in literary creativity and composition, to query its assumed subservience to the mind in such processes, and to compose a poetry which sought to bypass the cognitive. Its task was to inscribe the corporeal poetics that Walt Whitman observed in the gait of the “well-made man”: “To see him pass conveys as much as the best poem, perhaps more.” In these understandings of poetry, art collapses into the being and identity of the artist, a conundrum articulated in W. B. Yeats’s question, “How can we know the dancer from the dance?”

Eververse’s unique contribution to the tradition of electronic literature is its textualisation of biometric data. Past works of conceptual and media art have used biometric data as the basis for different modes of creative expression. Mark Boyle and Joan Hills’ Son et Lumière: Bodily Fluids and Functions (1966) featured the projected oscilloscopes of electrocardiogram (ECG) and electroencephalogram (EEG) devices, revealing the real time heartbeats and brain waves of a couple making love. Christian Boltanski’s Les Archives du Cœur (2008-) is a museum of recorded human pulses where a single lightbulb illuminates in chorus with an audio recording of a heartbeat playing at a given moment. Eververse shares some of the conceptual thrust of these works, combined with a conscious echo of the long history of influence between physiological science and art and design. From Greek physician Herophilos’s (335-280 BCE) comparison of the heartbeat’s rhythms to the metrical rhythms of music, to Étienne-Jules Marey’s (1830-1904) experiments to invent a device to inscribe the rhythm of the pulse, persons throughout history have observed bodily rhythms and theorised or actuated their expression in media outside of the body. While continuing and drawing upon these customs, Eververse also engages with historical and contemporary traditions of electronic literature. Different aspects of the project’s generation of poetry acknowledge different forms of electronic literature, at once revealing its attempts to synthesise diverse models of text generation and demonstrating the difficulties of establishing immutable typologies within the vast universe of electronic literature. This article seeks to examine in more detail Eververse’s novel contributions to the textualisation of biometric data and to elucidate its position within the field of electronic literature, while describing and analysing the different stages of its practical development and implementation.

Development and analysis

In this part of the paper, we describe and analyse the development of Eververse. Sections below examine the project conception; collection of biometric and poetic data; development of the text generator; and format and design of generated poetry.

Conception

Eververse was conceived as a project which would generate poetry from biometric data. As is the case with many creative endeavours, the precise moment of inspiration is hazy and remembered now as a coalescence of several influences and intentions. Biometric data has been used as the basis for other types of artistic and cultural expression, but this is the first time, we understand, that a work has sought to textualise biometric data. Each of the examples described above, whether directed by creative impulse or medical utility, sought to bridge a gap between physiological rhythms and their subsequent manifestation outside the body. Marey’s instrument (Figure 1) served the functional medical purpose of measuring a subject’s pulse and recording it in a legible form, while Boyle and Hills’ artwork used the same logic to produce more abstract and metaphorical inscriptions of live bodily rhythms.

The gaps bridged by these representative examples are the same, but the resulting inscriptions differ in their purpose. We know the normal range of heartbeats per minute and recognise a medical problem that requires attention if the rate falls outside this range. Such an aberration may register in the visual inscriptions created by Marey’s device and Boyle and Hills’ artwork, but a doctor would favour the relative precision of the former in a diagnostic situation. Son et Lumière is no less an inscription of bodily data—heartbeats and brainwaves—though that inscription is inscrutable. In this broad context of how physiological phenomena might be differently recorded and inscribed, Eververse seeks to place its conceptual and critical intervention.

Further development of the project’s concept emerged from discussions among an interdisciplinary team comprising a literature scholar and poet, a designer and developer, and two computer scientists. The initial outline for Eververse was to build a system which would gather live biometric data from the poet and to generate and publish poetry which corresponded in form and content to fluctuations in that data. This textualisation of biometric data would emulate the type of relationship between bodily data and inscription evident in Son et Lumière: expressing a correlation between heartbeat and creative output which is broadly legible after an interval of observation, but whose text is the noise that obscures the heartbeat’s functional numerical signal. Though obstructive to medical science, this noise amounts to the germ of artistic style. The difference between the numerical heartrate and its inscription in the peaks and plateaus of an electrocardiogram holds the potential for imaginative expression: the form of the inscription indicates its functional or creative purpose. As Boyle and Hills, Boltanski, and others used a visual inscription, Eververse’s would be textual. Its resulting utterance, and its conceptual claim, thus, would be: this is what the heart says.

Biometrics

The next task for the project team was to identify a device which would record the poet’s biometric data and make it available for access and development. After the proprietary data of clinical monitors proved intractable, the group settled on using a commercially available Fitbit fitness tracker, which has lower barriers to data-access. This device captures the wearer’s heartrate, sleep patterns, step count, and calories burned, among other metrics. Data gathered by the device synchronises with Fitbit’s web-based service, which has the primary function of making data-based insights available to the user. An additional service makes the raw data available via an API (Application Programming Interface), with the user’s permission; through this mechanism we extracted the poet’s biometric data in JSON format. Subsequently, the team reviewed the attributes of this data and created sets of rules for the output text based on mapping these data attributes to prospective elements in the text.

While our data requirements were satisfied by the Fitbit device and API, a conceptual gap between biometric data and poetry remained to be spanned. That connection relies on an understanding of poetry as a uniquely personal, individual expression. The individualism implied by this Romantic grasp of literary creativity fails to account for the perception of poetry as a necessary reckoning with poetic tradition. Developments in the corpora and text generator, discussed below, address this apparent conceptual discrepancy. However, Eververse also exploits the constructive ambiguity of that gap for its conceptual purpose. For, in that individualistic sense, it presents a solely bodily poetry: one that sidesteps the usual cognitive processes of poetic composition and foregrounds the expressive body. With Whitman, Eververse acknowledges the expressive multitudes of the human body: designed to allow the poetic body to speak, it allows readers to observe and deduce the body’s daily rhythms inscribed in verse.

Moreover, Eververse provides an oblique commentary on the increasing presence of the quantified self in our daily lives. While self-tracking functions as another means of voluntarily feeding surveillance capitalism, the poem routes personal data into something fundamentally valueless in market terms: a defiant example of how “poetry makes nothing happen.”

Once we had access to a source of data for a range of the poet’s bodily functions, the project team was able to begin the process of conceptualising how those data types might be mapped onto variable poetic forms. We decided that two types—heartrate and sleep data—were the most effective and appropriate sources. These were consistently collected throughout the day: the heart rate was recorded continuously, and sleep data was collected while the poet slept. Thus, the combined sources provided continuous daily data, while other biometric categories recorded by the Fitbit device, such as calories and step count, were intermittent and unpredictable. Before determining the precise relationship between biometric data and generated poetry, we needed to establish the method used to generate the text. Given our desire to generate a poetry which would include a recognition and acknowledgement of poetic tradition, we opted to use a number of NLG techniques that would generate a derivative, as opposed to a formulary work. Once established, that decision required the construction of a corpus from which to derive the poem’s generated language.

Corpora

While many works in the early history of electronic poetry relied on a formulary method of text generation which used dictionaries, lexicons, and vocabularies, more recent trends in poetry generation have depended on human-created text as an integral component in the generative method (Oliveira). To enable these methods, the Eververse poet set about building a corpus of extant poetry which matched the conceptual principles of the project. Since one of these principles was to circumvent the cognitive aspects of composition, none of the poems included in the corpus was an original work created by the Eververse poet. Chosen from across the span of five centuries of poetry in English, the content of the poetry was intended to enable the generation of a text which would reflect its somatic source. Instead of a single corpus, three corpora were created, each curated and built by the poet in accordance with a specific complementary theme. The corpus used to generate poetry during the poet’s waking hours comprised poetry on the theme of the body. A night- and sleep-themed corpus was used while the poet slept, and a corpus of surrealist poetry was the basis of the abstracted verse generated while the poet was dreaming. In the generated poetry, described in more detail below, phrasal traces of the corpora were preserved. Thus, the text of Eververse reflects the diversity of the corpora, which contain poems united by theme from across the span of modern poetry in English. In this fashion, the poem is derived from and alludes to poetic tradition: it is not generated from the broader storehouse of language, in the manner of some formulary electronic poetry.

Eververse is polyvocal, with different voices speaking in different personae, tenses, dialects, and persons of speech. This multiplicity might seem disconnected from the project’s conceptual intent to give poetic expression to the bodily variations and rhythms of a single authoring subject. However, it is intended to reflect and give expression to the natural polyvocality inherent in the development of the human being. Mikhail Bakhtin, theorist of the novel’s polyvocality, acknowledged such growth independent of the genres of poetry and prose: “The ideological becoming of a human being … is the process of selectively assimilating the words of others” (341). In modernist literature, too, we see a poetry that emulates the polyvocality archetypally associated with novel: most notably in The Waste Land (“He do the Police in Different Voices”). Whitman, again, may be the most faithful avatar for Eververse, giving poetic expression to the affinities of the fleshly body and the ineffable poetic self: “O I say these are not parts and poems of the body only, but of the soul, / O I say now these are the soul!”

Iteration

This point in the development of Eververse marks a pivot between fixity and iteration. As the project developed, the method used to generate the poetry evolved. The data gathering method and underlying corpora remained consistent, but we proceeded to explore ways of refining the textual expression of the data. Eververse iterated through multiple versions which appeared in different contexts, including live performance and exhibition. Each mode of presentation was refined to suit the context in which it was experienced and, consequently, different versions of the project’s generated text exist, though all derive from the same underlying data and corpora. The archived version represents text from the most recent evolution of the generator.

The necessity of highlighting these different versions and expressions of Eververse serves to emphasise another essential difference between printed and electronic poetries. The advent of electronic poetry represented a decisive moment for understandings of textual stability in literature, argued Jean-Pierre Balpe, where “e-poetry set the change and the variation as the central rule of its writing.” Balpe located the essence of the change in electronic poetry’s new relationship to time and movement, which manifested in the display of the poetic text and in its language, which “plays on variation, change, time, ambiguity, hesitation, infinity, and so on.” The notion of the poetic text’s primacy and stability is undermined, and, as we discuss below, the work of electronic poetry makes meaning through an expanded view of the relationships between its constituent parts: concept, data, algorithm, and text.

Generator

Given stable data and fixed corpora, the potential for generating poetry was vast. We determined the eventual form of the poetry through the design of the text generator: an iterative process implementing poetic ideas in code, testing, and refining until we reached the ideal output for the specific context, whether performance, exhibition, or web display. The use of a text generator seems to position Eververse within a certain view of electronic literature: it is computer poetry and not computer-assisted poetry. Bootz theorises this distinction in the difference that is realised by a text generator in its intervention between potential literature and its generation of a fixed and static instantiation of that potential. A work of computer poetry such as Eververse is potentially vast, expansive, and indeterminate in its conceptual state. Once its input biometric data has been passed to the generator (texte-écrit, in Bootz’s formulation), the resulting output (texte-à-voir) domesticates the unruly potential into fixed and static poetic text. In the interrelationship of the project’s different parts, their equivalent importance becomes apparent. For example, with the concept and the generator combined, we are still in the domain of potential literature. Only with the addition of the poet’s biometric data is literature realised. Similarly, situating that literature outside of the framework of its creation diminishes its meaning, underscoring its provisionality and illusion of fixity, as Bootz suggests: “A generator can only be given to be read on a computer and any display of a ‘generated text,’ outside of its generation context, is as significant an abbreviation and a deviation, as the display of a poster or a photograph instead of a film”  (126).

The Eververse text generator incorporates features which control aspects of the text being output in response to attributes of the input data: biometric zones, formal manipulation, and sentiment alteration. We discuss each of these below, but the initial process involved the generation of new text from the corpora. For this step, we used Poesy, an open source Python library developed by Anthony Federico. Poesy enables text generation by modelling a given corpus with reverse n-grams and building a generated poetic text from the end of the line, rather than the beginning. This feature is particularly beneficial for generating rhyming verse, and Eververse used its “generateCouplet” function to generate a single rhyming couplet for each biometric data point.

The use of reverse n-grams determines the nature of the relationship between the corpora and the output text. By this method of disassembling the corpora and reconstituting their parts into generated text, the poetic themes present in the corpora emerge in the phrasal traces preserved in the n-grams. The perpetuation of the three corpora’s themes—body, night, sleep and dream—is more fragmented in the generated poetry, arising from the juxtaposition of smaller pieces of language and their resulting images. Thus, poetry generated from the bodily corpus is replete with stray images of arms, legs, thighs, and breasts. The generated verse offers a literary game of recognition for the avid poetry reader in identifying notable phrases and images amputated from their original context and repositioned within a new poetic matrix: “stately pleasure dome,” “walks in beauty,” and so forth. The reproduction, citation, and modification of portions of poetry from the corpora represents the analogous process by which experience calls to mind fragments of the poet’s own storehouse of reading and poetic imagery. The process, thus, fulfils the project’s conceptual desire to acknowledge poetic tradition and engage with its anxieties and ecstasies of influence.

Biometric zones

The organising principle of the generator is a series of biometric zones defined by the project team and designed to act as the catalyst for establishing a legible correlation between the poet’s biometric data and the resulting poetic output. This key feature of the generator maps the heartrate and sleep data recorded by Fitbit to a set of pre-defined zones. Each zone assigns rules for generating the poetic text: these govern the corpus to be used, the length of the verse line, the use of punctuation and in-line spacing, and the intensity of sentiment in the poetic vocabulary (Table 1). Those variations in the form and content of the poetic output address more fine-grained conceptual purposes, discussed below, but the biometric zones provide the overarching architecture by which those purposes are achieved.

Table 1: Biometric zones

This biometric architecture is central to the project and to the ability of the system to generate poetry. However, as we note above, that poetry is only realised with the addition of an individual’s unique biometric data. The poetry generated and archived by Eververse is the system’s response to the biometric data of one individual: one poet. We recognise that the key biometric determinants of heart rate and sleep rate are variable between different individuals and are influenced by a range of personal, social, and environmental factors such as age, gender, fitness, health, and stress. The same heart rate might indicate different levels of exertion in persons with different levels of fitness. This variability might colour how the reader of Eververse interprets variation in the verse, but they should bear in mind that these shifts are relative. Given two people with identical degrees of physical conditioning, the system would still generate different poems owing to the native randomness of the generator. Eververse will generate poetry for any given individual through customisation of the biometric zones to the range of their heart rate, rather than using a fixed biometric architecture. Thus, the customisable nature of the Eververse system accounts for the different personal, social, and environment factors which can meaningfully impact a person’s heart and sleep rates.

Format

The linguistic content of Eververse’s poetic text is determined by the corpus that is operative at a given point of the day and by its arbitrary algorithmic reassembly by the Poesy module in the generator. Further changes are applied to intensify linguistic sentiment, which we discuss below, but the format of the poetry is an element which is more precisely controlled by our generator design. Thus, the form of Eververse’s poetry offers the most immediately legible correlation with its underlying biometric data.

First, the poem’s lines lengthen and shorten to correspond with the poet’s increasing and decreasing heartrate, in a visual correlate for the peaks and troughs of the electrocardiogram. This lineation is achieved by mapping the biometric data input to preconfigured line-length values, as illustrated in Table 1. Once the appropriate number of words for a line is identified for a given datapoint, this value is passed to Poesy’s “generateCouplet” function, which includes it in its text generation process. Thus, the work’s resulting visual form and mise-en-écran (Tether) are an immediate signifier of the poem’s conceptual message.

Poetic output is arranged in free verse. The choice of a verse form determined by the modulations of lineation, rather than an accentual or syllabic system, arises from a further conceptual connection of poetic form and physiology: specifically, that which links the length of the verse line and the poet’s breath. Olson’s theories which connect lineation and poetic breath also testify to a wider movement away from the perceived constraints of metered poetry in his time. Following this explicit correlation established by Olson, Eververse’s lines decrease in length as the poet’s heart rate increases and breath contracts. The correlations of biometric zones with word count represents a consistent principle of quantification that has been applied to the poem. Its reliance on words, rather than syllables, introduces a degree of rhythmic variation in the poetic output which acknowledges the wide array of potential causes for changes in biometric data. An increased heartbeat might occur as a result of physical exertion, medical emergency, or love. The rhythms of everyday life influence bodily response in ways that are not transparently legible in biometric data. Thus, the poetry produced by that data should contain subtle variations, which may be less apparent in the regularity of a system based on syllabic verse. Furthermore, the use of unmetered verse has crucial conceptual associations with spontaneity and unconscious expression. Too often defined in the negative, free verse is a form suited to spontaneity, giving shape to poetry that is unfettered by the deliberation and consciousness required to fit its language to prescribed metrical patterns. For a work whose conceptual objective is to bypass cognitive poetics, unmetered verse is the appropriate form.

Additional formal measures are implemented during the poet’s sleeping hours to reflect physiological associations with the sleep states that are recorded in Fitbit’s biometric data: light sleep, deep sleep, REM sleep, and interruptions to those cycles caused by periods of waking or fitful sleep. Line length varies during the poet’s sleep, as it does during the day, with longer lines reflecting the slower breathing characteristic of deep sleep and variable line lengths representing the increased heartrate and cognitive activity of REM sleep. The poetry is also punctuated by additional formal measures in these zones: increased ellipsis in deep sleep verse denotes the torpor of that state, where the poet is less responsive to stimuli. REM sleep poetry, produced from a dedicated corpus of dreamy and surrealist poems, is correspondingly fragmented in form by the addition of random indentation and interlinear whitespace. Periods of waking and interrupted sleep, the setting for nocturnal anxieties, are represented by an increased use of interrogative forms in the poetic output. Within the system, these formal measures are implemented as the final step in the generation process. Probability values for the manipulation of a sentence are preconfigured for each zone. As a line of text is generated, the system uses the appropriate probability with randomly generated data to decide if a manipulation of form should be introduced.

Sentiment

The generator includes one further correlative variable which operates on the poetic language as the biometric data varies. This feature maps sentiment thresholds to the biometric zones described in Table 1, in order to reflect changes in physical or emotional intensity—the poet is nervous; the poet climbs the stairs—which effect variations in the heartrate. Broadly speaking, when a variation in biometric data occurs, certain words in the poetic output are substituted for synonyms of a corresponding sentiment value. When a word is assessed for substitution, its sentiment value is compared to the appropriate sentiment threshold. If it fails to match, the generator looks at the sentiment values of that word’s synonyms and performs a substitution if a word matching the appropriate threshold is found. We used the intensity of the synonyms’ sentiment, as defined by SentiWordNet (Baccianella, et al.), as a basis for deciding on a substitution, without prescribing whether that sentiment was positive or negative. The generator performs sentiment substitution on adverbs and adjectives only, as substituting nouns proved detrimental to the poetry’s rhyme and sense:

Original:
Which souls, even souls, together ties
Up suddenly, needing still he tries

Sentiment substitution:
Which souls, still souls, in concert ties
Up on the spur of the moment, needing yet he tries (2019-11-02 10:41:00).

This function of the generator has a specific conceptual purpose in positing a link between variations in biometric data and variations in linguistic intensity. The connection this implies is of the contestable, Romantic kind described above: that poetic language is a faithful reflection of the poet’s states of consciousness or emotion. However, our concept asserts only some connection between biometric variation and emotional variation, rather than a more simplistic chain of cause and effect. Intense physical exertion can result in changes in emotion—tiredness, satisfaction, exhilaration—but the causal chain is not necessarily as simple as exercise heightens heartrate heightens emotion. In this sense, we use the heart and its rate as the real-time, measurable indicator and metonym for a range of possible physiological changes which contribute to variations in human consciousness and emotion.

The sentiment substitution method also serves a functional purpose by increasing variation of Eververse’s poetic vocabulary, which would otherwise contain more regular occurrences of repetition. This linguistic reiteration is an inevitable consequence of the project’s specific method of text generation—whose lexicon is confined to that of the source corpora—and of the imbalance between the extent of the corpora (approximately 10,000 lines of poetry) and the poetic output (approximately one million lines). In addition, this method also provides an occasional slant to the recognisable snippets of poetry from the corpus, preserving their semantic essence yet reconfiguring them in different linguistic terms (“noble pleasure dome,” “statuesque pleasure dome”). The attentive reader experiences a glimmer of recognition at these moments where a familiar poetic phrase has been subtly modified and, as in identifying the unmodified fragment, interpretive satisfaction results from recognition of the adulterated intertextual source. Finally, this element of word substitution alludes again to the anxiety of influence, where poets engage the same topics as their precursors, conscious of their obligation to acknowledge and extend those linguistic legacies.

User Experience and Visual Design

The final element of project design that we have not discussed is user experience and visual design. As we alluded to above, Eververse exists in different versions designed for different contexts, with each deriving from the same underlying corpora and data. Designing for an optimal user experience of Eververse in the separate settings of exhibition, performance, and (dynamic and static) web publication presented different challenges. We needed to consider the constraints and affordances of the environments in which users experienced the work. For example, during the yearlong generation of Eververse, our objective was to publish the generated poetry in real time on the project website (Figure 2). However, our ability to do this was constrained by the fact that Fitbit’s record of sleep data is only made available through the API after the poet has woken up. For that reason, we were forced to publish the generated poetry on the project website with a 24-hour delay.

As additional developments and refinements to the generator were made during the year, we decided to generate a final, archived version of the poem anew at the conclusion of the project. Thus, its text is different from that published in delayed real time during the year, though the corpora and biometric data are unchanged. To reinforce the stability of the underlying data and to aid the user’s understanding of the work, the website design incorporates visual cues—changing background colours and a display of data values—to indicate shifts between biometric zones as the user scrolls through the poem’s text (Figure 3).

These visual cues are implemented using a combination of React and P5.js, a JavaScript library for creative coding. The underlying biometric data is mapped to a set of pre-configured colour gradients. As the user scrolls through the text an appropriate gradient is applied to the page background.

In a similar fashion, different designs were implemented for Eververse in exhibition and performance. For reasons related to network access, the exhibition installation used an offline, standalone snapshot of data which ran in a loop to generate poetic text. The live performance, which featured the poet cycling a stationary bicycle on stage while the poetic text was projected on a screen above his head, generated text at a much higher frequency than on the project website owing to the relative brevity—10 minutes—of the performance (Video 1).

Our desire to design different user experiences for Eververse testifies to our awareness of the importance of how and where users encounter the work. The work aims to instantiate its conceptual intent in the relationships with the text of both the poet and the reader and to recognise, as Bootz argues, the central expressive importance of a work’s generation context (126).

Reflections: traditions and rhythms

Where does Eververse fit within the tradition of electronic poetry? While the project combines established methods of text generation, it prefers to emphasise the novelty of its textualisation of biometric data. Nonetheless, the work is conscious of articulating its relationship to the history of this field. The task of assigning Eververse to a particular typology of electronic poetry is not straightforward, because of the existence of a number of different, competing typologies. Not only that, but the boundaries between different categories within these typologies are often permeable. Certainly, though, Eververse is a derivative work, according to the categorisation of reference works (Milic)—that is, it relies on pre-existing text(s) for its means of constructing new texts—as opposed to formulary works which are generated from more piecemeal elements of language such as those found in lexicons and vocabularies. In M. Boot’s tripartite typology, Eververse appears to meet the characteristics of the filter model of computer poetry, which relies on pattern recognition in a text or corpus. A caveat applies, however: Eververse extracts arbitrary patterns from its corpora and combines them in the random fashion of Boot’s dice model of computer poetry (Mechelen). Thus, the determinism implied by pattern recognition is not present in the work’s decomposition of its source corpora. In the typology that Funkhouser uses to categorise historic computer poetry, Eververse aligns closest to permutational works, which “[recombine] elements into new words or variations” (36), though it includes aspects of the combinatoric as well (Rettberg 49-51). A further categorisation of computer-generated poetry which matches the process of Eververse is the text-matrix, which relies on the decomposition and reassembly of source text(s) which become the foundation of an “infinite, not eternal” sequence of derivative works (Balpe). Thus, in its objective to textualise biometric data, Eververse synthesises diverse models of text generation and different modes of electronic literature.

Ultimately, Eververse aims to literalise the abstract relations of self and art. Just as Henri Lefebvre attempted in his analysis of daily rhythms, the project aims to articulate the inner life: “If one could ‘know’ from outside the beatings of the heart, one would learn much about the exact meaning of his words” (47). Eververse seeks to realise Lefebvre’s plaintive wish by making the poet’s work correspond to his heartbeat. The reader, thus, is given the means of deducing the poet’s feelings. They are granted privileged access to observe the regular rhythms of the poet’s daily life and to observe the linguistic index of his social environment which, as Lefebvre argues, conditions human beings for legibility through their routines, movements, and interactions with their surroundings. With the advent of the quantified self, those hitherto obscure physiological rhythms are increasingly apparent and familiar to people, who modify and refine their quotidian activities to align their biometric figures with a personal or cultural ideal. Instead of functioning as a means to physical optimisation, the biometric data of Eververse seeks to expose the poet’s existing, unexamined bodily rhythms to the reader through the textual medium of poetry. The work emphasises and develops the relationship between poetry and rhythmanalysis that Lefebvre only briefly considers. While both perform verbal actions of “aesthetic import” (33), he argues, they are distinguished by different focuses: the poet’s concern is with words and the rhythmanalyst’s with temporalities. The durational nature of Eververse’s generation reveals the temporalities that are ordinarily outside of the poet’s ken. He is concerned with words, as per usual, but in his unrelenting production of words over the course of an entire year, his natural rhythms—through the mediation of the quantified self—are also revealed.

Conclusion

Though it bore the infinite prospects of potential literature, Eververse concluded at midnight on 13 February 2020, one year after the generation of its text began. During the course of that year, we developed and refined various aspects of the work and contemplated many others. For example, a persistent enticement was the possibility of using machine learning and neural net technologies to generate poetry from the corpora. Part of the pressure we felt in this regard arose from developments of this nature in the fields of electronic poetry and NLG, and a perceived responsibility to position our work at those vanguards. Thoughts and discussions about the conceptual integrity of Eververse eventually relieved us of those burdens: text generation is but one aspect of the project, and machine learning methods preserve a degree of operational opacity that would dilute, in particular, the conceptual engagements with poetic tradition we discuss above.

While our development of Eververse leaves the work in a satisfactory state, further possibilities for engagement remain. The poem’s text and its corresponding biometric data are published and available for potential use in text and data mining and analysis, visualisation, or for additional creative opportunities to develop the work’s original concept in new and unforeseen directions. What it means to read e-poetry is, of course, different from what it means to read poetry in print: and yet, the opportunity to read Eververse remains, with the current website design facilitating browsing through daylong segments of the work. The prospects of finding a reader for the entire million-line poem are remote, underlining the manner in which e-poetry and conceptual poetry eschew conventional methods of linear poetic reading. However, the fact of its preserved, archival existence is one of permanent meaningful potential, orbiting the two poles of attention for conceptual artworks—boredom and fascination—described by Kenneth Goldsmith and John Cage. Thus, it will remain: the textual and biometric record of a year; the poet’s body of work.

Works Cited

Arozqueta, Claudia. “Heartbeats and the Arts: A Historical Connection.” Leonardo, vol. 51, no. 1, 2018, pp. 33–39.

Auden, W. H. The Collected Poetry of W. H. Auden. Random House, 1945.

Baccianella, Stefano, et al. “Sentiwordnet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining.” Lrec, vol. 10, no. 2010, pp. 2200–04.

Bakhtin, M. M. The Dialogic Imagination: Four Essays. Edited by Michael Holquist, U of Texas P, 1981.

Balpe, Jean-Pierre. “E-Poetry: Time and Language Changes.” Transitoire Observable, 22 Jan. 2005, transitoireobs.free.fr/to/article.php3?id_article=39.

Bloom, Harold. The Anxiety of Influence: A Theory of Poetry. Oxford UP, 1973.

Boltanski, Christian. Les Archives Du Cœur. 2008. Teshima Art Museum.

Boot, M. Taal, Tekst, Computer. Servire, 1984.

Bootz, Philippe. “Poetic Machinations.” Visible Language, vol. 30, no. 2, 1996, p. 118-37.

Boyle, Mark, and Joan Hills. Son et Lumière: Bodily Fluids and Functions. 1966.

Cage, John. Silence: Lectures and Writings. Wesleyan UP, 2013.

Corn, Alfred. The Poem’s Heartbeat: A Manual of Prosody. Copper Canyon P, 2008.

Eliot, T. S. “Tradition and the Individual Talent.” The Sacred Wood; Essays on Poetry and Criticism, Methuen & Co., 1920, pp. 42–53.

---. The Waste Land; a Facsimile and Transcript of the Original Drafts Including the Annotations of Ezra Pound. Edited by Valerie Eliot, Harcourt Brace Jovanovich, 1971.

Eververse. 2019-20. eververse.nuigalway.ie/

Eververse Data. Zenodo, 24 Nov. 2020. doi:doi:10.5281/zenodo.4288804.

Federico, Anthony. Poesy. 2017. GitHub, github.com/anfederico/Poesy.

Funkhouser, C. T. Prehistoric Digital Poetry: An Archaeology of Forms, 1959-1995. U of Alabama P, 2007.

Ginsberg, Allen. Mind Breaths: Poems 1972-77. City Lights Books, 1978.

Goldsmith, Kenneth. “Being Boring.” American Poets in the 21st Century: The New Poetics. Edited by Claudia Rankine and Lisa Sewell, Wesleyan UP, 2014, pp. 361–66.

Johnston, David Jhave. ReRites. Anteism Books, 2019.

Lefebvre, Henri. Rhythmanalysis: Space, Time and Everyday Life. Translated by Stuart Elden and Gerald Moore, Bloomsbury Academic, 2017.

Lethem, Jonathan. “The Ecstasy of Influence.” Harper’s Magazine, Feb. 2007. Harpers, harpers.org/archive/2007/02/the-ecstasy-of-influence/.

Mechelen, M. Vincent van. “Computer Poetry.” TRINPsite, Apr. 1992, https://trinp.net/Poet/ComP/ComPoe.HTM.

Milic, Louis T. “Computer Poetry.” Princeton Encyclopedia of Poetry and Poetics. Edited by Alex Preminger, et al. Princeton UP, 1993. ProQuest.

Morris, John. “How to Write Poems with a Computer.” Michigan Quarterly Review, vol. 6, no. 1, 1967, pp. 17–20.

Oliveira, Hugo Gonçalo. “A Survey on Intelligent Poetry Generation: Languages, Features, Techniques, Reutilisation and Evaluation.” Proceedings of the 10th International Conference on Natural Language Generation, Association for Computational Linguistics, 2017, pp. 11–20. ACLWeb, aclweb.org/anthology/W17-3502.

Olson, Charles. “Projective Verse.” Poetry in Theory: An Anthology, 1900-2000. Edited by Jonathan Cook, Blackwell Publishing, 2008, pp. 288–95.

Orekhov, Boris, and Frank Fischer. “Neural Reading.” Orbis Litterarum, vol. 75, no. 5, 2020, pp. 230–46. doi:doi:10.1111/oli.12274.

Osterloh, Greg. “Helping More People with Wearables: Google to Acquire Fitbit.” Google: The Keyword, 1 Nov. 2019, blog.google/products/devices-services/agreement-with-fitbit/.

Rettberg, Scott. Electronic Literature. Polity, 2019.

Shelley, Percy Bysshe. The Major Works. Edited by Zachary Leader and Michael O’Neill, Oxford UP, 2009.

Simanowski, Roberto. Digital Art and Meaning: Reading Kinetic Poetry, Text Machines, Mapping Art, and Interactive Installations. U of Minnesota P, 2011.

Spahr, Juliana. This Connection of Everyone with Lungs: Poems. U of California P, 2005.

Tether, Leah. “Mise En Page, Mise En Écran.” Logos, vol. 25, no. 1, 2014, pp. 21–36. doi:doi:10.1163/1878-4712-11112036.

Thornton, Pip. “A Critique of Linguistic Capitalism: Provocation/Intervention.” GeoHumanities, vol. 4, no. 2, 2018, pp. 417–37. doi:doi:10.1080/2373566X.2018.1486724.

Whitman, Walt. Leaves of Grass. Edited by Jerome Loving, Oxford UP, 1990.

Wood, Jeremy. My Ghost 2000-2016. 2016.

Yeats, William Butler. The Major Works. Edited by Edward Larrissy, Oxford UP, 2008.

Notes

1. The Aeolian Harp, creating art from the influence of external forces, is the ur-symbol for this understanding of poetic inspiration: “Poetry, in a general sense, may be defined to be ‘the expression of the Imagination’: and Poetry is connate with the origin of man. Man is an instrument over which a series of external and internal impressions are driven, like the alternations of an ever-changing wind over an Aeolian lyre; which move it, by their motion, to ever-changing melody” (Shelley 675, emphasis added).
2. See https://eververse.nuigalway.ie. Full and open datasets of the project’s poetic text and biometric data are also available at doi:10.5281/zenodo.4288804.
3. Though Olson is the primary theorist of the breath as an organising principle in the poetry, the effect is also evident in the work of poets such as Allen Ginsberg (Mind Breaths, 1978) and Juliana Spahr (This Connection of Everyone with Lungs, 2005).
4. From section 2 of “I Sing the Body Electric” (82).
5. From “Among School Children” (115).
6. We are indebted to Claudia Arozqueta for her perceptive writing on the examples given in this paragraph.
7. Reviewing a range of typologies for electronic literature, C. T. Funkhouser concludes that “classifications are not impermeable—categorizations of digital poetry are still an open discussion. … the matter of category is not necessarily an either/or situation” (36).
8. Respectively, Justin Tonra and David Kelly, authors of this article, and Brian Davis (Dublin City University) and Waqas Khawaja (Insight Centre for Data Analytics). We gratefully acknowledge the European Association for Digital Humanities, whose grant initiated work on this project.
9. During the project, two different Fitbit models were used: Charge 2 (now discontinued) and Inspire HR.
10. Registration with Fitbit is required to gain credentials for access the data via the Web API.
11. Major critical views on this theory include those by T. S. Eliot (“Tradition”) and Harold Bloom.
12. From W. H. Auden’s “In Memory of W. B. Yeats” (50). As a critique of “linguistic capitalism,” this is a more limited expression than that achieved by Pip Thornton’s {poem}.py, which explicitly enumerates the value of language in the digital age. Eververse’s relationship to surveillance capitalism deepened during the project after Google announced its agreement to acquire Fitbit for a reported $2.1 billion in November 2019 (Osterloh).
13. Broadly defined, formulary works “consist of strings of sentences generated by means of a formula or syntactic rule,” while derivate poems “take an existing line or poem and alter it in some systematic way” (Milic).
14. The poems in each corpus were merged into a single text file. A portion of the poetry remains under copyright, so we are unable to publish the corpora.
15. “He do the Police in Different Voices” was a provisional title for The Waste Land, found on the typescript for the poem (Eliot, Waste 4, 10). It alludes to the character of Sloppy in Charles Dickens’ Our Mutual Friend who adopts different voices to dramatise his oral reading of the newspaper.
16. From section 9 of “I Sing the Body Electric” (88).
17. For example, the exhibition piece used an offline, standalone snapshot of the data, and the live performance generated text at more frequent intervals than each minute, as is the case with the archived web version. Details of these different versions can be found on the “Events / Publications” page of the Eververse website.
18. “Scoop a portion of unexpended thigh, arms, legs are the climb” (2019-11-01 22:01:00).
19. “Walks in beauty, like the map of a fawn” (2019-11-10 18:49:00, emphasis added).
20. Jonathan Lethem’s essay, “The Ecstasy of Influence,” is an update of the ideas of Eliot and Bloom in the context of appropriative and citational literature.
21. Though no term equivalent to mise-en-scène for the visual arrangement of a computer screen has gained widespread currency, Leah Tether’s mise-en-écran is a useful literal rendition.
22. Or, as Alfred Corn writes of unmetered verse’s affordances, “Language can be caught at its most spontaneous, with the implication that unconscious forces were more important in producing the poem than conscious ones” (152).
23. Sentiment values range between 0 and 1 with a higher value assigned by SentiWordNet to higher-sentiment words. For example, “furious” has a higher value than “angry.”
24. With its decision to gather biometric data over the course of a year, Eververse emulates durational digital artworks such as Jeremy Wood’s My Ghost, which creates pictorial representations of the artist’s movements around London, recorded in sixteen years of GPS data. It also alludes to older conceptual art, such as that by Bruce Nauman, which creates works from the documentation of particular routines over periods of time.
25. See https://reactjs.org/ and https://p5js.org/ for further details.
26. In the real-time version of the project website, we experimented with page backgrounds featuring both animated and static generative visualisations to represent data in the heartrate and sleep zones (visible in Figure 2 and Video 1). This approach applied a different generative algorithm for each zone, using the data value as an input to the algorithm. The animated and static images proved too distracting for users, however, drawing attention away from the text rather than providing the subtle reinforcement desired.
27. Boot’s work was published in Dutch and has not been translated into English. Thus, we rely on M. Vincent van Mechelen’s summary: “the computer is thus made to do, it is claimed, what poets themselves do when perceiving reality from a very personal point of view, that is, when ‘filtering’ it. … The data fed into the computer are not taken from a single poem … but from the complete oeuvre of the poet who is to be imitated.”
28. Thus, the Eververse poet becomes a successor to John Morris, who declared, after writing a programme which rapidly generated 4,000 haikus, “For one glorious summer month, I was the world’s most prolific poet” (17).
29. David Jhave Johnston’s ReRites is perhaps the most notable example in recent electronic literature, while Orekhov and Fischer adopt a formalist literary perspective in their analysis of poetry generated by artificial neural networks.
30. For an example of a critical work that draws attention to the process of reading electronic literature and media art, see Simanowski.
31. Goldsmith: “I am the most boring writer that has ever lived. … You really don’t need to read my books to get the idea of what they’re like; you just need to know the general concept” (361). Cage, paraphrasing Zen Buddhist advice: “If something is boring after two minutes, try it for four. If still boring, try it for eight, sixteen, thirty-two, and so on. Eventually one discovers that it is not boring at all but very interesting” (93).