To appear in IEEE Transactions on Visualization and Computer Graphics.

Poemage: Visualizing the Sonic Topology of a Poem
Nina McCurdy, Julie Lein, Katharine Coles, Miriah Meyer

Fig. 1. The Poemage interface comprises three linked views: (left) the set view allows users to browse sets of words linked through
sonic and linguistic resemblances; (middle) the poem view allows users to explore sonically linked words directly via the text; (right)
the path view shows the sonic topology of a poem.
Abstract—The digital humanities have experienced tremendous growth within the last decade, mostly in the context of developing
computational tools that support what is called distant reading — collecting and analyzing huge amounts of textual data for synoptic
evaluation. On the other end of the spectrum is a practice at the heart of the traditional humanities, close reading — the careful,
in-depth analysis of a single text in order to extract, engage, and even generate as much productive meaning as possible. The true
value of computation to close reading is still very much an open question. During a two-year design study, we explored this question
with several poetry scholars, focusing on an investigation of sound and linguistic devices in poetry. The contributions of our design
study include a problem characterization and data abstraction of the use of sound in poetry as well as Poemage, a visualization tool
for interactively exploring the sonic topology of a poem. The design of Poemage is grounded in the evaluation of a series of technology
probes we deployed to our poetry collaborators, and we validate the final design with several case studies that illustrate the disruptive
impact technology can have on poetry scholarship. Finally, we also contribute a reflection on the challenges we faced conducting
visualization research in literary studies.
Index Terms—Visualization in the humanities, design studies, text and document data, graph/network data

The use of digital tools across disciplines in the humanities has ex- and spatial field of the text, with each other and with the reader, to
ploded during the last decade. Popular projects such as the Google create meanings greater than the sum of the parts. As this description
Ngram Viewer [37] and Wordle [55] have harnessed the power of com- suggests, much of the work done in close reading is well beyond the
putation to look across huge corpora of texts, leading to insights that current capabilities of computation. Thus, the true value of computa-
had never been available before. Tools such as these are highly ef- tion to close reading is still very much in question and is the topic of an
fective in supporting what is called distant reading — a term coined ongoing dialogue in the digital humanities. While a handful of com-
by literary scholar Franco Moretti to describe critical approaches that putational tools have been designed to support close reading, much of
seek to understand literature and literary history by aggregating and the problem space remains unexplored.
quantitatively analyzing large text corpora. We conducted a two-year design study with poetry scholars and
Despite this new mode of scholarship, traditional humanities schol- practitioners to explore this gap. Our two primary collaborators, both
ars continue to engage primarily in a very different type of analysis of whom are co-authors on this paper, identify both as poets and as
called close reading. As its name implies, close reading involves a academics. We also engaged a network of practitioners, including two
detailed analysis of a text in all its complexity, encompassing an anal- professors and two students of poetry. Together, these collaborators
ysis not only of specific operations such as syntax, rhyme, and meter; have literary expertise in medieval, early modern, modernist, and con-
such figures as metaphor and allusion; and such linguistic effects as temporary poetry, and they analyze poetry from a range of traditions
affect, but also of how these operations interact across the temporal and periods. Furthermore, they write formal verse, free verse, and ex-
perimental poems, and thus bring a diversity of theoretical viewpoints
to their critical and creative work.
During this design study, we encountered several specific chal-
lenges that affected our design process. First, supporting close reading


decided to focus on sound. we de- sonance. allowing users to unclear as to what to visualize in a poem. and a reflection on the unique to create a tool that will be of use to the broader poetry community. Our formalism includes a language for expressing a broad range of vi- sented as isolated objects. Integrating computation into the practice of close reading is a wicked problem [6. as. by established computa- present a range of information about the poem. To quote a prominent critic of the digital hu- 2 P REVIOUS W ORK manities. participated remotely via video conferencing. such as context. faced was that our collaborators belong to a community that sees the The visualization tool ProseVis [9] allows users to interactively ex- integration of technology into their research practices as largely un. these The initial conversations were broad and open-ended: the poets tools treat a given text. monthly basis. We focused our visualization de. the level of analysis that was of particular interest to our col- facilitates exploratory analysis of literary text. detection of metaphor and imagery. were held at the University of Utah. revealing interesting patterns within and between and interactions within poems they had previously studied. but also disrupted our collaborators’ view of poetry. 49]. “You don’t know what you’re looking for and A number of highly effective tools exist in support of distant reading. we investigated established methods for computationally detecting and The Word Tree [57] present more complex patterns based on a and analyzing the devices that most interested them. collaborators would like. Myopia attends to a broad range of poetic of our formalism. and were often recorded for future reference. at a minimum. Their resis- individual poems. typically lasted allel Tag Clouds [15] provide a different kind of summary by focusing from two to three hours. Our collaborators presented examples of interesting features tion of expressions. For many of these range of relationships. In addition. pletely open. This software subsequently that is currently done manually by a poetry expert and thus limits the supplies the back-end to our visualization tool Poemage. Our primary collaborators had participated in previous visualization These new perspectives led us to consider the topology of a poem. pre. affect. The formalism describes sonic patterns as rhyme. who was non-local for the majority of this eral more sophisticated tools [52] [16] provide broad overviews while collaboration. The second challenge we and the exploration of the distance between the eye and the ear [9]. the definition of rhyme being one that is both broad and flexible — face does not capture the dynamism of a poem to the degree that our rhyme is a poetic device that varies in definition from poet to poet. and to us to use a highly collaborative and exploratory design process that a slightly lesser degree. which acted as a first step towards overcoming their re- complex structures formed from the interaction of sets of words across sistance to integrating technology into their own practices. on a cloud-based tools like Wordle [55]. pun. features that are critical to the interpretation of poetic text. with some limitations. personification. to test and improve our formalism and eventually as a tool for poets and emotion. that they look at in a close Diverging from this slightly. into questions of what makes a not only providing novel analysis insights but also enabling the cre. Even after determining what sonic data we wanted to explore. These challenges motivated patterns on different levels of granularity. Building on The Word Tree. nature of conducting visualization research in literary studies. PoemViewer attends to sound Once we. also allowing users to explore finer level connections. why you’re looking for it. In general. a tool called RhymeDesign. was designed to aid in the for poets. 3 D ESIGN P ROCESS vices within a poem work together to create a response in the reader. a task to explore custom sonic devices in poetry. however. how then do you proceed?” [26] Thus. Techniques such as Phrase nets [54]. TextArc [45] and the variant Par. must be coded a priori. Arc Diagrams [56]. the research [2]. they did not want a tool to “solve” a perform a range of analysis without regard to structural and semantic poem [48]. Stanley Fish. from traditional rhyme tional linguistics techniques. and for a broader range of sonic creating visual representations of poems and their features was com. on the frequency and distribution of individual words or phrases. left abstraction for visualizing sonic devices in poetry. These elements. but the design space for explore any poem of their choosing. since the use of technological tools as direct interven. from meter. and poems are portrayed as static systems. the qualitative experience of the poetic encounter and in part a skepti- sign efforts on capturing poetic topology and providing a canvas for cism that it would be possible to visualize the interaction of any set of poets to explore the complex structure of sonic devices within a poem. not only visualizing various turned to developing a system that would automatically sonify a poem. sound. as a bag of words. Myopia and PoemViewer. consonance. and alliteration but also providing information veloped a formalism for analyzing sonic devices in English-language about the physiology of sound production. of more complex sonic patterns from a given poem and visualizes the Our design approach not only enabled us to learn more about po. lel. Tag of a close reading. 49]: not only was it initially ited to sound. This initial project. as a group. Specifically. etry and close reading. and metaphor. poetry [40]. and Galaxies [58] employ semantic analysis to extract two primary poetry collaborators to determine what their goals were key concepts and allow users to gain quick overviews of one or more and how they imagined a visualization tool affecting their experience documents and to run comparisons across large bodies of text. The exception to this was sound. initially spent significant up-front time in joint conversations with our Compus [25]. While there was some remaining resistance. interaction of such patterns across the space of the poem. they approached this design study with “skeptical enthusiasm” to see called Poemage and shown in Figure 1. poem a poem and how a poem does what it does. imagery. on which they did not have specific goals. plementation of a tool for visualizing the sonic topology of a poem. and syntax to metaphor. WordSeer [43] devices. however. FeatureLens [20] includes the repeti. PoemViewer [2] employs rule-based visual mapping techniques to which is detectable. laborators was beyond current technological capabilities. types of phonetic repetition such as end rhyme. patterns to low-level sentiment analysis. poetic features at a level of complexity that would allow them to make The specific contributions of this work are a characterization and new and interesting observations. poetic elements are. pushing them to develop new perspectives on how poetic de. alize the individual components of sound. we developed an open-source implementation close readings of poems. validation of this design study if it is possible to use technology to probe more deeply. and they described a wide array of poetic devices. Myopia [8]. con- 2 . Their larger goal is ation of new poems and literary ideas.of poetry is a truly wicked problem [6. initially as a platform elements. Docuburst [13]. While PoemViewer pro. One primary collaborator. These sessions took place. sual and sonic rhyme types and an associated ASCII notation designed A second visualization tool. Although our analysis is lim. within a given poem we consider sets tance was rooted in part in an anxiety that the computer would inhibit of words with similar sonic patterns. for the most part. capture and visu- takes the same experimental and even playful approach that our col. such as the Several visualization tools exist for specifically analyzing poetry. Whereas ProsVis. an open-source im. Sev. Building on existing approaches for the sonification of text. its inter. sound. Poemage extracts a range laborators exhibit when reading and writing poetry. them deeply intrigued. we Synoptic text visualization tools like GistIcons [18]. internal rhyme. Poemage processes text automatically. tool’s usefulness to a handful of poems. or texts. In paral- documents. The sonic analysis aspect of our research is closely related to Tanya tions in close reading (as opposed to in pedagogy and instruction) is Clement’s seminal work on the analysis of aural patterns in text [10] still almost unknown to literary scholars. reading. plore the sonic transcription of a text and aids in the discovery of sonic necessary and potentially even intrusive. both structural and relational. with vides a wealth of information. beginning with through several case studies that illustrate the efficacy of Poemage for a high-level investigation into sound. our attention on a much deeper level than many tools. patterns.

which may or may not relate sonically. while also considering etic insights in our meetings on the fly. sound is arguably pervasive in every poem at all levels. the probes helped us create an experimental and playful research the probes over the course of several months based on extensive user environment that we maintained for the duration of the collaboration. literary device. allows many different points of entry. we refer to all sonic and linguistic devices as rhyme [40]. In addition. but in- tors a list of potentially interesting sonic devices that could be detected stead they sought to understand how different sonic patterns interact computationally and having them compile a list articulating the vari. prototype and presented to our primary collaborators. their literary to their feedback and critiques helped the poets become familiar with features interacting not only with each other but also with us as read- the technology and also resulted in an interface that reflected their in. for writing poetry. fol- lowed by a close reading of “Night” by Louise Bogan. and evolve across a poem. starting with the poem “Prayer” by Jorie Graham. would be useful to them. com- mitment. the poets and poetry scholars. Over- helping us better understand the problem space. The incremental steps and the adjustments we made in response Poets and scholars see poems as living and relational. we also had to cultivate their trust. In close reading. and attending nology probes to explore this notion of sonic topology. Other techniques for clarifying our entry point included studying thermore. as a fluid moving via its linguistic devices ter understand. In addition. to patterns involving probes. feedback. The ture data from a poem. we experimented metaphor of a poem as a flow [36]. and the results of. Users in homographs like wind and bow as well as in words with multiple would load a poem of their choosing into the interface. the flow metaphor captures three 3 . were not interested in exploring individual sonic relationships. our collaborators returned to “Night” and other poems and picked up close readings in order to illustrate particular concepts — in exploring in a poem. along Our collaborators consider a broad range of sonic and sound-related with written documentation. followed grams (desserts/stressed). devices in their close readings of poems: from traditional types of orators were given approximately one month to experiment with the rhyme such as rhyme/sublime and picky/tricky. the spellings of words. from our primary col- laborators as well as our extended network of poets and poetry schol. A highly collaborative and exploratory design process proved to be critical in helping us navigate these challenges. patterns surrounding the physiological production of speech sounds. what kinds of sonic relationship they were interested and figures through a defined space. understanding of rhyme. These ideas were implemented into an initial composition. Interviews in. 2. as opposed to how to visualize it. The first technique was an observation of a pair of close readings between our two primary collaborators. we ideated on a range of design possi. ers. Throughout many of our future conver- sations. simply in response to develop. siderable design challenges and open questions remained. these initial probes indicated to us that our collaborators an annotated poem from one of our collaborators. Furthermore. their previous visualization research. and general usability. a poetry scholar carefully attends directly to terests. Next. We iteratively refined all. Based on these activities. By approaching a poem. along with information about selected sets of the same poem. as well as the role that sonic Our collaborators are particularly interested in the conceptual analysis plays in close reading. the poets actually generated po. because our meetings were specific texts. encompasses varying styles of anal- ysis. unlike many devices which may or of sonic patterns. Using these probes. cause of its emotional power and the way it works directly on the body of the reader of the poem. and many poets do engage it as a prod to for the tool Poemage. in the literary canon. the details of which are provided in Section 7. after which formal interviews were conducted. Close read- ing covers a broad range of tasks. Thus. which dis. to by questions surrounding approach. In this de- was screen captured. bined into a single. These investi- public poetry readings to better understand the nature and practices of gations were instrumental to the development of our data abstraction. and syntax. between two or more people. we employed a number of different tech- niques in an attempt to clarify our point of entry. tracing the interactions among such literary features as highly conversational and interactive. and to define probes were successful both in engaging our collaborators and also in the space of what we could computationally detect in a poem. both casual and via formal interviews. which we discuss further in Section 5. primary collaborators. aesthetics. The technology probes also allowed us to establish a common vo- bilities to pursue. and to help our collaborators bet. we refined and improved existing features and added new of play and can deeply influence the interpretation of the poem. how a given poem explicitly or implicitly converses with other poems ing and imagining the tool. In addition. which we then developed into a set of technology cabulary with our collaborators. We began by dis- cussing the poets’ experience with. Interviews were recorded and transcribed. Interface for the first set of technology probes. What we found led us to develop a broader such as how sonic patterns can reinforce or undercut semantic mean. including eye rhymes (cough/bough) and ana- cluded brief observations of our collaborators using the tool. purposes of visualization. and one observation period such as the location of the tongue in relation to the lips. giving our collabora. Close readings can be performed internally by one poet or externally as a conversation Fig. rhyme and meter. The collab. presented in Section 6. to four of our collaborators. and accepts an extensive range of sometimes radically divergent interpretations. sound is an important source of poetic po- 4 T ECHNOLOGY P ROBES tency and can be used to reinforce or to undercut meaning conveyed The technology probes were implemented in Processing [27] and com. To appear in IEEE Transactions on Visualization and Computer Graphics. the experience of close reading often leads to the Results from the technology probes formed our initial design ideas generation of new poems. a The initial goal of the technology probes was to explore the many broad definition embraced by our collaborators. multi-tabbed interface. This gave them confidence that the work. sound provides poets with a rich source feedback. sound. for the broadly in order to better understand. our collaborators admitted resistance to integrating technology into their close reading. Be- features. to focus on understanding how to cap- probes [32] — we discuss details of these probes in Section 4. and values. Based on casual As a broad. shown in Figure 2. Although not viewed as an established technique and eventually the visualization tool. figures. sign study. We thus developed a second set of tech- ous sonic features that they were interested in exploring. and enthusiasm. via other poetic devices. pronunciations — also help generate multiple possible interpretations played the text of the poem. capabilities. Following an initial development period in which may not be present at a particular moment or even at all in a given versions of the probes were presented for informal feedback to our poem. Fur- ing. 5 P OEMS AND S OUND ars. the technology probes were deployed. sonic ambiguities — for example. different aspects of sound within a poem.

one. left to right. a the context of other poetic devices they identify in the course of their pop-up label indicates the specific rhyme pattern for the set. First. we describe the design and capabilities of the indi- different rhyme sets. rhyme set. we create paths from the rhyme sets. For the current version of the tool. comprises three into three data components: poemspace. and sonic topol.e. (c) diverging paths (also displaying overlap). maintaining a close connection with its original form. followed by top to bottom. vidual views and of the interface as a whole. From the conceptual metaphor of a poem as a flow. linked views: the set view (left). different poetic interpretations. energy. indicated as intersecting. is unique compared to other 2D spaces as the reading of the A user’s session with Poemage begins with the selection of a poem poem constrains the way that movement within poemspace can hap- of interest. In addition. results from sampling the ambiguous pronunciations as an ensemble tion. where each word in a rhyme set is connected by a link based upon their order in poemspace. The Poemage interface. We observed the ne- At a low-level. Not only did it come naturally to them. the words are ordered based upon their location much broader than those available in other poetry visualization tools. and the impact that individual flows and col. the use of our technology probes. we rotate through an adapted version of giCentre’s 12-class categorical paired colormap [1]. which in- • overlapping: paths intersect at multiple consecutive nodes. Collapse and expand but- • diverging: paths overlap and then split. The set view allows users to browse through the various detected For example. Section 8. calcite. For example. and the source code is freely available at http://poemage. expressed in terms of the extracted sonic patterns. The set view supports browsing of specific rhyming sets. and subsequently became emerging paths. visible. We discuss our data metaphor in Section 6 and our visual metaphor in Section 7. In this set. the set including machinations. Poemage pre- pen — left to right. and visual rhymes. the words cat. but it also provided (a) (b) (c) (d) a slightly abstracted way of exploring the poem while maintaining a close connection to its original form. multiple pronunciations. an example of which is presented in any sort of data we pull out from a poem computationally will (usu. we present two novel extensions of existing graph visu- rhyme sets across a poem and how those sets of words interact with alization techniques. i. depending on which rhyme schemes are defined. This feature was first requested by our col- displaying overlap). As part of the path view The sonic topology of a poem is represented by the distribution of description. A set of words linked by a rhyming scheme is called a source release of the software. and the path ogy. To capture the sonic 7. in ploring places of turbulence. we that line falls within a poem.. Each word in a poem can belong to none. in Figure 1. We note that the back-end of Poemage includes a at (underlined) are in a set together because they form a user-selected formalism for defining rhyme that can be modified through the open- perfect rhyme. When a stand the places of turbulence within the context of the poem. worked with our collaborators via the technology probes to define a set Words are related to each other not just spatially within poemspace. and where rhyme types built into the tool. each other. they want to under. and activity due of reading the poem. (d) and laborators in one of the technology probes. how the inter. shown in Figure 1. and in user hovers over or selects an individual set or entire rhyme type. however. verging. spectives and insights. and rhyme sets for a given poem. or not. a feature expressed repeatedly by our collaborators in importance of maintaining the spatial and textual context of the poem the technology probes as being highly effective for gaining new per- itself for two reasons. 4 . and within Poemage. and separately. that. tons located to the left of each category header allow users to omit • emerging: paths begin at a point of intersection. Second.1 Set View topology. either of these categories from their exploration. we worked to translate this conceptual metaphor first into a data of possible paths through a poem. For our collaborators. a word exists in multiple sets. is the 2D space of the poem view (right). our collaborators are interested in identifying and ex. we can describe the set of paths that to the interaction of poetic devices. Poemage was implemented in Processing [27]. lections of flows have on their surrounding region. Each circle represents an individual oblique is ordered from top to bottom. we translate this metaphor poem. Fig. At a higher level. and then in exploring this ensemble to probe for different sonic topologies and into a visual metaphor. Sets are organized by rhyme type that are of interest. nipulate and play with the text. The notion of browsing was some- thing that our collaborators responded very well to in the technology probes. the poem view (middle). In poemspace. merging. di. volve matching patterns in sound. Poemspace. (b) merging paths (also selects all rhyme sets. thus supporting a definition of rhyme rhyme set.2. The first component. which is loaded into the tool via a text file. which involve • merging: paths intersect and then overlap. tics captured in our data metaphor. the places of sonic turbulence in a poem exist where multiple rhyme sets intersect. in Figure 1. or many In this section. in poemspace. 7 P OEMAGE 6 A BSTRACTION Our second contribution is the design and implementation of Poemage. and emerging paths. sonic ambiguity exists for words with tic and sonic devices through the space of the poem. Color is used to link sets and rhyme types across views. Our collaborators are interested metaphor. the radius of which encodes the relative number of words Our collaborators noted several different types of path interactions participating in a given rhyme set. each of which is illustrated in Figure 3: and are ordered by decreasing set size to make the largest sets most • intersecting: paths intersect at a single node. poemspace. associated rhyme sets are also displayed in the poem and path views. a visualization tool for interactively exploring the sonic topology of a In order to visualize the flow of a poem.distinct levels of poetry analysis: the movement of individual linguis. and to view abstracted representations of the poem while ally) be meant to augment the reading of the poem itself. flow this ambiguity is a source of great joy as it represents possible alternate introduces the notion of sonic turbulence — a metaphor for locations or additional meanings. The implementation of our formalism for action among such devices contributes to the complex sonic-temporal describing rhyme captures this ambiguity and stores multiple versions structure of the poem. each processes the poem and creates the rhyme sets based on 24 different word has a location based on where it falls within a line. Our collaborators were adamant about the into the poem. matching patterns of alphabetic characters. The rhyme types are organized into sonic rhymes. cessity to explore visual and sonic rhymes together. a poet can play with whitespace and lay. multiple views allow users to ma- out to encode or enforce some sort of meaning in the poem. of rhyme types that captures the majority of interesting sonic patterns but also based upon their similarities to other words with respect to in a poem — Table 1 lists the types of rhymes currently supported some sonic pattern. When rhyming sets are represented as paths. As described in Section 5. of rhyme sets based on alternate pronunciations. several interesting Clicking the beautiful mess button at the bottom of the set view interactions can occur: (a) intersecting In addition. Over the course of our collabora. visually encoding the features and characteris. Multiple views provide users with multiple entrances as it is printed on a page. and so enriches the interpretive experience in a poem where there is increased intensity. As such. 3. For our color scheme. rhyme sets. and the close reading.

which in turn selects all the rhyme sets for which the word adjusted via the context slider. an abstract rhyme sets at once due to visual clutter. Although we explored for this view. Rhyme types implemented in Poemage for creating sets of sonically related words. This insight was particularly powerful to our collaborators given semantics of the involved words. view. similar to the concentric rings employed in LineSets [3]. Furthermore. starting with images we shared with our collaborators several different ways to encode selection. with heavier pebbles causing more our approach combines line-based with region-based overlay tech- ripples. the show ambiguity button highlights words with multiple pronuncia- tions and allows users to select alternative pronunciations. and even affective characteristics of our flow metaphor. The color of each ellipse corresponds to the assigned We developed a number of prototypes to explore design variations set color.2 Poem View to appear in this view. clicking We designed the poem view to support direct exploration of poetic the show words button displays the set words associated with a given devices in a poem’s original form. experiences. effectively rendering multiple paths at once. An example visualization employing is a member. Includes repeated words and homographs pair/pair. either the set or poem view. we designed a path visualization that words belonging to multiple selected sets. 7. Second. of poems. and incorporating sonic ambiguity.1 Rerouting Paths enabled for poems of longer length with a print-to-pdf keyboard option In our prototypes. Despite our initial hesitation to support showing all The sonic topology of a poem is visualized in the path view. Rhyme type Description Example Identical rhyme match in all sounds. our path visualization employs fill to emphasize regions en- sets by selecting specific words in the poem view. the selected set. was rooted in obser- by William Carlos Williams. we encountered problems with edges intersecting providing a complete view of the poem and visualization. Similar to hovering and selecting path. Our decision to map words to nodes. in which the spatial con- provides a quick overview of the set membership for a given word. ellipses are drawn around the words in We discuss each of these considerations in more detail below. clicking closed by sets intersecting at multiple nodes. bullseye. in the rhyme set is shown in the path view as a curve connecting the ships sharply heightened their sense of the poem’s ambiguities. Browsing through words in the poem was requested by these features is shown in Figure 5 (c). pebbles being dropped in a pond. words that were not included in the path. Similarly. examples like 5 . as it reminded them of their own annotation practices. siderations: rerouting paths to avoid ambiguous set membership of words. rather was the single. One such discovery location in poemspace. as shown in Figure observed close readings consistently tracing sonic devices back to the 4. Fourth. Rather than use convex hulls to delineate sets. and Kelp diagrams [19]. or directly in the set view. text of the set data is preserved and shortest path algorithms are em- One collaborator commented that this encoding appeared to her as ployed to determine routes linking set members. pare/pair Perfect rhyme: matching stressed vowels sound and all proceeding sounds Perfect masculine rhyme stress on the final syllable rhyme/sublime Perfect feminine rhyme stress on the second to last syllable picky/tricky Perfect dactylic rhyme stress on the third to last syllable gravity/depravity Semirhyme perfect rhyme with additional syllable on one word end/defending Syllabic rhyme perfect rhyme between stressed and unstressed syllables wing/caring Consonant slant rhyme matching trailing consonants of stressed syllables and/bent Vowel slant rhyme matching vowel sounds of stressed syllables eyes/light Pararhyme matching leading and trailing consonants of stressed syllables tell/tail/tall Syllabic 2 rhyme rhyme between initial stressed syllables restless/westward Alliteration matching leading consonant sounds of stressed syllables languid/lazy/line/along Assonance matching vowel sound (independent of stress) blue/estuaries Consonance matching leading and/or trailing consonant sound (independent of stress) shell/chiffon Forced rhyme perfect rhyme with imperfect match in final consonant sounds shot/top/sock Eye rhyme spelling indicates perfect rhyme but sounds do not match cough/bough Character clusters matching substring involving 1-4 characters restless/westward Mixed character clusters mixed substring involving 2-4 characters inlets/itself Anagram words formed out of the same set of characters nights/things Phonetic alliteration leading consonants of stressed syllable match in mouth placement pen/boy Phonetic assonance vowels of stressed syllables match in mouth placement edible/anchor one of their favorite features as well as a surprisingly valuable addi. Clicking the custom set button allows users to build custom rhyme however. and KelpFusion [42].3. users can hover over and select words in the surrounding a given path. Based the ellipse. First. Scrolling is 7. Third. as described in the previous section. supporting multi- When rhyme sets are selected either via word selection in the poem ple interpretations of poemspace. For on feedback about these tools. our collaborators early on and proved to be a very natural and effective Rendering paths in poemspace requires a number of design con- way for them to interact with the text. our collaborators preferred generated from off-the-shelf node-link diagram tools [23] [4]. We provide context in this abstracted view through several mech- anisms. as node-link diagrams and found this representation best captured the tics to elaborate readers’ potential interpretations. Table 1. isolated pronoun you in the poem “This Is Just to Say” than to smaller linguistic units such as syllables. our collaborators were able view that represents words in a poem as nodes at their corresponding to make interesting discoveries with this feature.3 Path View tion to the tool. the addressee and recipient of the os. The beautiful mess revealed that you was vations of our collaborators during the technology probes and in the the only sonically unconnected word in the poem. Like KelpFusion. The extent of surrounding words can be poem view. a nod towards a visual metaphor of the flow of a poem. when a user hovers over a node in the path view the corresponding word appears as a pop-up. concentric ellipses form a resembles other line-based overlay techniques such as LineSets [3]. excluded from the poem’s many sonic relation. When a user selects a rhyme set in the poem’s occasion: to see you. the path view is linked with the set and poem views such that selection and highlighting in the other views causes paths 7. clicking the show context button displays nearby words rhyme sets in the set view. This associated nodes. niques [14] [7] [30]. To appear in IEEE Transactions on Visualization and Computer Graphics. We explored a variety of ways to represent the paths example demonstrates how sound can work with and against seman. the associated path connecting the words tensive poetic apology.

divergence. we de. lation (left). Rendering the rhyme set paths in poemspace can lead to am. Path order is computed on-the-fly as users select and deselect paths of interest. This rerouting produces a meandering curve that avoids all words that are not included in the path. We place a new control point at the cen- ter of these whitespaces and render the edge as an interpolating cubic bezier curve. 7. looping backwards. This rerouting technique integrates several different aspects of the conceptual metaphor of a poem as a flow: the notion that adjacent flows tend to aggregate. rather than obscures. sometimes pulling cess. intersection. out added whitespace helped form interpretations as to why an author may have formatted the text in the way he or she did. and emerge. For these edges. In the path view. Our path rendering algorithm supports visualizing these locations through two mechanisms: maintaining a consistent ordering of paths to help users trace paths across a poem and a fill-function that emphasizes mergence. Fill is computed interactively on a pairwise basis. our visual representation 7. cided to reroute edges such that they explicitly avoid words not in. intensifying the same path. which increased the overall efficacy whitespace. In places where paths intersect portance of having isolated nodes appear isolated. The inherent grid-like qualities of our data make our stein who use whitespace to augment the shape of their poetry. of the tool for our collaborators. 6 . at multiple nodes. We Fig. The beautiful mess highlights the sonic isolation of the word “you” disable ordering for the beautiful mess. A visualization from a technology probe obscures this iso. fluence their surrounding region (and vice versa). (right) and even spacing. we determine the closest whitespace to the edge intersection. The ordering task closely resembles the metro-line minimization problem [46] with a unique combination of constraints — rerouting. as well as the idea that flows can behave like eddies. This shuffle is meant to visualize and inspire different in- technique to disambiguate words (b). bezier splines. as it causes significant visual in this poem. reg. As such. illustrated in Figure 5 (b) and (c). bending and diverging domi- nant courses. the fill spans the area enclosed by the two paths. Inspired by poets like E. surrounding words closer together. and emergence. A side-effect of this rerouting technique is that similar paths are naturally bundled together. our rerouting technique a more intuitive approach and allows us to avoid collaborators found that comparing the shapes of paths with and with- issues associated with standard grid-based methods. This fill tech- We experimented with several different rerouting [39] and nique is inspired by our collaborators’ belief that interacting flows in- bundling [17] [24] techniques before finalizing our design. the rerouting Fig. Cummings and Charles Bern- patterns [35] . separated by more than one line in the poem. merge. and the you anomaly that we discussed in Section 7.3. and evenly spaced nodes. we support three different deformations of poemspace: the orig- inal form. Our rerouting technique is most similar to techniques Another design concept that we experimented with was that of added that use grid-based rerouting to bundle edges and reveal high-level whitespace. clutter and obscures isolated nodes left exposed via rerouting. Buttons along the top of the path view allow users to toggle between these deformations. In the pro. We adapt an existing technique to address these differences [44] by first calculating ordering for pairs of paths sharing common subpaths and then iteratively adding one path at a time such that the pairwise orders are maintained. other times pushing them apart — ular spacing of poemspace to establish a simple and general rerouting the fill seeks to reveal possible regions of influence formed via the in- technique.. as shown in Figure 5 color that is dependent on the involved paths. at each line of the poem that the edge intersects. 6. The three modes of poemspace shown in the context of the poem generates much more organic. we discovered that we could take advantage of the vertical.e. In addition to interactive path ordering.2 Drawing Multiple Paths As we discussed in Section 6. of particular interest to our collaborators was exploring places in the poem where paths overlap. or at single points of cluded in a path. In addition to allowing users to select alternative pronunciations for B C D E homographs and other multi-pronunciation words in the poem view. a shuffle button reruns the entire program based on randomly selected Fig.E. dis- sipating. divergence. as these are the edges The user has the option of setting the fill to a constant color or to a that may intersect words not in the path’s set. we include a semi- transparent fill function to emphasize path mergence. compressed whitespace to just a single character width. 4.1 highlighted the im. diverge. aesthetic curves than those generated “Parsing” by Charles Bernstein: (left) original form. 7.3. and intermediate nodes.3. as shown in Figure 5 (d). We illustrate these deformations in Figure 6. the fill approaches the intersection point. pronunciations. i. disrupting their surroundings. Our technique reroutes edges connecting words that are tersection of paths. Poemage incorporates a path rerouting Section 6. as well as context information for terpretations of the same poem. (middle) compressed in our previous implementations. 5. (d) Fill function between two intersecting paths.3 Deforming Poemspace of poemspace. (a). emergence. While we appreciate that minimiz- ing wiggles is a common constraint in graph drawing. and in places where two paths merge. diverge. thereby sampling the ensemble members described in biguous set membership (a). we claim that our approach improves. or developing in new directions. the nodes (c). whereas rerouting in Poemage reveals the anomaly (right).4 Ambiguity The final feature in the path view addresses the concept of ambiguity. In addition. the closest space between words.

selecting and deselecting collaborator experimented quite a bit with the final implementation of rhyme sets almost arbitrarily. indicating to hover over and select particular rhyme sets in the set view. Her cento is shown in Figure line of the final stanza. This third approach allowed her to specifically investigate view shed new light on a poem with which she was deeply familiar: how introducing new sonic flows changed the sonic structure of the “In other words.1). and transcribed to gather user feedback.” is shown in in Figure 7(a). after which interviews were flexible paths through the poem allow users to shift their focus quickly.” vealed. She said that this path view. For less that there is typically about the same number of words per line and familiar poems. the second following very closely on the first. While the placement of nodes in this poem is mostly emage. hovering over different words in these words are of similar length. In a typical close reading. assonance. Curves and complementary soft colors. leading this collaborator to ing words and rhyme sets in the poem view and text view. in all three cases the open. there was one line that had only two the poem view. An example of how Poemage changed A cento is a poem composed entirely of lines or passages taken from this procedure for her occurs in the previously described observation other authors. These demos were either given vite engagement without visually overpowering the user. movement is left to right and down the page. The poem in person or recorded.” hovering over one sonic feature af. her exploration regular in that there are generally a similar number per line (around was guided by previous observations and investigations. pretive readings are made.3 A Cento in the (Re)making page again as far as the title. ing is absolutely conclusive. Such se- explore the rich sonic turbulence at that location and its connection lections were based on their visual impact on the visualization in the and reinforcement of the semantic flow of the poem. went “straight to the pleasure center of [her] way than she was accustomed. given a week to experiment with the tool. thereby revealing a word’s various sonic connections nodes. Poemage ing sequence of questions “Can you walk me through how you used the not only reveals patterns within the poem. the collaborators were itself remains central both figuratively and literally while the multiple. critical explorations in new ways — which in turn spur still further maining questions were answered and many additional topics were ap. or in rela- semi-structured interviews were planned. rather than at the beginning.” proached. and that noth- emage was introduced to the poets via demos highlighting the vari. She loaded the cento into Poemage and proceeded. 7(b). She said her greatest successes and in. We describe her experience using Poemage A specific example of this was an insight gained when glancing to generate erasure poems here. This ter another in the set view and poem view. one of which through idle play — as she says. this specific encounter with the poem began with the first Times written in honor of Pi day [53]. and recently exhibited several of her era- sights came in every case when she happened on something indirectly. ‘Is’ shows up in the cento sets. helped her to quickly gain entrance to stracted view of poemspace revealed an immediately visible anomaly the text. which immediately revealed some interesting results potentially as the dominant sound on exact rhyme. their own interpretive work and me how you arrived at them?” propelled a dialogue in which the re. she next turned her attention to that category of rhyme sets in tences of a journalist tend to rely heavily on the verb ‘to be. her usual practice.’ a verb I the set view.2 Erasure Poetry which provided direct narration for three of these case studies. and if so. sure poems generated using the tool in a local art gallery. She also compared the cento to other poems entirely to see what overall patterns appeared in the path view.4) — we argue that disrupting the thinking of these poets is an concept of an erasure was introduced by a collaborator from our ex- important mark of success for this work. users to explore all the possible erasures formed from single or com- lyzing a poem as “noodling. according to Another collaborator chose to explore Jorie Graham’s poem “Read. exploration. she generated new erasure poems by en- but there’s just a whole lot going on.1 Close Reading with Poemage words and show context features in Poemage. leading her 4) and they are mostly at similar distances from each other. tended network of poets in response to one of our technology probes. and especially those She commented that Poemage took her into the poem in a different using the fill function. As a second form of validation. 7 . a lot I wouldn’t necessarily have abling the show words feature and selecting one or multiple rhyme considered in quite this way without the tool drawing my idle eye — a sets based both on the shape of their paths and the subpoems they re- lot I hadn’t in all these years considered up to this moment. poem. Po. First. we discuss the impact that our col. with other words in the poem. mixed and blended through interaction. can you show see their own spontaneous choices. but in which a specific poetic event might send her back to the beginning of a line or up the 8. on the other hand. at the placement of nodes in the path view for the poem “Night” by This collaborator took several different approaches to using Po- Louise Bogan. which occurs late in the poem. Because of that she had composed by taking lines from an article in the New York observation. Because she felt “I noticed that my cento is in some ways as sonically intense as my least likely on her own to discern specific examples of complex visual poems built from scratch — with the notable difference that the sen- rhymes. By scrolling of her own making. connote changeability and in- iar to them from earlier prototypes. to noodle around. which she felt was one of the biggest bene- title and first line and forges a slow. she was able to piece to which she had managed to make another’s text her own — how the together a composite sense of the sizes and shapes of these patterns poem “looked” sonically like one of hers. She reflected on her exploration: case studies with our collaborators illustrate how Poemage not only “The multiple-view interface felt engaging and responsive and it re- supports novel analysis insights (Section 8. resulting in a new poem with potentially new meaning.” She began by hovering through words in the poem view lack of density. In a similar fashion. and consonance. She commented that the visualizations.2 and 8. she also experimented in the spatial distribution of words. bined rhyme sets and their surrounding regions in poemspace. Two of the poets kept journals of their experimentation [51]. While between minute details and single patterns in isolation. recorded. “almost out of the corner of my eye. looking for various densities or ing Plato. its pivot and crisis. and that this occurred via both the poem brain.” She also commented that Poemage encouraged her to spend view and the path view. but also how the tool flects the sensibility that I experience when reading a poem: that inter- supports making and remaking new poems (Sections 8. the ab. choice by responsive choice.3). To appear in IEEE Transactions on Visualization and Computer Graphics. Thus. Iterating on the idea subsequently led to our inclusion of the show 8. One collaborator used Poemage to explore a cento that of the node anomaly. What Poemage helped reveal to her was the extent up and down through the length of the poem. A visualization of her se- We present two forms of validation for this design study. these features in Poemage. but also enables users to tool?” and “Did you gather any new insights. 8 VALIDATION relevant to her interpretation of the poem. 8. conducted. she begins with the more time with the poem. recursive path in which the overall fits of the tool. In these ways. ous features and interactions. four lections is shown in Figure 7(c). For poems she was deeply familiar with. She recounts this: as they appeared and developed in the path view. These features allow One collaborator described her approach to using Poemage in ana. This anomaly coincided with a with building interesting shapes in the path view by randomly select- powerful semantic moment in the poem. She browsed through the mixed 4-character cluster rhyme use quite rarely when left to my own devices. The tion 8. ing text. tion to each other and the poem as a whole. many of which were previously famil. Erasure is a form of poetry generated by erasing words from an exist- laboration has had on the scholarship of our direct collaborators (Sec. not only is this the poem’s turn. Taking a slightly different approach. Following the demos.

reconsider how to include it in Poemage. gladly beyond” created by one of our collaborators and exhibited in a local art gallery. not like me? How do I unconsciously select sentences for the cento not only for their meaning but also for their sound? How does a cento in 9 R EFLECTIONS the making become more like me as I make it?” Working with poetry scholars has been a delightful.’ ‘out.” at the 2014 Modern Language Associa.1 Breaking Convention One interesting. into data and visual more important poetic insights of this work. ambiguity and visual [50]. User visualizations.’ which feels just like me. our collaborators led us to regard ambigu- tices through the lens of computation and visualization. This col- which they reflect on the impact that visualization research has had on laborator also reported that in times of feeling overwhelmed by the their poetry scholarship. We offer a number of insights we gained as visualization sentation in Poemage’s path view. which they identified more as chaos. one of our collaborators commented tion’s annual conference. and it revealed one of the conceptual metaphor of a poem as a flow.’ and ‘would. We wonder. The author came cluttered. and the computer scientists.’ ‘clouds. lem domains. however. These collaborators also published significant articles in parts of a poem work together to form complex meaning. but also In general.’ ‘you. however. tors developed new perspectives on their domain and research prac. never in my own drafts.E. combined with other observations made using Po.’ ‘source.’ new extensions of Poemage that probe more deeply into the literary and even encompassed ‘circle. was of another of these articles describes how this collaborative research a challenge for us as visualization designers. This messy view.” value clarity and readability. Since these other view of the poem shows me a wonderful set that included ‘now. our poetry collabora. and not restricting the tool to avoid such clutter. “somewhere I have never travelled. This revelation.’ ‘mouth’ ‘course. caused us to entails [36]. clutter. and discusses ways this research has ergizing and was a space that they felt very comfortable exploring.4 A Disruptive Technology 9. printed to pdf. coming to terms with how quickly the visualizations be- more complex questions of language and imagery [11]. as well as into new concepts such as question: how is my own cento (I have several of them) like me and sonic depth and the role of technology in promoting creativity. beautiful mess. In an interview. our collaborators’ excitement about ambiguity as an aspect literary questions as how poetic time operates and even what reading of the data that enhances meaning. Here we reflect on some of the issues that we believe make lit- emage. measure of success for a de. tasks from “find everything interesting in a poem” to “visualize the The beautiful mess appends established visualization principles that sonic topology. if there is a degree of novelty in the nology. Asking them ity as a fundamental source of insight. namely how the constituent time [12]. designers and provide several suggestions for future work. yet difficult to capture. which in turn enabled them to envision to investigate this topic in greater detail in our future work. more traditional visualization prob- arranging the lines and passages based on the resulting visual repre. she turned to the beautiful mess to ground and re-energize not on the technology itself but on how the need to teach the tech. and specifically on the notion of turbulence and poetic other poets seek to understand in a poem. (b) A visualization showing a cento created by one of our collaborators. what to look for and visualize in beautiful mess that may wear off in time — we plan to revisit the util- poems forced her to be much more precise in her own thinking about ity of this view in the future. In one of these articles. in a rhyme. was inviting and en- software was ready to explore. 8. Cummings poem. shown in Figure 4 . (a) An example of an erasure of E. process. re. In this research. In the field of visualization. As part of a session entitled “Things My Computer collaborators told us they would not have made this insight without the Taught Me About Poems. we plan a sound led to new thinking. our two primary collaborators presented on that the beautiful mess was completely representative of what she and this metaphor. avoiding ambiguity is the norm. and challenging. was consis- One such insight was the re-articulation and development of the tently one of our collaborators’ favorites. new ways of approaching a poem and to narrow the scope of their This work also challenged us to embrace a degree of visual clutter. Through the course of this design study. While Poemage includes some to define precisely what is interesting in a poem.’ interviews. 8 . This initially made me a little despondent — We put forth these results as an important validation of the impact should I give up making centos from the New York Times? — but an. this work has had on our collaborators’ poetry scholarship. inspired this collaborator to “sonically reload” her cento. rather than clouds it. This research challenged us to embrace concepts that visualization sign study is the act of disrupting the thinking of a domain expert conventions tell us to do otherwise. led her to re-evaluate and re-imagine her theoretical positions on such Similarly. however. herself. "OPOZNJ[FE B C D Fig. and in features that allow users to explore ambiguity within the data. specifically. we have continued to work with these poets to explore ‘down. erary studies different from other. not only sonic devices and how they signify within poems. To our collaborators. Our metaphors. 7. the author focuses technology. (c) A visualization of the poem “Reading Plato” by Jorie Graham. prompted her to re-read familiar poems in new ways long before our this clutter. This raises a concepts discussed in this paper.

etc. and M. et al. C. We validate our design through several motivated a somewhat informal branch of text interpretation delight. from our approach to conducting research to the design of tech. our research fits in the evaluating vi. In our undertaking to visualize poetic sound. University of Chicago Press. 22. and M. In Computer Graphics Forum. betic signs as a basic digital technology that operates. Bastian. Although we are continuing to improve software for exploring and manipulating networks. H. M. G. dering its sonic patterns as visual forms does not diminish sonic plea- sual data analysis and reasoning scenario [34]. we believe that developing a deep understanding of the in.. Therefore. We believe that such ers learned to do the same. Chen. linesets. search outcomes and play an important role in guiding exploration ventional visualization if explicitly. In our case. gether. Czerwinski. a fundamental component of our collaborators’ research. this project partici- plore a range of possible outcomes on the part of our collaborators pates in several persistent lines of critical inquiry. which can create experi- 9. browsing. 31]. Design study of In a similar vein.2 A Screwmeneutic Approach We present a two-year design study exploring the role and influence Within the digital humanities. and thereby lead to fuller aesthetic and termining how to measure our success. which deliberately engage cross-modal perception [5. user-driven selection of words. pages 237–248. the range of valid inter. For now. to generate an indefinite number of unique and sound in poetry. include a problem characterization and data abstraction of the use of ther literary criticism. nology probes and Poemage. in this research.” [47]. literary positioning more fully. Johanna Drucker’s observation that poetic texts activate both sonic and visual elements as “intersecting codes” [21]. Maguire. encouraging richer exploration and greatly Throughout this research. N. Sound as subject: Augusto de campos’s poetamenos. 28. Visualization and Computer tions presented in this paper place a stronger emphasis on the role Graphics. Amanda Hurtado. or at the same time [31]. tersection of technology and the humanities is fundamental to creating [5] A. Dworkin.gicentre.3 Measuring success ences that are synesthetic in the etymological sense of perceiving to- An underlying challenge throughout this design study has been de. This experience taught us to be willing to put some of our own Finally. we also evaluate new kinds of insights and how the Gleicher for their contributions to this work. Rule-based visual mappings– what is the role of technology in a creative pursuit like poetry? We with a case study on poetry visualization. The Sound of Poetry/The Poetry of tices — extending beyond close reading. connects back to our investigation of the role and disruptive impact of This work was funded in part by NSF grant IIS-1350896 and NEH technology on close reading. The visualizations that can emerge when poets are given a tool that supports free-form could contribute to discussions regarding concrete or synesthetic’ po- exploration. 8:361–362. Riche. and we encourage others to do the same. 9 . in addition to highlighting specific insights gained We would like to thank Craig Dworkin. we call for probed their resistance to integrating technology into their practices in an adapted set of evaluation guidelines for conducting research in the the hopes of advancing their research. In reflection. We attribute this to several interpretive engagement with poems. First and foremost. K. This is a lesson that we plan to bring to guidelines could extend to other arts as well. and inter- research findings. M. M. and Michael using our tool. Poemage the text of the original poem remains always in view. 2011. Wynne. pleasure and enjoyment are productive re- design principles aside and to be open to experimenting with uncon. This term comes from an influential and illustrate the disruptive impact that technology can have on poetry paper by Stephen Ramsay entitled “The Hermeneutics of Screwing research. and specifically in literary studies. of poems promises to exploit this engagement in rich ways. Abdul-Rahman. [1] gicentre utilities. J. R. ICWSM. 10 C ONCLUSIONS 9. A. the question arises: R EFERENCES could randomly selected sets of words yield equally valuable and equally abundant insights? Our collaborators have hinted that the an. this method ACKNOWLEDGMENTS seems less indicative of success than it perhaps might be for a different domain. One consequence of this. Ramos. Wiley Online Library. We plan to explore machine-learning algorithms that We embraced this philosophy whole-heartedly throughout this design would enable a more free-form specification of rhyme rules through study. 2013. on and because our collaborators gained so much from this aspect of the visual and aural/oral registers [5. we see the various components of Poemage addi. In future papers. 29]. or implicitly. In the immediate future. but it has the insight-based evaluations. a visualization tool for interac- sometimes radical interpretations and to guarantee continued mean. Johnson. humanities.. creating a visually pleasurable research environment goes cited about and also helped us to better understand the problem space. Since in pretations makes comparing against ground truths fairly unproductive. swer to this question might be yes. and impact of technology on poetry scholarship than on the support of [4] M. pages 381–390. we wanted to allow users to conduct similar in. terns by a poet. Tools created with a screwmeneutics sensibility en. as well as Poemage. or regarding how electronic design study process. In addition to being a culmination of our logical questions through coding languages. Poemage. Heymann. an adventurous wave of research re. and we as visualization design. Trefethen. tion of Poemage. IEEE Transactions on. http://www. The computational visualization different complications. Our validation results in Section 8 ex. face [38. which encourages sure readers might experience apart from visualizations. Doing so [36] [11] [47]. Concepts of text deformation [41] and tamperings [33] was heavily informed by a series of technology probes iteratively de- have energized members of the literary criticism community and have veloped with our collaborators. beyond general aesthetics. Sound. a novel set visualization technique. are continuing to explore these ideas with our collaborators and are volume 32. case studies that highlight the kinds of insights gained using our tool fully termed screwmeneutics. To appear in IEEE Transactions on Visualization and Computer Graphics. [3] B. re- Because a significant component of this research was investigating garding reading and writing about text and image and about alpha- the role and impact of technology on the experience of close reading. editors. in which gathering insights is strong potential to enhance it. tively exploring the sonic topology of a poem. Jacomy. future collaborations. Meyer. 17(12):2259–2267. grant HD-229002. we will articulate Po- tionally as technology probes for end-users — opportunities to explore emage’s theoretical. We would also like to insight gathering process may have changed using Poemage. digital (re)mediation further complicates semiotic and phenomeno- vestigations using our tool. which is the specification of interesting rhyme pat- courage a certain amount of playfulness and creativity from their users. of technology on the close reading of a poem via an investigation of jects the notion of using computation to solve a text or to verify sound and linguistic devices in poetry. 2009. designing user studies to test our hypotheses. Alper. Bessa. our collaborators actively challenged and increasing the overall efficacy of the tool. 31]. we note the impact of technology on their individual reading of a poem. ren- In terms of proper evaluation. The results of our research existing hypotheses. and instead focuses on using computers to fur. 2009. S. we acknowledge that the findings and contribu. etry. hardware. and as we found through our led us to include features that our collaborators were genuinely ex. Perloff and C. S. In reflecting on the outcome of this research. Gephi: an open source close reading with Poemage. we plan to address a current limita- Around. Coles. requested. and play. In truly effective tools supporting a broad range of literary studies prac. validation. The design of Poemage ing making. M. which thank our reviewers for their feedback in improving this manuscript. already. which leads to a second question: [2] A. Lein.

H. Muralidharan and M. B. meaning in large bodies of literature. editors. 2012. 2009. van Kreveld. Digital Studies/Le champ num´erique. [9] T. E. with wordle. Jigsaw: supporting investigative analysis Skeleton-based edge bundling for graph visualization. Chicago. and M. Feinberg. into bundles. 11(4):155–162. 19(11):1846–1858.-L. In Visualization Symposium. Hayles. G¨org. [24] O. Gregory. C. uncertainty. research. IEEE Pacific. Visualization and Com- 2009. Boyd. Don’t expect math to make sense. Frid-Jimenez. Computer Graphics. Not sound. [26] S. 2012. Kelp diagrams: Point set membership visualization. A. and D. Sedlmair. Telea. [46] S. Van Ham. [31] P. Chaturvedi. Plaisant. Wattenberg and F. Don. 30(1):25–56. Penn. P. pages 136–147. Lieberman. pages phrase nets. A. Wise. Chen. Meulemans. [17] W. ACM. visualization tools: multi-dimensional in-depth long-term case studies. Collins. Stasko. Cui. Kelpfusion: A hybrid set visualization technique. time. Ersoy. context. Strategies for evaluating information [22] J. 2002. Supporting exploratory text analysis tagged text for literary scholarship. B. [40] N. Routledge. M. Coles. In Proceedings of the 2006 AVI workshop on BEyond time and errors: [23] J. and J. In MLA2014. Visualization and tics for Literature. Poetics Today. Plaisant. Compus: visualization and analysis of [53] M. 5–21. Alper. S. Plaisant. Ramsay. Print is flat. Pupyrev. In Graph Drawing. 132. S. J. Wattenberg. 449(7163):713– [12] K. IEEE Transactions [27] B. tion. Collins. 2002. [55] F. Zheleva. E. Seven ities. 14(6):1277–1284. volume 29. Gleicher. 1995. Schur. Digital Human. Heer and D. 1999. bliss: Close reading. IEEE Transactions on. pages [29] N. The word tree. Vizster: Visualizing online social networks. 2. att. Hayles. N¨ollenburg. A. Telea. B. Westenberg. or what you do with a ics Forum. 2010. E. [32] H. Wester Humanities Review. Univer- B. analysis. pages 1–7. IEEE Symposium on. Auber. C. L. Nature. and K. 1998. Beaudouin-Lafon. and D. An improved algorithm for the metro-line crossing min- clustering for graph visualization. [54] F. R. Liu.-B. pages 51–58. 2004. Hutchinson. Reading machines: Toward an algorithmic criticism. Lantrip. Tampering with the text to increase awareness of poetry’s ficVis’ 09. Woodhull. In Proceedings of Computational Linguis- lations with isocontours over existing visualizations. Digital Humanities Quarterly. Nachmanson. 2011. Clement. 2006. Graphviz. [49] H. In 8(2):pp. Information visualization. [25] J. Tang.-J. Reas. 2010. Poster Compendium. and M. H. Proceedings of the SIGCHI conference on Human factors in computing [7] H. Slippage. University of Chicago Graphics. Tarkan. J. M. pillage. 2008. 1995. D. 25(1):67–90. tion with information from text documents. 2013. volume 31. pages 731–740. 2007. Pennock. B. Witmore. Ellson. N. Crow. Bourqui. structured documents for understanding social life in the 16th century. 2014. G. 1973. In Proceedings of IEEE Symposium on Information Visualiza. Humanities approaches to graphical display. G. C. M. V. E. [48] S. 32(5):399– and V. Isenberg. 2005. McCurdy. 2011. Visualization and Computer Graphics. B. Exploring collections of [43] A. [20] A. INFOVIS 2005. Viegas. Tcheng. [52] J. In Proceed. Processing. Discovering interesting usage patterns sity of Illinois Press. Koutsofios. 4(2):155–169. and poetic conse- patterns in texts. History. Hearst. Visualization and Computer structure. J. A philosophy of computer art. Meyer. Qu. Collins. Munzner. Mapping text with In Proceedings of the fifth ACM conference on Digital libraries. ACM. Docuburst: Document content visualization using language bundling for interactive graph visualization. 2013. 5(1):1–21. 2012. 16(2):263–290. Gansner. [51] B. A. [50] M. Wiley Online Library. 2010. 14(6):1221–1228. New Literary [15] C. Correll. Sounding the surfaces: Computers. 2000. and Z. [45] W. Distant listening or playing visualizations pleasantly with the [35] A. [57] M. pages 39–65. Lopes. Coles and J. [34] H. L. J.(theory and practice with a hispanic perspective).-L. [39] S. Drucker. 2008. on Information Visualization. ware architecture diagrams. E. B. A. Shneiderman and C. and M. In Information Visualization. Visualization and Computer Graphics. Wicked problems in design thinking. M. Ma. [38] D. M. Byelas and A. Geometry-based edge [44] M. pages 237–248. Plaisant. Carpendale. Bubble sets: Revealing set re. [41] J. [28] N. [33] E. Bederson. Zhou. Lein. in literature study. Visualizing metrics on areas of interest in soft. 2013. Perloff and C. 2011. pages 91–98. Visualizing the non-visual: spatial analysis and interac- 404. T. Dinkla. an interactive visual [30] J. Rittel and M. Edge routing [19] K. 2009. Wiley Online Library. org. [18] P. Winding roads: Routing edges eyes and ears. 2002. IEEE Transactions Information Visualization. 3(2). Visualization and Computer of Poetry/The Poetry of Sound. 2015. In concordance. Dwyer. 23(1). Gist icons: Seeing In Poster presented at IEEE Symposium on Information Visualization. IEEE. and E. H. vol. McGann and L. Vi´egas. Hansen. B. IEEE. imization problem. Evans. [16] M. Jackson. IEEE. S. Meyer. million books. Hurter. Samuels. and A. T. novel evaluation methods for information visualization. VAST 2009. F. 2011. with ordered bundles. pages 381–392. URL: http://www. K. In Visual Analytics Science and [42] W. Turbulence and temporality: (re)visualizing poetic 716. 2005. 2008. Speckmann. Arc diagrams: Visualizing structure in strings. Wattenberg. North. spillage. DeCamp. Westerlund. Digital Human. ACM. B. guiding scenarios for information visualization evaluation. Dufournaud. on. et al. and M. IEEE Transactions on. Deformance and interpretation. IEEE Transactions on. and S. 18(5):810–821. In Computer Graphics Forum. A. Lein. 2009. Fish. quence. In Graph Drawing. 2009. Participatory visualization New York Times. Design study methodology: [21] J. 2009. Fall 2014. 2006. In Computer Graph. Irizarry. pages 853–862. Visualization and Computer Graphics. W. 10 . Dilemmas in a general theory of plan- ings of the sixteenth ACM conference on Conference on information and ning. Speckmann. Drucker. Guiness. J. In Computer Graphics Forum. mation Visualization. and F. and T. and T. [47] S. 2012. and M. Mackay. Quan- and machines. Ambiguity-free edge- [13] C. IEEE Transactions on. Hertz. Springer. Viegas. Druin. Literary and linguistic computing. and M. H. ACM. J. and G. puter Graphics. Wattenberg. Visualization and Computer Graphics. Carpendale. Buchanan. 2005. Srikumar. D.-Y. Hodgson. [58] J. M. Riche. Textarc: Showing word frequency and distribution in text. In M.. plore and analyze faceted text corpora. Webber. lyzing sonic devices in poetry. 15(6):1009–1016. Auvil. S. INFOVIS 2002. 2012. 15(6):1137–1144. pages 33–40. 2009. Bereg. Graphics. code is deep: The importance of media-specific 110–116. B. Mandell. Parallel tag clouds to ex. The New York Times. Processing. Suri. The Sound Reflections from the trenches and the stacks. Wester Humanities Review. Liu. guistic computing. 2009. L. Pottier. IEEE. V. Synesthetic art—an imaginary number? Leonardo. K. Dworkin. 2003. Technology probes: inspiring design for and with families. M. Conversy. Springer. IL. Auvil. Wattenberg. Translating media: Why we should rethink textuality. L. P. Roy. G. IEEE. A. and X. Technology. Clement. IEEE Transactions on. tifying the evolutionary dynamics of language. Lambert. 7(2):118– Computer Graphics. on.-D. things my computer taught me about poems. Mind your ps and bs: the digital humanities and interpretation. org. Monroe. pages 17–24. Armstrong. C. Stephen Ramsay. Fall 2014. 2011. B. Sounding Wiley Online Library. 2007. Holroyd. [11] K. Cantareiro. M. pages 875–884. IEEE Transactions on. Bertini. H. and S. 7. knowledge management. The hermeneutics of screwing around. for meaning: using theories of knowledge representation to analyze aural [36] J. F. on. C. 28(2):283–295. 17(12):2364–2373. IEEE Transactions 47–55. [37] E. 2003. art. Jan. Thomas. Literary and lin- [8] M. Fekete and N. Fry and C. systems. B. Li. Michel. Capitanu. Clement. Paulovich. B. H. 2012. Gannod. and A. Policy Sciences. and C. ities Quarterly. 17:103. In Infor- Yale Journal of Criticism. pages 84–109. 1999. ume 30. Paley. com/sw/tools/graphviz. Ramsay. Nowak. Lam. 2012. C. [6] R. Press. Myopia: A visualization tool in support of close reading. 1996. pages 32–39. M. 2009. [10] T. Paci. W. Shneiderman. 2010. Wong. The [56] M. K. Design Issues. IEEE Symposium on. Proceedings. B. Luo. Vi´egas. pages 213–222. in text collections: integrating text mining with visualization. Rhymedesign: A tool for ana- [14] C. 18(12):2431–2440. IEEE Symposium on. In Proceedings of IEEE Symposium volume 2002. 15(6):1169–1176. 1992. Visualization and through interactive visualization.