Digital Chaucer Pedagogy and Editing

Probing Generative AI’s Reproduction of Hegemony

Sophie Whittle, University of Sheffield

DOI: 10.55520/ZXSXERGH | Scholarly Editing, Volume 42

§Introduction

The rush to adopt AI technologies and “tools” in pedagogical and editorial practices is a current challenge facing scholars in higher education. These concerns are particularly felt in arts and humanities disciplines, where the current funding landscape demands further communication about the value of this research. For (digital) medieval studies, the concern is that automated editorial processes may take scholars further away from close, nuanced appreciation of text, a longstanding characteristic of the field’s work. This paper investigates and probes the use of generative AI to determine if its proposals to interpret and (re)imagine literary works can be substantiated within digital pedagogy and editing. More specifically, I reflect on the importance of recovering marginalized stories within “canonical” texts, using queer and intersectional approaches to pedagogy and digital humanities, within the current age of AI. The analysis I present arises from the development of a prototype digital teaching edition of Geoffrey Chaucer’s Pardoner from The Canterbury Tales, for the C21 Editions project. 1 I center the insights of current students, educators, editors, and practitioners to situate my analysis of artificially generated responses to Chaucer’s literature.

Generative artificial intelligence (generative AI, or genAI) uses large language models (LLMs) to perform a wide range of natural language processing tasks, which scope large training datasets for statistical patterns and probabilistically generate content, based on these identified generalized patterns. To counter AI outcomes deriving from dominant discourse, I also draw on research from a recent issue of the Critical AI journal,2 which critiques the origin and development of LLMs without resorting to techno-determinism. Using these different research and educational insights to inform my approach, I propose how hegemonic flaws within LLM responses can be interrogated within the environment of a digital edition, to empower students to make decisions about their use within pedagogical spaces.

§The C21 Editions Project

C21 Editions initially investigated the use of state-of-the-art methods in scholarly editing to determine whether machines can improve the speed and accuracy of editorial tasks (e.g., annotation, glossing, and linkage of text), all while maintaining the detail and granularity of editorial curation. This curation within the digital humanities (DH) entails both distant and close reading of texts, bringing together the “curatorial and statistical” aspects of the analysis of literature, dimensions that are often treated disparately.3 There is little research at present highlighting if and how LLMs can serve scholarly and pedagogical editing, and there are few investigations into its limitations for producing and representing work in the humanities, particularly for medieval literature. Given the heightened hype and fear around generative AI and its impact on different communities, the project posed an opportunity to uncover how LLMs treat medieval texts, particularly stories of marginalization within literary “canons.”

Focusing on Geoffrey Chaucer’s literature draws attention to the shifting conversation about white male authors’ place within the university curriculum. For Chaucer, his position as a central figure head or “father” of English literature is questioned.4 This debate comes from recently discovered legal documents detailing Cecily Chaumpaigne’s release from her raptus, and Chaucer’s role in formations of rape culture through time. Baechle and Harris urge scholars to not overlook how the raptus discussion is part of wider systemic aggression against groups of people within medieval texts.5 In pedagogical contexts, the focus on Chaucer necessitates an interrogation of attitudes toward groups at the intersections, by recovering and centering stories of women, the queer community, racialized people, and the working class. Because of the multiple voices layered throughout the Tales, the frame narrative provides intriguing starting points for students to explore humanness without “deifying” white authors.6 The character of the Pardoner and his tale presents a route into this exploration; a pilgrim whose gender identity and sexuality has been viewed as fluid in recent years, which inherently challenges the normative structures of the medieval Catholic Church.

§Research Questions

The use of generative AI “tools” for learning about marginalized stories in Chaucer’s literature presents tensions between the investigation of these human voices and the concerns these technologies raise. There are two main questions I address in this paper, based on recent calls to probe LLMs: (1) How do generative AI tools limit student access to digital medieval approaches with their homogenized responses, particularly approaches that are inherently queer and intersectional? And (2) how can research and teaching within Chaucer studies, and the role of the digital humanities in society, benefit student learning as AI technologies continue to “evolve”? Both questions highlight the need for continued dialogue with students, teachers, researchers, practitioners, and the public across multiple disciplines about how best to develop technology for the benefit of all, with the view to challenging cultural hegemony. These pressures are in addition to those faced by educators when making space for in-depth critical discussion of medieval texts and their socio-historical and -political context.

§Queer and Intersectional Approaches to Chaucer Studies

Intersectional frameworks owe their approach to the work of Kimberlé Crenshaw, who theorized the term “intersectionality” to draw attention to the ways in which discrimination affects communities of multiple, overlapping identities in the legal sector.7 Along with many other Black feminist writers and activists, such as Sojourner Truth, Audre Lorde, Angela Davis, Patricia Hill Collins, and bell hooks, Crenshaw explains that intersectionality underpins and cultivates research within many disciplines, and is not a “closed critical framework.”8 This paper draws on the foundations of this research in highlighting injustices at intersections of race, gender, class, and sexuality, through the recovery of these stories in medieval “canonical” texts within digital education contexts. Conversation is often facilitated for students to critique top-down social and political approaches within the medieval period, to reach the “dynamics of difference” and “solidarities of sameness”9 within well-known authors’ works. For the field of digital humanities, there are queer, feminist, and critical race approaches for challenging methods of text analysis, and I position an analysis that probes generative “tools” for pedagogy within these frameworks.

As Boyd and Ruberg theorize, digital humanities as a practice is fundamentally queer; it challenges long-assumed norms of investigation, interrogating the methods we use to further research, teaching, and learning. Two of their descriptors central to the current investigation are “estranging” and “skeptical.”10 Study of the fluidity of expression of pilgrims in Chaucer’s works lends itself to uncanny and unexpected outcomes for challenging medieval ideologies of sin, provoking discussion which was initially divorced from analysis of religion in the Middle Ages. On top of this, researchers must question and be skeptical of large models that promise to explore untapped areas of study, especially where they promote heteronormative narratives. Gairola identifies the existence of AI algorithms modeled on normativity, despite their commitment to center queer people.11 These technologies are not neutral apolitical tools, and while they hold the capacity for social change, they have their “own endogenous interests and value-structures” that “cannot just be naively ‘applied’ to the humanities.”12 As this knowledge and understanding around LLMs is brought to light, researchers must challenge the use of generative AI in every context and discipline, given the uncertain basis around which it is built.

§The Pardoner

The character of the Pardoner in Chaucer’s Canterbury Tales, a medieval preacher raising funds for the Church, tells an intriguing story for students entering digital and medieval textual research. The tale is built around the Pardoner’s confessions, where he makes his transgressions obvious to the reader and his fictional audience, namely his selling of false relics to innocent people and his promises to absolve their sin.13 The Pardoner’s identity also challenges medieval normativity when the narrator announces, in the “General Prologue” to the Tales, the unknown nature of the Pardoner as a geldyng or a mare, translated in many works as “eunuch” or “homosexual.”14 Burger and Kruger’s early focus on “queer Chaucer in the classroom” calls to extend the disruptiveness of the Pardoner to analyses of the entirety of the tales, including other sexualized figures such as the Wife of Bath and the Summoner, and facilitate teaching that is “rigorously historicist and yet urgently attentive to the current moment.”15 They urge readers to consider a society before modern categorization, and how these characters disturbed the construction of white and masculinized forms of pilgrimage, as asserted by pilgrims such as the Host and the Knight. In addition to Burger and Kruger’s piece, many researchers have highlighted how the Pardoner, while silenced by characters such as the Host, continues to question the morals of the Church through his own transgression. There are accessible perspectives for students to digest and take further when it comes to newly generated “interpretations” of Chaucer’s tales, providing them with the skills to continue to challenge the cis-heteronormativity of the past and present.16

Further decolonial movements can be seen through modern adaptations used to learn about the flexibility of the form of Chaucer’s work and decenter whiteness in the university curriculum. There are no racialized characters in the “Pardoner’s Prologue and Tale,” so authors such as Patience Agbabi have created identities embodying the subversiveness of the Pardoner in new cultural contexts. Agbabi’s Telling Tales introduces Yves Depardon in a modern London, while taking inspiration from the prosody and vernacular of Chaucer. Her Pardoner is “recast as a motivational speaker,” using the same Latinate phrase Radix malorum est cupiditas (“Love of money is the root of evil”—Telling Tales translation) to closely link Yves Depardon to the sin he speaks against: “I stand before you, / guru of Gordon Gekko, ‘Greed is Good,’ / a liar, forger, thief: thigh-deep in sin.”17 In addition, poet Jean “Binta” Breeze and novelist Zadie Smith have produced The Wife of Bath in Brixton Market and The Wife of Willesden, respectively, to celebrate the lives of modern Jamaican-British women.18 Marion Turner’s students, as shown in her recent exhibition “Chaucer Here and Now,” recognized the relatability and joy of digging for human empathy and humor within the adapted works of Chaucer’s literature. Students in Chaucer studies therefore develop skills for inspecting medieval texts and retrieve their own connections with the works—a method of close reading that requires human cognition, in stark contrast to the way in which LLMs scrape their data.

§Approaches to Pedagogical, Critical, and Automated Editing

Digital scholarly editions with a pedagogical aim provide an opportunity to enrich the types of skills development of a medieval literature and Chaucerian university course with further avenues for independent research and collaboration. Having noted the rich body of work adopting queer and intersectional approaches to Chaucer studies, one way of empowering students to continue to disrupt normative standards of interpreting text within institutions is to experiment with various digital approaches.

There are a wide range of digital pedagogical editions that empower students to train within digital humanities and take up active roles in the creation of resources. For instance, Taylor’s open approach to critical pedagogy within a digital Shakespearian edition led to students bringing in their lived experiences of queerness to connect with the text.19 Similar researcher-student collaboration and solidarity can be found with Sutherland’s analysis of the Seward Family Digital Archive, where students communicated on transcription conventions through helpful glossaries in response to potentially inaccessible traditions of editing and the barriers of nineteenth-century archaisms.20 Bryant and colleagues also contemplate how students might come to find and engage with research “on their own terms, in their own good time, through their angles of vision,” emphasizing that editions are places for discourse, whether that be about original sources, historical contexts, interpretation, or adaptation.21 If generative AI were to be introduced into such a teaching and learning space, the question is how these technologies might impact its dynamic, and what additional support is required for students to interrogate the content and origin of its outputs.

There has been a shift in critical edition-making in the twenty-first century, with van Zundert calling for digital editions to represent “textual fluidity and text relations in a scholarly viable and computational tractable manner.”22 There is a simultaneous risk, with the rise of machine learning and LLMs, that we move away from the critical and reflective nature of edition-making in favor of increased automation.23 Projects such as Transkribus, a program offering AI-powered automatic text recognition (ATR), allow for increased focus on the expansion of the number of voices that are considered within editions, while lifting some of the “procedural barriers regarding transcription, formatting, and hosting of digital editions.”24 These initiatives, which involve AI for specific tasks and purposes, can bring individuals together from multiple disciplines to understand how digital editing can cover a more diverse range of narratives. In contrast, might larger language models than those used by ATR promote further homogenization of textual analysis and interpretation? Editors wishing to increase automation via LLMs must thus consider a cooperative framework for probing outputs if we are to encourage the development of key reflection skills in future student generations. As bell hooks emphasizes, “engaged pedagogy necessarily values student expression.” To fully commit to centering an authoritative student voice, educators must be willing to share their experiences of failure and risk-taking in order to resist and highlight the biases of dominant structures.25 The incorporation of generative AI within the space of a digital edition would need to facilitate the ability to raise questions and make mistakes on the part of both the teacher and student, especially as more room for error is introduced with LLMs.

§Methods

In the following sections, I present multiple insights from students, lecturers, researchers, and editors in Chaucer and medieval studies, digital humanities, and English literature and linguistics.26 These perspectives shed light on the skills developed within these fields, to bring to the fore the marginalized stories and identities within medieval texts, while new technologies continue to impact digital editing and pedagogical practices in different ways. Their insights also came at a time when there were growing conversations and concerns about how machine learning and AI may begin to impact their way of working. Looking back at these discussion points retrospectively shows how these technologies impact the study of medieval texts in the present day.

I begin the analysis by first exploring the preliminary user focus groups and interviews with students and teachers in the UK and US, which were conducted from May to September 2023 to understand contemporary Chaucer teaching, the resources that work in these communities, and any requirements for digital editing practices. In addition, I draw on data from interviews with theorists and practitioners on the future of digital editing from C21 Editions from 2021 to 2022.27 Building on these perspectives, I then probe the generative AI responses that formed an initial development phase of the prototype of The Pardoner’s Prologue and Tale (henceforth, PPT), where I attempted to draw out queer and intersectional lenses for analyzing Chaucer’s texts from the LLMs. Here, I build on recent critical approaches to AI to exemplify why unsatisfactory and unfocused responses occur, placing this scholarship within the medieval teaching context. This is particularly important given the uniqueness of learning around the Middle Ages and its complex textual tradition, dissemination, and ability to conjure and inspire multiple characters and stories. The final focus is on suggestions for collaboration, reflection, and critical thinking on the effect of LLMs both within and around digital pedagogical edition frameworks, to ensure generative AI “tools” are not hastily entrenched into our day-to-day teaching plans.

§Pedagogical and Scholarly Editing Insights for Focusing on Marginalized Histories

Digital pedagogical editions continue to foster and build on many of the skills developed in Chaucer studies, in courses with and without digital humanities provisions. Our interviewees referred to several of their skills as cultivated by educators and researchers, which are crucial for countering hegemonic discourses through texts and new technologies. For example, Derksen and Fokken, focusing on Dutch colonial history, locate marginalized voices within colonial and patriarchal sources by unsettling these works through various textual, visual, and material lenses.28 For Chaucer studies, a similar approach can be taken where students view characters such as the Pardoner as disruptors and nonconformers in the medieval period.

Even though Chaucer adopted “fence-sitting” strategies by writing purposely ambiguous characters, the existence of these characters reflect the murky waters of attitudes toward gender and race in medieval England.29 Later adaptations also highlight how contemporary writers, artists, and activists reclaim the form of Chaucer’s texts to celebrate the diversity of England today. Students in our focus groups expressed their ability to locate multiple interpretations in medieval literature, as encouraged by their lecturers. S2 found that his peers were more engaged in the literature if prompted by their teachers to develop their own interpretations, and S3 found that newer directions in medieval literature courses can be supported by digital editions that provide different options for independent research. S9 explained how she could adopt creative approaches to her research, where she delved deeper into experiences of different women as they navigated medieval society. Their teachers confirmed that they aim to empower students to find their voice within a particularly dense and complex medieval literary tradition. One lecturer reflected the sentiment of the other lecturers we interviewed:

I try not to underestimate how curious and intelligent my students are because often they come up with brilliant questions and really interesting insights, and that really happens when you give the floor to them and when you give them things to work on. (ID: T1)

T3 also explained the importance of choice when it came to assignments and independent study, while providing digital resources such as the OED, corpora, and historical thesauruses to track language development, especially where students needed to uncover editorial and translation decisions within a text. At the same time, T6 provides direction for students to discover hidden meaning and interpretation within The Canterbury Tales, acknowledging that as well as these more open approaches to study, students must be supported to avoid disorientating them. What came through strongly in these pedagogical conversations was the motivation to constantly explore different ways of approaching Chaucer’s works in ways that build on a large body of research but also provide fresh perspectives. Students’ eagerness to investigate medieval works must be maintained alongside digital tools promising increased access to knowledge to avoid dominance of one particular discourse.

There are clear foundations within Chaucer and medieval studies for students to explore areas that mean something to them. In their conversations with peers in the focus groups, they expressed their interest in different interpretations of medieval texts, particularly research emerging at the intersections. In the first group, S1 and S2 considered analyses of the “subaltern” and those excluded from medieval narratives. They referred to their desires to seek out queer gazes, complex discussions of bodies and anatomies, and nondominant religious discourses within the tale. S1 expressed the difficulties of analyzing discrimination across all time periods, especially as it still exists today, ensuring to focus on cultures outside his own, including Buddhist and Jewish parallels with the Old Man of PPT. Similarly, S2 was interested in androgynous perspectives of the Pardoner, stating that “humans have always been very nuanced and there’s always been 50,000 shades of grey.”

In the second focus group, the idea of queer pilgrimage was raised. For instance, S3 noted the contemporary readings of the Pardoner as a queer figure, a type of medieval transgression which he viewed as more interesting than characters such as the Knight “who are admired and validated for all the wrong reasons.” S6 also described the pilgrims of the Tales as rebellious against norms of religion at the time:

It’s going against the grain of the idea of pilgrimage. There are religious characters, there’s a prioress, there are nuns, and there’s a parson—so there are religious representations—but they’re going against the norm of pilgrimage. Doesn’t that immediately put them into a queer space? (ID: S6)

The third focus group was interested in decentering patriarchal interpretations of Chaucer’s works and focusing on feminist perspectives. They discussed the unexpected variety of women characters in Chaucer’s works, while simultaneously acknowledging the development of language perpetuating sexual assault and rape culture. Similarly, the final focus group emphasized the need for guidance for teachers and students around experiences related to gender and sexual fluidity, in the face of rising discrimination in the present day and the shifting focuses to underrepresented communities within medieval texts. Referencing Carissa Harris’s work on medieval authors’ perpetuation of rape culture, one student stated:

Not that the conversation shouldn’t centre on the implications of rape culture, because that’s part of a long legacy starting with Ovid and several other authors who perpetuate this idea of rape culture or sexual assault as the butt of the joke. I also think it’s part of a larger historical conversation about this movement from investigating people that are not typically investigated and making them part of the narrative. (ID: S10)

In our interviewing phases, conversations with students and their teachers occurred separately, yet students’ interest in uncovering many of the topics mentioned can be traced to the values instilled by their teachers. For example, T6, the teacher of students in the final focus group, encouraged them to consider narratives from lenses of consent and make links between Chaucer’s appreciation of the vernacular to the work of Black educators, who question the need for white male-oriented forms of poetry in current works, such as iambic pentameter. Today’s lecturers are highlighting how Chaucer’s language and literary themes introduce instability to the ways we normally view medieval texts and fostering these conversations is crucial for probing any source, regardless of its provenance.

Nuanced consideration of all areas of medieval textual history is also crucial for digital editing practices. Alex Gil and Aodhán Kelly speak to the capture and recovery of marginalized histories and the need to reach wider audiences through digital editing practices. In particular, Gil emphasizes the need to record and showcase the people’s history as opposed to the elite few, to avoid continuing to represent “a history of founder years.”30 In a similar vein, Kelly explains there is still a place for traditional and historical roles of scholarly editors, but that editions must open up pathways for scholars, students, the public, and people from diverse walks of life to contribute as editors and users.31 New digital tools are referenced in the interviewees’ discussions, but there is a sense that one must not forget to return to messages of erasure, as we approach a rise in black boxes that hide the internal workings of language models. There is perhaps a dissonance between increasingly automated practices, the outcomes of this automation, and claims that they center a diverse range of communities.

§Machine Learning Approaches to Digital Editing and Teaching

New approaches are always developing which consider Chaucer’s literature in different lights, appealing to communities that are not centered in medieval text interpretation and subsequent editions of these works. With increasing knowledge about the potentials of machine learning and natural language processing tasks, there is greater responsibility to avoid undoing the body of work forefronting a plurality of voices within the literary canon. Before generative AI was widely known in its current form, we received a range of opinions about how machine learning will affect our editorial and pedagogical practices in positive, negative, and neutral ways. I will refer to these insights here before showing elements of LLM response analysis conducted later in the project.

Many of the interviewees referred to automated transcription processes, particularly Optimal Character Recognition (OCR) and Handwritten Text Recognition (HTR). In terms of automation for the critical apparatus of a digital edition, interviewees noted where automation might create opportunities for editors to spend more energy on creative-critical practices. For instance, Joris van Zundert argues that the development of AI technologies might result in editors spending time understanding how deep learning methods meet their aims for the resource, with some tasks requiring manual labor regardless of the advancements.32 Likewise, Elena Pierazzo has been working with computational methods of editing, taking the same approach to representation as editors did when the digital dimensions of editorial practices grew—that of a puzzle that can be overcome by asking further questions of the technology.33 James Smith also recognizes that scholars must maintain the human aspect of the editorial work, “devising more interesting, and intensive, and hybridised, and creative ways of interpreting the text,” and viewing the tasks of the machine and human as “two parallel tracks of innovation.”34

Students were also cautious, yet intrigued, about the adoption of generative AI by their peers. Some referred to their teachers experimenting with critical approaches to LLMs for assignments, encouraging comparisons between the work they produced and that of the model, with the aim of developing students’ ability to be reflexive and reflective (S12 and S13). S3 similarly explained “a valuable way of learning in reverse engineering”; if students are aware of the faults of LLMs, the likelihood of using generative AI from a purpose of interrogation is higher. As part of a critical AI approach, Hanlon argues that generative AI cannot be a useful research tool when there are mismatched expectations about its potential—its outputs must therefore be treated as fiction, in order to account for what they have misrepresented.35 Following only a brief discussion about the impact of genAI in education, the conversation among students turned toward realistic benefits of AI tools, alongside existing strengths of teaching practices. They noted translation exercises may be best conducted within peer and whole group settings (S3) and ideas about representation are best developed by reliable human critical thinkers (S6). Promoting these attitudes to technology is pertinent because, as a collective, we know little about its capabilities and actual harms in some contexts. Instead, recognition of its limitations from the outset—before it is embedded into our pedagogical and editorial practice—can allow us to understand what is needed from the learning environment before critiquing its limitations. Only after this approach can opportunities for enriching the learning of medieval texts, through probing these “tools,” be assessed.

§Probing AI-Generated Linguistic Analysis and Literary Interpretation

This section provides evidence of LLMs’ capabilities and limitations for uncovering stories of marginalization in Chaucer’s PPT, both from linguistic and literary perspectives. LLM responses are analyzed with the view to creating pedagogical critical apparatuses to increase student engagement with Chaucer’s literature at multiple overlapping axes. I thus provide a snapshot of the testing phase conducted to create elements of the digital edition, including translation, annotation, topic guide generation, and representations of the textual tradition. The prompts were a combination of general and more specific tasks, to see more precisely the patterns LLMs would retrieve regardless of the user’s prior knowledge.36 While there was no one method for probing responses, given the disparity in length between responses and their changing structure over a short period of time, I analyze the content and language to describe the term, phrase, or theme of focus, followed by any sources referenced, and evaluate whether the response met the digital editing purpose the prompt set out to achieve.

Example 1: LLM Responses and Their Prompts for the Line “And haue a ioly wenche in euery toun”

LLM: ChatGPT, version 4

Date: October 4, 2023

Prompt:37 Produce an activity for university-level undergraduate students which allows them to investigate the use of the following words from the Chaucer’s “Pardoner’s Prologue and Tale.” Each of the words and their translations relate to referents in the text with different perceived characteristics or roles, all with unique etymologies. Please also include a brief description of each word in relation to data in the Oxford English Dictionary and Middle English Dictionary. Finally, include a summary for each translation which serves as a prompt for the students, explaining some of the reasons behind the choices of these translations:

ME “ioly wenche”

Response [smaller extract from response]:

Words and Descriptions:

4. ME “ioly wenche”: In Middle English, a phrase for a pretty or cheerful young woman. Modern English retains the positive associations of this phrase even as “wench” has taken on more negative or derogatory connotations.

LLM: Bing Chat (now Copilot)

Date: October 9, 2023

Prompt: Please provide the meaning and etymology of the following words from Chaucer’s “Pardoner’s Prologue and Tale,” to a university-level standard, using information from the Oxford English Dictionary and the Middle English Dictionary. Ensure to include information on its meaning in the 1300s and within the context of the tale, and details on the origins of the word and whether it was borrowed from another language. ioly wenche (And haue a ioly wenche in euery toun)

Response: I’ll do my best to provide the meaning and etymology of the words you have requested from Chaucer’s “Pardoner’s Prologue and Tale.”

Ioly wenche: In Middle English, “ioly wenche” referred to a pretty young woman. In the context of the “Pardoner’s Tale,” it is used in reference to a woman who is kept by one of the three rioters as a mistress. The word “ioly” has its roots in Old French, where it was used to refer to something that is cheerful or merry.

The first prompt initially sought to determine whether etymologies of a predefined list of words from the tale, particularly those assigning a category or descriptor to further exclude voices from the narrative, could be retrieved from an LLM, whose access to paywalled resources is limited. One word I was drawn to in the Pardoner’s prologue was “wench.” At this point in the text, the Pardoner explains his lack of care for those in poverty and how he will continue to indulge in sin while selling his false relics. Carissa Harris has written extensively about medieval wenches, stating that “the label of wench is capacious enough to include many, yet binding enough to compel to coerce and to constrain.” She notes the ways in which Old English wenchel extended to children, servants, and enslaved people, in addition to women.38 The term is steeped in misogynoir from the seventeenth century, where the word was used to dominate and oppress Black women during the transatlantic slave trade.39 It is necessary to incorporate this formation of coercion into the earliest meanings of “wench,” including how the word has changed over time to the present, where there is perhaps lower usage but in equally discriminatory ways. However, LLMs do not achieve this.

The prompts in Example 1 request that the LLM teases apart these etymologies, within the space of a student activity and annotation. ChatGPT’s consideration of the present-day usage of the phrase is problematic because the model (1) associates the entire phrase, regardless of the descriptor, as positive, when the explicit use of “wench” can only result in negative outcomes, unless reclaimed by those directly affected by its use in modern contexts; and (2) assumes “pretty,” “cheerful,” and “young” are positive descriptors, yet the clear benefactor would be men who relish in objectifying women in medieval taverns. The response from Bing Chat—now Microsoft Copilot—is inaccurate; “wench” is used by the Pardoner to move further away from the ideals of the Church and express his lack of cautiousness around sin, and it does not form part of the tale he tells including the rioters/taverners. The “source” provided links to a synopsis of the tale on the Harvard Geoffrey Chaucer Website, with no explicit mention of the line to be annotated, making the web link redundant for notifying the user of the source. Thus, both responses fail to annotate the word accurately and lack detailed acknowledgment of the word’s origins and their complex semantic histories. Students are likely to gain more knowledge from using the OED, and the LLM does not deny it has not been able to utilize paywalled web resources such as this.

Example 2: LLM Responses and Their Prompts for Discovering Themes of Gender and Sexuality in the “Pardoner’s Prologue and Tale”

LLM: ChatGPT, version 4

Date: August 14, 2023

Prompt: Can you summarize the theme of gender and sexuality in Chaucer’s “Pardoner’s Prologue and Tale” to a university-level standard?

Response: Gender and sexuality in Chaucer’s “Pardoner’s Prologue and Tale” are complex and multifaceted. The Pardoner himself is a controversial figure whose ambiguous gender and sexual identity have long been subjects of critical debate.

The Pardoner’s ambiguous gender presentation is apparent in his physical description in the “General Prologue,” where he is described as having hair as yellow as wax, hanging as smoothly as a hank of flax, and a voice as small as a goat. He is also said to be unable to grow a beard, which was a significant marker of masculinity in the Middle Ages. These descriptions suggest a certain androgyny or effeminacy, which is further emphasized by the fact that he carries a bag of relics, an occupation typically associated with women in the Middle Ages.

Sexuality is also a significant theme in the “Pardoner’s Prologue and Tale.” The Pardoner’s sexuality is hinted at in the text, with some critics interpreting him as a eunuch or a homosexual, based on his effeminate appearance, his apparent lack of sexual desire for women, and his close relationship with the Summoner. “The Pardoner’s Tale” itself also deals with sexual sins, as the three rioters in the tale are guilty of gluttony, gambling, and lechery.

Moreover, the Pardoner’s hypocritical preaching against the very sins he himself indulges in can be seen as a commentary on the corruption of the Church and its officials, who often violated the vows of chastity, poverty, and obedience they were supposed to uphold.

In conclusion, the themes of gender and sexuality in the “Pardoner’s Prologue and Tale” are intricately linked with the broader themes of hypocrisy, corruption, and moral decay in the Church and society. The Pardoner’s ambiguous gender and sexual identity not only challenge the rigid gender norms of the time, but also serve as a critique of the moral failings of religious officials.

LLM: ChatGPT, version 4o

Date: November 18, 2024

Prompt: You are a scholarly editor of Geoffrey Chaucer’s text “The Pardoner’s Prologue and Tale” from The Canterbury Tales, with an interest in pedagogical approaches to the text. Your task is to identify and annotate words from the tale from an intersectional feminist and queer lens. Ensure the output is suitable for undergraduate university students.

Response [smaller extracts taken from the response for analysis]:

“By this gaude have I wonne, yeer by yeer, / An hundred mark sith I was Pardoner.” (“Prologue,” lines 403–4)

Annotation: The Pardoner openly acknowledges his deception and manipulation of others to achieve financial success. From a feminist lens, this could be examined as a critique of patriarchal systems where exploitation and power dynamics are normalized. Additionally, his candid admission of his fraud can be interpreted as a queer disruption of sincerity and “honest labor” narratives.

“I trowe he were a geldyng or a mare.” (“General Prologue,” line 691, describing the Pardoner)

Annotation: This line has sparked debates about the Pardoner’s gender and sexual identity. The ambiguous reference to the Pardoner as possibly a “gelding” (castrated male) or “mare” (feminized term) destabilizes fixed notions of gender and sexuality. Feminist and queer interpretations highlight this ambiguity as a space for questioning medieval norms of masculinity and the rigid binary of male/female.

Discussion Points for Pedagogy:

Queering the Pardoner: How does the Pardoner’s ambiguous gender presentation and sexual identity invite discussions about medieval perceptions of queerness? How do these perceptions align with or differ from contemporary understandings of gender fluidity and nonconformity?

Gendered Morality: How are sins like greed, gluttony, or avarice gendered in the text, and what might this suggest about medieval attitudes toward morality and gender roles?

Institutional Hypocrisy: What parallels can be drawn between the Pardoner’s critique of sin and his embodiment of it, particularly regarding feminist critiques of institutional power?

3. Greed and Colonial Narratives

Textual Reference: “For though myself be a ful vicious man, / A moral tale yet I yow telle kan.” (“Prologue,” lines 459–60)

Analysis: The Pardoner’s sermon against greed underscores wealth’s corrupting influence. This critique can be connected to medieval Europe’s burgeoning commercial expansion, which laid the groundwork for colonial exploitation and racialized systems of labor, especially during the transatlantic slave trade. Students might consider how such moral tales perpetuate the idea of European superiority while ignoring systemic greed on a broader geopolitical scale.

The “Three Rioters” and Stereotypes of Disorder

Textual Reference: “Sworn brotherhood in filthee rage and pride.” (“Tale,” line 709)

Analysis: The three rioters embody moral and social disorder, engaging in drunkenness and violence. Medieval European narratives often associated social chaos with groups perceived as “other,” such as non-Christian or non-European peoples. The alignment of moral decay with “outgroups” can be interrogated for its proto-racial undertones.

The prompts in Example 2 task the LLM with identifying themes that represent gender and sexuality throughout the tale. I compare a more general prompt—which students are more likely to have at their disposal at the beginning of a Chaucerian course—with a focused prompt placing the LLM in the role of a scholarly editor and analyzing the Pardoner from a queer and intersectional lens. These responses are also over one year apart, to highlight the development of the LLM versions and the change in structure of its responses.

ChatGPT version 4 reproduces some of the common scholarly themes on the Pardoner and his tale, with a heavy focus on translations of geldyng and mare in the “General Prologue,” his description as a marker of “effeminacy,” a potential relationship with the Summoner pilgrim, and an emphasis on the wrongdoings of the church. The response does not move away from discussions on categorizing his gender and sexuality and his status as controversial, as opposed to disrupting norms of the church. There is no developed argument comparing the “sexual sins” articulated by the Pardoner to describe the acts of the taverners, and the guilt arising from committing these sins. Lastly, more is insinuated about Chaucer’s ability to critique the Church through the character of the Pardoner, rather than about medieval characters’ identities disrupting and challenging normativity. While the later version “4o” response is prompted to explore these issues using a specific framework, it only loosely connects fluidity and nonconformity with the rigid historical and colonial context. For instance, the Pardoner’s announcement that he is a “ful vicious man” is linked to conversations about colonial expansion, but it is unclear how that argument can be developed for a student, especially as they require support to uncover colonial narratives with opaque language such as “vicious.” The social chaos of the rioters in the moral tale is loosely linked to racialized groups, yet non-Christian and non-European people are not mentioned in this aspect of the tale. Despite recognition of gendered and racialized labor, lack of power, and fluidity of experience with specific examples quoted from the tale, there is no development of these ideas through a clearly identified link to the text. Without clear embedding of this intersectional framework in a classroom environment, a response such as this would mean little to a student who is just beginning to learn the intricacies of Middle English language.

Example 3: LLM Responses and Their Prompts for Uncovering Analogues and Adaptations of the “Pardoner’s Tale,” Past and Present

LLM: ChatGPT, version 4

Date: August 14, 2023

Prompt: Can you provide examples of analogues of Chaucer’s “Pardoner’s Tale” in ancient and modern literature?

Response: Certainly! “The Pardoner’s Tale” has several analogues in both ancient and modern literature. Here are a few examples:

Ancient Literature:

1. The Story of King Midas. In Greek mythology, the tale of King Midas serves as an analogue to the “Pardoner’s Tale.” Both stories revolve around the theme of greed and its consequences. King Midas, driven by his insatiable desire for wealth, is granted a wish by the god Dionysus to turn everything he touches into gold. However, he soon realizes the negative impacts of his greed as he turns his daughter into gold and suffers from the inability to eat or drink.

2. The Story of Icarus. Another ancient analogue to the “Pardoner’s Tale” can be found in the Greek myth of Icarus. In this tale, Icarus and his father Daedalus attempt to escape imprisonment by constructing wings made of feathers and wax. Icarus, overwhelmed by the exhilaration of flying, ignores his father’s warning not to fly too close to the sun. His wings melt, and he falls to his death as a result of his reckless ambition.

Modern Literature:

1. “The Monkey’s Paw” by W. W. Jacobs. This short story serves as a modern analogue to the “Pardoner’s Tale.” “The Monkey’s Paw” revolves around a magical talisman that grants its owner three wishes. However, the wishes come with severe and unintended consequences. The story explores themes of greed, temptation, and the danger of tampering with fate.

2. “The Tell-Tale Heart” by Edgar Allan Poe. Poe’s famous short story can be seen as a modern analogue to the “Pardoner’s Tale.” It delves into the psychological consequences of guilt and the destructive power of one’s own conscience. The narrator, driven by guilt, confesses to the murder he committed due to his obsession with the old man’s eye.

These examples demonstrate how the themes and moral lessons found in the “Pardoner’s Tale” have resonated throughout literature, both ancient and modern, by exploring the consequences of human flaws and vices.

LLM: Gemini

Date: November 28, 2024

Prompt: Discuss how Patience Agbabi’s book Telling Tales and Baba Brinkman’s rap “The Pardoner’s Tale” are adaptations of the “Pardoner’s Tale” by Geoffrey Chaucer, and how they are modern takes on issues from the Middle English period.

Response [smaller extract of response]:

Unique Perspectives and Modern Issues:

Agbabi’s Telling Tales:

Brinkman’s “The Pardoner’s Tale”:

In conclusion, both Telling Tales and “The Pardoner’s Tale” offer fresh and relevant interpretations of Chaucer’s classic tale. By addressing contemporary issues such as materialism, consumerism, violence, addiction, and mental health, they demonstrate the enduring power of storytelling and the timeless nature of human vices and virtues.

The aims of the prompts in Example 3 are to determine whether LLMs retrieve and link between stories of influence at different moments in history, whether that be a reclamation of Chaucer’s works for a new purpose, or showing how Chaucer himself was impacted by literature occurring prior. This may form the core critical apparatus of an edition, in highlighting textual transmission through time. At the same time, the first prompt seeks a more general response about possible avenues for students to explore, and the LLM obliges by listing ancient and modern literature inspiring, or inspired by, the tale. There is little diversity in the textual suggestions; they are Eurocentric, Greek mythology, or modern literature written by white male authors. These responses ultimately contribute to the erasure of marginalized identities, past and present, and reproduce widespread whitewashing leading educators and students to overhaul university curricula. The literature provided in the response is again loosely connected to the tale, and it is therefore no clearer to a student whether the literature factors into the adaption of these moral tales. The second response, provided by a different chatbot, Gemini, takes the prompt requesting a discussion of more specific creative outputs in the present day, and provides an overview with little connection to the character of the Pardoner telling the tale. Broader themes are explored, including objectification, multiculturalism, and mental health (bringing in the character of the Old Man in Agbabi’s Telling Tales as only one clear example). There is little by way of comparison, and while not explicitly requested in the prompt, an editor or student may wish to compare the medium of poetry versus rap and hip-hop music, while drawing on adaptation and origin (e.g., the African American origins of hip-hop) and their link to the Chaucerian tale. There are areas for students to begin their study, but at a potential cost of the heterogeneity of the suggestions, depending on the prompting at their disposal.

LLMs were initially derived from automatic speech recognition in the late twentieth century, resulting in the probabilistic scoring of text as spoken language gave rise to temporary ambiguity. It is thus to be expected that many of the generative AI responses today produce repetitive, unrepresentative, and non-human-like outputs, and while newer versions of chatbots led to lengthier responses as shown in Examples 1–3, the concepts and themes raised are similarly disconnected to examples and provide limited analysis. Stone and colleagues discuss these origins, with reference to RLHF (reinforcement learning from human feedback); any time an LLM is rewarded for producing an accurate, human-like response, the likelihood of sanitized responses that do not reflect the diversity of experience is increased, further erasing marginalized voices from the response but also within the developmental considerations of these technologies.40 Schneider also analyzes the language of certainty in responses, as LLMs cannot recognize where knowledge in their training data is lacking, resorting to humanistic responses such as “I’ll do my best” or an affirmative “certainly!”41 This language of certainty not only contributes to anthropomorphistic views of AI but also, as shown in the above responses, may disguise inaccurate and vague content for student learning. These do not mirror students’ experience of teaching during which they learn to explore the nuances of textual traditions, transmissions, and disseminations. As Conrad and Goodlad argue, LLMs create a greater need for students to probe technologies as opposed to training in effective prompt engineering.42 LLMs do not have any interpretive abilities given their statistical pattern-finding, but they are not neutrally created; their data relies on these generalizations, leading to greater chances of responses promoting hegemony. Student skill development instead lies in developing solutions within pedagogical environments and for real-world applications, and in the final section I highlight further ideas for training in digital editing, digital humanities, and Chaucer studies that do not rely on retrieving “perfect” responses from LLMs.

§Communicating the Value of (Digital) Humanities and Pedagogies among AI Hype

The analysis presented in the previous section represented a small part of the work the C21 Editions project conducted to probe the suitability of generative AI in automating aspects of the critical apparatus for digital editions. There is also the pedagogical impact of AI-assisted editions on its learning community, and how responses that lack detail, are overly repetitive, and present loose connections between topics might affect students who come to use these technologies for their independent study. As demonstrated, such responses can only reproduce the main narratives of a topic in an opaque manner, particularly with regard to its sources and training data. There is little support built into these “tools” for beginners in an area, even when a particular pedagogical, queer, and intersectional lens may be prompted. How can similar probing techniques be adopted in classroom environments, particularly where there are risks of reproducing the dominant discourse within learning environments? The answer to this question comes back to understanding how educators can best transform debate around the capabilities of these technologies in a way that truly serves the community.

An activity that anticipates flaws within specific outputs and sources, and later critiques them, can assist students to consolidate their understanding of and engagement with the overall medieval textual tradition. By doing so, students consider the voices at danger of erasure, whether that be within the text itself or within societies where technologies are implemented to communities’ detriment. For example, university courses where there are challenges to condense and cover not only the learning of the Middle English language and its evolution through time, but also the sociohistorical and cultural context and the history of the book, all while centering marginalized voices within these “canonical” texts, might turn to digital editing. These approaches can instead provide practical applications of medieval text analysis for real-world considerations in the digital humanities and in the galleries, libraries, archives, and museums (GLAM) and heritage sectors. The critical literary perspectives that can be developed by incorporating generative AI technologies, within knowledge on the transmission of text, have wide implications for student understanding of shifts in narratives, and how and why voices at the intersections are systemically erased from the record. There is a bidirectional relationship that can be fostered within digital pedagogical editions, as part of a training program centering students as editors, between learning about the challenges of the medieval period and the challenges faced by users of LLMs today.

Conversations within the C21 Editions project returned to ideas for iterative community editions where members of the teaching and learning community are consulted and assume positions of authority on matters of AI in education. Equally, minimal computing and beginning with smaller language models trained for specific tasks, with smaller training datasets, must have a place within digital pedagogical editing. Stone and colleagues explain that the main weakness of LLMs is a “reflection of developers’ failure to conceptualize and pursue their ambitions for intelligent assistance as responsible to and engaged with a broader public.”43 Bottom-up approaches are thus crucial for incorporating transparent and accountable developments that invest time in empowering students to apply aspects of their course to everyday scenarios. For the future of digital editing, O’Sullivan and Whittle highlight that the Text Encoding Initiative (TEI), alongside infrastructure hosted on services such as GitHub and Zenodo, and close and distant reading using smaller models of textual analysis, preserve time and energy to counter techno-deterministic solutions.44 The TEI element <respStmt> (statement of responsibility) is one example where this type of conversation, about language models’ place within pedagogical editing, might be encoded within a text to highlight author contribution.45 Like the slow digitization approach proposed by Prescott and Hughes, a slower and more nuanced editing approach within classroom environments is necessary for remaining close to the medieval text and the multiple stories shaping its enduring interest for students today.46

A final point lies in the community solutions for promoting the value of the digital arts and humanities within a continually evolving AI age, centering student expression and authority within what bell hooks termed “engaged pedagogy.”47 I have shown that students are engaged in these discussions, and educators, researchers and practitioners of varied expertise must cooperate to resist and disrupt the prevailing hype placing generative AI on a pedestal. The Design Justice Network and scholars such as Sasha Costanza-Chock are heading the way for prioritizing participant-centered design practices for digital technologies, with a focus on what already works for the community, and nonexploitative, sustainable outcomes. There is a dual aim of providing pathways for marginalized communities to head design initiatives, while assuming the role of designer as facilitator rather than expert.48 One relevant community initiative adopting similar principles is Beowulf by All, which held space for a range of expertise and lived experience to shape translations of the Old English text.49 Likewise, there are direct benefits to making space for students to share their insight within pedagogical editing, especially where their voice may not have been considered in the creation of university curricula. Living documents such as those created by the Design Justice Network and the Critical AI team at Rutgers, provide practical advice for educators wishing to explore the impact of generative AI on their classroom and practice, along with alternative digital resources.50 It is only with continued knowledge sharing around the development of generative AI that we can commit to producing digital pedagogical outputs of direct value for all in the learning community, particularly those typically pushed to the margins within a highly competitive, profit-making AI market.

§Acknowledgments

I would like to thank the many students and educators whose insights I have had the pleasure of hearing over the last two years, during the focus groups and interviews we conducted, at conferences and workshops, and during day-to-day conversation about the changing climate of higher education. I have also received much support, advice and guidance from my colleagues on the project, Mike Pidd, Matt Groves, James O’Sullivan, Bridgette Wessels, and Órla Murphy. Along with my colleagues, I would like to thank the AHRC and IRC for their generous funding on C21 Editions, and the flexibility they have provided for me to take multiple approaches to the edition as conversation on AI shifts.

§Appendix 1: Student and Teachers Consulted in Focus Groups and Interviews

§Focus group 1 (UK university, in-person)

§Focus group 2 (UK university, in-person)

§Focus group 3 (UK university, in-person)

§Focus group 4 (US university, online)

§Interview 1 (UK university, online)

§Interview 2 (UK university, in-person)

§Interview 3 (UK university, in-person)

§Interview 4 (UK university, online)

§Interview 5 (US university, online)

§Footnotes

  1. This research was funded by UKRI-AHRC and the Irish Research Council under the UK-Ireland Collaboration in the Digital Humanities Research Grants (grant numbers AH/W001489/1 and IRC/W001489/1). ↩

  2. Critical AI is a journal based at Rutgers University’s Center for Cultural Analysis, published by Duke University Press. It takes an interdisciplinary approach to “building and implementing accountable technology in the public interest.” “About the Journal,” Critical AI, accessed November 20, 2024, https://read.dukeupress.edu/critical-ai. ↩

  3. Katherine Bode, “Computational Literary Studies: Participant Forum Responses, Day 2,” In the Moment, April 2, 2019, https://critinq.wordpress.com/2019/04/02/computational-literary-studies-participant-forum-responses-day-2-3/. ↩

  4. See Christopher Cannon, “The Myth of Origin and the Making of Chaucer’s English,” Speculum 71, no. 3 (1996), https://doi.org/10.2307/2865797, on the myth of Chaucer as the “father” of English literature, the reliance on the Chaucerian agenda, and overestimation of his “excellence” in comparison to other authors and poets. ↩

  5. Sarah Baechle and Carissa M. Harris, “The Ethical Challenge of Chaucerian Scholarship in the Twenty-First Century,” The Chaucer Review 56, no. 4 (2021): 316, https://muse.jhu.edu/pub/2/article/804395. See also Robyn A. Bartlett, “On Chaucer, Raptus, and the Physician’s Tale,” Exemplaria 35, no. 4 (2023), https://doi.org/10.1080/10412573.2023.2280724, which offers a parallel to examining Chaucer’s raptus within his own writing and mind, and the perpetuation of rape culture. ↩

  6. Baechle and Harris, “The Ethical Challenge of Chaucerian Scholarship,” 318. ↩

  7. Kimberlé Crenshaw, “Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics,” University of Chicago Legal Forum 1, no. 8 (1989), https://chicagounbound.uchicago.edu/uclf/vol1989/iss1/8/; Kimberlé Crenshaw, “Mapping the Margins: Intersectionality, Identity Politics, and Violence against Women of Color,” Stanford Law Review 43, no. 6 (1991), https://doi.org/10.2307/1229039. ↩

  8. Sojourner Truth, Audre Lorde, Angela Davis, Patricia Hill Collins, bell hooks, and Kimberlé Crenshaw, “Black Feminist Thought and Intersectionality,” in Roads to Decolonisation: An Introduction to Thought from the Global South, ed. by Amy Duvenage (New York: Routledge, 2024), 82. This textbook serves as an accessible introduction to the work of Black feminists and researchers furthering decolonial movements in societies across the globe. ↩

  9. Sumi Cho, Kimberlé W. Crenshaw, and Leslie McCall, “Toward a Field of Intersectionality Studies: Theory, Applications, and Praxis,” Signs: Journal of Women in Culture and Society 38, no. 4 (2013): 787, https://doi.org/10.1086/669608. ↩

  10. Jason Boyd and Bo Ruberg, “Queer Digital Humanities,” in The Bloomsbury Handbook to the Digital Humanities, ed. by James O’Sullivan (London: Bloomsbury Publishing, 2023), 63–66. See also Amy E. Earhart, “Feminist Digital Humanities,” in O’Sullivan, Bloomsbury Handbook, 80, which highlights that much of the work in DH exists outside clearly defined norms of analysis, crossing many of the boundaries and intersections identified by Black feminists, along with the long histories of work in fields such as queer theory, critical pedagogy, postcolonial studies, etc. ↩

  11. Rahul K. Gairola, “Race, Otherness, and the Digital Humanities,” in The Bloomsbury Handbook to the Digital Humanities, ed. by James O’Sullivan (London: Bloomsbury Publishing, 2023), 57. ↩

  12. David M. Berry, “Critical Digital Humanities,” in O’Sullivan, Bloomsbury Handbook, 126–27. ↩

  13. In the Middle English Dictionary (MED), the word pardoner is linked to senses (1) “a seller of indulgences, pardoner” and (2) “rascal, scoundrel.” This latter sense is attested in the poem Cursor Mundi (Cambridge, Trinity College R.3.8 manuscript), and the development of this meaning is interesting given the overall portrayal of the Pardoner in The Canterbury Tales as “deviant.” The religious poem, which tells biblical stories from creation to doomsday, states that the two pardoners (rascals) were ful of gile “were full of schemes/tricks/plots.” “pardonẹ̄̆r n.,” Middle English Dictionary, accessed November 22, 2024, https://quod.lib.umich.edu/m/middle-english-dictionary/dictionary/MED32410/track?counter=1&search_id=2410761. ↩

  14. The translations of geldyng and mare into “eunuch” and “homosexual” can be found on the Harvard Geoffrey Chaucer website, although they are widely used translations and definitions of these Middle English words (e.g., as shown in the MED and OED). Homosaurus, a linked data resource of LGBTQ+ terms, notes “homosexual” as outdated, which emphasizes the need for translations to incorporate flexibility into the process, something that digital editions may be able to accommodate. “General Prologue,” Harvard Geoffrey Chaucer Website (accessed November 22, 2024), l.691, https://chaucer.fas.harvard.edu/pages/general-prologue-0. “Homosexuals,” Homosaurus (accessed November 22, 2024), https://homosaurus.org/v3/homoit0000648?from_q=homosexual. ↩

  15. Glenn Burger and Steven F. Kruger, “Queer Chaucer in the Classroom,” in Teaching Literature: A Companion, ed Tanya Agathocleous and Ann C. Dean (Basingstoke: Palgrave Macmillan UK, 2003), 32. ↩

  16. See Carolyn Dinshaw, Chaucer’s Sexual Poetics (Madison: University of Wisconsin Press, 1989); Richard E. Zeikowitz, “Silenced but Not Stifled: The Disruptive Queer Power of Chaucer’s Pardoner,” Dalhousie Review 82, no. 1 (2002); Tison Pugh, “Queering Harry Bailly: Gendered Carnival, Social Ideologies, and Masculinity under Duress in the Canterbury Tales,” in Sexuality and Its Queer Discontents in Middle English Literature, ed. Tison Pugh (Palgrave Macmillan US, 2008), https://doi.org/10.1057/9780230610521_3; Mohamed Karim Dhouib, “De/Stabilizing Heterosexuality in the Pardoner’s Tale,” International Journal of Language and Literary Studies 3, no. 4 (2021), https://doi.org/10.36892/ijlls.v3i4.759. ↩

  17. Patience Agbabi and Candace Barrington, “Conversations: Candace Barrington Interviews Patience Agbabi, Author of Telling Tales,” New Chaucer Studies: Pedagogy and Profession 5, no. 1 (2024): 181, https://doi.org/10.5070/NC35064187; Patience Agbabi, Telling Tales (Edinburgh: Canongate Books, 2014), 70. ↩

  18. Jean “Binta” Breeze, “The Wife of Bath in Brixton Market (Poetry),” posted November 2009, by iPoems, YouTube, 4:15, https://www.youtube.com/watch?v=MiyKat1QzbQ; Zadie Smith, The Wife of Willesden (Penguin Books, 2021). ↩

  19. Whitney B. Taylor, “The Pedagogical Possibilities of Editing a Digital Text in the Shakespeare Classroom,” Early Modern Culture 14, no. 10 (2019): 135, https://open.clemson.edu/emc/vol14/iss1/10. ↩

  20. Serenity Sutherland, “Pedagogies of Scholarly Editing and Digital History in the Seward Family Digital Archive,” Scholarly Editing: The Annual of the Association for Documentary Editing 39 (2022): 6, https://doi.org/10.55520/6ZH06EW2. ↩

  21. John Bryant, Mary Isbell, Christopher Ohge, and Mary Erica Zimmer, “Digital Editing and Pedagogy: Making Editions / Building Arguments,” Scholarly Editing: The Annual of the Association for Documentary Editing 41 (2024): 2, https://doi.org/10.55520/KBS01GGN. ↩

  22. Joris J. van Zundert, “Barely Beyond the Book?,” in Theories and Practices: Digital Scholarly Editing, ed. Matthew James Driscoll and Elena Pierazzo (Open Book Publishers, 2016), 106. ↩

  23. James O’Sullivan and Sophie Whittle, “Conclusion: The Future of Digital Editing and Publishing,” in Digital Editing and Publishing in the Twenty-First Century, ed. James O’Sullivan, Michael Pidd, Sophie Whittle, Bridgette Wessels, Michael Kurzmeier, and Órla Murphy (Edinburgh: Scottish Universities Press, 2025), 363–81. ↩

  24. Melissa Terras, Joe Nockels, Sarah Ames, Paul Gooding, Andy Stauder, and Günter Mühlberger, “On Automating Editions: The Affordances of Handwritten Text Recognition Platforms for Scholarly Editing,” Scholarly Editing: The Annual of the Association for Documentary Editing 41 (2024): 9, https://doi.org/10.55520/W257A74E. ↩

  25. bell hooks, Teaching to Transgress: Education as the Practice of Freedom (New York: Routledge, 1994), 20–21. See also Katherine D. Harris, “Play, Collaborate, Break, Build, Share: ‘Screwing Around’ in Digital Pedagogy,” Polymath: An Interdisciplinary Arts and Sciences Journal 3, no. 3 (2023), https://hcommons.org/deposits/item/hc:55513. ↩

  26. These studies received ethics approval from the University College Cork and the University of Sheffield Research Ethics Committees, and all participants gave their informed consent to take part and have their responses published by the DHI Data Service and in future publications. The data from students and teachers in our late 2023 user focus groups and interviews have been anonymized and assigned an ID number, in line with this particular ethics procedure and the sensitivity of the topics they raise in relation to ongoing learning and teaching. This number is linked to details about their degree course and year of study, or career stage and research area, but without identifying information about their specific institution (see Appendix 1). Interviewees for the future of scholarly digital editing are referenced with their full names, given their insights were made open access via the DHI Data Service as part of their consent to taking part. ↩

  27. See the DHI Data Service for the C21 Editions transcribed interviews and the biographies of the interviewers: James O Sullivan et al., “Interviews about the Future of Scholarly Digital Editions,” C21 Editions, May 19, 2023, https://www.dhi.ac.uk/data/c21editions. ↩

  28. Maaike Derksen and Margriet Fokken, “Editorial: Gender and (Post)Colonialism: Locating Marginalised Voices,” Tijdschrift voor Genderstudies 18, no. 3 (2015), https://doi.org/10.5117/TVGN2015.3.DERK. ↩

  29. Malek J. Zuraikat, “Chaucer’s Gender-Oriented Philosophy in The Canterbury Tales,” Literature Compass 20, no. 4–6 (2023): 1, https://doi.org/10.1111/lic3.12706. ↩

  30. Alex Gil, “Interviews about the Future of Scholarly Digital Editions,” interview by James O’Sullivan, C21 Editions, January 20, 2022, transcript, https://www.dhi.ac.uk/data/c21editions. ↩

  31. Aodhán Kelly, “Interviews about the Future of Scholarly Digital Editions,” interview by James O’Sullivan, C21 Editions, February 17, 2022, transcript, https://www.dhi.ac.uk/data/c21editions. ↩

  32. Joris J. van Zundert, “Interviews about the Future of Scholarly Digital Editions,” interview by James O’Sullivan, C21 Editions, January 31, 2022, transcript, https://www.dhi.ac.uk/data/c21editions. ↩

  33. Elena Pierazzo, “Interviews about the Future of Scholarly Digital Editions,” interview by Michael Kurzmeier, C21 Editions, April 14, 2022, transcript, https://www.dhi.ac.uk/data/c21editions. ↩

  34. Claire Connolly and James Smith, “Interviews about the Future of Scholarly Digital Editions,” interview by Michael Kurzmeier, C21 Editions, March 8, 2022, transcript, https://www.dhi.ac.uk/data/c21editions. ↩

  35. Aaron R. Hanlon, “LLM Outputs Are Fictions,” Critical AI 2, no. 1 (2024), https://doi.org/10.1215/2834703X-11205210. ↩

  36. The outputs retrieved from LLMs are particularly lengthy in places, so these responses have been separated and shortened where I focus on one word or line within the text and subsequent response. ↩

  37. Some of the prompts and responses have been reduced due to space; in prompts referencing extracts, the entire folio around the relevant words was included. ↩

  38. Carissa M. Harris, “Reproducing Wenches: Histories and Futures of Intersectional Disadvantage,” The 2023 Jankofsky Lecture, 2023, by University of Minnesota Duluth, https://hdl.handle.net/11299/254512. Oxford English Dictionary, s.v. “wenchel (n.),” July 2023, https://doi.org/10.1093/OED/5682590818. ↩

  39. Oxford English Dictionary, s.v. “wench (n.),” July 2023, https://doi.org/10.1093/OED/7242023491. ↩

  40. Matthew Stone, Lauren M. E. Goodlad, and Mark Sammons, “The Origins of Generative AI in Transcription and Machine Translation, and Why That Matters,” Critical AI 2, no. 1 (2024), https://doi.org/10.1215/2834703X-11256853. ↩

  41. Britta Schneider, “A Sociolinguist’s Look at the ‘Language’ in Large Language Models,” Critical AI 2, no. 1 (2024), https://doi.org/10.1215/2834703X-11205168. ↩

  42. Katie Conrad and Lauren M. E. Goodlad, “Teaching Tips for Navigating AI in the Classroom,” The AI Hype Wall of Shame, March 7, 2024, https://criticalai.org/2024/03/07/katie-conrad-and-lauren-m-e-goodlad-on-teaching-tip-navigating-ai-in-the-classroom-03-07-2024/. ↩

  43. Stone, Goodlad, and Sammons, “The Origins of Generative AI.” ↩

  44. James O’Sullivan and Sophie Whittle, “Conclusion: The Future of Digital Editing and Publishing,” in Digital Editing and Publishing in the Twenty-First Century, ed. James O’Sullivan, Michael Pidd, Sophie Whittle, Bridgette Wessels, Michael Kurzmeier, and Órla Murphy (Edinburgh: Scottish Universities Press, 2025), 363–81. ↩

  45. “<respStmt>,” “TEI: Guidelines for Electronic Text Encoding and Interchange,” last modified November 1, 2024, https://www.tei-c.org/release/doc/tei-p5-doc/en/html/ref-respStmt.html. ↩

  46. Andrew Prescott and Lorna Hughes, “Why Do We Digitize? The Case for Slow Digitization,” Archive Journal (2018), https://www.archivejournal.net/essays/why-do-we-digitize-the-case-for-slow-digitization/. ↩

  47. hooks, Teaching to Transgress, 20. ↩

  48. Sasha Costanza-Chock, “Design Justice, A.I., and Escape from the Matrix of Domination,” Journal of Design and Science (2018): 11–12, https://doi.org/10.21428/96c8d426. See also “Design Justice Network Principles,” Design Justice Network, last modified 2018, https://designjustice.org/read-the-principles. ↩

  49. Jean Abbott, Elaine Treharne, and Mateusz Fafinski, eds., Beowulf by All: Community Translation and Workbook (Arc Humanities Press, 2021). ↩

  50. Lauren M. E. Goodlad and Sharon Stoerger, “Teaching Critical AI Literacies: ‘Explainer’ and Resources for the New Semester,” Critical AI @ Rutgers, last modified October 20, 2024, https://docs.google.com/document/d/1TAXqYGid8sQz8v1ngTLD1qZBx2rNKHeKn9mcfWbFzRQ/edit?tab=t.0#heading=h.kgds7i8l6uca. ↩