Challenges of Collaboration

Research in the digital humanities ranges across a number of fields and implements both qualitative and quantitative methods. As Neilson et al. observe, however, “Digital humanists cannot claim a shared body of literature or theory that orients their work” (“Introduction: Research Methods,” 6). Rather, the digital humanities includes the academic outputs of collaboration between “people with different disciplines, methodological approaches, professional roles, and theoretical inclinations” (Spiro, “‘This Is Why We Fight,’” 16). Similarly, “disciplinary difference in digital humanities is not a binary” but is rather “a spectrum of emphasis, with varying degrees of interest in methods, tools, and values” (Robertson, “Differences between Digital Humanities,” 291). Collaboration between different disciplines can make up for the deficiencies of any one particular research method. Humanities scholars, however, are still accustomed to working alone, which is very different from working collaboratively in a lab (Giannetti, “Against the Grain,” 259). Humanists, therefore, can clash with practitioners, professionals, and theorists based in other fields over research methods in the digital humanities (DH) and over questions such as how computational approaches should be deployed, if practice should be emphasized over theory, and how workloads should be distributed. As a PhD candidate who holds not a research assistantship under a DH center or lab but a teaching assistantship in the English department at Texas A&M University, I am pursuing a focus on computational literary studies in my PhD dissertation. But I have found it difficult to collaborate while pursuing computational research in the humanities due to the different cultures and environments between departments and institutions, a lack of proper technical support and funding, and my inability to access a DH lab. In this chapter, I explore the limits of current institutional systems in serving the needs of humanities graduate students pursuing computational literary studies based on my empirical experience as a current graduate student. I argue for the value of providing graduate students with opportunities to participate in virtual and physical DH labs.

In my PhD dissertation, I mainly draw upon computational research methods using deep learning models for literary analysis. Though I am a humanities graduate student, I deal with a broad range of tasks for my dissertation, from literary analysis to fine-tuning deep learning models and creating deep learning datasets, all of which require extensive time, effort, and funding. In his chapter “The Life Aquatic: Training Digital Humanists in a School of Information Science” in chapter 29 in this volume, Ted Underwood argues that it would be challenging for those who pursue computational research in the humanities to finish their PhD dissertation within five or six years if they had to start from scratch for their digital projects, a statement that applies to my situation. In my dissertation, I deploy deep learning models, namely, the BERT-Base model developed by Google Research and a conditional Generative Adversative Networks (GANs) model, which stems from a GAN model created by Goodfellow et al. (“Generative Adversarial Nets”). Although I use the deep learning models created by others, I have to fine-tune them for each task to tailor them to my research purposes, along with creating my own datasets. As a graduate student who pursues computational research in the humanities but is not a machine learning engineer, my dissertation is highly interdisciplinary; thus, collaboration, which Katherine D. Harris describes as the “lynchpin” that supports “productivity, learning, experimenting, and knowledge acquisition,” is key to my research (“Play, Collaborate, Break, Build,” 8).

I tentatively identified a coauthor, a PhD candidate in computer science at an R1 university, through one of the communities that I belong to for a project aimed at developing a new deep learning model for sentiment analysis. However, we reached an impasse for several reasons, all of which stemmed from the different systems in place at our respective academic institutions. To develop a deep learning model and run repeated tests on it with cloud services, we needed a research grant, which we did not obtain. According to Jessica Webster’s survey on digital collaborations, 10.29 percent of participants answered that their projects “would be more successful with increased funding” (“Digital Collaborations,” para. 34), and this was the case for our potential collaboration. Without funding, it is challenging to pursue deep learning projects, especially with colleagues from different institutions. Another difficulty arose when attempting to establish equal workloads between coauthors from different departments. Although one of us considered writing to be a minor task and experiments to be a major task, the other believed that both were equally important. Compared to measuring the workload of experiments, there are difficulties in measuring the workload of writing and research, which can create barriers to receiving credit for one’s work. Graban et al. contend that humanities scholars’ labors are “often made invisible by the writing, much of their own work disappear[ing] in the redistribution of digital labor,” and humanities scholarship can therefore be undervalued during collaboration due to its potentially invisible character (“Introduction: Questioning Collaboration,” para. 23, 10). Webster mentions Posner’s argument that “information professionals face different workloads than other DH scholars often do, especially those scholars for whom teaching and research is a primary responsibility” (“Digital Collaborations,” para. 19). In engineering and computer science, lab research with experiments is a major part of scholars’ responsibilities, whereas the absence of lab research in the humanities means that the work of research and teaching tends to be distributed differently. Dissimilarities in both research environments and research methods can give rise to differences in scholars’ understandings of the workload that will be expected to complete an academic paper. In other words, epistemological difference is an obstacle to collaboration across fields.

According to Webster’s survey on digital collaborations, the number of information professionals who collaborated with graduate students from another institution was only three out of 197 (“Digital Collaborations,” para. 27). Although my case is slightly different from her survey question, I assume that graduate students from different institutions face difficulties when attempting to collaborate due to differences in the availability of funding. Graduate students in engineering and computer science are mostly tied to their professors under research assistantships that restrict them from working with other graduate students from other departments or institutions. Collaborations between graduate students in the humanities, engineering, and computer science have the potential to be fruitful combinations yielding outputs that a single scholar could not produce. Without changes to the exclusive collaboration system in the engineering and computer science fields, however, there will continue to be obstacles to collaboration between graduate students from different departments.

Collaboration comes in forms other than teamwork. Participating in discussions with various communities by sharing experiences and suggestions is another way to collaborate. Through community activities, new ideas or solutions to issues can arise. For instance, while working on my project about deep learning-based colorization and sentiment analysis, I often received feedback on online communities from expert users of specific technologies. Stack Overflow, one example of such a community, is a question-and-answer space for both novice and professional developers, and its archives offer some of the best resources for learning coding and fixing bugs. A plethora of questions, however, remain unanswered on its pages, in part because the community is so comprehensive and includes questions about a large number of computer languages, both managed and unmanaged. For this reason, I draw mainly upon three communities: the TensorFlow KR and Computational Humanities Research (CHR) communities to get feedback and technical support for my DH projects and the Pseudo Lab for collaboration.1 The TensorFlow KR community, which has around fifty-four thousand members on Facebook, consists of a variety of deep learning users, from a high school student who created an automatic recycling system at his school by using deep learning models based on technical support from the community to senior engineers and computer science professors. The community is responsive, passionate, and benevolent: when posting a question about deep learning, it often takes less than thirty minutes to start receiving comments. Most members of the deep learning community, however, are not from the humanities and therefore lack disciplinary knowledge in that area. To make up for this, I also take advantage of the CHR community to get feedback for my DH projects. The CHR has a discussion forum where digital humanists across a range of subfields freely share opinions on DH projects, theory, and coding, and it is a mutually beneficial space where it is possible to get specific feedback and collaborate, participate in ongoing discussions, and share recent information in the field of computational humanities research. Lastly, I have actively participated in the Pseudo Lab by collaboratively writing deep learning tutorials and helping with the implementation of deep learning models as a collaborator. The Pseudo Lab is a virtual nonprofit lab for those who need a space to learn, create, and share knowledge of deep learning. It brings together several groups with varying focuses, such as Kaggle AI competition groups, a study group for deep learning theory, a tutorial group, and a coding group for the implementation of deep learning models based on academic articles. All resources created by the Pseudo Lab are open to the public, including reviews of academic articles, code reviews, and tutorials. The Pseudo Lab is an example of how a virtual lab can act as a valuable space for graduate students who need technical assistance and who wish to work and learn collaboratively.

As a graduate student who pursues computational humanities research, being able to get feedback from professionals and scholars in the field is significant. Not only is feedback from my committee members from the humanities helpful for improving my work in terms of the H (humanities) in DH, but technical advice from my committee member in computer science also validates the D (digital) aspect of my work. Due to the different conventions between departments, however, it is necessary to maintain communication with committee members from outside of the department to avoid misunderstandings. A faculty member who supervises a graduate student in engineering and computer science, for instance, is usually included in the student’s paper as a second author by providing feedback or research environments, whereas this is uncommon in the humanities. Understanding how different disciplinary cultures result in differences in funding systems, citation practices, and other important practicalities is key to the relationship between graduate students and faculty members from outside of their departments. It is important to clarify that I am speaking based on my knowledge of one specific PhD system, that of the United States. In some countries, such as the United Kingdom, it is still possible to have a second supervisor from outside of the home department, but it remains uncommon.

Digital humanities centers and specialist DH librarians are another source of support that can provide consultation, instruction for project assistants, and access to potential collaborators. Through DH centers, graduate students and faculty members can access resources and advice for their projects and “develop [the] digital competencies” needed for their work (Fraistat, “Data First,” 83). The benefits of DH centers, however, are only available to digital humanists affiliated with institutions that are sufficiently well-resourced to have such initiatives. In addition, getting the right kind of specialist technical support can be challenging. As Bobby Smiley points out, a DH librarian is not “a messianic unicorn” who can support every digital project with robust technical assistance (“From Humanities to Scholarship,” 416). Similarly, DH centers are usually not places primarily oriented toward troubleshooting but rather are more likely to be helpful in setting up projects with technical support or funding. For instance, when developing the Victorian400 dataset, a deep learning dataset of nineteenth-century illustrations created for the colorization of black-and-white illustrations, I had difficulty finding experts in computer vision and deep learning datasets at my own institution who could provide feedback on my colorization project (Kim, “Victorian400”). Instead, I received technical feedback from my committee member from computer science, from deep learning engineers through my personal networks, and from the online communities mentioned above. I also spent my personal funds to initiate the project on my own device, though this was later supplemented with a project development grant from the Center of Digital Humanities Research (CoDHR) at TAMU and graduate funding from my department so that I could continue the project’s development. These project development grants, however, are required to receive assistance from the CoDHR. Without the initially successful prototype of the project, it might have been challenging to secure the grant and funding, in addition to assistance from the CoDHR. Although it is technically possible to get CoDHR funding for a project before starting it, many DH projects do not qualify and require personal funding in the incubation stage, as my project did.

Though a multitude of environmental difficulties exist for graduate students pursuing computational humanities research in the humanities, these obstacles can be mitigated by communicating with a variety of experts in different communities, collaborating with practitioners and professionals outside the academy in addition to scholars, and getting technical and funding support from DH centers and departments. Digital humanities centers and DH institutional programs, such as the Mellon Graduate Program that Erin Francisco Opalich, Daniel Gorman Jr., Madeline Ullrich, and Alexander J. Zawacki introduce in “A Tale of Three Disciplines: Considering the (Digital) Future of the Mid-doc Fellowship in Graduate Programs” in chapter 25 in this volume, are helpful for graduate students pursuing digital projects, but collaboration between individuals is one of the most valuable ways of pursuing interdisciplinary digital work, especially for projects of any significant magnitude. The DH community has organized a supportive system for communication, for instance, with the Association for Computers and the Humanities (ACH) mentorship program and a DH Slack channel, which create opportunities for collaboration.2 By contrast, opportunities for collaboration within individual institutions such as DH centers and DH mentoring programs are offered to few graduate students. Therefore, having a virtual DH lab across departments and institutions such as the Pseudo Lab is one way to facilitate collaboration on digital projects. In addition, as Gorman et al. point out, it is important to have institutional support from departments for both teaching and practicing DH. More specifically, departments in the humanities might consider whether it is possible to set up a DH lab without a principal investigator, following Amy Earhart’s argument that a “neutral laboratory space” is needed to provide a research environment for both students and faculty members, without any initial requirements, which can serve as an unhampered space for individual and team research projects as well as for collaboration (“Digital Humanities as a Laboratory,” 397). It might take time to build an institutional system that provides graduate students with a lab space that facilitates collaboration with students from different institutions, but efforts to improve the system even in small, incremental ways will make it easier for future students to become digital humanists who can freely collaborate with other scholars on digital projects, without being stymied by environmental research restrictions.