Carthusian Citation Credits

Tags: musings, research

Published on
« Previous post: A Simple and Cheap Recording Setup for … — Next post: Societal Donatism »

Many of the conflicts in academia are to some extent related to giving proper credit where credit is due. Sometimes, feuds arise over which papers of co-authors to cite in the ‘Related Work’ section of a manuscript; sometimes, researchers are accused of intentionally ignoring related work; sometimes, people even get removed from the author list of a work to which they contributed.1 Another type of conflict arises when such works are to be presented. Given the tendency of science to require ever-larger collaborations, how should one present the contributions of everyone in a fair and unbiased manner?2

Let’s take a look at a few strategies that are already in place:

Strategy I: Author Order

As a naive undergraduate, I always thought that everyone contributed equally to all papers. Heck, most of my reading consisted of one-author or two-author papers and some rare three-author collaborations. It was only later on that I learned that certain academic domains have come up with their own rules. In computer science, for instance, the rule of thumb seems to be that the first author is usually a Ph.D. student, while the last author is usually a principal investigator providing the funding and supervising the whole work. This leads to a double-cone shape of the importance of authors: importance decreases from first to middle authors, while it increases again towards the end.

Upon first learning this, it struck me as interesting that this sort of arcane knowledge is typically not directly communicated to the reader of a paper. You have to be ‘in the know’ in order to grasp the relevance of a certain author position. Moreover, I immediately wondered where the cut-off was. In other words, if you are a middle author, did you contribute (almost) as much as the first author, or was your contribution more humble, such as providing additional references, etc.?

Author names can also be adorned with additional typographical marks, such as an asterisk* or a dagger to indicate shared authorships. While I find this to be an absolute must in order to communicate the fair sharing of credit, most bibliographies have no built-in support for this—citations will only show up as a list of authors, and of course, there can only be one first author. This has led to interesting discussions as to whether a first-first position is to be preferred to a second-first position (meaning a shared authorship, without being listed as the first author). I was surprised and stumped by the fact that people were actually looking at these minor distinctions. My inner mathematician cries foul because either the two authors contributed equally, thus making them indistinguishable so that the order should not matter, or they did not, in which case equal contributions are not justified.

The root of the problem seems to be that we tend to cite a list of authors as First et al., thus hiding other authors that participated in a project. We have to change this if we want to avoid discussions of this sort! Below, I will mention a few ideas that might help, but I do not have a good solution for this yet.

Strategy II: Contribution Statements

A nice addendum to Strategy I involves paragraphs that state the contributions of individual authors. This is an excellent way to get readers of your paper to understand who did what—at the same time, it has become a kind of rote template to fill. I often wonder whether it can really do justice to the excellent insights all of my colleagues contribute to a project. Sentences of the form ‘Alice conceptualised the work, Bob prepared the experiments, and Charlie ran them’ do not really tell about the heroic deeds of Alice, Bob, and Charlie—who knows how hard it was to come up with everything, to figure out a good experimental setup, and then finally to run it?

Despite these misgivings, the contribution statement is a useful thing to have, and I wish that machine learning venues would encourage it more. At the end of the day, it is up to the authors whether they take such a statement seriously, though, and it is hard to know whether all authors agree with all formulations of such a statement. The role of power dynamics should not be misunderstood here. I am lucky to be in a situation in which this is not a big deal, but I am learning that this is not the norm, unfortunately.

Strategy III: Everyone Is Equal

A simple strategy, not often followed, is to make everyone contribute equally to a publication. In pure mathematics, this seems to be the norm. This strategy is in fact reminiscent of the famous Hardy–Littlewood collaboration, which followed four axioms:3

1. When writing to one another, it was completely indifferent whether what they said was right or wrong.
2. When one received a letter from the other, he was under no obligation whatsoever to read it, let alone answer it,
3. It was preferable that they not both thought about the same detail, if possible.
4. All papers would appear under their common name, regardless of whether one of them had contributed anything.

What a marvellous set of rules! In particular the last one removes all quarrels and potential conflicts concerning co-authorship! Most interestingly, these rules were known to others as well, so they knew about this type of ‘contract’.

The remarkable thing about such rules is that they are quite pragmatic: both Hardy and Littlewood knew very well that it would be helpful to be linked to certain works; at the same time, their collaboration rules were fundamentally about trust, making other arrangements unnecessary.

The fundamental hitch of this strategy is that it does not scale, nor is it applicable to all types of collaborations or all types of scientific inquiry indiscriminately. Nevertheless, it would be an interesting strategy to follow in smaller projects, and I would definitely like to be part of such a collaboration in the remainder of my career. I trust all my collaborators to a sufficient extent for this to work.

Credit Where Credit Is Due

The fundamental issue with all of these strategies is that they impede true unfettered collaboration:

• Students know that their name needs to be attached to a certain number of single-author papers.
• Some universities have rules in place that prohibit joint authorship.
• Other universities only ‘recognise’ certain author roles on a paper, such as (joint) first/last author.

When applying to academic positions, positions on a publication are often used to judge the merits of a candidate—contribution statements etc. notwithstanding. A major issue with this sort of judgement is that it is not transparent. This gives us academics all kinds of wrong incentives,4 in the worst case forcing us to be ‘opportunistic’ when it comes to seeking collaborators, or making us jaded5 and turn away from academia altogether. I think all of us are sufficiently smart to figure out a way to ‘game the system’, but is this really how we want to spend our waking hours?

Again, I unfortunately do not have a fully-fledged solution for this yet, but I urge everyone to rethink their strategies for giving credit where credit is due. Here are some of the themes I often reflect upon:

1. Do I give enough credit to the team when presenting joint work? For instance, I always use the ‘Collaborative We’ when talking about joint work; I also always briefly introduce all authors of a paper. Is there more that I can do on a regular basis?
2. Do I perceive contributions sufficiently well? Do I overlook a certain type of contribution, or a certain type of personality?
3. Can I codify or abstract certain of my rules to make them more generalisable?
4. Is my deliberation process fair? Or as close to fairness as it can get?

As a reflection about a very extreme case of crediting people, I want to leave you with my ‘Carthusian Way’.

The Carthusian Way

I have always been fascinated with monastic communities and my (idealised) understanding of them. Devoting yourself utterly to a certain kind of task while shutting out the rest of the world has a certain kind of appeal to me—and reading works such as The Glass Bead Game or Anathem only fuelled that desire to a certain point. I have since come to the conclusion that fleeing the world is not necessarily a good answer, but I learned a very interesting set of practices from different religious orders.

One of the most remarkable ones is the Order of Carthusians. Their way of life combines the way of the hermit with living in a community, and I found this to be an excellent situation to strive for as an academic. The Carthusians tend to eschew worldly honours even more than other orders, and one of their strategies involve the conscious seeking of anonymity. As Thomas Merton writes in The Silent Life on the Carthusians:

When a monk of exceptional virtue dies, the highest public honor he receives in the Order is a laconic comment: laudabiliter vixit. In good American we would translate this as: “He did all right.” Finally, the Carthusian does not even have the personal distinction of a grave marked with his own name. He is laid away in the cemetery under a plain unmarked cross, and vanishes into anonymity.

If my legacy was laudabiliter vixit, I would certainly be pleased! This desire for remaining unknown also expresses itself in the way the Carthusians communicate their knowledge. If they feel the need to publish anything because it might serve others, they often sign their name as ‘A Carthusian’, eschewing any form of commendation from others. That way, their work truly speaks for itself.

We have more technology at our disposal than the Carthusians when they were first founded, so I wonder if it would not be possible to prevent attaching work to one name, while still reaping the benefits of being an author. I would love to see the following strategy being implemented:

1. Everyone publishes under the name of their discipline. If you send a paper to a machine learning conference, let the author be ‘A Machine Learner’, ‘A Group of Machine Learners’, ‘A Group of Scientists’, or just ‘Scientists’.6
2. Calculate a hash of the original paper source or title using SHA-256, for instance.
3. Tweet about it or use some other type of system to distribute/announce the hash to others. See Terence Eden’s Blog for a brief tutorial on this. The neat thing about using Twitter as a distribution medium is that, theoretically, your tweets can also tag all co-authors of the paper or at least list their names.
4. For such a tweet to be useful, it would ideally not come from a Twitter account tied to your person, but maybe an institutional account or another type of service. The idea is to keep all names and identifiable information out of this for as long as possible.

This would solve certain types of priority disputes, similar to how anagrams where used by early scientists. It would also be massively beneficial for peer review, which is known to favour the ‘bigshots’ over the ‘underdogs’, provided some identifiable information is available—in the form of a preprint, for instance. Most importantly, this strategy would alleviate the issues with referring to publications only by a single name; we would have to refer to it by its title, or its abbreviation, or some other kind of identifier. Plus, since you have control over the hash, you can still use it to get the recognition for the work in certain areas, such as faculty applications! I concede that this algorithm probably works best for ‘Everyone is Equal’ publications, but maybe it would also make shared authorship in papers more attractive again.

I know that such a solution is neither practical nor feasible for many other reasons, but I like to use it to challenge myself a little bit by asking myself the following question on regular basis: ‘Would I still like to work on the things that I am working on if my name was not attached to them?’

My answer, so far, is a resounding ‘Yes’. I hope it stays that way.

1. I myself have seen a poor justification for this once or twice now; the other party claimed that the journal had a limit on the number of authors on a paper. I will give the physicists of my audience a few seconds to catch their breath after finishing laughing. Just imagine how these publishers would fare if you present them with a paper from the ALICE Collaboration, for example. Let’s move on, though, and discuss useless publishing rules in another post. ↩︎

2. I operate under the assumption that even if such a state of affairs is not achievable, we should at least strive towards it. ↩︎

3. The axioms appear in slightly altered wording. See here for a longer paragraph with a source. ↩︎

4. As Goodhart’s Law goes: ‘When a measure becomes a target, it ceases to be a good measure.’ With the irony befitting this type of blog post, the expression was actually coined by Marilyn Strathern↩︎

5. This is my daily reminder that I should resist the urge to believe in external measures of academic success such as the $h$-index. ↩︎

6. The collective noun for a group of people doing machine learning research should maybe be ‘A Batch of Machine Learners’, or a ‘A Gradient of Machine Learners’, or some such thing. ↩︎