9.2 Philosophers’ Imprint

I’ve often speculated in the book about what would happen if we ran the model forward beyond the end of JSTOR’s moving wall. The purpose of this section is to look at what happens if we run the model forward in one very particular way. I downloaded all the articles published in 2019 by the open access journal Philosophers’ Imprint. I extracted the text from these articles using Jeroen Ooms’s package pdftools. And then I applied the model to these articles using the posterior function in topicmodels. Here are the primary topics for each of the 54 articles.

For each article I’ve shown which topic the model thinks the article most likely falls in. And these are, I think, mostly sensible classifications. But I’ve also shown the probability the model gives to that article being in just that topic. And given that these are maximal, they might look surprisingly low.

As we saw in the previous section, the model is extremely cautious when making out of sample classifications. So even for what looks to us like a fairly clear case, like say Janum Sethi’s Two Feelings in the Beautiful: Kant on the Structure of Judgments of Beauty, the model is only 30% sure of its judgment. Now I guess it isn’t surprising that the model is a little hesitant about whether to classify this with article on Kant or articles on Beauty, but the uncertainty you see in out-of-sample applications goes well beyond this. Here, for instance, is the list of all the topics it thinks Sethi’s article might be in. (As has been the standard through this book, I’ve cut the table off at 2%.)

Table 9.14: Janum Sethi, “Two Feelings in the Beautiful: Kant on the Structure of Judgments of Beauty”, Philosophers’ Imprint, 2019, vol. 19, number, 34
Subject Probability
Kant 0.2090
Emotions 0.1427
Beauty 0.1188
Dewey and Pragmatism 0.0964
Norms 0.0852
Psychology 0.0415
Arguments 0.0320

The top three look reasonable, though the numbers are too low, especially for Kant. This is just what happens in these out-of-sample applications. I have no idea why the model thought Dewey and Pragmatism was there. But the story about Norms is a bit more interesting.

It turns out that the language of twenty-first century philosophy is rather different from the language of twentieth-century philosophy. This is like the way in which the language of mid-century British philosophy was rather different to the language of other places and times. And the Norms topic picks up some of these newly fashionable words. I’ll come back to this at much greater length in section 9.3.

It isn’t only Sethi’s paper that is part of this linguistic change. There are many papers that are primarily in Norms, even though they aren’t really all about normative philosophy. Some of that is washed away when you use the weighted counts, but far from all of it. Here is the table of the weighted sum of each of the topics over the 54 articles. (That is, it’s the sum, over the 54 articles, of the article being in that topic.) The table is long, so it’s split up, sortable, and searchable.

How does Norms get to 4.8 like that? Lots of small pieces it turns out.

This is really surprising, and a sign that something’s gone wrong. It tells us something, I think, about the language of 21st century philosophy. I’ll return to this in the next section. What I’d rather look at, because I think it tells us something more about where philosophy is heading, is the topic that was second on the above list: Other History.

This topic had looked rather dead in recent journals, but it turns up second overall here. And just eyeballing the article list that might not be too surprising. There are articles on Suárez (two of them!), William King, Amo, and Mary Shepherd. But this isn’t entirely why Other History pops up so high here.

This makes the lack of attention to ‘Other History’ in the journals up to 2013 even more striking. James, Wittgenstein and Nietzche are hardly obscure figures. If articles on them are enough to keep the Other History scoreboard ticking over, it’s really telling that that scoreboard had ground to a halt.

Apart from this topic, the results are largely as I expected. The contemporary topics mostly keep growing. Vagueness, which peaked in the early 2000s, is an exception. And Imprint does less philosophy of science than some journals, so the recent philosophy of science topics aren’t showing up much. But otherwise the new topics are still growing. And the old topics are still mostly shrinking, but with two big exceptions. One, the resurgence of interest in historical figures outside of the huge names like Plato, Descartes and Kant, I’ve already mentioned. The other is that there is a paper on Idealism.

Table 9.15: Miri Albahari, “Perennial Idealism: A Mystical Solution to the Mind-Body Problem”, Philosophers’ Imprint, 2019, vol. 19, number, 44
Subject Probability
Idealism 0.1013
Self-Consciousness 0.0715
Perception 0.0646
Norms 0.0612
Other History 0.0489
Cognitive Science 0.0462
Kant 0.0402
Composition and Constitution 0.0360
Physicalism 0.0358
Depiction 0.0329
Faith and Theism 0.0295
Modality 0.0232
Classical Space and Time 0.0211

Now on the one hand, that’s only a 10.1% probability of being in Idealism. On the other hand, it doesn’t look like a mistake for the model to put this article in Idealism. And to my eyes, in the twelve journals I’m focussing on, there are no articles from about 1995-2013 that were correctly classified as Idealism articles.

I don’t want to say that one article in Imprint is a trend, even if it is something we hadn’t seen in the last 7000 or so articles in the original data set. But it is interesting, and something to watch for over upcoming years. It is so striking to me that Idealism went from being so big to so small, and some regression to the mean seems like it wouldn’t be a surprise.