Charles Travis published a revision of one of his seminal philosophical essays. When authors revise their work in print, the need for comparison arises. Can we visualise what Travis changed?
The Silences compared
Charles Travis published ‘The Silence of the Senses’ eleven years ago in the philosophy journal Mind. Travis thought he saw a fundamental weakness in a dominant theory of perception, ‘representationalism’. His essay has proven a landmark in discussions of perception up to today (Google Scholar currently counts 241 citations).
in 2013 Oxford University Press published Perception: Essays after Frege, a collection with Travis’ recent work on perception. ‘The Silence of the Senses’ is reprinted as the first essay. But it’s reprinted with a catch: modified, clarified, and supplemented with new material. In the collection’s introduction Travis writes about the original, as it appeared in Mind:
“Here my perhaps idiosyncratic terminology provoked what I thought avoidable misreadings of the text. I revised the terminology accordingly. Revision is extensive.”
Were his revisions for the better? No doubt that is for a reader to decide. But it does invite curiosity: What exactly did Travis change?
To answer this question, I tried the comparison algorithm GitHub rolled out last year. Its results are tidy presentations of changes — ‘rendered prose diffs’, as the GitHub team calls them. It makes manifest the often small but pregnant revisions of Travis’ essay. Visualising them may help us retrace some of the philosophical action. At the same time, it sets an interesting example of scholarly visualisation (though I’ll close with some caveats).
What an experience was like
The first revision we encounter occurs in the essay’s title. The revised version’s title now contains the plural ‘Silences’ instead of the earlier singular noun. This is likely to prevent bibliographic confusion.
Interestingly, Mark Kalderon has suggested to me that this particular change was unintended:
That notwithstanding, speaking of the silences of the senses is suggestive. For, there seems no reason to assume there is only one way for our senses to speak and, therefore, to be silent. I think Travis’ discussion in the essay is sensitive to this potential plurality, and it is perhaps not bad now to find this reflected explicitly in the title as well.
Two other significant changes stand out on the opening page. First, a revision of what could motivate philosophers to think of perception as representational.
The second change that occurs early on is the omission of a footnote. As we browse further, it becomes clear that Travis has removed all but one of the original twenty-four notes. Many of them seem to have disappeared without a trace. Only his expression of debt to the (then) Editor of Mind survives the sterilisation. Travis has modified that note to identify him more explicitly.
Auto- & allo-
A distinction central to Travis’ argument against the representationalist is that between ‘autorepresentation’ and ‘allorepresentation’. He must have noticed that these custom-made terms do not wear their meaning on their sleeves. To be sure, the prefixes auto- and allo- do suggest a contrast between representing something to oneself and representing it to someone else. But in turn these locutions are not entirely clear. It is not at all obvious how such an opposition bears on the fundamental question of perception.
To explain autorepresentation more fully, Travis subtly tweaks his presentation in the revised version of the essay. Significantly, in the section titled ‘The position,’ he now seems to emphasise the activity of representation, as he introduces the main idea as that of autorepresenting.
Autorepresenting, as the term suggests, is a ‘representing something to oneself’. That’s what Travis gave his reader in the first version of the essay. In the new version, he unpacks this in terms of a stance one takes towards things being a certain way (or, “thus and so”). For example, my autorepresenting that you’re cooking dinner tonight is a stance I may come to take after reading a note held fast by a magnet on the fridge; my representing things in that way is not a representing of the sort the note itself performs. Travis mentions the note on the fridge in parenthesis to enable a better grasp on what autorepresenting is not. The note does not auto-represent anything; it instances a different form of representing, a form Travis introduces as allorepresenting. It represents something to someone else. (Or, to avoid over-animating Post-It notes, it was attached to the fridge in order to represent something to someone else.)
The original discussion of allorepresentation is expanded considerably. Also here Travis now prefers to speak of allorepresenting. He goes through some length to clear up misunderstanding, emphasing that, like the Post-It note, all allorepresenting requires a ‘vehicle’ that makes a way for things to be recognisable to a perceiving thinker. The changes here are significant enough for GitHub to treat them as a full replacement of paragraphs:
Travis’ clarification in these lines helps to bring out how bold his contention is. We asked: How does the distinction between autorepresenting and allorepresenting bear on the fundamental question of perception? I venture to say that Travis answers that the representationalist is committed to regarding perception a species of allorepresenting. It has that in common with the Post-It note. If that is right, then it must be possible for us to identify some vehicle that makes perceptual representing recognisable. To me this seems Travis’ central move. It goes to show that misunderstanding the concept of allorepresenting is tantamount to misunderstanding that move. It’s not surprising, then, that we find some further specification of what allorepresenting is.
What Austin saw
Why can perception, if representational, only be allorepresentational? The rough idea is that, according to the representationalist perception has a ‘face value’ — we ‘read something off it’, something which we can believe or not. This implies it is a potential source of error or mistake. Travis is critical of that point, and the section devoted to it — ‘Misleading’ — remains largely in its original form.
Two small but interesting observations are brought to the attention when we compare the two versions. The first concerns an historical matter.
In this first modification, Travis imports from a longer footnote an insight about what Austin saw and Descartes missed: the way occasion-sensitivity applies also to our perceptual reports.
The second observation Travis slightly modifies is about how to understand visual illusion. Some have taken optical illusions to be a reason to give representation a place in perception. (In his essay Travis discusses McDowell, but the trend goes back to Peter Aureol.) Travis bucks the trend by suggesting we may conceive of the Müller-Lyer lines as actually having a misleading look. There’s nothing wrong with our visual experience of those lines, given that so presented they indeed have the look of lines of unequal lengths.
If the senses speak, so to say, then we must somehow be able to pick up on what they have to say. You can’t lie to me if I can’t hear you. And so perception, as the representationalist sees it, is only possible if there is some vehicle of representation, something that encodes for us what the senses have to say — something that makes their message recognisable. That’s why allorepresenting is so important. Just as your Post-It note needs to be legible to convey anything to me, perceptual experiences, if they are representational, need to make what they represent manifest. But how could they?
Travis’ central claim is that in perception per se there simply is no plausible candidate to fulfil a role that parallels that of the Post-It note. Of course this commits him to consider at least a likely candidate. Travis thinks that the representationalist’s best bet is that, at least in vision, it is the look of things that constitutes our being represented to in perception.
In his discussion of how there is no understanding of ‘look’ that could help the representationalist, we find a few significant changes. One notable thing is that Travis no longer uses the term ‘demonstrable look’, and instead opts for the more straightforward (though somewhat pleonastic) idea of a visual look.
It is because the way something looks is a merely visual condition it would be rash in general, Travis writes, to conclude from something’s looking like a lemon that it is a lemon. Soap manufacturers are known for their skill in copying just that look.
A further change is at the level of organisation. The older essay had two sections on the way things look. The first, ‘Looking like’ has been retitled to ‘Visual appearance’. It still focuses on the way some of our statements about “how things look” qualify an object’s visual features — features the object can share with a variety of things.
The second section, originally titled ‘Looking as if’, has in the revised version been split in two: ‘Thinkable appearance’ and ‘Hybrids’. It focuses on a different, ‘epistemic’ or ‘evidential’ class of looks-statements. Travis reinforces his earlier exposition by tying it to something he thinks Frege said:
What we are talking about when we say that “it looks as if…” Travis now calls a ‘thinkable look’. Those are not themselves things we can perceive. They are not visual looks. The notion of a thinkable look is novel, and it contrasts with the earlier idea of a visual one.
Epicycles of representation
In the final section of the essay, ‘Responses’, Travis discusses Gilbert Harman’s version of representationalism. His aim in doing so is to explore if, in some way, the representationalist has something in mind other than allorepresenting by means of visual looks, when they maintain that perception is representational. The section meanders past various alternative formulations and theoretical epicycles of the idea that perception is representational.
It is perhaps significant that Travis did not change the target of this section. Harman put forward his view in 1990, and so is hardly ‘recent literature’. Given that the last decade has brought us a tidal wave of articles and books defending or relying on some form of representationalism, Travis’ seeming conviction that these have not significantly altered the landscape should be noted. It’s a case where the absence of a revision speaks for itself.
But in his discussion of Harman, Travis did choose to add some further clarification. In commenting on Harman’s use of ‘awareness’, Frege makes another appearance:
Travis also introduces various small changes in the presentation of the way his opponents may wish to qualify their views, without however changing much about his own objections against these qualifications.
Comparing the two versions of Travis’ essay in this way brings to light a number of significant changes (you can see the full diff here). What he changed was not merely terminological, and often digs into matters of substance. To get sight of that, GitHub’s ‘track changes’ algorithm is very helpful, as it allows one quickly to make the difference between the two versions visible.
Because of other features of the Git version control system, and GitHub as a way of managing Git repositories, I think GitHub may present us with a powerful tool to represent versions and changes of scholarly output. This invites some day-dreaming: Travis’ essay is a simple case, but it is easy to extrapolate. Imagine documents of which we have a number of different versions or copies, such as the numerous copies in manuscript of an ancient text scholars have in their possession. A Git repository with different branches could represent the entire history of an important historical text, identifying minute changes and modifications over the course of its history. This would make the various versions and their interconnections visible, searchable, and easily comparable.
Three caveats: First, documents need to be availabe in one of the markup languages GitHub supports (for this reason, I had to convert both versions from html/pdf to Markdown).
Second, there is also a limit to the number of lines GitHub will render at once. This surprised me somewhat — Travis’ essay isn’t that long.
But that’s how it is. “When the changes take too long to analyze,” their help page tell us, “GitHub can’t always produce a rendered view of the changes.” This is why I had to split up the document into smaller ones.
There is a third shortcoming of the current situation that is significant. At present it does not seem possible to embed the the visualisation of the changes between two (states of) documents directly into a website. That’s why I had to seek recourse to using screenshots in this post. Embedding of other material hosted on GitHub is possible; the service allows embedding of code snippets (‘gists’) or (I think) even entire files in a repository. Adding to this range the functionality of embedding prose diffs, or line-by-line fragments thereof, may be an attractive feature.