Geoffrey R. Loftus Ph.D.
Editor, Cognitive Psychology
University of Washington
Seattle WA 98195

November 16 2000

Dear Dr. Loftus,

Although I am of course disappointed that my paper, "Computational Implications of Gestalt Theory" (99-068R) has been rejected for publication by Cognitive Psychology, I cannot say that I am entirely surprised. After a long series of unsuccessful submissions of various papers to a variety of journals, I am beginning to get fatalistic, and almost expect rejection from the outset. I selected Cognitive Psychology for this particular submission specifically because of this journal's sympathy with the Gestalt viewpoint. Unfortunately Gestalt theory is largely misunderstood today, even by its supposed proponents. As usual, I consider this rejection to be thoroughly unjustified. As with my previous rejections, I have made this paper, the detailed reviews of it, and my rebuttals to each objection of the reviewers, available at my web site. I provide hyperlinks to my rebuttals of the reviewer's critiques at the bottom of this page, in case you should be interested in hearing my defense. I am confident that one day this paper will receive the recognition it deserves as a significant contribution to our understanding of the computational principles revealed by Gestalt theory. I am sorry that Cognitive Psychology did not have the vision to recognize its significance.

I do not fault you personally for this rejection. In fact I was deeply gratified that you found it clearly written and informative, given the complexity of the subject matter. I find it regrettable that clarity of exposition is a rare commodity in scientific papers. (Have you have ever tried to read one of Grossberg's papers?) My problems with publication highlight a broader more general failure of the anonymous peer review process as a whole. Astute "career scientists" have long recognized that the secret to success is to propose only minor modifications of well established ideas, sufficiently novel to be considered original, but not so novel as to threaten their chances for publication. Authors like myself, who wish to probe the more fundamental issues underlying their science, and to investigate alternative paradigms, are tilting at windmills, for their papers are unlikely to survive the peer review process.

The present paper is only one of a number of submissions I have made on a variety of subjects ranging from detailed models of specific perceptual phenomena, to more general discussions of consciousness and behavior. However all of these diverse papers have been guided by the larger vision of Gestalt theory. For I believe that the original Gestaltists were very much on the right track, and contemporary psychology has been diverted to a theoretical dead end, seduced by the lure of an atomistic paradigm of neurocomputation, inspired by the point-like recordings of the single-cell electrode, and by the robotic logic of the digital computer. The time has come to re-evaluate the implications of Gestalt theory for our models of neurocomputation.

I have discovered however that it is extraordinarily difficult to communicate a paradigmatic message through the peer review process, because an alternative paradigm, by its very nature, has implications in so many diverse areas, that it can never possibly be presented in a single paper. If on the other hand, the larger concept is carved up into smaller portions which are submitted for publication individually, those papers are promptly rejected by reviewers who judge them from within their own narrow paradigmatic framework. Such is the case precisely with the present paper. Believe it or not, this paper is my most conservative and reserved attempt at publication, limited to a single general principle of proposed perceptual computation, clearly presented and supported by evidence, and actual computer simulations to prove that I am not just making it up. Can you imagine my naivite'- I actually thought this one would get accepted!

I expressly avoided defending the specific computational stages for the model, which were presented only as a concrete example of the concept in order to focus on the more general principle proposed by the paper, which is that whatever those stages might be, the transformations between those stages should be accompanied by inverse transformations, in order to unite the entire hierarchy into a single coupled system. This is the only way to resolve the hierarchical architecture observed neurophysiologically, with the principle of emergence identified by Gestalt theory. This however led reviewer A to comment that "I am not saying the inverse transformation approach is wrong, it may very well be right. However it cannot, by itself guide the development of models of visual perception; it needs other computational constraints to guide it." However had I proposed those specific constraints, they would surely have become the central focus of the criticism, distracting attention from the principal message of the paper.

I specifically avoided objections of "neural plausibility" by proposing a perceptual modeling approach, investigating the information processing evident in perception in a manner that is agnostic to any specific neurophysiological assumptions. Reviewer B complains that stripping away the identification of computational stages of the model with specific neural mechanisms "undermines the justification for the types of filtering and feedback processes that comprise the model". It so happens that I have a theory of the neurophysiological implications of Gestalt theory, generally consistent with the present theory. But that theory is so far beyond the scope of contemporary concepts of neurocomputation that it requires a whole paper devoted to that specialized topic. But that theory is far more radical than the tame proposal presented here, which is why I left neurophysiology out of the discussion.

I have specifically avoided discussion of the philosophical or epistemological implications of the present model. But this led reviewer A to complain about the core ideas of the paper, saying "I am left wondering why the visual system would be built with the levels proposed by the MLRF approach ... the visual system surely did not evolve to allow it to see Kanizsa triangles!". Again, this is a topic that is so broad as to require an entire paper devoted to that subject, which I have also produced. Ironically, those other papers have encountered vigorous opposition, in part because they leave unspecified the computational principles presented in the present paper.

It seems therefore that the standard peer review process is unsuited to presentation of larger paradigmatic hypotheses which cannot be expressed in a single self-contained paper. Perhaps this explains why our psychology journals are clogged with great numbers of papers on minor variations on the tired old tried-and-failed ideas, and why we see so little discussion of the larger issues which are so relevant to perception and consciousness.

Disappointed but undaunted,

Steven Lehar

Reviewer A comments & authors responses

Reviewer B comments & authors responses

Geoffrey Loftus responds?