TransitiveInferenceAndTheHippocampus

There is a hypothesis that the hippocampus is essential for the function of TransitiveInference.

Evidence for

• Human functional MRI studies directly show task dependent hippocampal activity at study and at test (retrieval) for Transitive Inference. See Greene, Gross, Elsinger & Rao July 2006, J Cognitive Neuroscience. The fMRI evidence completely obviates claimes (see below) by Frank, O’Reilly and their colleagues. The Franks claim is derieved almost entirely from computer simulations, not upon direct evidence. Human hippocampal activity is observed both in implicit and explicit versions of the task. There may be other possible strategies for solving the TI task, but under normal circumstances, TI is hippocampal dependant in humans (whether implicit or explicit) and other animals. Performing idiosyncratic manipulations that bias alternative strategies hardly makes a strong case.
• Animals with hippocampal lesions have problems with transitive inference (Dusek and Eichenbaum ‘97) (but this was one study; another with hippocampal pigeons showed normal transitive performance Strasser, R., Ehrlinger, J.M. & Bingman, V.P. (2004).
• Imaging studies show hippocampal activation during transitive inference (Heckers et al '04) (Preston et al ‘04) (Nagode and Pardo '02
• Behavioral studies with schizophrenia patients, who often show hippocampal (and prefrontal) dysfunction, show a deficit in BD performance (Titone et al '04)

Evidence against

• A closer look at the task used in (Dusek and Eichenbaum ‘97) shows that it is possible the rats are using a non-inference strategy. If so, each stimulus item may obtain an associative weight. See (Van Elzakker et al '03) for details. You may also be interested in (Frank et al '03) for a computational model of the results of (Van Elzakker et al '03).
  • (note: Frank et al’s model claims to utilize the “pattern completion” abilities of the hippocampus, which sounds like sequence learning; but it seems to me that they actually just use supervised learning (learning to pair an appropriate with a given stimulus).
  • Note: modeling is not direct evidence.
  • (note: Frank et al’s model allows “transitive” inference, but with only one or two hops; theirs is sort of like the diffusion of value in the value function in a.i.-style reinforcement learning)
  • Interestingly, humans show the same psuedo-transitivity judgements as rats do when tested appropriately. See When Logic Fails: Implicit Transitive Inference in Humans
    • Rebuttal This is a dubious claim based on superficial similarity of results. What does “tested appropriately” mean? Who decides what the “appropriate strategy for a task should be other than the way its done usually? Do they really mean when tested in such a way as to promote an idiosyncratic strategy?
  • Rebuttal How does this “possibility” provide direct evidence against hippocampal involvement? This bullet merely argues there may be an alternative esplanation but doesn’t establish it as non-hippocampal.
  • Rebuttal to Rebuttals: The point of the Frank et al ‘05 paper was that if you prevent people from employing explicit logic, they can still learn associatively in a TI task, and when doing so, their test performance looks a lot like the pattern seen in rats, pigeons, etc. More recent modeling from that group suggests that the associative part of learning depends on the striatal dopamine system, and evidence with Parkinson’s patients performing a TI task on and off their dopamine medication is consistent with this view (Frank, Seeberger & O’Reilly, 2004). Also a recent study with human subjects taking midazolam (a benzodiazepene known to transiently disengage the hippocampus) showed explicit memory deficits, while their implicit transitive performance was actually (and substantially) improved (Frank, O’Reilly & Curran, 2006, Psychological Science). This group also argues that the hippocampus may still sometimes be involved in TI, especially in humans performing the standard logic-based reasoning (the subject of most human imaging studies). Also, while fMRI data such as the Greene study is useful, it can never (as suggested above) “obviate a claim” that a brain region (in this case, the hippocampus) is or is not “essential” for a given task (in this case transitive inference). The Greene ‘06 paper shows hippocampus activation for both explicit and implicit TI, but several other brain areas including striatum were also activated. Clearly more studies are needed, and one might want to investigate whether any of these brain area activations actually correlates with successful transitive performance. -mjf
• Some papers argued that a closer look at some of the imaging studies show that the activation is during encoding, not retrieval. However, this is no longer a good argument because other imaging studies do show activation at both test and retrieval: Greene et al 2006, Heckers et al 2004. (the second was clearly explicit, though)
• One imaging study instead shows activation elsewhere (Acuna et al ‘02). This study didn’t image the hippocampus or medial temporal lobe at all

Discussion

Thank you to the anonymous contributor for updating this page. I moved some of your rebuttals into separate bullet points to make reading easier. I removed the “possible (but not probable)” because, first, this is already what “possible” means, second, it makes it easier to read, and third, to someone who likes the other explanation I doubt a naked assertion that it is improbable would convince them. I removed “The existing facts must be contorted and misrepresented substantially in order to accept the psuedo inference view.”; because this seems like it may be a controversial judgement/opinion, I don’t think it will convince anyone if it comes from an anonymous source. But feel free to put it back as your personal opinion and sign your name.

I wrote the original page here but I haven’t done work in this area myself; I just read some papers and summarized what I thought their arguments were. It’s perfectly possible that the Greene et al study does indeed make the counter-arguments obsolete, but I am wary of making such a judgement myself. Maybe I should ask some of the authors of the other papers to put in their 2c.

– BayleShanks

The Greene 06 paper does NOT? show striatum activation.

• Both caudate and putamen, together forming the striatum, are activated in some (but admittedly not all) of the same conditions that activate the hippocampus, including a critical one involving successful BD inference (see Tables 2 and 3). Still I don’t take this as evidence for striatal mediation of TI, as there may be a number of factors that lead to activations in those areas here. While both caudate and putamen have activated to reward based feedback, the caudate is often also activated in conjunction with prefrontal areas associated with explicit working memory. I do think the hippocampus activation is interesting and should be further explored. -mjf

The idea that Midazolam is a reversible hippocampal lesion is based on flawed premises. First, lots of areas are put to sleep by normal doses of midazolam, including frontal and parietal areas known to be involved in TI. The Frank 06 paper should have never made it past review. Apparently Rudy and O’Reilly have good connections at the editorial office of Psych Science. That paper is just bad science.
– Anonymous

Ouch. Right, this point is discussed in the paper. But by far the best and most extensive evidence for midazolam effects, based on both psychological and neural evidence across multiple methods, is in the hippocampus – see also another paper by Curran et al 2006, JNeurosci which used the identical drug procedures and found that the drug selectively impaired brain potentials previously associated with hippocampal based recollection, but not cortical based familiarity. Moreover, how would midazolam “putting to sleep” other areas involved in TI enhance TI performance, as was found in that paper? The key point is that neither the striatum nor striatal-based learning seem to be impaired by the drug.. anyway, I think we agree that there may be multiple ways to solve TI tasks. -mjf

Response: There’s a distinction between necessary and sufficient conditions here: Take an analogous sittuation: we know that damage to the optic nerve results in blindness, but does that mean that the observation of blindness necessarilly indicates damage to the optic nerve? Likewise, we know from Korsakov’s and other amnesia research that damage to areas other than the hippocampus can result in dense amnesia. There is no direct evidence that intravenous midazolam selectively innactivates the hippocampus, and evidence that midazolam causes amnesia is not adequate.

As to how midazolam could enhance TI performance, I’d like to wait and see if anyone is able to replicate the effect before commenting on that…

Refs

• Greene,A.J.; Gross,W.L.; Elsinger,C.; Rao,S.M. (2006) An fMRI analysis of the human hippocampus: Inference, context and task awareness. Journal of Cognitive Neuroscience, 18 (7) 1156-1173.
• Acuna BD, Eliassen JC, Donoghue JP, Sanes JN. Frontal and parietal lobe activation during transitive inference in humans. Cereb Cortex. 2002 Dec;12(12):1312-21.
• Jeffery A. Dusek and Howard Eichenbaum. The hippocampus and memory for orderly stimulus relations. Proc Natl Acad Sci USA? 94:7109-7114. (1997)
• Frank, M.J., OReilly?, R.C. & Curran, T. (2006). When memory fails, intuition reigns: Midazolam enhances implicit inference in humans. Psychological Science, 17, 700-707.
• Frank, M.J., Rudy, J.W., Levy, W.B. & O’Reilly, R.C. When logic fails: Implicit transitive inference in humans. Memory & Cognition (2005)
• Frank MJ, Rudy JW,O Reilly RC. Transitivity, flexibility, conjunctive representations, and the hippocampus. II. A computational analysis. Hippocampus 13:3:3341-354. (2003)
• Frank, M.J., Seeberger, L.C. & O’Reilly, R.C. (2004). By carrot or by stick: Cognitive reinforcement learning in Parkinsonism. Science, 306, 1940-3.
• Heckers S,Zalesak M,Weiss AP,Ditman T,Titone D. Hippocampal activation during transitive inference in humans. Hippocampus 14. (2004)
• Jennifer Nagode and Jose Pardo. Human hippocampal activation during transitive inference. Neuroreport. 13(7):939-944, May 24, 2002.
• Alison R. Preston, Yael Shrager, Nicole M. Dudukovic, John D.E. Gabrieli. Hippocampal contribution to the novel use of relational information in declarative memory. Hippocampus 14:2. 2004.
• Peter R. Rapp. Who's the Fairest of Them All? Role of the Human Hippocampus in the Relational Organization of Memory. HIPPOCAMPUS? 14:141-142 (2004) rel. besttermpaper.com
• Rapp, Peter R.; Kansky, Mary T.; Eichenbaum, Howard. Learning and memory for hierarchical relationships in the monkey: Effects of aging. Behavioral Neuroscience. 1996 Oct Vol 110(5) 887-897.
• Strasser, R., Ehrlinger, J.M. & Bingman, V.P. (2004). Transitive behavior in hippocampal-lesioned pigeons. Brain, Behavior and Evolution, 63, 181-8.
• Titone, D., Ditman, T., Holzman, P. S., Eichenbaum, H. B., & Levy, D. L. (2004). Transitive inference in schizophrenia: Impairments in relational memory organization. Schizophrenia Research, 68, 235-247.
• Mike Van Elzakker, O’Reilly RC, Rudy JW. Transitivity, flexibility, conjunctive representations, and the hippocampus. I. An empirical analysis. Hippocampus 13:3:334-340. 2003.

CategoryCog

CategoryHippocampus