7

An article on The Laughing Squid shows a German shepherd seemingly following written commands.

An impressively intelligent German shepherd named Rambo, who’s learning how to read with the help of his human Lily, deftly demonstrated his comprehension by correctly following written commands like “sit” and “bow” on signs held up for him to see.

The article is mainly just a wrapper around a video on YouTube showing the same. Rambo's owner holds up a card with a written command such as "bow" or "sit", which Rambo then obeys.

Is Rambo following these written commands, without any additional clues?

Has this been tested independently? Or is he just the latest Kluger Hans?


The claim in both the article and (the title of) the video is that the dog is learning how to read.

I would be satisfied with evidence that the dog is able to recognise (the words on) the cards in any way. For instance a blind test (which could even be conducted by the owner, as long as she doesn't know which card she's holding up) with one or more added control words (commands the dog hasn't learned to recognise, so he should do nothing).

2
  • 7
    Note that even if true this doesn't prove he can read, any more than responding to a spoken "sit" means he understands English. Mar 3, 2018 at 3:36
  • How does one know if the video hasn't been selectively edited? Mar 6, 2018 at 18:46

1 Answer 1

6
+100

Summary:

  • We don't have good evidence that let's us conclude precisely what Rambo's "reading" abilities are. The experimental design and setup in the video does not allow to judge this, and there isn't that much more material on amount Rambo's "reading" abilities publicly available.

  • In general, dogs can be trained to distinguish visual patterns. This has been done e.g. to measure their visual acuity.

  • There are several sources in the internet on training dogs to recognize words on flash cards.

  • Yes, Kluger Hans effect as in the dog in fact evaluating clues from the owner/trainer instead of from the flash card regardless of whether he'd be physically and mentally able to properly learn the required level of patten recognition may be a problem.


Long version:

Rambo

Searching for Rambo's reading ability, I found two youtube videos

and no whatsoever further information/actual tests as suggested about his "reading abilities".

From these videos, it is not even remotely possible to judge whether he actually recognizes the patterns on the flash cards or uses clues by the owner (Kluger Hans effect) – without even considering the possibilities of video editing and selection. We may say that the Kluge Hans effect is closely related to bad experimental design, so as a reference for the following points take a textbook on experimental design of your choice.

  • as OP says, we have no indication that the test is performed blindly, and
  • no measures have been taken to reduce the risk of (unconscious) clues by the owner
  • we have no indication whether the order of the commands is (pseudo)randomized (including random as opposed to arbitrary choice by human) or whether the dog may possibly have learned the order of commands.
  • some kind of negative control, and or "placebo" e.g. empty card, unknown command would be important for calculating the probability on accidentally getting the trials correct.
  • Reading would usually be interpreted at least as opposed to pure shape recognition, e.g. differentiating horizontal from vertical stripes.
    We'd usually also require a certain generalization ability in the pattern (recognize the commands when printed on a different paper, off-center, in a different font, size, …)

So slightly sarcastic: if we call the performance we're seeing in the video reading, we may as well go on and call him bilingual since there's the German Platz on one of the cards in the 2nd video.</sarcasm>

So we have 2 short videos with two different flash cards each (sit/bow and sit/platz).

The first video has 4 trials: BOW - BOW - BOW - SIT, and he gets them all right. However, 4 trials are not much. 4 successes in 4 independent trials yields a 95% confidence interval for the %-of-success ranging roughly from 45–100%. And there's a whole lot of potential confounding factors here that could make the behaviour dependent – just not on the printed command (Kluger Hans). E.g., he may have learned that commands are usually repeated.

Also consider the 2nd video. It shows 4 trials: SIT - PLATZ - PLATZ - SIT. Both times, when shown the flash card "PLATZ" (German for down) he sits. When this doesn't yield a treat, after a while he goes into down and is rewarded.

This observation is in accordance with a wide variety of explanations:

  • This may be a perfectly fine recording of Rambo being introduced to/learning the PLATZ card after already knowing SIT.

  • This may also be a textbook example of a very obvious Kluger Hans effect, the clue being "sit is not sufficient to get a treat".

  • sit and down are not the best choice to check the dog's pattern recognition ability, since they may be confounded.

    • Many dogs learn Platz/down as an extension of sit.
    • In consequence, some training may be needed to make the commands independent afterwards. We don't know whether this has happened.
      I for one am perfectly fine with my dog going on to see down as a "stronger sit" – that perfectly suits my needs. I.e., on "sit" I accept the dog sitting or laying down. Only, on "down" sitting is not sufficient. And the dog has certainly tried many times whether he can get away with only sitting down (just as he's tried whether he can get away with his ass not truly on the ground. Particularly on wet grass + cool morning temps).

    Rambo may be perfectly recognizing the difference between SIT and PLATZ, but tries to get away with sitting.


Dog visual acuity and a well-designed experiment on pattern recognition in dogs

One possible concern with such reading experiments, in particular with handwritten flash cards as shown in source 2 below is that dogs have a lower visual acuity than humans (they don't see as sharp). The pattern would need to be very roughly magnified 3x to be seen as well by the dog compared to a human.
The printed commands in the Rambo videos should be OK, but the handwritten commands in source 2 may be physically difficult depending on how far away the dog starts.

source: Lind et. High visual acuity revealed in dogs, PLoS ONE 12(12): e0188557, 2017

This paper is a very nice example that dogs can be trained to differentiate horizontal from vertical stripe patterns. It contains a nice experimental section showing one possible setup that avoids many of the experimental design pitfalls brought up by OP and above. (Though there are experimental differences since the dogs in the paper only face a binary decision, whereas Rambo is asked to perform one particular action out of his large repertoire of possible actions, or maybe rather out of the smaller repertoire of actions that may yield treats according to the flash-card play rules ;-) )

(Personally, I'd probably go for a somewhat less elaborate setup and e.g. train the dog to "read" from a computer monitor and write a program to take care of proper randomization and place myself so the dog cannot get clues from me)


Teaching dogs to "read"

Teaching dogs to recognize words on flash cards and associate them with various commands seems to be a fairly well-known trick to teach a dog. Variety 1 Variety 2

  • from source 1:

    They do not learn the words like a human would, they learn the shape of the word. They would need to be retrained if the font changed.

  • Note how the method discussed in source 2 is IMHO asking for Kluger Hans effect, unless deliberate measures are taken to avoid such wrong associations. They outline only 1 such concern: not associating the hand with the right answer. But of course dogs are quite good at reading body langue and the trainer may thus guide the dog while it's on its way to the desired card.

  • All these lay experiments could comparatively easily be transformed to become far less prone to Kluger Hans effect. E.g. by presenting the visual clue via computer screen, in (pseudo)randomized order and with the owner well away e.g. treat being paid out "mechanically" (Arduino, Raspberry, …) or with acoustic feedback, observation could be done by owner in next room via web cam.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .