A “granny summary” of the article by Dror Dotan
Top-down number reading: Language affects the visual identification of digit strings
Please read the following text from Lewis Carrol’s Alice in wonderland:
There was nothing so very remarkable in that; nor did Alice think it so very much out of the way to hear the Rabbit say to itself, “Oh dear! Oh dear! I shall be too late!” […] but when the Rabbit actually took a watch out of its waistcoat-pocket, and looked at it, and then hurried on, Alice started to her feet, for it flashed across her mind that she had never before seen a rabbit with either a waistcoat-pocket or a watch to take out of it.
After reading, take a look at this picture and think what you see:
Some of you must be already familiar with this picture; but still – which animal did you see?
I’m guessing you saw the rabbit. And maybe you also saw the duck immediately after (if you haven’t: the rabbit is looking at the carrot. The duck is looking the other way, left and up, and its beak is the rabbit’s ears).
Why did you see the rabbit first? Well, there are many factors that could affect which of the two animals we see first. For example, the way the picture is drawn and even its orientation. Another factor is that we let you read Carrol’s story about the rabbit, and that the picture also includes a carrot. This is the point of this demonstration: that the context could affect how you interpret the picture.
Think about it – this is not trivial. I think that our intuition about our visual system is that it works in a “bottom up” manner: the light returned from the screen enters your eye and hits the retina; the visual information is transferred to low-level visual processing modules in the brain, which detect the basic elements in the picture (intensity, lines, dots, etc.); then, in higher processing levels, we recognize perhaps a picture similar to something we’ve already seen, and finally we come up with the word “rabbit” and its associations. Right?
Wrong. Vision, as well as our other perceptual systems, do work this way, but not only this way. Yes, much of the visual information flows in the “bottom up” way I described above, from the eye up to the brain’s high processing levels, but there’s also a lot of information flowing back, from high-processing brain areas down to the lower-level perceptual processes. This feedback information helps tuning the visual perception according to our expectations. It’s interesting because it tells us something about how our brain works: it’s not like a super-smart CPU that receives information from several “dummy” sensors; it’s more like a distributed computer, in which both the CPU and the sensors are smart, and the interaction between them is bidirectional, not unidirectional.
OK, how is all this related to the way we read numbers?
In the domain of vision and perception, we already know that the story is complex and that perception involves a combination of bottom-up and top-down processing. But ironically, complex cognitive operations such as language and reading numbers are often described using the simpler story, of an only bottom-up processing.
Even here on this website, we told you – in a previous granny summary – a simplistic story of this kind, which says the following thing: when we read numbers aloud, the process begins with the visual analyzer, which encodes the visual digit string we saw (e.g., the identity of the digits and their order); and it continues with several processing stages, at the end of which we say the number words corresponding with the digits we saw. This story is true, but like in the rabbit-duck example, it’s not the whole story. Today I want to tell you about a study of ours, which showed that this is not the whole story.
The basic question was: are there top-down processes in number reading? i.e., on top of the visual analyzer sending information to the verbal-production processes, are there feedback channels via which the verbal-production processes send information back to the visual analyzer and tune its behavior? To examine this question, we focus on a particular characteristic of the visual analyzer – it scans the digits from left to right. We asked why it scans the digits in this particular order. One possible answer is that this is an intrinsic behavior of the visual analyzer. But another answer, which for our present story is more interesting (and it also happens to be true), is that the digit scanning order is not determined only by the visual analyzer but also by the verbal production processes, via top-down channels. Why should verbal production processes care about the digit scanning order? Because they may prefer getting the digits’ visual information in the same order in which we say the number words. For example, we say 123 as “one hundred twenty three”, and a left-to-right order of digits is a perfect match to the word order.
To arbitrate between these two answers, we examined number reading in speakers of Arabic. In Arabic, unlike English and Hebrew, the ones word precedes the tens word. For example, 32 is “tnen wa-thalatin” (two and thirty), and 54,321 is “four and fifty thousand, three hundred one and twenty” (by the way, ancient English too allowed for this word inversion). If the top-down story is true, we should expect the visual analyzer of Arabic speakers to scan the unit digit before the decade digit.
How can we tell in which order the visual analyzer scans the digits? We used a method that we already used in a previous study: the participants read aloud numbers that were presented very briefly (100 milliseconds). Under these conditions, the visual analyzer occasionally fails to reach the digits it scans last, so there are more errors in these digits. So if we observe more errors in digits in a particular position in the number, we can tell it was scanned late.
So now, our argument goes as follows: if the visual analyzer is affected by the order of number words in a given language, we expect a difference between the speakers of Arabid and Hebrew. Compared to Hebrew, the Arabic speakers will succeed a little more in the unit digit, and a little less in the decade digit. Note that we don’t care about the overall difference between the units and decades, or about the overall difference between Hebrew and Arabic – these could be affected by additional factors. We only predict different patterns between Hebrew and Arabic. And this is precisely what happened:
The error rate in each digit according to its decimal position, separately for Hebrew and Arabic. The critical finding concerns the inter-language differences in the accuracy levels in the decade and unit digits. In both languages the unit digit was identified better (presumably because it’s at the end of the number), but this effect was magnified in Arabic. This indicates that Arabic speakers processed the unit digit before the decade digit, but Hebrew speakers did not.
And that’s not all. We also found the same effect in Hebrew-Arabic bilinguals, who read the same numbers in the two languages. This is interesting because it means that the effect of language on the visual analyzer is not just a global effect, which determines the modus operandi of a particular person’s visual analyzer. Language affects the visual analyzer even momentarily, according to the language we currently speak.
This study shows that top-down channels operate also in the case of reading numbers. It also shows that language affects our thought, including cognitive mechanisms that we don’t view intuitively as “linguistic”, such as the visual parsing of a digit string. We’re also happy that this study taught us a little more about how precisely the number-reading pathways operate. As we’ve told you in another “granny summary”, reading numbers is hard, and in Arabic – with its number-word inversion – it’s even harder. The more we understand the number reading mechanisms, the closer we get to make it easier for children to learn.
Interested in more details? The full article is here.