A granny summary of an article by Hadar Efodi-Klerman and Dror Dotan (find it here)
A phonological input buffer for numbers
How hard is it to understand numbers form hearing?
It might seem like something easy, but the truth is—it’s not that simple. Even in a seemingly straightforward task such as dictation – hearing a a number (like “thirty-seven thousand and five”) and writing it down on paper (37,005). People, even those without any learning disorders, make many errors in this task – about one error per 20 numbers.
You won’t be surprised to hear that most errors occur with long numbers – those with many digits or words. Why are these numbers harder? There are several reasons, a central one being that long numbers are harder to remember. In the dictation task, there is a delay between the moment we hear the number and the moment we write it down. During that delay, we have to maintain the number in short-term memory—and short-term memory is a tricky business. It can store only a fairly small amount of information, and it’s highly prone to confusions.
Okay, so short-term memory is involved in number dictation. That by itself is fairly trivial. The more interesting question—and the one we examined in the present study—concerns the kind of short-term memory involved in number comprehension, how it works, and how it is integrated into the flow of number-processing.
To answer these questions, we examined NANI and BIMA—two individuals in their 30s who had a short-term memory deficit – low memory capacity. For example, whereas an average person can hear a sequence of about 7 digits and repeat them without errors, NANI and BIMA can manage 5 digits at most.
Unsurprisingly, NANI and BIMA made many errors in number dictation. We administered several other number-processing tasks, and we uncovered several interesting patterns that helped us diagnose NANI and BIMA’s precise impairment. For example:
- They made a lot of errors in tasks in which they heard the number. In contrast, they had no difficulty in tasks where the input was not auditory – e.g., reading numbers aloud, stating their birthdate, etc. We concluded that their impairment lies somewhere in the auditory input mechanisms.
- They made many mistakes in auditory-number tasks as long as those tasks imposed load on short-term memory. In contrast, they performed well in similar tasks with no short-term memory load. We concluded that the errors do not stem from hearing or comprehension deficits, but specifically from a short-term memory deficit.
These result patterns—along with some additional findings not detailed here (for the sake of brevity)—indicated that NANI and BIMA’s deficit lies in a short-term memory mechanism at the auditory-verbal input stage. We called this mechanism the phonological input buffer. As is typical in studies like this, there are two interesting conclusions: one, that the phonological input buffer exists (if it didn’t, it couldn’t be impaired); and two, that a selective impairment can occur in this buffer without impairing other number-processing mechanisms.
The next stage in the study was no less interesting: we tried to understand how exactly this buffer works—specifically, when and for what it is needed. We used a simple trick: if our participants have a phonological input buffer impairment, they should have errors in tasks that require this buffer, but they should perform well in tasks that don’t.
NANI and BIMA had many errors when they heard structured multi-digit numbers (“two thousand three hundred forty-five”), but considerably fewer errors when they heard sequences of isolated number words (“two, three, four, five”). This indicates that the phonological input buffer is a bottleneck for structured multi-digit numbers, but not for single digits. The conclusion is that one of the key roles of the phonological input buffer is to assist in parsing the structure of auditory multi-digit numbers.
To understand intuitively why this makes sense, consider the following simple example: if you hear the word “four” as part of a list of isolated number words, you don’t need to hold it in memory for long—you can immediately interpret it as 4. But if you hear that word as part of a multi-digit number, you have to wait for the next word to know whether it’s four, four hundred, or four thousand; and wait even longer to know whether it’s 400 or 400,000. While you’re waiting, you must retain the word “four” in short-term memory.
The above conclusions—along with several others not detailed here (e.g., from Zohar Cohen’s MA thesis)—allowed us to map out in detail the steps involved in verbal-auditory number comprehension:
Stages involved in auditory number comprehension: (1) Identification of each number word. (2) Decomposing each word into features. For example, the word “twenty” is decomposed into two features: a 1-9 value (in this example: 2) and a decimal class (here: tens). Decimal words like “thousand” or “hundred” are represented as a single feature each. (3) Phonological input buffer: a short-term memory component that holds the features. (4) Creating a representation that captures the multi-digit number’s full structure, and essentially reflects the role of each word relative to the other words. This representation is used by the output mechanisms to generate the corresponding digit sequence and write it down.