The Man Who Could Read Only Numbers
We were not very surprised that Mr Harvey - not his real name -
could still calculate accurately, even though brain damage had
so deprived him of speech that he could not name even the commonest
objects. Three patients had already been reported whose maths had
survived brain damage, but whose language had not. We were mildly
surprised by his ability to read aloud the numerals 1, 2, 3, and
even four-digit numbers such as 7495 without error. Given his
enormous difficulties with speech, this was unusual but not
Was it relevant that Mr Harvey, before his retirement, had been
a banker and his hobby was gambling? This would have given his
mental mathematical muscles more exercise than most people's,
and we knew that with brains, as with real muscles, you either
use it or lose it. Practice rewires the brain. It was likely,
therefore, that the mathematical part of his brain had recruited
more brain cells and strengthened the connections between them.
So if he started with more tightly interconnected brain cells
devoted to mathematics, a reasonable level of number skill
might well have survived three years of an insidious disease
eating away at his brain.
What astonished us, though, was his word reading. He was quite
unable to read even the simplest and commonest words - for, tree,
then, take, give, you - but he had no trouble at all reading number
words. Though he couldn't read you, he could read two and three,
ten, eight, and five, not to mention thousand and million.
This was quite unprecedented in the annals of neuropsychology.
What could be the explanation of such a weird anomaly, and could
it tell us anything about the nature of normal reading, or indeed
about normal mathematics?
Patients with these very specific and unusual patterns of preserved
and defective abilities are like rare archaeological specimens. We
can't go into a laboratory and make them. We just have to find them,
and to do this, like the good archaeologist, we need to know what to
look for. And, perhaps more importantly, we need to know what we've
got when we find it.
This takes both acute observation, and a wide knowledge of theory.
"Theory" in this case means having a view about how different
mental abilities are organised in the brain. In practice, this
translates into having an idea of what can go wrong by itself
and what cannot.
These specific disturbances of cognitive function have been
responsible for major advances in our understanding of the mind.
We now know, for example, that amnesiacs - the "lost weekend"
type of amnesiacs - do not forget everything. Even the most
severe cases still remember how to speak, to read, that Paris
is the capital of France, that kangaroos are not indigenous to
Southend-on-Sea, and how to calculate.
The part of memory that is responsible for this preserved
information is called "semantic memory". Remembering who you
are, what you did yesterday or last year - "autobiographical
memory" - is a function of a brain structure called the
hippocampus, while semantic memory uses other brain regions.
In the thriller The Code To Zero by Ken Follett, the hero
tries to recover his life by using his semantic memory.
Finding that he understood the technicalities of a rocket
launch told him that he must have been a rocket scientist.
I won't ruin your enjoyment of the ingenious plot by revealing
more. There is also "procedural memory" for skills such as
driving or skiing.
A few years ago, a distinguished Italian neurologist and
expert on memory disorders, Dr S, was skiing too fast, as
usual, and took a tumble. When the rest of his party caught
up with him, he stared at them strangely. He was surprised to
find his wife and his best friend suddenly looking very old.
And he failed to recognise some of his younger colleagues.
The neurologists in the party did a piste-side examination,
and diagnosed autobiographical memory loss, probably of the
past 25 years or so. This is why his wife and friend looked
old, because his memory of their faces was 25 years old, and
of course he had come to know his younger colleagues within
the 25-year blank period. However, he was still able to ski
back to the bottom station, where he declared, "See, my
procedural memory is still intact!"
He was taken to the local hospital where he asked for a brain
scan, a procedure that he would have learned about during the
blank quarter of a century. A scan was carried out, and Dr S
looked at the results and made his own diagnosis. Seeing his
name and age on the scan, he noted, "There, the brain looks
in very good shape for my age; no signs of atrophy. I must
have transient global amnesia."
This was accurate, and showed that the semantic memories
created in this period were still accessible. Global amnesia
meant that not only was he unable to remember the past, but
he could not create memories of new events. Fifteen minutes
later he said, "Shouldn't I be having a brain scan?" This
condition was, thankfully, transient, and after 20 hours
he recovered completely. My impression is that he now skis
a little more carefully.
Mr Harvey is the exact opposite of Dr S. His autobiographical
memories are reasonably well preserved. We asked him to read
the word "theatre". He couldn't pronounce the word, but he
did say, "I delved into that every year in Oxford from 1950
to 1960". His wife confirmed that, indeed, he had been a
keen amateur actor and director during this time. His problem
lay in his semantic memory.
He had very few words left - "delved" came up very often in
his speech. In a standard clinical test, he failed to name
a single common object, such as table, chair, clock, or glove.
And he couldn't point to the right picture when you said these
words. He was unable to classify pictures of animals into those
living in England and those living outside England. The
combination of good autobiographical memory and poor semantic
memory caused by a neurodegenerative disease is called
For two years Mr Harvey had been in the care of one of the world's
leading experts on semantic dementia and other memory disorders,
Professor Michael Kopelman of St Thomas's Hospital in London.
Kopelman and I were interested in how numerical knowledge and
calculation skills relate to other aspects of memory, and we
had noticed that in reports of other semantic-dementia patients,
these abilities could be preserved even when much else had been
lost. On our team, which is funded by the Wellcome Trust, was an
enthusiastic Italian neuropsychology graduate, Marinella
Cappelletti, who was to do most of the testing of Mr Harvey,
even going to his home an hour and a half away when we needed
to collect more data.
We devised a battery of numerical tests and systematically
tested Mr Harvey more or less once a month for over a year.
Mr Harvey always arrived at the clinic immaculately dressed,
usually in a pinstripe suit and tie; and he was always smiling
and affable, but it became apparent that his condition was
getting worse. He was losing his vocabulary, and was
performing more and more poorly on our tests of semantic
memory. However, for most of this period, his calculation
remained at a very high level. His arithmetic, both oral
and written, was usually without error. He could do long
multiplication flawlessly. He could read long numbers and
write them to dictation with no difficulty.
We had now demonstrated conclusively, for the first time,
that numerical skills could be preserved when most of the
rest of semantic memory, including words, were lost. This
is important because it refutes the idea, popular with both
psychologists and laymen, that adult calculation is carried
out in language and that arithmetical facts are stored just
as verbal formulae.
However, we still did not know why he was able to read only
number words, and, as we soon discovered, these were also
the only words he could write. It wasn't just that number
words were the only words he could remember. After all, he
chatted with us, albeit hesitantly, about things other than
numbers. So, we asked him to read words that he had used
correctly in conversations, to be sure he knew them. He
was slightly better with these words than assorted common
words; but since he was reading number words 100% correctly
every time, this could only be a tiny part of the explanation.
Memory research wasn't going to provide the answer.
We turned to research on reading. In 1973, two Oxford
neuropsychologists, John Marshall and Freda Newcombe,
had published a study of three patients whose reading
had been affected by brain damage. In this study, they
did something that had never been done before. Previously,
reading disorders, called "alexia", had been classified
into two types: alexia with agraphia (writing disorders)
and alexia without agraphia.
Marshall and Newcombe took the revolutionary step of not
just counting the words correctly written but analysing
the kinds of errors that the patients made when they
attempted to read a word. They found that one of their
patients, a skilled reader before his brain damage,
frequently mispronounced one or two letters in the
word in rather the same the way that a child who didn't
recognise the word might. For example, he read "insect"
as "insist", softening the C; and "listen" as "Liston,
the boxer", sounding out the T, which, unusually, is silent.
Two other patients made characteristic errors that had
never before been noticed, reading "bush" as "tree",
and "ill" as "sick". Obviously, they couldn't make
these errors simply by sounding out the letters in
a childish way. They must in some way have read the
words correctly, and retrieved their meanings.
Somehow in the process of trying to say the words
they had come up with a word that was similar in
meaning, but not in sound, to the target.
Pondering on this puzzle, Marshall and Newcombe
came up with a theory that became the standard
in reading research (though nowadays it has a
few more bells and whistles). They argued that
when we see a word, we use two "reading routes"
automatically and simultaneously: we try to sound
it out letter by letter, a kind of mental phonics
process, and at the same time we try to retrieve
the meaning of the whole letter string without
bothering about its pronunciation. The patient
who said "Liston, the boxer" was using the
letter-to-sound route only, presumably because
the other route had been damaged. The other two
patients were using only the reading-via-meaning
route, again presumably because the other route
had been damaged.
Of course, bush doesn't mean the same as tree,
so why did these errors arise? It seems that we
need both routes. We need to be able to recognise
words as wholes, since many have irregular spellings,
or pronunciations that depend on their context -
such as lead and wind. And the letter-to-sound route
is needed in case we have never seen the word before.
With two routes operating together, one can check the
output of the other. This may be particularly important
given that we read so fast, far faster than we normally
hear words - 200 to 300 words per minute, three to five
words per second. With one route dysfunctional, errors
will arise, with the kind of error depending on which
route is affected.
This idea gave us the crucial clue to Mr Harvey's
reading. Suppose he wasn't able to use the letter-to-sound
route at all, then he would have to rely exclusively on
reading by meaning. We tested this by seeing if he could
read letter strings that he had never seen before, such as
"zind" and "yead". He couldn't. We also looked to see if he
made "regularisation" errors, such as pronouncing the T in
listen, or reading pint to rhyme with hint. He didn't.
Normal readers using both routes are more accurate reading
regularly spelled words than irregular words because the
two routes interact. Mr Harvey showed no advantage for
regular words because the letter-to-sound route wasn't working.
The words he could and could not read were defined solely
by their meaning. If they meant a number (or an ordinal -
first, second, eighteenth) he could read them. If not,
he was unable to. This reinforced our view that he was
reading exclusively by meaning.
We knew from our other tests that his semantic dementia
had severely affected his knowledge of the meaning even
of common words for everyday objects. But it was clear
that he understood the meaning of number words, since
he was still able to carry out flawlessly tasks that
depended on knowing these meanings. So he could say
which of two numbers was larger, even up to four-digit
numbers. And, of course, he could calculate. So it was
not just that he knew that that "seven" denoted a number;
he knew precisely which numerical value it had - not six
and not eight, and that, added to five, it makes 12. In
other categories of knowledge, even when he wasn't
completely at a loss, his grasp of concepts and meanings
was much vaguer, as is usual in these cases.
As his disease progressed, he found that he was unable to
recall some of the facts he had learned in school, such as
multiplication tables. But he compensated well enough by using
successive addition to solve the problems we gave him. This was,
in a way, even more impressive than being able to retrieve the
product of, say, eight times seven, which could be mere rote
memory, as it showed that he still understood both number
meanings and the concepts of arithmetic.
Now Mr Harvey's case is not just a medical curiosity.
Nor is it just an example of one man's struggle to overcome
an unpredictable and crippling disease, or a pointer to how
far neuroscience has progressed by 2001. It solves two
problems (although it raises more questions). First, we
have confirmed what theorists had previously only speculated
about, that reading via meaning is sufficient for accurate
pronunciation, since Mr Harvey can read number words with
This may give us a clue as to how to help dyslexic children.
One of their main problems is in learning the sound of each
letter. This is particularly hard in English, where letters
can have very different sounds depending on the word they
are in. Just think of G in the following words: tug, tough,
A dyslexic young woman we studied some years ago suffered
six years of trying to learn to read through phonics, and
was almost classified as "educationally subnormal" when her
mother sent her to a different school which taught reading
by the whole-word, look-and-say method. This worked
excellently for her. We found her reading to be efficient,
but unusual. She read most words so well that on standard
tests she would not be dyslexic. However, she was quite
unable to read new words. She always had to ask someone
to read them to her and then she would try to remember how
I think many dyslexics would benefit from more emphasis
on a whole-word, meaning-based, approach to reading, and
less on phonics. Our study of Mr Harvey shows that this
single route can work effectively (though not as effectively
as the two routes together, of course).
The second result was this: many years ago the great British
neuropsychologist, Elizabeth Warrington, discovered that
neurological patients could have selective impairments of
single categories of knowledge within semantic memory -
they might still know about living things and foods, but
had lost information about furniture, for example. In these
studies she did not specifically consider numerical knowledge.
The case of Mr Harvey, along with converse cases where language
is spared but numerical abilities are severely damaged, shows
that numerical knowledge is separate in the brain from our
knowledge of language and the rest of semantic memory. This
is somewhat counterintuitive given that most of what we know
about numbers is learned through language, and we can all
remember hours of tedium reciting our tables in a singsong
voice. But however we learn about numbers, they end up in a
region of the brain known as the parietal lobes. Knowledge
of language is mostly in the dominant frontal lobe
(Broca's area), and Wernicke's area in the temporal lobe,
near where most of the rest of semantic memory is located.
Brain scans showed that disease was ravaging Mr Harvey's
temporal lobes, but had left the parietal lobes intact,
showing that different categories of knowledge may be in
quite separate lobes of the brain. Why number knowledge
should be in the parietal lobe and not in the temporal
lobes is a question that we are currently pursuing.
Our study left us with one further puzzle. Mrs Harvey had
told us that her husband, an inveterate gambler on the horses,
now only bet on dogs. Why, we wondered, had semantic dementia
driven him to the dogs? This is where my misspent youth came
in useful. I remembered that one bets on horses by giving the
horse's name. "Ten pounds to win on Galileo, please." If you
can't read the name, this is going to be difficult. But since
all dog races have six starters, it's normal to say,
"Ten pounds to win on number six." This Mr Harvey can do.
I hope he has better luck with the dogs.
This is an extract from Frontiers 01: Science and Technology,
2001-02, edited by Tim Radford, Guardian science editor,
and published by Atlantic Books.
To order your copy priced £10.99 (inc free p&p) please call
0870 727 4155.