Slashdot Mirror
The Science of Word Recognition
neile writes "I stumbled across a fascinating paper over at the Microsoft Typography site today that provides a really nice overview of the different theories on how humans read. If you thought we read by recognizing word shapes, think again! With the assistance of fancy eye-tracking cameras researchers have been able to devise several clever experiments to give us new insight into how reading works." We've linked to some of Larson's work previously.
Does anyone else think that merely analyzing how english is read is very closed minded? I'm pretty sure only a very small percentage of the world speaks and reads english.
I would love to see a study comparing how english is read to how chinese is read by native speakers. Very interesting i would gather.
A Fatal OE Exception has occurred, Sig will now reboot.
when are they going to repeat these experiments in let say China or Japan? I'm *very* interested in what would the conclusions be there. ...
For what i know abaout japanese, they don't use spaces between 'words'. A single kanji represents the whole word and their outline is always more or less square. So the whole bouma theory fails here, as he finds out.
I'm sure they could leard more interesting things in other writing sysmtems
Read it, it's interesting. It does get a wee bit weird when it's describing how you read as you read... a sort of super-conciousness about my eye movements. It's like when you become aware of your breathing or something and then have to conciously pay attention to it for a while to make sure it doesn't stop!
-- *~()____) This message will self-destruct in 5 seconds...
While some of the results here are interesting (but old), the fact that the entire study focuses on exactly 1 script and 1 language basically renders the conclusions worthless (as conclusions about cognition in general... I suppose they still have value as conclusions about English and the Latin script).
What has happened here is:
1 -- Observe people reading a given language/script
2 -- See how they make use of features of that particular language/script, such as tall letters, case, and the occurrence of 'skippable' words such as articles
3 -- Describe the way they use these local features, and call that a theory of reading in general.
I don't really understand how to apply a theory of reading based on word and letter shapes when there are so many people reading text in which:
--There are no letter boundaries, and/or
--There are no word boundaries, and/or
--Letters all have the same form factor
The experiments described would probably generalize very well to arabic and greek scripts, pretty well to cyrillic (no tall/short letters to speak of), badly to devanagari-type scripts, very badly to Chinese and Japanese, and not at all to hieroglyphics (though I agree that there may never have been a reader of hieroglyphics who was fluent by modern standards).
To pretend that these experiments apply to humanity in general rather than the author's own language/script choice is silly. It's an interesting article and I'm glad the research was done but unfortunately a certain failure to 'get' the multilingual nature of humanity, which I don't really expect to find in MS work, is in evidence here.
Whence? Hence. Whither? Thither.
It's not bloody funny! The parent, in a true Slashdot style, didn't even get what the subject of the paper was!
The question pondered is whether _experienced_ reader reads by, in the first place, recognising the word shape, or by recognising the letters.
P.S. yes I know that psychologists are great for stating the obvious, but not here...
P.P.S. to parent: read the article properly, I'm sure you'll find a nice funny case of stating the obvious.
The internal representations for these models convert the letter information to phonemic information, which is seen as a mandatory step for word recognition. It is well known that words that have a consistent spelling to sound correspondence such as mint, tint, and hint are recognized faster than words that have an inconsistent spelling to sound correspondence such as pint
I can not believe this is in a serious paper. Mandatory? Please. What about people born deaf? Are they all unable to read?
I'm wondering about the competence level of the readers used in the various tests. People have, in my experience, a wide range of reading ability levels ranging from those who still have to "spell out" many words letter by letter up to speed readers who can read entire phrases (or even whole sentences) as easily as most people read single words. If we divide them into three groups (phonetic readers, whole word readers, and "cognitive chunkers"), would these results be consistent from group to group?
:)
Learning to read, like learning higher math, is a process of internalizing certain reflexes. Most people alive today do not understand calculus. Most will also never learn to read much faster than they can speak aloud.
Ultimately, 80% of the people voting in the next presidential election are of average or lower intelligence.
The rest of us are Slashdot members.
TLR
A man no more knows his destiny than a tea leaf knows the history of the East India Company
Actuallythesplittingintowordsisnotnecessarytounder standwhatiswritteniftheorderoflettersiscorrect.Thi s"proves"thatyouarereadingbytheletter,notbytheword .(relyingonslashcodetoinsertameaninglessspaceevery nowandthen:-))
The Tao of math: The numbers you can count are not the real numbers.
The relationship is probably a lot more complicated than "thought comes before language". I suspect they are both highly dependent on each other.
For instance, it is clear that many non-verbal animals are able to think, in at least some limited fashion. Larger rodents, for instance, are able to build models of their world and solve simple problems (not limited to learning by trial and error). It is exactly this kind of modelling that concepts like one object being inside another stem from -- spatial reasoning is almost certainly the most deeply embedded and instinctive part of thought, and therefore the least likely to depend on language.
However, the ability to form complex theories and plans may or may not be entirely dependent on our ability to express them. Could primitive man, for instance, have looked at the weather and decided whether it would be best to go hunting today or finish building that shelter first, if he didn't have words for 'rain', 'shelter', and 'later'? The question might be too complex to approach without some kind of symbolism that can be internalised. Or it might not. Its very hard to tell.
I am sure that we've seen this e-mail floating around. Doesn't it seem like we read in shapes?
I cdnuolt blveiee taht I cluod aulaclty uesdnatnrd waht I was rdgnieg The phaonmneal pweor of the hmuan mnid Aoccdrnig to a rscheearch at Cmabrigde Uinervtisy, it deosn't mttaer inwaht oredr the ltteers in a wrod are, the olny iprmoatnt tihng is taht the frist and lsat ltteer be in the rghit pclae. The rset can be a taotl mses and you can sitll raed it wouthit a porbelm. Tihs is bcuseae the huamn mnid deos not raed ervey lteter by istlef, but the wrod as a wlohe. Amzanig huh? yaeh and I awlyas thought slpeling was ipmorantt!
I notice MS Research doing lots of basic research that has never been productized. Its rare to see corporations being so liberal with their resources. Even Google's very imaginative projects seem to be directed towards a commercial goal.
This suggests an interesting contradiction in MS product strategy. MS has a long history of "clone and conquer", e.g. Excel copies off VisiCalc and Lotus 123. Just this week MS cloned Apple iTunes. Yet MS Research is conducting some very interesting basic research. Go figure!
"Evidence from the last 20 years of work in cognitive psychology indicates that we use the letters within a word to recognize a word."
Very strange because if y_u r____r we d__'t n__d a_l those l_____s.
I'm no linguist (elec eng w/ neural net studies), but I would argue that the ability to perceive concatenated sentences like that is a function of the ability of the brain/eye to focus on a particular range and filter out "distractions" (letters to the left and right). Padding our words with spaces helps the brain to quicker define the focus boundaries, after which we can process the text range for meaning...
I imagine the brain's focus as little perception boxes, scanning up and down the concatenated sentence until enough symbols are aligned to fire a recognition signal... As I read your post above, I find my eyes darting about a little more, actually darting to the center of the "word" once recognition is made.
runonsentencewithlowercase -- here's your letter by letter scan "mode"
runonsentencewithcoloring -- slightly easier to define word boundaries by color
runonSENTENCEwithuppercase -- it's easier to locate the word SENTENCE because we perceive a boundary beween small letters and upper letters.
runo nsente ncewit hbads pacing -- pain in the ass, but we still comprehend
run on sentence with lowercase -- whitespace speeds compehension.
One problem with deciding "word shape" vs. "letters" as the method a reader uses to recognize a word comes from treatment of the reader as "atomic". I am a proficient reader. When I read a word, whether written by another, or by myself (as I type), I have multiple subcurrents of consciousness. A typo in a word might leave me with recognition of the word, and a sense that "something's wrong", simultaneously - it sometimes takes me several seconds to detect the typo, especially if it's one I often make myself. Likewise, some spelling mistakes derive from the difference my spoken accent makes with the written conjugated spelling, most often in the case of syllables separated by an "e" that is pronounced as a "schwa", easily confused with some pronunciations of "i" or "a", and sometimes "y".
Reading words silently, I sometimes notice an inner chorus pronouncing the words, with one or two discordant notes, even from poorly organized structure or unparseable punctuation. Deciding how people recognize words must also account for how people's minds are organized. The myth of the "undivided self" gets in the way of understanding not only how complex we are "under the hood", where media is digested, but denies credit to our grand integrator, which juggles these partial selves into one face with which to confront the world. As machine intelligence benefits from multiple simultaneous processing, why should they have all the fun? As we mimic our own minds in computer simulations, why should we have all the fun?
--
make install -not war