Studies challenge legitimacy of speed reading

[drkitten]

Ok but don't you so still need semantic understanding to resolve the anaphoric cases? I can't see how you could do this in general without checking for matching properties.

But what you're talking about goes beyond "semantic" understanding to pragmatic understanding -- to world knowledge. It's not enough, for example, to check the properties of entities already mentioned in the discourse; you need to be able to check the properties of entities already mentioned or inferrable from the previous discourse. So you're talking about the general inference-from-language problem, i.e. story understanding.

Here's a sample story, a modification of Schank's restaurant script....

"John was hungry, so he went to a restaurant.

It was so bad, he didn't eat any of it, and stormed out without paying."

What was so bad? What didn't he eat? What needed paying for?

As this example shows, anaphor resolution is generally considered to be a necessary component of story understanding. Therefore, you've just made the problem AI-complete. If we can solve story understanding, we can use that to solve anaphor resolution --- but we need to solve anaphor resolution in order to solve story understanding.

 

[Jekyll]

But what you're talking about goes beyond "semantic" understanding to pragmatic understanding -- to world knowledge. It's not enough, for example, to check the properties of entities already mentioned in the discourse; you need to be able to check the properties of entities already mentioned or inferrable from the previous discourse. So you're talking about the general inference-from-language problem, i.e. story understanding.

Here's a sample story, a modification of Schank's restaurant script....

"John was hungry, so he went to a restaurant.

It was so bad, he didn't eat any of it, and stormed out without paying."

What was so bad? What didn't he eat? What needed paying for?

As this example shows, anaphor resolution is generally considered to be a necessary component of story understanding. Therefore, you've just made the problem AI-complete. If we can solve story understanding, we can use that to solve anaphor resolution --- but we need to solve anaphor resolution in order to solve story understanding.

Gottcha. Thanks, for taking the time to explain.

 

[Wudang]

I spent a good part of my final year in 1982 studying this stuff and most of it is lost to history. DrK - can you recommend some reasonably recent texts?

 

[drkitten]

I spent a good part of my final year in 1982 studying this stuff and most of it is lost to history. DrK - can you recommend some reasonably recent texts?

Actually, the "Probabilistic Linguistics" text from which you clipped Jurafsky's chapter is about as definitive as you can get. And well-written, too. It's five years old or so, but that's better than twenty-five, yes?

 

[Dave1001]

What's the hard demonstrated current human limits for reading/skimming speeds and comprehension levels, for materials at various levels of familiarity, complexity, and technicality. Any studies parsing this type stuff out? Sort of like how we have hard demonstrated current human limits of lifespan, running speed for 100 meters, etc.

 

[drkitten]

What's the hard demonstrated current human limits for reading/skimming speeds and comprehension levels, for materials at various levels of familiarity, complexity, and technicality.

That is not an answerable question. It assumes a degree of precision in measuring "various levels of familiarity, complexity, and technicality" that doesn't exist. You might as well ask what the hard demonstrated human limits in how fast people can produce "fine art," controlling for genre and materials.

 

[Dave1001]

http://www.slate.com/id/74766/

When Carver set out to find the world's fastest readers and see for himself how quickly they read, he ended up with 16 test subjects—four groupings of four. The first group were the highest scorers on a test Carver administered to a large group of college students. The second group were demonstrably successful at jobs "that required a great deal of reading." (They included a New Yorker writer named Fred Shapiro and the editor of a medical journal.) The third group scored "exceptionally well" on standardized reading tests administered by a colleague. The fourth group had "reputations and supporting documentation for being speed readers."

At Carver's direction, the 16 brainiacs read passages from Reader's Digest condensed editions under controlled conditions: None of them could read faster than 600 words per minute and retain more than 75 percent of the information contained in the texts.

http://en.wikipedia.org/wiki/Speed_reading

According to some speed reading advocates, the World Championship Speed Reading Competition stresses reading comprehension as critical, and that the top contestants typically read around 1000 to 2000 words per minute with approximately 50% comprehension. The 10,000 WPM claimants have yet to reach this level.

Much controversy is raised over this point. This is mainly because a reading comprehension level of 50% is deemed a failure by normal reading teachers, the public at large, and reading experts (Carver 1992). Speed reading advocates claim that it is a great success and even state that it is a demonstration of good comprehension (Buzan 2000).

Informative. It seems to me demonstrated current human reading/skimming limits for unfamiliar, nontechnical material of median complexity is about 600 wpm with 75% comprehension, and about 2000 wpm with 50% comprehension, with the latter being a more weakly demonstrated claim. I'm curious what the comprehension level has been determined to be of the elite 10,000 wpm readers, more specifically than under 50% comprehension.

 

[Dave1001]

That is not an answerable question. It assumes a degree of precision in measuring "various levels of familiarity, complexity, and technicality" that doesn't exist. You might as well ask what the hard demonstrated human limits in how fast people can produce "fine art," controlling for genre and materials.

I'm also curious about to what degree of precision we can measure such things. Both in terms of reading speed and comprehension, and in terms of fine art production.

 

[Dave1001]

It seems a bit suspect to me that that the "World Championship Speed Reading Competition" apparently doesn't have a google-discoverable website, and the few websites that mention it all seem to be quoting the wikipedia entry on speedreading, or each other.

Is this a real thing? Or is Phil Glass editing wikipedia these days?

 

[jimecallaw]

referring to the original article whilst the human eye might only be able to focus sharply on a small area. reading is about pattern recognition. whilst you can't read individual letters that are blurred very easily, lots of letters together can form a recognisable pattern through any blurring. especially after you have been exposed to them millions of times, also surely the area of focus must be relevant to the distance of the material from the eyes. there was no mention of this in the article.

 

[Ben Tilly]

Informative. It seems to me demonstrated current human reading/skimming limits for unfamiliar, nontechnical material of median complexity is about 600 wpm with 75% comprehension, and about 2000 wpm with 50% comprehension, with the latter being a more weakly demonstrated claim. I'm curious what the comprehension level has been determined to be of the elite 10,000 wpm readers, more specifically than under 50% comprehension.

What do 75% comprehension and 50% comprehension actually mean?
Does it mean that you understood 75% of the plot? Does it mean that you remember 75% of the facts and figures that might be in the material? Does it mean that you'll notice 75% of the typos? Speaking of the material, what kind of material are we talking about? Some passages are going to be a lot slower than others. For instance I can read science fiction with adequate comprehension much faster than I will a chess book.

In any case I know that I read very quickly. I once measured my speed at reading a fantasy book at 900 wpm, with sufficient comprehension that you could start anywhere in the book, read a few sentences, and I could tell you what was going on in the book there, and easily find that spot in the book. (I already described this incident earlier in the thread.) My comprehension of the story was flawless. However I doubt I remembered 75% of the minor characters. I would be shocked if I noticed 50% of the typos.

In the sample reading tests posted earlier in this thread I read at about 600 wpm.

It wouldn't surprise me if I read unfamiliar textbooks at 200 wpm or less.

In reading technical math papers, 20 wmp might be a good goal.

What is my reading speed? What is my comprehension?

Cheers,
Ben

 

[Dave1001]

What do 75% comprehension and 50% comprehension actually mean?
Does it mean that you understood 75% of the plot? Does it mean that you remember 75% of the facts and figures that might be in the material? Does it mean that you'll notice 75% of the typos? Speaking of the material, what kind of material are we talking about? Some passages are going to be a lot slower than others. For instance I can read science fiction with adequate comprehension much faster than I will a chess book.

In any case I know that I read very quickly. I once measured my speed at reading a fantasy book at 900 wpm, with sufficient comprehension that you could start anywhere in the book, read a few sentences, and I could tell you what was going on in the book there, and easily find that spot in the book. (I already described this incident earlier in the thread.) My comprehension of the story was flawless. However I doubt I remembered 75% of the minor characters. I would be shocked if I noticed 50% of the typos.

In the sample reading tests posted earlier in this thread I read at about 600 wpm.

It wouldn't surprise me if I read unfamiliar textbooks at 200 wpm or less.

In reading technical math papers, 20 wmp might be a good goal.

What is my reading speed? What is my comprehension?

Cheers,
Ben

Follow the link I posted earlier. The 600 wpm at 75% comprehension hard demonstrated limit was reached using old reader's digests under controlled conditions, according to the slate article.

I think that's a reasonable approximation for fresh, non-technical material at median complexity, although I think a lot more experiments could and should be done fleshing things out and using a greater variety of materials and types of materials.

Still, it's the best current research I'm aware of looking at the limits of capability for top performing humans.

 

[drkitten]

Follow the link I posted earlier. The 600 wpm at 75% comprehension hard demonstrated limit was reached using old reader's digests under controlled conditions, according to the slate article.

Unfortunately, the link/article does not answer the question, precisely because the question illustrates a fundamental problem that no researcher has yet been able to answer, and arguably will never be able to answer.

75% of what? I'm a college professor -- I write tests for a living. Given a single passage that the students are required to read and learn from, I can write a test at almost any level of detail, from the completely gross ("what kind of animal was Moby Dick?") to the unbelievably specific. ("Where is Mr. Starbuck from?") and all points in between ("What was the name of the ship on which Ishmael sailed?") I can write a test on which my worst students get 80%, and one on which my best get 20% -- and they're all valid tests.

Part of the reason that companies like ETS need to do such extensive testing of proposed new test items is to determine what level of comprehension is necessary to answer that particular question.

I think that's a reasonable approximation for fresh, non-technical material at median complexity,

It's not. Not in the slightest. In the same way that "about thirty" isn't a reasonable approximation of speed. It's a dimensionless quantity that can be twisted to almost any interpretation.

 

[Dave1001]

Unfortunately, the link/article does not answer the question, precisely because the question illustrates a fundamental problem that no researcher has yet been able to answer, and arguably will never be able to answer.

75% of what? I'm a college professor -- I write tests for a living. Given a single passage that the students are required to read and learn from, I can write a test at almost any level of detail, from the completely gross ("what kind of animal was Moby Dick?") to the unbelievably specific. ("Where is Mr. Starbuck from?") and all points in between ("What was the name of the ship on which Ishmael sailed?") I can write a test on which my worst students get 80%, and one on which my best get 20% -- and they're all valid tests.

Part of the reason that companies like ETS need to do such extensive testing of proposed new test items is to determine what level of comprehension is necessary to answer that particular question.



It's not. Not in the slightest. In the same way that "about thirty" isn't a reasonable approximation of speed. It's a dimensionless quantity that can be twisted to almost any interpretation.

Right, and you can't prove your own existence to any of us without allowing for incalculably large margins of uncertainty. You make valid points, but in the end I think they're recursive to our lack of ability to discern anything in apparent reality. So I'll cede the general point to you and let's move on to discuss what exactly Professor Carver thinks he's measuring and how reasonably accurate or useful do we think he is with regard to those measurements, as well as the current work other scientists and researchers are doing in this field.

 

[drkitten]

So I'll cede the general point to you and let's move on to discuss what exactly Professor Carver thinks he's measuring.

Good. So discuss it. Or perhaps, dig up evidence that would raise the question to the level of discussable. Because there's absolutely nothing in either the Slate article or the wikipedia article that you cite that allows you to say anything about that. The information just isn't there. THe closest we have is a quotation that says "None of them could read faster than 600 words per minute and retain more than 75 percent of the information contained in the texts."

Show me some actual procedures from Dr. Carver, and I will be happy to analyze it for you in detail.

how reasonably accurate or useful do we think he is with regard to those measurements,

At the moment? I don't think he's accurate or useful at all, since there's literally no basis for me to make any other determination.

There's a reason that psychology papers provide a "materials and methods" section.

 

[Dave1001]

Good. So discuss it. Or perhaps, dig up evidence that would raise the question to the level of discussable. Because there's absolutely nothing in either the Slate article or the wikipedia article that you cite that allows you to say anything about that. The information just isn't there. THe closest we have is a quotation that says "None of them could read faster than 600 words per minute and retain more than 75 percent of the information contained in the texts."

Show me some actual procedures from Dr. Carver, and I will be happy to analyze it for you in detail.



At the moment? I don't think he's accurate or useful at all, since there's literally no basis for me to make any other determination.

There's a reason that psychology papers provide a "materials and methods" section.

Anybody able to look up Dr. Carver's publications on this topic or more recent ones? I'll try to look it up, but any other thread participant is welcome to post a link to the paper and similar research more quickly.

 

[Ben Tilly]

Right, and you can't prove your own existence to any of us without allowing for incalculably large margins of uncertainty. You make valid points, but in the end I think they're recursive to our lack of ability to discern anything in apparent reality. So I'll cede the general point to you and let's move on to discuss what exactly Professor Carver thinks he's measuring and how reasonably accurate or useful do we think he is with regard to those measurements, as well as the current work other scientists and researchers are doing in this field.

I'd respond to this in detail, but Dr. Kitten has said pretty much what I'd want to say. I'll just say that from my point of view I have no real idea what is being claimed until I can see an example passage and what kinds of questions the reader should be able to answer after reading it. I know I read quickly. I know that I can vary my reading speed depending on how much detail I need to retain. My reading speed naturally varies depending on context, material, font, lighting, how alert I am, etc, etc, etc.

The articles claim an upper limit of 600 wpm. I'm curious where I am relative to that limit. In one exercise I was half again as fast. According to http://gandalf.psych.umn.edu/~gellab/MNREAD/DEMO_RS/ I only comfortably go at 2/3 that speed. The truth is presumably somewhere in between, but that's a pretty giant range. So from my point of view, "600 wpm at 75% comprehension" is a virtually meaningless phrase.

(My guess is that if there is a top speed, I'm probably fairly close to it.)

Cheers,
Ben

 

[king catfish]

deleted by author

 

[Dave1001]

I'd respond to this in detail, but Dr. Kitten has said pretty much what I'd want to say. I'll just say that from my point of view I have no real idea what is being claimed until I can see an example passage and what kinds of questions the reader should be able to answer after reading it. I know I read quickly. I know that I can vary my reading speed depending on how much detail I need to retain. My reading speed naturally varies depending on context, material, font, lighting, how alert I am, etc, etc, etc.

The articles claim an upper limit of 600 wpm. I'm curious where I am relative to that limit. In one exercise I was half again as fast. According to http://gandalf.psych.umn.edu/~gellab/MNREAD/DEMO_RS/ I only comfortably go at 2/3 that speed. The truth is presumably somewhere in between, but that's a pretty giant range. So from my point of view, "600 wpm at 75% comprehension" is a virtually meaningless phrase.

(My guess is that if there is a top speed, I'm probably fairly close to it.)

Cheers,
Ben

You mentioned a list of factors that can affect reading speed and comprehension. That's not much different from the sort of factors that can affect results in lots of different experiments, and don't seem to particularly confound measuring reading speed and comprehension compared to many other less controversially measured experimental phenomena. I put lighting and font and perhaps alertness into that category.

There are also concerns about if it's even possible to measure things like passage complexity and familiarity, and if it's possible to mention % comprehension.

I think like all measures it's beyond currently known human capacity to create perfect measures for all these things, but it might be possible to create useful measures, at least in comparison to other measures of other phenomenon that are uncontroversially considered useful to most of the thread participants here.

I'll post more on my thoughts about how useful measures for these things can be constructed -in my next study break!

 

[Ben Tilly]

You mentioned a list of factors that can affect reading speed and comprehension. That's not much different from the sort of factors that can affect results in lots of different experiments, and don't seem to particularly confound measuring reading speed and comprehension compared to many other less controversially measured experimental phenomena. I put lighting and font and perhaps alertness into that category.

Fair enough. Let's just consider reading speed and comprehension under ideal circumstances. So I've got good lighting, a font I could read, and I'm fully awake and functional.

There are also concerns about if it's even possible to measure things like passage complexity and familiarity, and if it's possible to mention % comprehension.

Those are honestly my bigger concerns.

I think like all measures it's beyond currently known human capacity to create perfect measures for all these things, but it might be possible to create useful measures, at least in comparison to other measures of other phenomenon that are uncontroversially considered useful to most of the thread participants here.

I'll post more on my thoughts about how useful measures for these things can be constructed -in my next study break!

I'm sure that one can create useful measures of such things. I'm sure that such measures have been created. And I'm sure that studies have been done using such measures.

The problem is that when looking at the results of said studies, I don't know how to interpret what is said because I have no clue what the relevant measures are. Certainly a phrase like "75% comprehension" doesn't tell me the answer. I'd have a sense if I could see sample text and sample questions. But right now I have no clue.

Furthermore I believe that multiple such measures are out there. For instance after speed reading courses people really do read articles, and conduct tests. These tests are graded. Unlike the linked articles, I'm inclined to believe at least some of the stronger claims. After all people really did take tests, and they really did score something on them. But I think that the reading material and tests were far from equivalent, so the phrase "75% comprehension" or "50% comprehension" meant very different things.

Cheers,
Ben