ABC Hints&Tips Overviews Reference Glossary Support
Contents : Overviews
Incremental reading

SuperMemo 99 made the first step towards efficient reading of electronic articles by introducing reading lists and reading tools. Reading lists are prioritized lists of articles to read, and reading tools make it possible to accelerate the conversion of knowledge extracted from articles to knowledge that can effectively be used in the process of learning.

SuperMemo 2000 greatly increases the efficiency of reading by introducing the concept of incremental reading. The essence of incremental reading is in simultaneous reading of many articles at the same time, i.e. reading individual articles in increments.

At the moment of writing this documentation (October 2000), SuperMemo 2000 seems to be the only applications that makes it possible and encourages incremental reading.

Incremental reading may become a breakthrough in the efficiency of knowledge acquisition. It may be as important for the future of SuperMemo and learning as the original algorithm for spacing repetitions

Advantages of incremental reading:

  1. Massive learning: possibility of studying a huge number of subjects at the same time. In traditional reading, one book or academic subject might need to be completed before studying another was recommended. With incremental reading, there is virtually no limit on how many articles you can study at the same time. Only the availability of time and your memory capacity will keep massive learning in check
  2. Prioritization: improved flow of knowledge as a result of more accurate prioritization of material. For example, reading lists in SuperMemo 99 required assigning Value to each article. Estimations of Value could be highly inaccurate. An interesting material that would receive a low Value might indefinitely wait in the queue for processing
  3. Creativity: the association-creativity bonus. The essence of creativity is in association of remote ideas. By studying multiple subjects in stochastically unpredictable sequences, incremental reading dramatically affects association of ideas and immensely improves your creativity
  4. Understanding: the slot-in factor. One of the limiting factors in acquiring new knowledge is the barrier of understanding. All written materials, depending on the reader's knowledge, pose a degree of difficulty in accurately interpreting their contents. This is particularly visible in highly specialist scientific papers that use a sophisticated symbol-rich language. A symbol-rich language is a language that gains conciseness by the use of highly specialist vocabulary and notational conventions. For an average reader, symbol-rich language may exponentially raise the bar of lexical competence (i.e. knowledge of vocabulary required to gain understanding). Incremental reading makes it possible to delay the processing of those articles, paragraphs or sentences that require prior knowledge of concepts that are not known at the moment of reading. The processing will take place only then when the new information slots in comfortably in the fabric of the reader's newly acquired knowledge
  5. Order. The web is a goldmine of information. However, rarely do we find step-by-step articles that provide all information and entirely satisfy our needs. In scientific research, acquiring engineering knowledge, studying a narrow topic of interest, etc. we are constantly faced with a chaos of disparate and often contradictory statements. Incremental reading makes it possible to resolve contradictions and build harmonious models of knowledge on the basis of the information chaos drawn from the Internet. Incremental reading stochastically juxtaposes pieces of information coming from various sources and uses the associative qualities of human memory to emphasize and resolve contradiction
  6. Stresslessness. The information era tends to overwhelm us with the amount of information we feel compelled to process. Incremental reading does not require all-or-nothing choices on articles to read. All-or-nothing choices are stressful! Can I afford to skip this article? For months I haven't found time to read this article! etc. SuperMemo helps you prioritize and skip articles partially or transparently. Oftentimes reading 3% of an article may provide 50% of its reading value. Reading of other articles may be delayed transparently, i.e. not by stressful procrastination but by sheer competition with other pieces of information

Important! Only SuperMemo makes it possible to implement incremental reading. Incremental reading requires continuous retention of knowledge. Depending on the volume of knowledge flow in the program, the interval between reading individual portions of the same article may extend from days to months. Repetition spacing is the foundation of incremental reading which is based on stable memory traces in-between reading bursts

See also: incremental reading from user's perspective by Len Budney 

Incremental reading algorithm

Incremental reading requires a collection of skills that you will perfect only with passing time and growing experience. Additionally, incremental reading in SuperMemo 2000 is a work in progress and the optimum use of individual tools may not yet have been entirely defined. Consequently, you will meet a number of choices and optimization issues that may be overwhelming in the beginning. This overview will help you handle the most basic skills and help you make a start with incremental reading.

The following assumptions will help us simplify the incremental reading algorithm:

  1. a reading list is a queue of articles sorted for priority. Articles in the reading list are waiting for their reading turn. We will neglect the existence of the reading list and assume that all articles directly enter the process of review and repetition (skills related to reading lists are described in detail in: Reading the Internet)
  2. a pending queue is a queue of articles, extracts and questions waiting for the introduction into the learning process on the first-come-first-served basis. We will neglect the existence of the pending queue and assume that no articles, extracts or items are kept waiting for entry into the learning process (skills related to managing the flow of knowledge between selected knowledge pools are described in: Flow of knowledge in SuperMemo)

Most important skills/steps in incremental reading:

  1. importing articles to SuperMemo
  2. reading articles and decomposing articles into manageable pieces
  3. converting most important pieces of knowledge to question-answer type material
  4. repetitions of the material (to ensure good recall of the learned knowledge)
  5. handling the unavoidable knowledge overflow

Skill 1: Importing articles

To import an important article to SuperMemo, follow these steps:

  1. Select the imported text in your web browser and copy the selection to the clipboard (e.g. with Ctrl+C)
  2. Switch to SuperMemo (e.g. with Alt+Tab)
  3. In SuperMemo, press Ctrl+Alt+N (this is equivalent to Edit : Add to category : Article on the main menu). SuperMemo will create a new element, apply the reading template (i.e. a template with one scaleable rich text component), and paste the article
  4. Press Ctrl+R or click Remember to introduce the article into the learning process (this step follows from our early assumptions on ignoring the reading list and the pending queue, and can otherwise be considered optional) . Alternatively, instead of Remember, use Ctrl+J to specify the first review interval (e.g. one day for high priority material or 30 days for low priority material)

If you would like to import articles with pictures, you will need to paste pictures separately. Use Copy on the picture menu in the browser and then press Shift+Ins in SuperMemo to paste the picture (you will need to specify the name under which the picture will be stored in the image registry). As you should not keep more than 2-3 pictures per element in SuperMemo, you should paste pictures from multi-picture articles only to relevant extracts of a main article. In such a case, it is recommended that you review the article in SuperMemo, extract fragments associated with pictures to separate articles (e.g. with Reading : Remember extract on the article pop-up menu), and paste pictures to the relevant extracts only.

Instead of articles, you can also import longer pieces of e-mail and respond incrementally (simultaneously with incremental reading). For more see: E-mail in SuperMemo

Skill 2: Reading articles

You could precede reading articles with conveniently locating the reading toolbar on your screen. Choose Window : Toolbars : Read, place the toolbar in a convenient place on the screen and press Ctrl+Shift+F5 (to save the chosen layout as your default layout). You will need to enter middle or professional levels for Windows menu to be accessible (see: File : Level).

Reading toolbar:

This is a simplified algorithm for reading articles:

  1. Click the article (to make sure you enter the editing mode in which you can modify text, select fragments, etc.)
  2. Start reading the article from the top
  3. If you encounter an interesting fragment, select the fragment with the mouse and choose Reading : Remember extract on the menu available with a right-click (or click the green T icon on the reading toolbar). This operation will introduce the extract into the learning process
  4. If you would like to adjust the first interval after which the extract review will take place, choose Read : Schedule extract instead of Remember extract (checked T icon on the reading toolbar)
  5. If you read a fragment that does not seem important, select it and choose Reading : Ignore or click the stop-sign icon. The fragment will be marked with ignore font
  6. If the selected fragment does not include all the important reading context, you will need to add this context manually. For example, if you are learning history, you may extract the following fragment from an article about Lincoln:On Sept. 22, 1862, President Lincoln issued the Emancipation Proclamation, one of the most important messages in the history of the world. He signed it Jan. 1, 1863. If you would like to extract the fragment related to signing the Emancipation Proclamation, you will need to change He to Lincoln and it to Emancipation Proclamation so that your fragments retains all contextual clues: Lincoln signed the Emancipation Proclamation on Jan. 1, 1863
  7. Once you decide to stop reading the article before its end, mark the last processed fragment as the read-point (e.g. with Ctrl+F7). Next time you come back to this same article, SuperMemo will highlight your read-point and you will be able to resume reading from the point you last stopped reading the article
  8. If necessary, provide a meaningful title to your article. Occasionally, add some context clues at the place of the current read-point. Those actions may be needed to minimize time needed to resume reading the article without losing the context (context recovery overhead)
  9. After marking the read-point, choose Learn or Next repetition to proceed with review of other articles
  10. In incremental reading, interrupted reading is a rule, not an exception! (see advantages above)
    With a dose of practice, you will quickly get accustomed to this not-so-natural state of affairs and learn to appreciate the power of incremental approach. Use the following criteria to decide when to stop reading the article:
  11. Once you complete reading the article, press Ctrl+D (or Dismiss) to dismiss the article and store it in the archive. SuperMemo will warn you once all extracts and clozes of the article are processed and moved to various branches. Once this happens, you can decide to delete the source article to save space
  12. Occasionally, you may want to read the same article more than once. In such cases, leave a note at the bottom of the article (e.g. Read again, Oct 21, 2000). Most often this will happen with long articles which you want to quickly review for most valuable pieces of information before your read them carefully word for word. Once you arrive to the bottom of the article in incremental quick review, you can delete the read-again note and start reading incrementally from the top

Skill 3: Extracting fragments, questions and answers

In the course of reading, we often mark important paragraphs. In SuperMemo, those paragraphs should be extracted as separate elements that will later be used to refresh your memory. Each extracted paragraph or section becomes a mini-article that will be subject to the same reading algorithm as discussed above. Extract your fragments and single sentences with Remember extract. Remember to add necessary context clues to make sure the extracted fragment does not become meaningless with time. If you cannot recall the necessary context, use the reference source link button on the navigation toolbar to jump to the article from which the extract had been produced

SuperMemo will demonstrate to you that extracting important fragments and reviewing them at later time will have an excellent impact on your ability to benefit from the reading material at later times. However, it will also show that once the review intervals grow beyond 200-300 days, passive review will often become insufficient. At that time you will need to use Remember cloze (blue Z icon on the reading toolbar). This option will convert single sentences into question-and-answer items.

For example, if you have extracted the following fragment from your reading about the history of the Internet:

The Internet was started in 1969 under a contract let by the Advanced Research Projects Agency (ARPA) which connected four major computers at universities in the southwestern US (UCLA, Stanford Research Institute, UCSB, and the University of Utah)

you may discover than when review intervals become long enough, you may not actually be able to recall the name of the ARPA agency or even forget the year in which the Internet started. You can then select an important keyword, e.g. 1969, and use Remember cloze to arrive to the following item:

Question: The Internet was started in [...] under a contract let by the Advanced Research Projects Agency (ARPA) which connected four major computers at universities in the southwestern US (UCLA, Stanford Research Institute, UCSB, and the University of Utah)
Answer: 1969

In the course of learning, you will yet need to polish the above item by manual editing:

Question: The Internet was started in [...](year) under a contract let by the ARPA agency
Answer: 1969

The editing added the following benefits to the above item:

  1. emphasizing the fact that the question is about the year in which the Internet began
  2. removing superfluous information. In the course of learning, you will answer the question and never focus on which universities were originally connected by the early Internet. If you believe this information is important, you will use the original extract to produce more cloze items that will focus solely on the universities in question

Important! Your work on extracting fragments, producing cloze deletions and editing them should also be incremental. In each review, do only as much work on the learning material as it is necessary! Extracting and editing in intervals adds additional benefit to learning and is more time-efficient. Often your priorities change as you proceed with learning; hence, the incremental approach should not only refer to reading but also to the follow-up processing

Skill 4: Repetitions and review

SuperMemo is based on repetition. You make repetitions of the learned material in order to ensure that your knowledge retention reaches the desired level (usu. 95-98%).

In SuperMemo 2000, your incrementally processed articles will also be subject to repetitions. We will often use the more intuitive term review in reference to incrementally processed material; after all, when you resume reading an article after a certain interval of time, you are not actually repeating anything. You are simply delving into new sections of the same material and extracting newly acquired wisdom into separate elements (Remember extract).

The algorithms used to make (1) standard repetitions of question-and-answer material and (2) reviewing reading material are similar. Most importantly, all repetition and review is made in increasing intervals. In incremental reading, you will constantly see inflow of new material to your collection. Unprocessed material will need to compete with the newly imported material. Increasing review intervals make sure that your old material fades into lower priority if it is not processed quickly. Naturally, the speed of processing will depend on the availability of your time and the value of the material itself. Articles that are boring, badly written, less critical for your growth or work, will receive smaller portions of your attention and may go into long review intervals before you even manage to pass a fraction of the text. That is an inevitable side effect of a voluminous flow of new information into your collection and your brain. However, intervals and priorities can easily be adjusted. If the priorities change, you can modify the way you process important articles. Upon next review you can read the whole article, revert it to a short-interval review or even use search (Ctrl+F) to locate more articles on the subject you feel you have neglected.

The algorithm for repeating questions-and-answer (e.g. cloze deletions) is quite complex and you do not have much influence on the timing of repetitions (see: Algorithm SM-8). This stems from the need to keep a high level of knowledge retention, which can be compromised by manual intervention.

However, the algorithm for determining inter-review intervals in incremental reading is much simpler and is entirely under your control. Each article receives a number called A-Factor that determines how much intervals increase between subsequent reviews (the name A-Factor is used here for orthogonality; however, A-Factors here correspond to the extinct concept of E-Factor known from earlier versions of SuperMemo and should not be confused with A-Factors used by items in Algorithm SM-8). For example, if A-Factor=2, review intervals will increase twice after each review. A-Factors are determined heuristically on the basis of the length of the text. Long texts will receive low A-Factors (e.g. 1.1), while short extracts will receive higher A-Factors (e.g. 1.8). You can change the value of A-Factor associated with a given article by choosing Ctrl+Shift+P (A-Factors associated with items cannot be changed by the user).

You can also control the review timing by manually adjusting inter-review intervals. Use Ctrl+J to determine the date of the next review. If you want to reduce the interval with Ctrl+J at the moment of review, you will often need to close the current review (e.g. with Esc) to make sure that the interval before the next review will be equal to the current inter-review interval (if you do not close the review, the current interval will simply be incremented by the period that will pass before the next review).

Skill 5: Handling large volumes of knowledge

With incremental reading, your work with SuperMemo will freely combine and mix reading with standard repetitions of knowledge. Actually, randomizing the sequence of repetitions should be encouraged. Only random coverage of the material will provide you with a true sense of your progress. You can randomize your daily portion of repetitions and review with Learn : Random : Randomize repetitions (Ctrl+Shift+F11).

By using Randomize repetitions, your repetitions will not favor more accurate processing of material based on the length of the interval, element type (e.g. articles, extracts, question-and-answer items, etc.), contents (i.e. branch of the knowledge tree) or degree of processing. Random repetitions will help you better understand possible negative trends such as excessive inflow of new material, lowered retention (mostly as a result of frequent rescheduling), poor formulation of newly created cloze deletions, low quality or applicability of the acquired knowledge, excessive emphasis on certain subject at the cost of others, etc.

Your hunger for new knowledge may quickly result in substantial overflow of new material at the cost of the quality of knowledge and retention. For this reasons you may, but do not have to, decide to execute your repetitions in the following stages:

  1. random and indiscriminate review and repetition of all outstanding material
  2. random repetition of all material with the exclusion of newly created and not fully processed questions-and-answers (mostly cloze deletions) (see how)
  3. random repetition of all material with the exclusion of all articles, and unprocessed extracts (see how)
  4. random repetition of all question-and-answer material with selective delay of repetition in those branches of knowledge that can receive lower priority

In order, to make the above steps possible, you will need to learn a few new tricks available in SuperMemo 2000. Most importantly, SuperMemo 2000 introduces an option Postpone. Unlike Mercy, which can be used to reschedule all outstanding repetitions, Postpone makes it possible to reschedule only a subset of repetitions. For example you can opt to delay repetitions in these subsets:

Postpone uses a number called a postpone factor that is used to increase intervals of outstanding repetitions. Intervals are simply multiplied by the postpone factor. For example, if you choose to Postpone with the postpone factor of 1.1, all intervals will be multiplied by 1.1 and will increase by 10%. Postpone will always increase intervals by no less than one day from the present day. This way, all items on which Postpone is executed fall out of the outstanding subset. Postpone works in a slightly different way on topics, when the postpone factor is modified depending on the value of A-Factor. This way articles with low A-Factors will get postponed less than articles with high A-Factors.

You will execute Postpone with Ctrl+Alt+P in three ways:

  1. in the contents window: if you want to postpone a given branch of knowledge (e.g. new items)
  2. in the browser: if you want to postpone any subset of items that you can generate with browser operations (see: Using subsets)

  3. in the element window: if you want to postpone a branch to which the currently repeated element belongs

Here are some typical ways in which you will execute Postpone:

If you would like to make your repetitions using the above suggestions, you should add all your new material to a selected branch, and transfer individual items to target categories such as sociology, psychology, history, etc., only then when you are sure that these items have received their final wording and meet your stringent quality criteria. As a rule, you should not ever use Postpone on your mission critical knowledge. Postpone should only be reserved for dislodging the excess of newly processed articles, extracts and cloze deletions. It can also be used sparingly on branches with lower priority (e.g. Private, Hobbies, etc.).

To efficiently work with categories you should be familiar with the following subjects:

  1. using categories (adding knowledge to different branches)
  2. using templates (providing different branches with different look)
  3. flow of knowledge in SuperMemo (summary of skills related to the flow of knowledge between branches and knowledge pools such as memorized, dismissed, processed, pending, extracts, articles, etc.)


Frequently Asked Questions

Cloze deletion font cannot currently be customized
Use Remember Extract if you do not want to decide the first interval

Use Remember Extract if you do not want to decide the first interval
The need to specify the interval in Schedule Extract is annoying. I would like SuperMemo to just use the optimum interval
This is exactly what Remember Extract does

Cloze deletion font cannot currently be customized
(Walter G. Mayfield, Jr., Wednesday, July 04, 2001 12:37 AM)
Is there a way to do cloze deletions without SuperMemo altering the original text?
Currently you cannot customize cloze deletion behavior. Marking the keywords with a different font is very important in properly structuring knowledge for active recall. Usually, while at knowledge processing stage, your items will form a messy mix of various fonts and formats. However, once they assume their final shape, they will usually be moved to the target category. This will apply the default category template with a uniform category font (assuming space-saving plain text components are used in the target template). In the future, cloze formats are likely to be customizable