Knowing how Memrise functions with its algorithms, generating 6 questions against 1 answer (or flipping the question to be the answer) based on 2 different columns in each database, where the audio column (if clips exist for the line entry) are included in the testing sequence, wouldn’t it be possible to modify the coding/algorithms relatively simply to achieve what most of us have been wanting all along?
By example…
Level - 6 questions for each Line of the database testing against Column (text) 1 & Audio (if available).
Inclusion of a second text column…
Level - 8 questions for each line, generated from Column 1, Column 2 & Audio
Inclusion of a third text column…
Level - 10/11 questions for each line, generated from Column 1, Column 2, Column 3 & audio.
For each column added, there are 6 question generated, so for 3 columns we end up with 18 questions, testing us on each “packet” of information that has 3 “qualities” related to the target word.
The reality is, additional columns are meant to represent additional qualities or aspects of the original target word, not an entirely new target word & qualities.
For example…
Word “bus”, now its qualities
- der Bus
- masculine noun
- den Bus (akkusativ)
- dem Bus (dative)
Another example, verbs…
German Verb, and its qualities (represented by columns)
- Language 1 translation
- Language 2 translation
- Stark (strong) or Weak
- Transitive or Intransitive
- Present root/stem
- Simple Past root/stem
- Perfekt Participle & sein/haben conjugation
- A sentence using it
Excluding a second language translation, we have 6 x 6 = 36 questions to test/learn under current Memrise algorithms/logic. But if we think of what algorithms & logic are meant to achieve, we would likely see only 15, 18 or maybe 20 questions (maybe even less) would achieve the same goal.
I can’t code, but the logic behind this is relatively simple and would seem to be achievable by Memrise (to vastly improve the software/product).
Any course determines its testing style based on database columns. The vast majority are 3 columns (including audio), plus 1 informative, but not tested against.
Memrise already runs a test against the Audio column… if there is not an audio file for the line, it does not test against it. So this logic could equally be applied against a 3rd column. Blank field, no testing against it. Field with data, test against it.
If a database has 4 columns, increase the testing to 8 questions per row as a default. If a database has 5 columns, increase testing to 10 questions. 6 columns, 12 or 13 questions.
If the logic is recording data per row to know when to test again, or when a row answer has been answered wrong to increase the questions until it is right, then it should also be able to test backwards on a row having empty fields to revert to the proper number of questions accordingly.
In other words, if the database has 6 columns…
- has 2 blank fields - test for a 4 column row, 8 questions
- has 3 blank fields - test for a 3 column row, 6 questions
There are other examples of Memrise logic/algorithms on why could work, but I think you get my point. Many existing courses would be vastly improved by this change, and using current “work arounds” (adding levels, adding parts of speech to fields that don’t match column headers, etc) only degrades the quality of the courses, not improving them.