Quiz generation turned out to be a bit more complicated than I initially thought. The main problem
being, how does one automatically generate authentic possible responses to a multiple choice question.
The options a student has to choose from should have some thought put into them, as the student won’t
get much out of the quiz if they are not being adequately tested. This is fundamentally a problem with
multiple choice questions, but I prefer the simplicity of multiple choice in this first iteration of
the Kanji Study Tool. So the question is what are good candidate words for choices of any particular
kanji question. Or rather what are good distractors for a muliptle choice quiz. I have three
categories of distractors that should fit the bill. The first of which being…
Words from the same source (Book Chapter)
Students are typically learning kanji on a chapter by chapter basis. This means it is reasonable to think that a student may get words in kanji from the same chapter confused with each other as they are trying to learn all of them at once. These choices are pretty weak as once the student gets a decent grasp on the kanji of the chapter, they won’t have trouble distinguishing between words of the same chapter. Which takes me to the next best category…
Words that contain the kanji being quizzed
These are probably the more obvious candidates for choices as kanji can have multiple readings across different words, so a student is likely to get confused when presented with choices that all have that kanji within it. The problem with this solution is not all kanji are going to have enough words associated with it to generate enough choices for a question, so another category of words must be used with this category. As the system is implemented today, the categories of words mentioned are the ones used in generating a quiz, but the last category is…
Words that are either written similarly or have similar meaning
This category cannot be implemented at this point as an additional table must be added to relate words to other words that one would reasonably confuse with each other. An example, 右 and 左 which mean right and left respectively. Both of these kanji look similar and have related meaning, so it is reasonable that a student who doesn’t know these kanji very well might get them confused. Adding this data to the schema shouldn’t be too difficult. A table that contains two fields, both foreign keys to word.id, where a single record identifies two words that are similar to each other in either meaning or appearance. The difficult part here is coming up with those words that fit that description.
The first two have been implemented by the following stored procedure:
CREATE PROCEDURE `GET_QUIZ_ANSWERS` (IN quizKanji VARCHAR(1), IN quizWord VARCHAR(10), IN answerCount INT)
BEGIN
SELECT w.word, w.reading
FROM word w
JOIN kanji_in_word kiw ON kiw.word_id = w.id
JOIN kanji k ON k.id = kiw.kanji_id
WHERE k.kanji = quizKanji AND NOT w.word = quizWord
UNION
SELECT w.word, w.reading
FROM word w
JOIN kanji_in_word kiw ON kiw.word_id = w.id
JOIN kanji k ON k.id = kiw.kanji_id
WHERE k.source_id = (SELECT source_id FROM kanji WHERE kanji=quizKanji) AND NOT w.word = quizWord
ORDER BY RAND()
LIMIT answerCount;
END
Where quizKanji is the kanji to be quizzed in a question, and quizWord is the word that will be presented in the question. The first two categories of distractors are implemented as separate SELECT statements, then UNIONed together. The resulting set is randomized and LIMITed to answerCount. With this set up, it will be easy to add the final category once the necessary task have been complete.
As of this moment I see two issue, one of which is immediately relevant. There is no precedence made to either of the SELECTs. The latter SELECT will produce higher quality distractors, but the distractors returned are randomly selected from the whole resulting set. I believe I can fix this, but for now this will suffice. The other issue will be relevant once I explain the next query. It retrieves the kanji and words to quiz the student on:
CREATE PROCEDURE `GET_QUIZ_QUESTIONS` (IN username VARCHAR(30), IN quizLength INT)
BEGIN
SELECT qKanji, qWord, qReading
FROM
(SELECT k.kanji AS qKanji, w.word AS qWord, w.reading AS qReading, sk.retention_score AS qRetention
FROM student s
JOIN student_kanji sk ON s.id = sk.student_id
JOIN kanji k ON sk.kanji_id = k.id
JOIN kanji_in_word kiw ON kiw.kanji_id = k.id
JOIN word w ON kiw.word_id = w.id
WHERE s.name = username
ORDER BY RAND()) AS q
GROUP BY qKanji
ORDER BY qRetention
LIMIT quizLength;
END
Here the kanji with the N lowest retention scores are SELECTed with a random word that associated with it. To ensure only one kanji is SELECTed, GROUP BY is used. I mentioned that there was a problem with the previous query. This problem persist in this query as well. That is, words that contain kanji from future chapters that the student may not be studying could be SELECTed. This is problem since it would be unreasonable to quiz a student on a kanji they don’t know, either through the question itself or the choices presented. I am not sure at this moment how I will fix this, but I will find a solution.
There is also one more problem with the GET_QUIZ_QUESTIONS procedure. That it is possible for the same word to be chosen if the kanji contained within that word are both being quizzed. For example, if 八 and 千 are both being quizzed, it is possible for this query to choose 八千 for both questions. From the perspective of the student this is essentially asking the question twice. I am not immediately sure how to fix this problem either, but I predict it won’t be too difficult to solve.
Getting quizzing to this point, even with the minor issues, is a huge milestone for this project. As such I will putting further development on quizzing logic on hold for a bit while get a better picture of what the site as a whole will look like. This primarily mean figuring out a layout, but through this I will get a better picture of what is left to do, and I can better visualize what the final product will look like.