Anki vs. Gold Lists vs. Iversen-style Wordlists

Discussion in 'Learning Techniques and Advice' started by Peregrinus, Jul 13, 2014.

  1. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish

    As a somewhat experienced language learner I personally am not afraid of falling into a 1<>1 trap, but new learners may not realize that they in fact need to refine their initial one or two word meaning. But that wasn't my real point. I use very detailed definitions on my Anki cards, although as I have mentioned, I just try to hook a word or three initially, after which in subsequent review when I know that initial meaning well enough, I try to add further meaning from that longer definition.

    But I want that process of iteration and refinement to my English meaning of a L2 word to end as fast as possible, and for further refinements to be in L2 in my mind. With German for example, I want to drive on a road that only has German signs, insofar as possible, instead of prolonging my drive on a bilingual road.

    This is an issue deserving of its own thread perhaps, and not really central to the question of Anki vs GL vs Iversen Wordlists.
  2. Bob

    Bob Active Member VIP member

    Joined:
    Apr 18, 2014
    Messages:
    337
    Native Language:
    English
    Intermediate Languages:
    Cebuano, Greek_ancient, Spanish
    Basic Languages:
    Filipino, French, Hebrew_clasical, Mandarin, Portuguese, Russian
    I'll bring this over from the Goldlist thread. How long, overall, do these methods take, and what do you walk away with, for say 3000 words? Goldlist has an upper limit of 129 hours. You could NOT pass a vocab test doing only that, but words will be absorbed quickly when you look them up (or figure them out).
  3. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    I answered over there, but my guess for my hybrid Iversen-Anki experiment, is that I would get between 70-85% on a vocab test within 3 days, while Iversen's attrition rate of 15% suggests he would get around 85%. For Anki alone, the question is when. Only when a group of words become "mature" are they fairly well learned. Since I can check the mature rate of various subdecks in my German deck, I would say that eventually, maybe after 6 months +, it would be in 95% range.
  4. Montmorency

    Montmorency New Member

    Joined:
    Jul 28, 2014
    Messages:
    13
    Native Language:
    English
    Advanced Languages:
    German
    Intermediate Languages:
    French, Spanish
    Basic Languages:
    Danish, Welsh
    I've contributed today to the GL thread, but as I'm fascinated by this subject, I can't keep out of this one as well. :)

    A crude ball-park comparison: Peregrinus said he'd learned the very creditable number of 11,000 words in about a year with ANKI. In the other thread, I've mentioned that in theory, you could learn 10,000 words with GL in a year, at the work-rate of about 1 hour 10 minutes per day, in GL. This is using Davey's own quoted figure of about an hour per 25-word headlist. I just mention this to suggest that in ball-park figures only, the two methods are comparable for overall achievement.

    Now, Davey claims his method is more efficient than automated SRSs, because you never look at a word you "know" again. (I have a feeling that's also true with Iversen-wordlists, which I have not used lately, but used to a lot).

    One reason I prefer GL to wordlists is perhaps a rather silly one, in that it usually involves discreet sheets of paper, and I happen to be hopeless with those. They soon get out of order and untidy, whatever I do, and I just want to throw them away. (Actually I think that's what Iversen does). But then you've lost track of what you've learned. I now like the idea of working in a good quality hardback book, keeping it nicely in order, keeping track of what I've learned and having an idea of the number of words I consider that I know. It's actually not an accurate count of that, because as mentioned elsewhere, I don't GL words I know I know already, or "easy" words. But it's a start.
    (Perhaps I'm trying to get in touch with my inner chartered accountant).

    As also mentioned elsewhere I think Iversen has a much better handle on the science of all this than Davey does. But has Davey hit on something that science hasn't quite caught up with yet? Could be. It's never going to happen, but I'd like to see Iversen give the GL a try (and also maybe see Davey give wordlists a try). That'd make a good video series actually. Are you up for this gentlemen? :)

    One of the doubts I have about GL is Davey's oft-repeated claim that it doesn't matter that GL learning is all to passive vocabulary. He says that when you want to activate it, you just go to a TL country, and in 3 days it just happens. hmm. Well, Professor Arguelles did something like that with Russian, but it wasn't 3 days: it was 3 months, and he had intensive 1-to-1 sessions with a phonetician, if I remember correctly.
    This is not to condemn GL though, although personally I think you can be GL-ing and activating the language at the same time. I don't mean consciously using each word as you learn it, but in a more general sense.
    Big_Dog and Bob like this.
  5. Bob

    Bob Active Member VIP member

    Joined:
    Apr 18, 2014
    Messages:
    337
    Native Language:
    English
    Intermediate Languages:
    Cebuano, Greek_ancient, Spanish
    Basic Languages:
    Filipino, French, Hebrew_clasical, Mandarin, Portuguese, Russian
    I think what he meant is if you understand the spoken language, then you can start speaking in 3 days (but I doubt perfectly). The details are really fuzzy here though. He says reading will come very quickly (which I must say is true ), and the thing about 3 days, but when I asked about how long for oral comprehension, he said it depends.
  6. Iversen

    Iversen Member VIP member

    Joined:
    May 1, 2014
    Messages:
    91
    I have read a lot about goldlists, and I do think they work - but they work for the same reason Anki works: by making sure you are confronted with certain words in a sequence that with a bit of luck should ensure that you see them often enough to remember them. The one quibble I have with the system is that once you have discarded a word you don't see it again. However my own tests show that the words I know in a first repetition round are MORE likely to be forgotten in round two than those I missed in round one (because I memorize these once more, but just skip over those I think I know). I have normally been doing one and only repetition round because I knew that the subsequent loss would be partly compensated by the extra memorizing obtained in repetition round one ... and after that I have thrown my lists away right away or stored them in pile for possible later scrutiny (which rarely happened). But with my experiences from the Serbian study I think I'll add a second round, maybe a month later, where I simply copy all foreign words from the original list once again, adding translations when I don't remember a certain word (and only there). And that means that I'll have to be slightly more careful about preserving the wordlists.

    My system is more labour intensive at the start than the goldlists are, but personally I don't mind taking the time to think about each word - inventing associations, looking for etymological or other intralinguistic clues etc. I can not only feel, but measure that these activities help me to remember the words. The funny thing is however that it isn't necessarily a 'memory hook' I use to recall a word later. The point seems to be that you are more likely to remember a word the denser its network of relations to other things are, and establishing such a mesh right from the start must be more efficient than gradually building it while you plough your way through thousands of pages or hours of listening.

    As for activation I do think you need to get so much input that your head starts spinning - and for me that usually means I have to visit a suitable country. But watching TV or listening to podcasts are also useful. Reading is slightly less efficient because it is more of a selfdriven activity than listening, where it is the outside world that pumps content into your head. But with a good passive vocabulary you can make reading a pleasureable and not too tiring activity, and then it will eventually begin to have the same buzz inducing effect as listening. In an ideal world all my TV in foreign languages would have undertitles, and I would be watching it for hours while doing wordlists and reading extensively .. and THEN I would buy a ticket to Targetlanguistan to get the last boost.
    Last edited: Jul 29, 2014
    emk likes this.
  7. Montmorency

    Montmorency New Member

    Joined:
    Jul 28, 2014
    Messages:
    13
    Native Language:
    English
    Advanced Languages:
    German
    Intermediate Languages:
    French, Spanish
    Basic Languages:
    Danish, Welsh
    Thank you Bob and Iversen for your responses.

    "Targetlanguistan" :) :) :)

    Iversen was too polite to point out the obvious objection to my silly objection to WLs which is that it is of course, perfectly possible to do those in a nice hardback book if you want to, and indeed I did start doing that eventually. Sheets of paper or small notebooks do give you more flexibility (e.g. on a crowded tram or underground train, etc), so for someone like me, I'd ideally work with those while "on the hoof", but be careful to consolidate them into a hardback book later on, and throw away the worksheets, if I could be disciplined enough. :)

    A more serious question I might have is about the actual method of memorisation. Not everyone is good at inventing associations or find that it works for them; not all words have obvious etymological etc clues to offer, so this seems to all be a bit of a "black art". Clearly Iversen has become good at it over the years, but it's not easy to see how this can be taught to a neophyte. Davey's instructions are easier to follow in that respect, since there isn't any actual memorisation involved. (Whether it actually works or not is another question).

    Iversen has put his finger on the crucial difference between the two manual handwritten systems: Davey thinks it is a virtue that you never look at words you think you know again, while Iversen thinks this is a weak point, and appears to have the evidence to prove it. I would just add that this evidence only holds true while actually doing his wordlist system. For example, he is actively trying to memorise the word on the first pass, which Davey suggests turns off the long-term memory. Now if the same mind were to try out the GL system for a while (and unfortunately, it would have to be over many weeks to be significant), i.e. specifically not actively try to remember the words on the first pass, it would be very interesting to see what happened.

    Is the following a realistic (and not too time-consuming) experiment?

    Restrict it to 100 headlist words, i.e. 4 headlists of 25. Could be done one list per day for 4 days, or all in one day, or whatever fitted best.
    Then just leave each headlist for 14 days and do the 1st distillation.
    Then leave for another 14 days and do the 2nd distillation.
    Then leave for another 14 days and do the 3rd distillation.

    That would take 6 weeks and be the equivalent of doing one "bronze" book, albeit only for 100 words.

    If you like, you could take it further, and in a further 8 weeks, complete your "silver book" and the 8th distillation. (Shouldn't be very large by then).

    To do it in the proper spirit, I think you'd have to voluntarily refrain from testing yourself on the words you thought you knew in the early distillations, although it might be interesting to do a test a little while after the 8th (or after the last, if you don't get as far as 8).

    I suppose the best test would be to read a passage of text including those 100 words (which is what I think you sometimes do with wordlists isn't it?).

    Well, it can remain a thought experiment for the time being if you prefer :) , but if ever you decide to really do it Iversen, I'd be very interested to hear the result.
  8. Iversen

    Iversen Member VIP member

    Joined:
    May 1, 2014
    Messages:
    91
    So far I can't see that the use of active memorization turns off anything - if it did I would be in deep trouble. But the experiment outlined by Montmorency would not be too labour intensive to try out. It wouldn't take me long to make a list with 200 words from a dictionary, run 100 of those words through a wordlist (with 1 or 2 repetitions) and run a gold list based on the rest of the words- The final test would then be not whether the goldlist has dwindled to almost nothing in 8 weeks time (I suppose it will do so), but whether I remember more from the original 100 words on my wordlist or from the 100 words on the undiluted first goldlist in 8 weeks time from now (which will be right before Novi Sad).

    Maybe I should add that simply reading through the passage which delivered the words won't be a relevant test with this setup, so I'll simply do the kind of repetition which I use most these days: I copy all the words in question to a sheet of paper, and then I add an explanation or translation whenever I am in doubt about the meaning of a word. Afterwards I can simply count the number of comments, and that's my statistic.
    Last edited: Jul 30, 2014
    Montmorency likes this.
  9. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    I've read a bit more on Davey's blog, and while he does try to give a scientific basis to part of his method via referencing the forgetting curve, other stuff like switching on an off of long- and short-term memory seems far more specious. In fact he gives as one of many possible reasons for GL not working, one doing other stuff involving short-term memory close in time to the GL. Perhaps he is basing this on his interpretation of some study or other, but I didn't see it.

    Re the experiment above, it would also seem a more fair test to be unfair about the vocabulary chosen, namely to choose only relatively infrequent vocabulary lest even light general reading in the L2 reinforce too many of the words in the gold lists, the same as with Iversen's texts as he mentions.
  10. Iversen

    Iversen Member VIP member

    Joined:
    May 1, 2014
    Messages:
    91
    I have also read a bit more about the goldlists, and I noticed one small thing that changes my idea about the whole method drastically. I assumed that you didn't see the original resp. previous versions again once you had made a destillation - but it seems that the original version remains visible while you write the next three versions (which are distributed clockwise over two opposite pages in a notebook in A4 format). This explains why it isn't as big a problem to kick out 'wellknown' words - you can just confer with the original list with all 25 words.

    Apart from that: starting this afternon and until about an hour ago I have made 4 goldlists à 25 words - not in a pretty book, but on yellow A4 sheets. Apart from that I have followed the rules - except insofar that I have spent the pauses between the four lists on making a wordlist with 100 words, using my own system. And then I'll revise this list tomorrow and in 2 weeks' time, and I'll destill the goldlists in 2, 4 and 6 weeks time - and then I'll see after 8 weeks how much I remembered - or maybe rather forgotten. To prepare for this I have written the translations in the goldlists in a way that makes it simple to cover them while I make the test sheets, just as I do with my own wordlists.
  11. Bob

    Bob Active Member VIP member

    Joined:
    Apr 18, 2014
    Messages:
    337
    Native Language:
    English
    Intermediate Languages:
    Cebuano, Greek_ancient, Spanish
    Basic Languages:
    Filipino, French, Hebrew_clasical, Mandarin, Portuguese, Russian
    I've gone back out of curiousity to some old lists. I just tried to distill them again. The ones choose are most often the ones I chose before. Also the most solid one was the first one that I took from something I read.
  12. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    I wonder whether Davey mentions this facet in his voluminous explanations of the method, i.e. the chance to make a quick revision of the entire list at each subsequent distillation. Even without hiding a side and engaging active recall, reading them all again has some positive effect on retention. Which begs the question, why not just rewrite the entire list at some interval between 2 weeks and 2 months? Davey says the reason for the distillations is to avoid spending time on what is already known, which is of course why Anki sends better known words further into the future than less well-known words. Obviously the difference with a distillation is the careful rewriting of less known words, but how much time is actually being saved?

    It would seem to me that one might as well explicitly incorporate the rereading of the entire original list as part of the method, instead of just the most recent distillation, since that takes only seconds.
  13. Montmorency

    Montmorency New Member

    Joined:
    Jul 28, 2014
    Messages:
    13
    Native Language:
    English
    Advanced Languages:
    German
    Intermediate Languages:
    French, Spanish
    Basic Languages:
    Danish, Welsh
    @Iversen: Many thanks. This is very sporting of you. I hadn't really thought much about the original list of 25 being visible during the 2nd and 3rd distillations. (It has to be visible for the 1st distillation of course). Personally I wouldn't think I have much incentive to look across to the 1st list when doing them, since when doing the 2nd, you are working all on the right-hand page, and on the 3rd, you are doing the real work by looking at the bottom list on the right hand page, and only using the left hand page to transfer the distilled words to.

    And on the 4th, 5th, 6th and 7th distillations, you are working in a different book, the so-called "silver book", so the original list is not visible unless you go looking for it.

    @Peregrinus: I suppose that Davey would say that it is simply not necessary, and it may also smack of trying too hard, in the way he looks at these things. The not trying is quite important to the philosophy of the way this is supposed to work. In that sense, it is the antithesis of Anki, which, saving your presence, comes over as a tool for Streber. :)
    (I mean that in the more positive sense of striver, rather than some of the pejorative translations :) )
  14. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    LOL. Ich kenne diese Wort, "Streber". Vielleicht ist es aber doch noch passend in der pejorativen Bedeutung.
  15. Iversen

    Iversen Member VIP member

    Joined:
    May 1, 2014
    Messages:
    91
    Earlier today I have done my first destillation round for four goldlists à 25 words, and apart from a slight disturbance from the outside (my sister entered and started reading TV program listings aloud from a magazine - with comments) I have tried to follow the instructions, including the one about avoiding associations. I have also done a late second repetition round of the wordlist of 100 words I made at the end of July, at the same time as the original gold lists. But a few days ago it occurred to me that I also needed a totally 'do-nothing' baseline, so I made a list of 100 fairly random words which I intend to leave totally alone without any further attempt to learn them. At doomsday near the end of September I'll check how many of the words from the original goldlists, my three-colum wordlist and this new do-nothing list I know. Obviously this can't be called a scientific experiment (with me occupying all roles, including the one of sole test person), but it suggests how a more trustworthy experiment might be structured.
    Last edited: Aug 18, 2014
    Peregrinus and biTsar like this.
  16. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    So, a few days ago I completed my hybrid Iversen-Anki test. I did 11 lists, one every other day, of between 59 and 66 words, with the most common being 60. I did this digitally with a spreadsheet (smaller window so both open at same time) and my procedure was as follows:

    1) open up my Anki browser to the subdeck of about 60 words and copy 6 at a time onto the spreadsheet. I then worked with that group of 6 until I knew them all well enough to write the translations.
    2) I hid the column with the German words and went over the translations until sure I could write the German words, taking care per Iversen's instructions, that if I had to peek at the German word, I waited a while doing something else before writing the 3rd column with the German words, and even if I knew them all, I would type in the next group of words and review them until I knew the English translations (since I had the 1st column hidden, I had to cut and paste the group into that column later after finishing with the previous group).
    3) Then I continued in groups of 6 until finished, and separated them with empty rows.
    4) Several hours later in the day, I went over the words again in Anki, choosing 2 days, i.e. known, no matter if I missed a word or not, so as not to interfere with the next days review of the Iversen list. I noted the miss rate.
    5) 2nd day I went over the Iversen list again, noting the miss rate by changing the color of the 3rd column German word to blue for easy counting at the end. I worked with each group of 6 words until I could make all the translations into the 4th column, keeping the first and second columns hidden unless I needed to peek. Because the words are then partially known, I worked with 2-4 groups for the translation from English to German in the 5th column to add a little more time to the interval.
    6) 4th day the Anki group would come due. It takes 2 days instead of 1 because Anki intentionally separates the group into approximate halves so that the order is changed for the official first review, so I had to wait an extra day. I counted the misses, though I passed a word if I knew part of a two or three word definition. I then chose 15 days for each one that I passed, to get a good idea of how the experiment was working, i.e. as if I had no need for further review other than misses which Iversen loses but Anki does not.
    7) the 15th day review of known words from the 4th day is too hard to count accurately because the failures of the 2nd day have also been on a more frequent review track. My estimation is that I missed a further 5-10% of those.

    Tallying and averaging the stats from above I get the following miss/failure rates:

    1st day Anki review (same day as 1st Iversen review): 18%
    2nd day Iversen review: 23% (partial overlap with previous statistic)
    4th day Anki review: 25%
    estimated 15th day review rate (not counting fails on 2nd day of Anki): 5-10%

    If I recall correctly, Iversen estimates a 15% attrition rate from his lists, to a maximum of 25% or so for Russian. My guess for Anki alone on the first review or two is near 40-50%, because of less thorough initial learning. So my estimated miss rate for the experiment after 15 days is in the 30-35% range, but with no attrition since the misses will get reviewed again in Anki.

    That is a better rate than Anki alone, at the cost of more time up front, which generally was around 60 minutes for 60 words for the 1st Iversen review, and 20-30 minutes for the second. But it seems as if long term with Anki I can more quickly push them to the future, saving reviews. The only downside of that is for longer more complicated translations, where I first try to get just a toehold, and then try to add to that on subsequent review even though I pass them.

    I still am extremely weak at making mental associations, with some being clearly remembered every time but just as many not. And that is only for words I can quickly come up with an association based on spelling, sound or the English translation. For most words I could not come up with an association unless it was a morphological one, and the latter only because I already knew so many German words. For words where both the root and its derived or related forms are all very low frequency, I could make zero associations.

    All in all I view this experiment as a success, and learners with better (i.e. any meaningful degree) of ability with mnemonic associations should do much better. As I would expect myself to do with a related language after German, or maybe any language since I could then use the German translation as well as the English as a hook. This is a keeper for me and I have continued several lists past the ones I used in the count above, trying to up my daily rate by doing 60 every other day and 30-40 on alternate days. At the very worst it will be a wash time-wise, but it seems a more methodical method for initial learning, and one with the real possibility of doing much better for long-term time spent reviewing words.

    As with any hybrid experiment, one loses some benefits of each component method, but hopefully with the result that the sum is greater than the parts.
    Big_Dog likes this.
  17. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    While not strictly scientific, it seems as scientific as any of we self-learners could make it. That counts for a lot. I will be interested to see your results.
  18. Iversen

    Iversen Member VIP member

    Joined:
    May 1, 2014
    Messages:
    91
    I think Peregrinus is correct in assuming that the ability to make mental associations fast is the key to remembering wordlist words. It also explains why I sometimes have a higher loss rate - even when working with one specific language the rates fluctuate quite a lot. Outside distractions and time pressure will tempt you to skip the associations phase, but any lapse here will be cause grievance when you have to recall the words later. A word meaning is like a nerve cell with a fuzzy blot in the middle and a lot of synapses stretching in all directions. And the more 'synapses' a word has, the better chances you have for recalling the word - even at a stage where you don't consciously recall the associations you used.

    I have thought a lot about association making, and I think that it helps not to be too ambitious about them. Trying to find words or word combinations in your own language that cover a whole foreign word or expression may succeed, but you can't count on such 'super associations'. And then it is better to make up one or two associations that only cover the beginning or the root or some other part of the word/expression. Foreign sounds can also be a problem because you can't make associations in your own language containing those sounds. One trick is to mix a snippet of the foreign word with its translation or a suitable associative word in your base language or another wellknown language - like German "Schneider" and English "taylor" --> "schnaylor" (or "schnaelor", pronounced with an open a and reinforced by the image of a snail on a taylor's table). I know this sounds artificial and contrived, but when you make these associations yourself they work. They don't work nearly as well when somebody else imposes their associations on you.
    Big_Dog, Peregrinus and biTsar like this.
  19. Peregrinus

    Peregrinus Active Member

    Joined:
    May 27, 2014
    Messages:
    613
    Native Language:
    English
    Intermediate Languages:
    German
    Basic Languages:
    Spanish
    Question for Iversen: Perhaps you have discussed this elsewhere and I missed it, but how do you determine if you have already studied a word in a word list in the past or not? Surely there are words you think you know and have studied but are not sure of, especially those that account for the 15% or so attrition rate with your method. With Anki, I simply search for a word if I am not sure I have already added it in the past (more necessary because of my future queue that I have not actually studied yet). Thus are there words that you might add to a current list that have already been in a list previously?
  20. Iversen

    Iversen Member VIP member

    Joined:
    May 1, 2014
    Messages:
    91
    Good question, and not easy to answer - mostly because it is impossible totally to exclude that you actually have seen a given word but just not memorized it. The words I use in my wordlists most certainly come from some kind of source , and I have at least seen them there. So there it doesn't matter whether I have seen a word before or not - I have. However it is a real problem for word counts based on dictionaries where I have to decide in a flash whether a know a given word or not. In my first counting operations (say, up to 2010 or 2011) I just separated known from unknown words, but in many cases I felt that I certainly hadn't seen a word before - often because something in its form surprised me, even though the word still was comprehensible. So I added a middle category "guessable", but I have since realized that this name only covers those cases where the form somehow surprised me (mostly because of unexpected derivation patterns), not other cases where I am sure fairly I haven't seen a certain word - like loanwords belonging to themes which I KNOW I haven't studied in the target language. And even there I can't exclude that it has been present in a text about something totally different, or someboday may have said it on TV. So in fact this middle category has just become a place for all kinds of dubious cases, but I haven't really found a better name for it.

    The consequence of dropping the original definition of 'guessable' is that I don't need to prove to myself that I never ever have seen a given word - which obviously is difficult or impossible. But there is still another criterion which can be used. In many cases I classify a derivation as known because I know the base word and the derivation pattern, and if I needed the word for some purpose I would make a correct guess. However with most words in the middle category I would have made the wrong guess - like choosing a wrong affix or verbal ending. The result would probably be comprehensible to a native reader or listener, but according to my dictionary they have just chosen to favor another form. So in these cases I get surprised at seeing the correct word, and I know I have found a candidate for the middle category.

Share This Page