Iversen's second multiconfused log thread

[Iversen]

I painted this painting in 1981, so I don't remember what I thought - but Sigurd Orm-i-øje ("Sigurðr ormr í auga") is actually a good guess. I definitely knew the tale at that time since I had participated in some courses at the 'Vestnordic' institute at the university in Århus (the one that dealt with Icelandic and Old Norse), and the human figures in the background have something medieval about them.

Sigurd was according to Saxo Grammaticus one of the sons of Regnar Lodbrog, and when Regnar was killed in a snakepit in the year 865 on the orders of the Anglosaxon king Ella he and his brothers took revenge on Ella by 'carving' the socalled blood eagle on his back ("riste ørn" in Danish) - i.e. Ella's ribs were cut loose from the spine and turned outwards as wings. The smartest among Regnar's brethren was Ivar 'Benløse' (legloose), and to this day nobody knows how he got that name - maybe he was lame.

Later Sigurd allegedly became king of Zealand and the other Danish Islands, and his son was Hardeknud, who was the father of Gorm den Gamle (Gorm the Old) - known in his own time as Gorm "den Løge" (the dull). And this king (father of Harald Blåtand ('Bluetooth') is commonly quoted as the first in the royal line that ultimately led to Margrethe II (our present queen), mostly because he is mentioned on the Jelling stones, but his predecessors are attested fairly shortly after so they probably did exist - though their lives were embellished by people like Saxo who wrote several hundred years later.

However there are people who don't accept that the Regnar who was father to Sigurd Orm-i-Øje was Regnar Lodbrog. You can see an alternative table of descent here. And now we are at it: Regnar Lodbrog's wife is said to be a certain Aslaug Kráka, daughter of - ye Wagnerians listen up now! - Sigurd Fafnersbane (=Siegfried) and a lady called Brunhilde - not to be confused with the Merovingian queen of the same name.

Kunst057.JPG

[Jar-Ptitsa]

It is also great fun to paint silhouette people surrounded by colours, like in the following picture - however even I have to admit that Jip and Janneke with a bit of good will could be characterized as marginally cuddlier than my models:

Jip and Janneke aren't cuddly in my opinion, they are very unattractive. I wouldn't read this books, but maybe it's different for the Ducth or Flemish who are little children when they see them. I prefer your paintings. and yes, I completely see why they are not abstract, but they are a combination of views in each picture, not only one. Anyway, horror is not the same as ugly, like Jip and Janneke are not frightening or like this, but simply ugly. your pictures are often frightening, or they show the frightening images, but for me, this is much more meaningful or realistic than the stupid Jip en Janneke's design. I only put them in your log because the black and white view of the family and computer vignette made think of them, but I haven't seen many vignettes.

[Iversen]

I think thoseJip and Janneke are cuddly, but for me it is not a positive thing. They remind me of Mars bars, which are so sweet that they put your teeth on edge if you are foolhardy enough to munch them. I have made vignettes for the evening school where I was employed for a while in the 80s and for the institution 'Huset, where I painted my paintings, and I like the genre for the concise and demanding format where you can't allow yourself any superfluous elements - I guess that the literary equivalent would be writing haikus. But I like(d) making weird and surrealistic paintings even more. I stopped making paintings because the 'House ('Huset) had less and less free places where I could sit down with my stuff - especiallly in the evenings. And because I had got a job from 1986 onwards I couldn't just come in the daytime. I tried to paint at home (in spite of having even less free space there), but didn't like the smell of turpentine - and then I just had to stop.

2009 Greek:	29% (7900)	… ( … )	71% (19200)	total: 27100 - Pataki
2009 Greek:	32% (10600)	… ( … )	68% (22400)	total: 33000 - new Langenscheidt
2013 Greek:	23% (8000)	11% (4000)	66% (23000)	total: 35000 - Pataki
2013 Greek:	18% (7000)	5% (2000)	77% (30000)	total: 39000 - Old Langenscheidt
2014 Greek:	51% (14000)	13% (3000)	35% (10000)	total: 27000 - Pataki
2015 Greek:	43% (10000)	20% (4500)	35% (8500)	total: 23000 - Pataki

Έγραψα λίστα ελληνικών λέξεών σήμερα, αλλά με μια παραλλαγή που έκανε από τη λίστα την πρώτη λέξη-καταμέτρηση του 2015: συμπεριέλαβα όλες τις λέξεις από τις επιλεγμένες στήλες, και έγραψα την άγνωστη λέξη με κόκκινο, απλά-εικασία λόγια με πράσινο και γνωστές λέξεις με μπλε . Και τότε θα μπορούσα να εκτιμηθεί ο αριθμός των γνωστών λέξεων στο λεξικό.

Το ποσοστό και ο αριθμός των γνωστών λέξεων ήταν ελαφρώς χαμηλότερα από ό, τι ήτανε στον 2014, ενώ το ποσοστό και αριθμός των μισώ γνωστών λέξεων ήτανε υψηλότερες. Αλλά δεν πρέπει να εκνευρίζομαι πάνω αυτό - με πολύ μικρά δείγματα έχουμε μια μεγάλη αβεβαιότητα. Και όπως μπορείτε να δείτε, εγώ εκτιμάται ότι ο αριθμός των λέξεων στο λεξικό 23.000, ενώ το 2014 ήταν 27.000 λόγια για το ίδιο λεξικό (το "σύγχρονο Ελληνοδανικό Λεξικο" του Rolf Hesse).

Kunst107.JPG

Ο τίτλος της ζωγραφικής είναι: κοντινό πλάνο της ελληνικής χελώνας (Testudo hermanni)

[Iversen]

Yesterday I did a (small) word count, and even though I have written about that kind of activity at HTLAL I haven''t explained my use of it here. So here goes..

Passive vocabulary is one of the few things in language learning that can be measured, but not without a pinch of salt. FIrst it should be stated in as clear terms as possible that I don't believe that a large vocabulary alone is enough to become fluent in a language. There are a few words which are very common in ordinary texts, and your ability to use these is the basis for using a language actively. But the basis isn't enough to build a house, and to cover enough words to read ordinary texts or listen to native speech you need a surprising amount of words. Here we have to distinguish two kinds of understanding. If yu just want to 'get the gist' you can make educated guesses on the international words and a few of the common words and see what the text is all about. But to understand it in the sense that you can follow it closely, word for word, and in principle be able to reproduce or even translate it into your own language, you need to know thousands of words. So word counts do measure something relevant, and as I said they represent one of the few measurable parameters of language learning, which has been enough for me to make them. But I don't expect others to do the same - you'll definitely know whether you know words enough to deal with ordinary novels or newspapers without counting them.

Below I have first some graphs that show the proportions taken up by the most common words in English as represented by the Kilgariff corpus, and after that a diagram that shows understanding (content recall measured by cued recall) as a function of the proportion of unknown words - and it is evident from this graphic that you need at least a coverage of around 90% of the vocabulary to have any chance of guessing the remainder correctly:

011-Kilgariff-frequencies,curve.jpg

Hu-Nation.jpg

[Iversen]

So there is a rationale behind word counting, but there are also serious pitfalls. The first is that you have to work on samples. Internet tests usually present you with with a series of words combined with multiple choice answers. Multiple choice allows for precise calculations, but it only measures recognition. When you read or listen, the context give you something to compensate for the lack of multiple choices so with suitable calibration multiple choice does give you a answer that says something abut yur ability to guess meanings - but unaided elicitation (as you get with RSR methods) will always be harder. Another problem is that 'academic' vocabulary is commonly assumed to be more difficult than ordinary words, but for foreign learners with an academic background they actually are easier, which means that the vocabulary of such a person will be overestimated. Still another problem is that you have to compensate for pure guesswork - guessing wildly when you know absolutely nothing about a word shouldn't be rewarded - but the more sophisticated tests have methods to deal with this problem.

The other basic method is to take all words in a series of samples from a dictionary and see whether you know them - but this also is fraught with problems. The first problem is of course to define what you mean by 'word'. The easy solution is to let the lexicographers decide that for you by counting everything printed with bold typeface - also known as 'headwords'. These actually represent 'the middle of the road', so that's my solution - although I have mostly excluded proper names if they occur in a given dictionary. If you use automated counting you get wordforms (i.e. "am", "are" and "is" will be counted as seperate entitities), but from a purely linguistic point of view you ought to be counting word families - i.e. all derived words from one stem whould form one entity (so the substantive "being" would be part of the family of "to be") - but this is obviously hard to do in a consistent way, especially across languages. So I stick with the forms in the dictionaries. But even here there are problems. Not all dictionaries separate fixed compounds (like "mute swan") from more or less loose combinations and examples. Some language use separate words in the writing, others (like German) use one word compounds.

My solution to this problem is to do calculate not only absolute numbers, but also percentages. This also solves the problem with different sizes of dictionaries. It is obvious that you can't prove that you know 20.000 words with a dictionary that only contains 2.000 - but you may be able to show that you know, say, a third of the words in that tiny thing, and then you might conceivably also know a third of the words in a dictionary with 10.000 words. But wait, a larger dictionary would contain more rare words, so the percentage should then be lower. This is a reasonable assumption, but the results from my own word counts demonstrate that there isn't such a simple relationship. You need very large dictionaries to get the expected large fall in the percentages, whereas across small to midsize dictionaries the percentage fluctuate with little.

I have published my results up to 2013 at HTLAL, so there you can check out this claim - but a look at my results for Spanish will show that it holds water. I own one dictionary in the megaton class (Bratli) , and the percentage here is definitely lower - but the estimated total number of known words is still higher than with any other dictionary. NB: you can see different figures for the total estimated number of words in both Langenscheidt and Gyldendal. The reason is the small sample sizes - sometimes I hit pages with few, but 'big' words, and sometimes almost every line has a new headword. The same amount of uncertainty is of course found with the estimed numbers for known, 'guessable/dubious' and unknown words.

(the figures represent estimated 'known' resp. estimated middle category words - the rest consists of the 'unknown' words)

2009: 34400 - 17 % - - Spanish - Bratli (total 200.000 off.)
2009: 19900 - 41 % - - Spanish - Gyldendal (total 48000)
2009: 17600 - 44 % - - Spanish - Langenscheidt (total ca 40000)
2013: 19000 - 42 % - 4000 - 8 % - - Spanish - Gyldendal (total 46000)
2013: 17000 - 60 % - 2000 - 6 % - - Spanish - Berlingske (total 28000)
2013: 21000 - 64 % - 2000 - 7 % - - Spanish - Langenscheidt (total 33000)

When I did my first word counts around 2007 I just noted down the known words and afterwards counted the total number of words on the pages in my selections. If a dictionary has 300 pages, and I use 5 pages with a total of 200 words on them including 50 known ones, then my estimate is that I know 300 / 5 * 50 = 3000 words out of 300 / 5 * 200 = 12.000 words - which is 25%.

With later counts I have actually written all the words down in different colours - and only with translations of the unknown ones. This means that I have a much safer basis for my calculations, but it also takes longer time. One added benefit of this is that I have been able to measure the 'possible cheat factor' - it is just a matter of returning to the sheet later to check that I actually do know the meanings of the known words even without having the translation. Of course I could have learned the words when writing them, but then they should still count as known when I do the check later. And by and large I have found that my first impressions are confirmed if I do a later control.

But there is still one problem? How precisely do I have to know a word? I definitely know the word "geranium" in Danish, but I'm not quite sure what the thing looks like, except that it has flowers. Do I count this as a known word? Of course I do - this is a vocabulary test and not a botanical exam. However if I wasn't sure whether it was a flower or a tree or some kind of furniture I would have a problem - you can be totally sure that you have seen a word 100 times and still don't have a clue about its meaning. In this case I would have to count it as unknown even know the word itself was known. Another problem is words which I almost certainly haven't seen, but which are derivations from something I know (maybe from another language). To deal with such problems I have introduced a middle cathegory in my latest word counts. I haven't found a good name - "guessable" covers a large part of the words in question, but not all. But in practice I can put most words either into the known or into the unknown basket.

[Iversen]

I have added two small word counts to my list at page 428 in my HTLAL log, one for Greek and one for Indonesian. When looking into the last Greek word count from 2013 I noticed that the estimated number of words in my Greek-Danish dictionary by Rolf Hesse (Pataki) was quoted as 35.000. But this was clearly an error - I got 27100 words in 2009 and 23.5000 this time - so I checked the spreadsheet with my historical stats, which I have kept in a spreadsheet. And then I saw that the true number should be a total of 27000 words estimated for the whole book, which of course meant that the other figures had to be corrected. The difference between between approx. 23.500 and twice 27000 doesn't bother me - I have only counted 161 words this time so the figures should be taken with a pinch of salt. However as the stats look this time I definitely knew 69 words (43%), 32 were so-so or guessable (20%) and only 60 were definitely unknown (37%), corresponding to estimated totals of 10000 known, 4500 lala and 8500 unknown words in total (out of an estimated 23500). The percentages are the most interesting because they show little progress since 2013, but my gut feeling is that I understand Greek better know. This is however not something you can read from the numbers..

2009: 7900 - 29 % - - Greek - Pataki (total 27100)
2009: 10600 - 32 % - - Greek - Langenscheidt (total 33000)
2013: 7000 - 18 % - 2000 - 5 % - - Greek - Gl.Langenscheidt (total 39000)
2013: 9500 - 35 % - 3500 - 13 % - - Greek - Pataki (total 27000 (not 35000 as previously indicated))
2015: 9000 - 37 % - 4500 - 20 % - - Greek - Pataki (total 23500)

The Indonesian word count is only the second one I have done, and I only have one dictionary. But here the progress is clearly visible - albeit from a very low level in 2013:

2013: 2700 - 22 % - 500 - 3 % - - Indonesian - Tuttle (total 12000)
2015: 4500 - 44 % - 1500 - 12 % - - Indonesian - Tuttle (total 11000)

As always the percentages are the most relevant because they aren't as strongly dependent on the size of the dictionary as the absolute figures. If you want a figure for your total number of known words in a language you should in principle use the largest dictionary you can find - but I only have two monster dictionaries, one for English and the other for Spanish. In both cases the percentages go down, but the estimates for known words nevertheless remain above the figures obtained with smaller dictionaries:

2009: 34400 - 17 % - - Spanish - Bratli (total 200.000 (officially))
2009: 19900 - 41 % - - Spanish - Gyldendal (total 48000)
2009: 17600 - 44 % - - Spanish - Langenscheidt (total ca 40000)
(...)

2006: 35000 - 78 % - - English - Gyldendal (gl.?) (total 45000)
2009: 51600 - 31 % - - English - Webster unabridged (total 165900)
2009: 43500 - 91 % - - English - Gyldendal (total 48000)
2009: 27600 - 92 % - - English - Oxford Advanced (total 30000)
(...)

The highest figure I have obtained for English in an internet vocabulary test was 51.000. I have quoted it somewhere in my log thread in the HTLAL forum, but heaven knows where - it has just become too big to look through in a pinch. But based on the results with the two large dictionaries it is clear that the percentages taper off with dictionaries that have headword numbers above 100.000 or so, so I doubt that using an English dictionary with twice as many words or more would push the estimated 51.600 words much further up.

And the vocabulary numbers shouldn't hide the fact that even a any native Spanish or English child speaks Spanish resp. English better than I do. But I speak Danish better than the Spaniards or Brits do.

[Teango]

The percentages are the most interesting because they show little progress since 2013, but my gut feeling is that I understand Greek better know. This is however not something you can read from the numbers.

Since your knowledge of Greek vocabulary has had lots of time to settle (and proverbially and biologically "grow roots" since 2013), I'm sure you've now developed a deeper contextual understanding of many of these words and their collocations. Gut feelings may be difficult to quantify but they can indeed be very revealing!

The highest figure I have obtained for English in an internet vocabulary test was 51.000. I have quoted it somewhere in my log thread in the HTLAL forum, but heaven knows where - it has just become too big to look through in a pinch.

Had a quick search just now using my own gut feeling and found these pages on HTLAL (not sure if these are the posts to which you were referring, but here they are just in case):
http://how-to-learn-any-language.com/fo ... =0&TPN=106
http://how-to-learn-any-language.com/fo ... 3200&PN=21

[Iversen]

Teango's second link was spot on, but for once my memory has deluded me downwards::

"I got estimates* of 77717 known, 16973 inferred and 8933 familiar words (but I don't acknowledge words whose meaning I have forgotten, so those 8933 socalled familiar words are irrelevant).

My own estimates - based on counting known headwords on sample pages in dictionaries af different sizes - would suggest a much lower number. In Webster's Unabridged Monster Dictionary I got 51.000 known words (around a third of the more than 160.000 headwords in that sublime creation of unfathomable industriousness), while I hovered slightly above 90% of the words in smaller dictionaries. I have written the actual numbers in this post in my log. But from those numbers you should subtract more or less automatic derivations because you get those for free, not by laboriously learning them.

I don't know why the numbers are higher in this electronical test.

EDIT: And an estimate of 35867 French words (out of 43.000). My one and only dictionary count gave 23.100 in a dictionary of the same size, but this time I think that the 'truth' is somewhere in the middle, maybe around 30.000 words"

(end of quote)
* results obtained from the vocabulary test at Plenilune

I wrote this Oct. 1 2010, and the number I remembered was apparently the number from my own manual test, based on my Webster monster. The number from the online test was 77717, and I have to say that I trust my own, much lower figures more than the online test resultat. A couple of other HTLALiens also reported English word counts in the same range.

But other tests yield lower results. According to a post from 2012 I 'only' got 38.000 words with testyourvocab.com and 22.500 wordfamilies with my.vocabularysize.com - i.e. numbers that are more in line with my own test results.

At the site of testyourvocab.com there is a quite interesting blog with remarks and statistics, and among other things they show this interesting graph that shows the benefits of reading voraciously:

Testyourvocab_NativeAnglophones_readinghabits.jpg

... but for non native learners the results are miserable - so much effort and so poor results:

Testyourvocab_nonnatives.jpg

[Teango]

Tarvos' second link was spot on...

I think that's the first time someone's mistaken me for Tarvos - I'll take it as a complement!

But other tests yield lower results. According to a post from 2012 I 'only' got 38.000 words with testyourvocab.com and 22.500 word families with my.vocabularysize.com - i.e. numbers that are more in line with my own test results.

A score of 38,000 words is very impressive, even for a well-read native speaker of English!! If I recall correctly from the video "Extensive Reading and Vocabulary Range", Alexander Arguelles scored 28,300 English word families on the vocabularysize.com test, and my own score on testyourvocab.com just now was marginally lower than yours at 37,800 words. In fact, if you were to score 100% on the vocabularysize test, the maximum value would still only come to 45,000 words, so there seems to be a glass ceiling for rarer less predictable words where statistical validity starts to break down.

[Iversen]

Argh ... my apologies to you both.

As for the numbers, vocabulary learning has always been one of my priorities, and it would be somewhat disappointing if it didn't have an effect on my results. Besides I spend a lot of time reading and hearing English, and it has been like that since I was 8 or 9 years old and we got our first TV set in my family - I normally say that my first English word was "yabba dabba doo", enunciated by the immortal Fred Flintstone. By the way, one of the few errors committed by Fry & Co. in the quiz QI, was asking for dinosaurs starting with the letter b - and accepting Barney. Yesterday evening I watch a similar program on Swedish television (named something like 'the interest club') - and it was actually quite interesting too.

Kunst091.JPG

Right now I'm halfway through a book in Danish named "Persisk kronologi og længden af jødernes babyloniske fangenskab" (Persian chronology and the length of the Babylonian captivity of the Jews). The main point is that there roughly are three sources to an absolute chronology for the Middle East region: an immense hoard of Akkadian/Babylonian cuneiform tablets, similar Persian tablets and the Bible. Seeing the Bible as a source text normally sends shivers down my spine, but here we are speaking about a period where at least some of the authors of the Bible actually lived, so their writings can be seen as just another historical testimony - which also implies that it should be used with caution. The author, Rolf J.Furuli from Oslo, can read most of the relevant languages, but he stresses that he isn't an astronomer so for the identification of the astronomical events mentioned on the tablets he has to rely on specialists. But to cut matters short, his main point is that there is a series of 14 Persian tablets called the Saros tablets which indicate moon occultations AND intercalary months for a period from 747 to 315 AD, and they have been seen as sacrosant and infallible by the leading authorities in the field. The Persians and earlier cultures in the area used a lunar calendar which had to be corrected periodically to keep up with the course of the seasons, and this was in principle done by adding intercalary months in a pattern that reflected a cyklus which was repeated roughly every 19 years. But Furuli shows that some of the intercalary months simply don't fit into the regular 19 year cycle - and if they don't then the calculations based on those tables crumbles.

He furthermore points at a possible cause for the shortcomings of the Saros tablets, namely that they fit TOO well with the theoretical system proposed by a certain Ptolemaios - the same man who was seen as an infallible source of wisdom up through the Middle Ages, but who was debunked as a crook by no less than Newton*, who pointed out that his astronomical data simply don't hold water. It is a reasonable assumption that Ptolemaios didn't make observations himself, but relied on his theories to invent the data which were supposed to support the theories - or at least we today see it like that, but the Greeks were not known for observations of facts. Already the French scholastic philosopher Abelard is said to have given a lecture called "Everything Aristoteles said was wrong", and judged against the physical realities of the world this is true. His predecessor Platon Plato didn't even try to do natural science, but just told boring philosophical tales invented in his own brain (my evaluation, not Furuli's). So if the Saros tablets partly presented fraudulous pseudo data based partly on calculations in the footsteps of the crook Ptolemaios, partly on the belief that the 19 year cycle had been observed meticulously since its invention, what then can we rely on? Well, the Babylonian astrologers did mention occultations on their tablets which they acknowledged that they couldn't have seen because of cloudy weather etc, so they must have been using some kind of calculation to extrrapolate those, but in between they DID observe the sky, and once in a while their observations can be linked to historical events and things like the name of a reigning of kings, so even if the Babylonians didn't give a systematic encyclopedic overview in the way of the Saros tablets they may actually be more reliable than those.

And that is where the Bible becomes relevant. The Babylonian captivity lasted according to the Bible 70 years, but with the duration of reigns of the Babylonian and Persian kings which can be calculated on the basis of the Persian tablets there is only room for 50 years of exile. Furuli isn't a fundamentalistic Bible thumber who believes everything in that book, but in this case he chooses to believe in the canonical 70 years because they fit rather well with the Babylonian cuneiform tablets once you put the things in that material together without trying to fit it to the Persian Saros tablets.

All this could have been superfluous if we just had had some reliable tables of rulers with exact durations of each reign, but it turns out that those we have got from antique authors - mostly as isolated quotations in hardly trustworthy secondary sources - do not agree so they need corroboration and sometimes more than that. For instance Furuli claims that he can prove that the reigns of Xerxes and Artaxerxes overlapped, but I haven't reached the part of his book where the arguments for this are presented, so I'll not comment on that claim. I'll just say that my belief in ancient chronologies has been undermined by the part of this book which I already have read. The data are more contradictory than most scientific and popular scientific literature indicates. We are not speaking of pseudoscience of the kind that accepts Oannes or the 'Annunaki' as real living beings coming to invent Sumerian culture or states that the Sphinx and the Pyramids were built by aliens 10.000 years ago. It is simply just a matter of problematic, but genuine data being pushed aside because they didn't fit a canonical system which ultimately was built on dubious sources.

* I just checked the list of references in back of the book. "Newton" here isn't sir Isaac Newton, but R.R. Newton.