[B-Greek] 2nd year NTG on my own / followup

James Tauber jtauber at jtauber.com
Sun Nov 4 13:34:17 EST 2007


On Nov 3, 2007, at 11:14 PM, James Bowick wrote:
> Thanks for the thoughts.  I readily agree that my numbers are pretty  
> rough and rugged.  Interestingly, they do seem to point in the same  
> general direction as yours, that there is a droppof in return for  
> words invested somewhere between 1,000 and 1,500 words.

Also significant in my numbers is the drop off from 50% coverage to  
75% coverage to 90% coverage. Learning the top 200 words might mean  
knowing more than half the words in 96.9% of verses but you'd only  
know 90% of the words in a mere 9.8% of verses.

> My numbers are also misleading in that they do not predict success  
> evenly across books.  Some have much richer vocabularies than  
> others.  I suspect this fact is burried in your numbers as well -  
> some books have more sentences that are unreadable at every level  
> than others.

Absolutely. For example, here is a table, as before, but the numbers  
are just for John's Gospel. Note that, to enable comparison, the  
*vocabulary* isn't restricted to the words in John, this still assumes  
you learn the entire GNT vocab in frequency order, but the percentages  
are just versus within GJohn.

		any		50%	75%	90%	95%	100%
100 	100.0% 	97.7% 	47.6% 	8.4% 	3.3% 	2.6%
200 	100.0% 	99.6% 	76.1% 	25.2% 	10.9% 	7.8%
500 	100.0% 	100.0% 	95.6% 	67.7% 	44.9% 	34.8%
1000 	100.0% 	100.0% 	98.6% 	86.2% 	68.3% 	58.2%
1500 	100.0% 	100.0% 	99.5% 	92.0% 	78.1% 	69.7%
2000 	100.0% 	100.0% 	99.8% 	95.5% 	85.9% 	79.9%
3000 	100.0% 	100.0% 	100.0% 	98.2% 	93.9% 	90.6%
4000 	100.0% 	100.0% 	100.0% 	99.4% 	96.9% 	94.8%
5000 	100.0% 	100.0% 	100.0% 	100.0% 	99.3% 	98.1%
all	 	100.0% 	100.0% 	100.0% 	100.0% 	100.0% 	100.0%

In contrast, here's the same table for Hebrews (again with full GNT  
vocab):

		any		50%	75%	90%	95%	100%
100 	100.0% 	83.1% 	11.5% 	0.6% 	0.3% 	0.3%
200 	100.0% 	94.0% 	31.3% 	2.6% 	0.6% 	0.6%
500 	100.0% 	99.6% 	62.7% 	13.8% 	4.2% 	3.6%
1000 	100.0% 	100.0% 	86.7% 	38.6% 	15.5% 	11.5%
1500 	100.0% 	100.0% 	94.7% 	54.4% 	32.0% 	23.1%
2000 	100.0% 	100.0% 	98.0% 	72.9% 	45.2% 	38.2%
3000 	100.0% 	100.0% 	99.6% 	89.1% 	65.3% 	58.7%
4000 	100.0% 	100.0% 	100.0% 	96.6% 	80.5% 	77.8%
5000 	100.0% 	100.0% 	100.0% 	100.0% 	96.3% 	94.0%
6000 	100.0% 	100.0% 	100.0% 	100.0% 	100.0% 	100.0%


> So let me ask you the practical question.  If frequency is not the  
> best way of learning vocab, as you demonstrate in your blog, what  
> is?  I think part of the reason we use frequency is that it is  
> readily availabe.  It seems to me that you may have the basis for a  
> more immediately rewarding vocab list, which I would find very  
> valuable.  Myself, it seems at this point that learning words by  
> frequency to somewhere over 1,000, then switching to another method,  
> would be optimal.  If possible, as it is with some programs,  
> focusing on words from the book or corpus you are going to be  
> reading may be helpful.

It's early days, but my current work has the following noteworthy  
characteristics (some are pros, some are cons):

- it is entirely inductive
- it is restricted to John's gospel (vocab *and* reading)
- it is based on clauses (from the OpenText analysis) rather than  
verses or sentences
- it involves learning inflected forms, not lexemes
- it does not yet take into account predictability of one form if you  
know another (the key topic of my PhD research)
- it currently assumes all words are equally easy to learn
- it does not currently take into account any other aspect of reading  
comprehension other than knowledge of inflected forms

Note also that there is no easy way to find the optimal ordering of  
vocabulary (it's an NP-complete problem for the computer scientists  
amongst you), however there are techniques for finding pretty good  
ones. One technique, applied within the constraints listed above gave  
the following as the first 10 inflected forms (in order) to learn to  
read clauses in John's Gospel:

APEKRIQH
IHSOUS
KAI
EIPEN
AUTWi
LEGEI
AUTOIS
hO
LEGW
hUMIN

Learning APEKRIQH lets you read one clause. Then with IHSOUS you can  
read another 22. With KAI and EIPEN you can read an additional 13. And  
with AUTWi, 18 more are added for a total of 54 clauses.

Despite the limitations and caveats, it's an interesting approach. I  
hope to have something available online soon but I want to address a  
couple of the limitations first.

Incidentally, I plan to be talking about this sort of stuff (and a lot  
more) in my keynote talk at BibleTech 2008[1]

James Tauber

[1] http://www.bibletechconference.com/



More information about the B-Greek mailing list