Biblical Hebrew Forum

[James Read <J.Read-2 AT sms.ed.ac.uk>]

Hi Randall,

it just so happens that this summer I was working on the optimisation  
of the EM algorithm for the unsupervised training of bilingual  
dictionaries. Essentially, this is a probabalisitic operation.  
Basically, you start with two sentence aligned texts (a source text  
and an equivalent target text) and the program aligns words with  
probability scores of being translations of each other.

In any language we get what is known as a Zipf distribution. This is  
basically a phenomenon whereby there are a handful of words which make  
up the majority of any text and then at the other end of the extreme  
there are many words which will only happen once in any considerable  
corpus. Of course, in between these two extremes you get a whole  
spectrum of gradual behaviour. Research has shown that the EM  
algorithm produces better alignments when you apply smoothing. So,  
what is smoothing.

Smoothing, in this case, is when we acknowledge that if we increase  
the size of the corpus the relative frequency of the frequent words  
will not change dramatically but the number of rare words we encounter  
will increase steadily. Smoothing is when we factor this loss of  
probability space into our probabalistic calculations to account for  
all those rare words we have not encountered because of the limited  
size of our corpus.

How do you think the principal of smoothing affects your probability  
based arguments?

James Christian


Quoting Randall Buth <randallbuth AT gmail.com>:

> vayyixtov Karl
> > > Speculation. Indicated by the term “probability”.>
>
> Speculation?
> Maybe we could communicate better in biblical Hebrew. probability is
> based on evidence and in this case by an intersecting of three probabilities
> for any word deemed 'highly probably Late Biblical Hebrew". The low side
> of this probability gets diminished with each similar patterning word with
> which it is joined. One stick may be breakable, ten sticks become very
> difficult to break. Thirty sticks become like an iron rod. Speculation would
> be to ignore the reinforcing confluence of pattern and to posit, to
> speculate, that somehow in a parallel universe it is all irrelevant.
>
> > This is especially relevant in that ZMN has a semantic
> field not shared by any other Hebrew term, unlike your sheep = mouton
> example above, so it cannot be used as a replacement for another (“earlier”)
> term.>
> > You have not come to grips with the fact that it has a different meaning
> than your earlier examples, in fact was even used as a modifier to one of
> your examples, Therefore it is not a replacement for “earlier” terminology.>
> >
>
> Maybe you have not come to grips with the fact that zeman remained in
> the language along with `et and mo`ed, causing a semantic field adjustment.
> You will note that mouton caused a meaning adjustment in English, it does
> not mean 'sheep' but 'sheep flesh as food'. It entered with the Normans and
> metamorphicized.
>
> and you analysis is poor and unreliable.
> It shows an etymoloigical fallacy as if a word's meaning is it root.
>
> > MW(D is from the same root as Y(D meaning *to convoke, call out to a meeting
> * so as a noun it has almost the same meaning as εκκλησια in Greek.>
>
> This is not lexicography but etymology and leads to faulty comparisons.
> totally unreliable.
> mo`ed is a time word. You will find the LXX translates with a word like
> KAIROS 'specific time' and others.
> While EKKLHSIA is word closer to qahal and `eda in Hebrew.
>
> On zeman, Eccl 3:1 is an example showing how much `et and zeman
> overlap. there is a short parallelism:
> la-kol zeman  'for everything there is a time'
> ve-`et le-xol Hefets 'and a time for every item'
> taHat ha-shamayim
> zeman covers all of the `ittim that follow
>
> In Neh 2:6 uses zeman for a length of time for a trip to be made.
> In Esth 9:27 and 31 use zeman as a tight synonym to Hebrew mo`ed.
>
> I would conclude that zeman is a word that covers mo`ed and is wider,
> more generic, so as to reach to include `et. (Of course, in principle
> one should
> trace how the word continued to be used throughout the history of the
> language. 4 biblical ocurrences are not a lot of examples.)
> This analysis receives some confirmation
> in Daniel and Ezra where the idiom beh-zimna "at that time, in the very hour"
> is common in Aramaic for these authors. And as a word that is equivalent and
> more generic than mo`ed, its lack in First Temple Hebrew is quite remarkable.
> You may scoff, but many of the places in torah where we have mo`ed or
> pa`am or a time phrase in Hebrew we find zeman in Onkelos. This is not proof
> but is supportive of the overall picture.
>
> > Ps: Just out of curiosity, what does “vayyixtov” mean? Why don’t you write
> it in Biblical Hebrew?>
>
> I really don't know how to answer this question. A language does not depend
> on a script. vayyixtov is biblical Hebrew, written in a latin-
> based script.
> Here in BH, 'v' can be sounded as a 'w' or 'v' as you please.
> 'x' is the standard IPA for a voiceless velar fricative, the sound of a soft/
> fricative kaf according to masora. 'v' at the end of the word is the  
>  fricative
> bet. the vowels are according to masoretic tradition. The word is
>
> 'and [Karl] wrote' . . .
>
> the word is so common that it is difficult to conceive how someone would
> not readily grasp it. A couple of other scripts
> ويختب كارل
> ουαγιχτοβ καρλ
> ויכתב כארל
> all the same language,
> but its easier and faster to write vayyixtov Karl ...
>
> umeqavve ani ki-yesh le-xa Hedva ba-`et ha-zot
> ומקוה אני כי יש לך חדוה בעת הזאת
> הוי שלום  heve shalom
>
> Randall
> --
> Randall Buth, PhD
> www.biblicalulpan.org
> randallbuth AT gmail.com
> Biblical Language Center
> Learn Easily - Progress Further - Remember for Life
> _______________________________________________
> b-hebrew mailing list
> b-hebrew AT lists.ibiblio.org
> http://lists.ibiblio.org/mailman/listinfo/b-hebrew



--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.