[b-hebrew] Machine translation of the Bible, was: heno- v's poly- theismin the torah

Peter Kirk peterkirk at qaya.org
Fri Jul 29 12:54:23 EDT 2005

On 29/07/2005 16:39, Read, James C wrote:

>In fact, I firmly believe that grammar is more of a hindrance 
>than a help to learning a language and I intend to demonstrate 
>this with my AI project.
>The purpose of my project is to develop algorithms which analyse 
>any input of language. The alphabet it will use will be based 
>on phonemes and not the inadequate arrays of alphabets that 
>modern languages use.
>The algorithms will completely avoid grammar and rely entirely 
>on examples and contexts to learn the meanings of words (a word 
>is a series of sounds/phonemes and may not necessarily agree with 
>the classical division of words we are accustomed to).
>It is my aim that this tool will produce a translation of the 
>hebrew bible, which is completely devoid of doctrinal influences, 
>as a demonstration that the natural way of learning languages 
>is the superior one. It's a big project and the algorithms have 
>many challenges but I am benefitting from the thoughts posted 
>in the various threads.
James, I wish your project well, but I am not very hopeful for you.

After all, researchers have been working on machine translation between 
major world languages for many decades, and the best results that they 
can come up with are appalling. For example, a few years ago I used 
Babelfish to translate Psalm 23:1 from English to Chinese to English and 
from German to English, and in both cases got the result "The Lord is 
not my shepherd"! (The "not" was wrongly transferred from the second 
clause.) And even when used on actual web pages the results were 
frequently incomprehensible.

To come up with a useable Bible translation from Hebrew you will have to 
do very much better than such programs can do even from German. And yet 
the only way you can do this in a way which "is completely devoid of 
doctrinal influences" (a laudable aim, although I don't think an 
achievable one) is to avoid "teaching" the program with ANY 
human-generated translations of Bible passages. But then, as I 
understand it (although I am not at all an expert), the only way that 
the current generation of machine translators can make any real progress 
is by matching to patterns found in real examples of human-generated 
translation. So you will need to find a new approach here.

Sorry to be discouraging. I fear that your aim is unattainable. But I 
hope that it is not.

Peter Kirk
peter at qaya.org (personal)
peterkirk at qaya.org (work)

No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.338 / Virus Database: 267.9.6/59 - Release Date: 27/07/2005

More information about the b-hebrew mailing list