Biblical Hebrew Forum

[Yitzhak Sapir <yitzhaksapir AT gmail.com>]

Hello James,

You have suggested several lines of arguments over the past few days.  In
addition, you provided a link to a 30-page article that did not answer any
particular point that I asked of you.  It does however provide background to
the list of corpora that you linked at Europarl, as well as to one particular
point in a discussion by Cynthia Miller, "Methodological Issues in
Reconstructing Language Systems," where she deals specifically with
Biblical Hebrew.

The article you provided deals with training automatic translation algorithms.
The algorithms have minimal linguistic content.  In the authors' words, "it is
limited to one set of rules for analyzing a string of characters into
a string of
words, and another set of rules for analyzing a string of words into a string 
of
sentences."  The goal, is that using a large data set for training the
algorithm,
the algorithm will be able to automatically translate an arbitrary sentence in
the language.  You are quoting figures on the order of 2 million.  According 
to
this, with 2 million sentences and their respective translations, the 
algorithm
is still short of being able to provide completely adequate translations.

Actually, 2 million is a low number.  Consider Google, which has automatic
translations for web pages.  It probably has way over 2 million examples.  In
addition, the Google interface allows any user to suggest an alternative
translation for a particular fragment.  In this way, Google amasses together
a very large input data set.  The data set is still not good enough as you can
tell if you try to read any Google translation of a web page.

If we apply this to epigraphy, we would be saying that given only the 
particular
attested inscriptions and their respective translation in, say, English, we
would like to be able to translate any other sentence in epigraphic Hebrew
that we come upon.  Furthermore, we want to do this only knowing how to
join letters into words, and words into sentences.  We would like to do this
without knowing anything about Hebrew, about Semitic languages, or about
linguistics in general.

Of course the data set is too small for such a task!

How much is the linguistic knowledge that we dropped worth?  Well, again,
Google has (a guess) way over 2 million inputs, and still often generates
ridiculous translations.  But a linguist who has only a dictionary, of some
few thousand entries, a grammar, and some passages in the input
languages, as well as in cognate languages, will probably be able to create
a reasonable translation that is superior to a machine translation that
has no linguistic knowledge.

There is probably also an issue of diminishing returns.  Given so many
input phrases, the computer program can still achieve some degree of
effectiveness.  However, the programmer would like to achieve a much
higher degree, and so is forced to feed it more input.  With each newer
input, the algorithm gets better, but by a lesser degree than the previous
input.  This means that to get a truly high rate, a simply ridiculous amount
of input has to be fed in.  But the program/algorithm already was at a
high rate (comparatively speaking) with a little input (comparatively
speaking).  I am just making a guess here regarding the diminishing
returns.  You would probably be able to verify it if so.

Evidently, then, linguistic knowledge counts for a lot.  But there's another
issue here.  The question at hand is not about translation, not about
semantics, not about syntax, not about morphology, not about
phonology.  It is about orthography.  Compared to all the above,
orthography is a much more manageable study.  For orthography, the
data set is reasonable.  Even for some other aspects it is reasonable.

The data set does not allow us to effectively determine if there was
a dual number in pronouns in epigraphic Hebrew.  It does not allow
us to know if the 2fs suffix was sometimes spelled -ty.  This is because
dual numbers and 2fs suffixes are not attested.

But we can make conclusions about the 3ms pronoun which is
attested.  In Lachish 3 (that I quoted before to you and Karl), we find
the direct object אתה corresponding to אותו in the Bible.  The argument
that it is the laziness of the author does not help much.  Writing the
letter he for waw is not an issue of laziness.  Furthermore, such
examples are rarely attested in the Bible as in 2 Ki 9:25 שלשה vs
2 Ki 15:25 שלישו, or Jer 2:3 תבואתה vs Lev 19:25 תבואתו.  At times,
the sentence uses both as in Gen 49:11 עירה אתנו לבשו סותה.  But
whereas the Bible predominantly uses waw, the epigraphic
inscriptions use predominantly he.  Waw is used in epigraphic
inscriptions very rarely.  It may be used in Siloam (although there
is much debate on the subject, and on the development in this
particular case), and it is used in the 6th century Ketef Hinnom
amulets.  Whereas in the Bible the -w predominates in thousands
of cases, the epigraphic evidence shows -h in almost all the
cases.  In the epigraphic orthography, the -w suffix is used for
the 3ms pronoun when affixed to a plural noun (the examples
above deal with the singular noun).  The corresponding
orthography in the Bible is -yw.  So we see two systems
being used:

Bible -
  singular noun:  -w
  plural noun: -yw
Epigraphy -
  singular noun: -h
  plural noun: -w

Linguistically, the he is probably the original.  First of all, it is found
in Hebrew itself in verbal suffixes -hu, -ha.  Furthermore, it is found
in cognate languages, such as Aramaic.  Also, it would logically
appear to be a development from the 3s independent pronouns
which start with h.  The waw form can be explained as a
development: hu > wu > o.

The epigraphic evidence is varied.  Multiple geographic locations
are represented, multiple topics, multiple occupations of the authors,
and we even have one on papyrus.  In addition, we know that
religious texts always tend to a more archaic form of orthography
and language.  We see this in the DSS where the authors use the
Paleo-Hebrew script.  If the scribes used -h almost exclusively in
day to day correspondence, it makes no sense for them to use
the "innovative" forms with -w for official religious texts.  Either
the same spelling will be used, or the more innovative one for day
to day.  But not the other way around.

Yitzhak Sapir