[b-hebrew] Pronoun resolution
kwrandolph at gmail.com
Wed Sep 26 16:50:09 EDT 2007
Sorry for the delay on this one.
On 9/11/07, biblical hebrew <jcr.bhebrew at gmail.com> wrote:
> Thanks Karl. This is kind of along the lines of what I was thinking. I'm
> trying to invent a system of digitally representing human understanding of
> a text. I want the digital understanding to represent everything that a
> human reader would understand when reading a text. My n-gram approach takes
> you quite far to such an understanding but n-grams of the type 'I saw her'
> present a problem. Is the 'her' a female person? Or an object of feminine
> declension? See the following link for examples of the kind of structure I'm
> talking about.
I did a spot check on some of the examples listed, and found them for
the most part what I could read right off. A few I had to check up a
larger context than included in the verses themselves, but once the
context was clear, could read it.
Now how did I do it? Well, for the most part it was so automatic that
it came without conscious thought. But I understand what was simple
for me may be devilishly difficult to program for.
The number one clue is the context, and how does one teach a computer
to recognize that? Particularly in Job, the context that explains a
verse may be as much as 20 or more verses earlier, or the context
built up over several verses. Even in Isaiah 37:20 which your list,
MYDW refers back to a subject that is a few verses earlier, but
defined by the context.
The next step is to recognize what parts of the sentences for each
word: verb, subject, object, and so forth, and again context helps.
For example if direct speech is indicated, then it helps in
interpretation when a form that could be 2nd pers. sing. masc. is
found, it is quickly put in understanding that here is the verb,
unless later in the verse another form that could be a verb is found
(though that is rare) or there is an indication that here there is a
change in subject (also rare). Seldom do I have to go back and say
"Hey, I don't have a verb" or other part of speech and analyze more
closely what I have just read. In fact, having to reread a passage
happens almost always in poetry, with its frequent allusions, literary
speech, poetic structures, euphemisms, and other non-direct speech.
(Did I say that I hated poetry back in school?)
>From this, the meanings of words is very important to recognizing what
is what, and is directly tied in to context. So when looking at a form
that could come from two or more roots, the form that fits the context
is the chosen one. How do you program for that?
> From what you've said and my own thoughts on the subject I'm thinking
> something like this: The human brain makes use of mental variables to track
> the objects referred to in a narrative. When we encounter pronouns we refer
> back to the mental variables we have stored and start to employ logic to
> resolve which object the pronoun is referring to. As you have noted the
> first use of a pronoun is likely to not be too far removed from the last
> explicit mention of the person/object it is referring to and examples where
> resolution presents difficulty for a human are rare (see the much discussed
> example with Moses, his son, his wife, Yhwh's angel and the circumcision
That example in Exodus 5, WTG( could have either of two subjects, both
of them fem. sing. nouns. Because I don't subscribe to what I consider
a weird meaning for "feet", the subject of the verb that best fits the
context then becomes (RLH. Meaning is an important part of deciphering
what is what.
> I think the logic we use to resolve pronouns may have similarities in all
> languages but the problem is in specifying just how exactly we do it and
> more specifically just how would a native Hebrew have done it. I think one
> of the tools in a native Hebrew's toolbox could be the instinctive knowledge
> of masculine and feminine nouns and in many cases pronoun resolution could
> be as simple as referring back to the only mental variable which matches.
> What do you think?
> I suppose the real problem comes about when more than one of the mental
> variables available matches the pronoun to be resolved. What mechanisms do
> we use to make this resolution?
Again much of it goes back to the meanings of words. Here is where
careful lexicography is needed. Context, meaning and form all work
together to resolve these questions, the only one easily computed is
> James Christian Read - BSc Computer Science
> http://www.lamie.org/hebrew Thesis 1 - Aleppo codex machine translation
> http://www.lamie.org/lad-sim.doc Thesis 2 - language
> acquisition simulation
Karl W. Randolph.
More information about the b-hebrew