From: Ulrik Petersen (firstname.lastname@example.org)
Date: Thu Sep 18 1997 - 20:29:59 EDT
Clayton Bartholomew brought up some issues of teaching NT Greek morphology
and syntax in a new and bold way. What I want to add is my penny's
worth of computer science information (I am a student of computer
science), as well as to comment on the feasibility of this
method. Clayton wrote:
> I have been thinking for about a decade about a rule-based approach
> to teaching NT Greek morphology and syntax. This is not a system
> based on memorizing rules, it is quite the opposite. It is a system
> based on writing rules.
> The student would be give a simple form of graphic notation based on
> predicate logic to use as a basic tool for doing rule construction.
> (Examples of this notation can be found in a standard text book on
> Compiler Design, by Aho and Ullman. My memory is slipping but I
> think one form of this notation is called Backus-Nauer Form. Some of
> you younger software engineers need to help me out at this point.)
> Writing rules forces a student to think about the structural
> relationships in the morphology and syntax of the Greek language. It
> is much more fruitful activity than rote memorization. A certain
> amount of memorization will take place as a by-product of rule
You are right about the Backus-Naur Form, in that it is a notation for
describing the _syntax_ of a language, given its lexical elements,
called _tokens_ in computer science. A token would, in Greek,
correspond to a surface word. Backus-Naur Form is one notation for
writing Context Free Grammars (CFGs) largely invented by Chomsky in
the 50's. A Context Free Grammar is basically a set of rules by which
one rewrites a string of tokens, forming a tree which would, in the
case of Greek, ideally reflect the "right" linguistic parse-tree, as
anyone familiar with generative grammar will have seen, with the
sentence at the top, the clauses below that, the phrases below that,
and the actual words at the bottom, branches connecting everything.
Here is the bad news: More than 30 years of research has not produced
a single CFG for any natural language that is general enough to parse
all of the syntax of the language correctly. Why? Because there is
_ambiguity_ in any natural language at the syntax level, and thus the
"right" answer simply cannot be found by looking only at syntax and
morphology. I realise I'm talking about linguistics now, and I'm not a
linguist, so I may very well be wrong :-).
Here is the good news: It is _possible_ to formulate a CFG that parses
a _subset_ of a natural language, and this has been done for a rather
large subset of English (as any SIL man or woman will tell you -- note
their publicly available PCKIMMO program).
Clayton goes on to say:
> Again this is not a hypothetical concept. I did it myself, using a
> predicate logic language called Prolog. Syntax rules are actually
> fun to write because you can get instant feed back. You write the
> rule and then feed it a sentence. You keep working on the rule until
> you get the sentence to parse correctly. You are learning NT Greek
> syntax all the time you are doing this.
> There is a snag to this system. Most of the logic notations that I
> have encountered are intended for L-R languages. An L-R language is
> highly word order dependent and can be read from left to right (L-R)
> in a single pass. NT Greek is not an L-R language. So the notations
> that are currently available are not ideally suited for NT
> Greek. However we have a lot of bright, young and highly motivated
> technical types on this list. Why cannot one of you take on the
> task of designing a rule notation system which can accommodate
> languages like Greek and Russian. I have given the problem a lot of
> thought but the solution is beyond me.
You are confusing things here, I am afraid. Backus-Naur Form or CFGs
are normally not used to parse strings in a single pass, though most
parsing is done left-to-right, as you say. The term "single pass", I
take to mean: When one has seen something, one never considers it
again, in one form or other. The parsing technique used for CFGs
involves the concept of a Finite Push-Down Automaton, the details of
which I will spare you for, but I will say that it does look at things
more than once, at least in the sense that it remembers a number of
tokens in order to be able to parse a sentence.
You might be confusing the device for parsing _syntax_ with the device
for parsing _morphology_. It is true that parsing morphology will look
at a sentence, taking each word and probably only looking at it
By the way, LR stands for "Left-to-right-parse Rightmost derivation",
the techicalities of which lie beyond the scope of this mailing list.
Yes, there has been some indication in the research that the languages
that can be parsed by CFGs are word-order dependent, but it has, to my
knowledge, not been shown to be an intrinsic property of those
languages (commonly called CFLs for Context Free Languages).
My guess would be that to write a CFG for even a small subset of NT
Greek would be very impractical for a course on NT Greek, for several
1) The CFG paradigm is not easy to grasp, and especially not for
people who are not trained in mathematics,
2) There is not a straightforward, "obvious" way of going about
constructing a CFG for incrementally larger subsets of NT Greek (or
any natural language) without getting it wrong several times in the
process. I know from painful experience that wrong CFGs are not
easy to rewrite, such that they contain neither ambiguities, nor
over-productions in the sense of being able to derive or produce or
parse something which is not in the language.
3) An even small but still interesting subset of NT Greek would
require a huge CFG, taking a long time to write from scratch, and
with the possibility of introducing errors at every point in the
process of building this CFG.
What you have probably done, Clayton, is describing a subset of the _morphology_ of Greek in Prolog, though please correct me if I am wrong. I know that this can be done with about a third of a year's full-time work by a computer science student (it was attempted by two students at my university, and they described quite a substantial subset of the morphology of a text by Sophocles in Prolog). Their experience was this: It was great fun and very rewarding in terms of learning Greek to write these rules. In fact, one of them said to me that this ought to be _the_ way Greek morphology was taught in a modern world.
Clayton then says:
> Given a rule notation system that is well adapted to NT Greek one > could build a course on NT Greek based on rule writing. A student's > set of rules could be tested by feeding it real paragraphs from the > NT and seeing where the rules fail. This would require that the > rules be implemented as a program on a PC/MAC. This is not a > insurmountable problem.
No, you are right! It is not an insurmountable problem. It might even be a job for me in my future work for the Lord.
One could have several levels of difficulty, one in which, for example, the student was given the part-of-speech of each word, then being required to write the rules for that part-of-speech. Or, one could give the student only the surface word, they being then required to figure out its morphological functions from its morphemes. Or, at another level, one could opt between given the student each word already split into morphemes, and perhaps tagged with information on long and short vowels, substitution elongation (or whatever the term is in English), and other sound laws, and, on the other hand, having the student write rules to split a word into morphemes, as well as parsing hiatus, sound laws and what have we.
I would be interested in getting other ideas for this sort of program to assist in learning NT Greek. I might be trying to write a program in Prolog myself next spring (on the northern hemisphere), which would be able to parse the morphology of some text from the NT.
But before I get too carried away, maybe someone with lots of teaching experience could comment on the feasibility of teaching NT Greek morphology with this method? As Clayton so succinctly puts it:
> This suggestion is a *major* departure from traditional NT Greek > instruction.
Ulrik Petersen Student of computer science and mathematics, University of Aarhus, Denmark e-mail: email@example.com
This archive was generated by hypermail 2.1.4 : Sat Apr 20 2002 - 15:38:30 EDT