[b-hebrew] Generation grammar and b-hebrew

Eric Inman eric-inman at comcast.net
Thu Jun 17 20:21:03 EDT 2010


I'm happy to see that there has been work in the area of viewing grammar as
something other than a binary operation.
 
I agree that a generative grammar can be used for something other than
defining the set of sentences that are valid in a given language. What I've
seen described in the responses makes sense.
 
My next question is about word order. It seems like generative grammars
would be easier to manage in languages where word order is fairly
constrained, such as English, and harder to manage in languages where word
order is more free, such as Greek. Where does Hebrew fall in this spectrum
and how do you think that will effect the difficulty of coming up with a
grammar?
 
Eric Inman

  _____  

From: James Christian [mailto:jc.bhebrew at googlemail.com] 
Sent: Thursday, June 17, 2010 4:10 AM
To: Eric Inman
Cc: Pere Porta; b-hebrew
Subject: Re: [b-hebrew] Generation grammar and b-hebrew


This is about the most intelligent and informed response I've had yet. You
are spot on in every respect.  

Yes! Unfortunately, attempts to define generative grammars computationally
are predominantly binary operations i.e. accepted or rejected. Recent
psycholinguistic experiments which I have heard of but cannot provide
references for at this moment view grammar not as a binary operation but as
a ranking operation. Let's look at an example:

1) I am James
2) I is James
3) I are James

The common meaning behind each of these three sentences is easily extracted
by a native speaker. A binary operation defined by standard English grammar
would only accept sentence 1 but reject 2 and 3. In spite of this the data
of spoken English would show that, while considered incorrect, forms 2 and 3
are attested in the corpus. Probability scores would assign 1 as the most
likely in certain genres. In the genre of gang A member speaking to rival
gang B member in the back streets of Detroit maybe form 2 is the most
likely.

One thing I would disagree with is the assumption that a generative grammar
must define sentences. This is, as you observe, the de facto but there is no
reason why couldn't first start with smaller chunks and work our way up to a
definition of paragraph grammar and even eventually to an entire discourse.
Certain sentences, however grammatical they may seem in isolation, are not
likely to be accepted when they follow an irrelevant context.

I would suggest we start with smaller and most frequent units. I suspect
that the category of PP's can be better defined if we split this over
generic category into more useful functional units like phrases of location
of action, phrases of direction of motion etc. This is my theory. Only
experimentation with the data would show if my theory is worth anything. I'm
not here to push theories. I would like to start experimenting so that we
can see together what the data says for itself.

Back to the subject of grammatical ranking. I have no doubt that there are
many utterances in the b-hebrew corpus that some speakers of b-hebrew would
have ranked low in psycholinguistic experiments of ranking for grammatical
correctness. It would be interesting if we could spot patterns in the data
indicative of this but with no informants I really don't see how we could be
sure of any interpretation of the data we offer. I suggest we start by
attempting to define a grammar which can produce what is in the corpus. This
in itself is already far more complicated and interesting a task than it
would at first seem

James Christian


On 17 June 2010 04:08, Eric Inman <eric-inman at comcast.net> wrote:


I have a question.

My understanding is that such a grammar is for defining the set of sentences
that can be considered to be valid Hebrew sentences. This would imply that
any sentence is in one of two states: either a valid Hebrew sentence or not.
My concern is that typically there are sentences which are technically valid
but which would never or rarely be used in practice. In some cases this
would be because the sentence makes no sense semantically. In other cases it
would be because for whatever reason a different sentence would normally be
preferred for expressing the given content. In other words, the question of
whether or not a sentence is a valid Hebrew sentence sometimes is neither
yes or no but rather more of a probability or set of probabilities of how
likely the sentence is to be used in practice and/or considered valid.

Therefore I'm not sure how useful this kind of grammar is as a model of the
language. It also seems that trying to build one would get bogged down in
discussions of these gray areas. I think it would be better to use a model
that had a way of taking into account these gray areas.

What do you think?

Eric Inman


-----Original Message-----
From: b-hebrew-bounces at lists.ibiblio.org
[mailto:b-hebrew-bounces at lists.ibiblio.org] On Behalf Of James Christian
Sent: Wednesday, June 16, 2010 2:56 PM
To: Pere Porta
Cc: b-hebrew
Subject: Re: [b-hebrew] Generation grammar and b-hebrew

Hi all,

further to Pere's question I'll explain with an example of why I believe
joint participation from list members on an experiment of this kind could
make for good linguistic discussion on this list.

Let's assume that in defining our grammar we want to start with high
frequency phenomenon to increase our chances of developing a wide coverage
grammar. Doing a simple word count on words that occur in the Tanakh we see
that with the exception of Yhwh the most common words (as in most languages)
are prepositions. This would suggest that prepositional phrases are high
frequency linguistic chunks and a good place to start developing a wide
coverage grammar.

Following a theory of constituency it is pretty much standard to create high
level definitions of prepositional phrases (PP) with DCG rules like the
following (for English):

PP --> P NP

NP --> (Det) N

Obviously, in practice such an over generic rule set is bound to over
generate because of the lack of fine grained control on the rule set. In
certain contexts certain pseudo-legal PP's are ungrammatical at some level
of grammar that goes beyond simple constituent structure. This is perhaps a
partial motivation of formalisms like LFG (lexical functional grammar) which
add a functional structure to the generation grammar.

In any case, the general idea is to get list members involved in an
experiment of developing a generational grammar so that we can discover the
finer nuances of b-hebrew together by experimenting with rule sets on real
data. I predict that such a practical experience will be eye-opening to the
intricacies of b-hebrew grammar and will stimulate many a linguistically
based discussion (something we keep aiming at on b-hebrew but always seem to
end up straying away from).

The reason I suggested we do something like this using Prolog is simple.
Prolog (IMO) is a shoddy programming language which is inefficient and I
would never dream of using for a serious linguistic technology. However, it
has inbuilt DCG support and so no programming skills required. If you can
understand what PP --> P NP means then you can start working straight away
with Prolog to build a generation grammar. It really is that simple. Totally
linguist friendly.

http://en.wikipedia.org/wiki/Definite_clause_grammar

In any case, I've already had some positive responses off list and I would
welcome some more and would like as many to get involved as possible. Please
keep writing to express interest if you haven't already done so. I've
already decided the positive response was large enough to attempt to get
something going. I'll go ahead and make a simple toy grammar to play with,
experiment with, build on etc. Those who are interested in participating the
best thing to do while waiting for me to post further information is to
download and install Prolog and perhaps play around with it.

If you're running Unix or Linux then you may already have it installed. If
not there are plenty of distributions available. GNU Prolog is probably the
most common standard.

www.gprolog.org

Obviously, GNU Prolog can be run in Windows using an emulator like Cygwin.
I've been told there are other good Prolog distributions that run on windows
but can't recommend any as I've never used them. If you can get Prolog
running on your system and your installation can interpret DCG's then your
ready to make a start.

I'll let you all know when I've got a toy grammar ready to start playing
with. It shouldn't take too long and I should be able to fit it in around
other things going on. If no-one has any objections to discussing b-hebrew
grammar via experimentation with a generational grammar on the b-hebrew list
then I'll continue to communicate via the list. If not, then we can always
take it off list amongst those who have expressed an interest.

Again, thanks for your responses. Keep them coming.

James Christian

On 15 June 2010 12:01, Pere Porta <pporta7 at gmail.com> wrote:

> (As a replying to James Christian's mail about his project on
> generation grammar, which I mistakenly sent only to him and not to the
> list)
>
> In this case, James, start, please.
> Now, I do not know whether this can be done within b-hebrew list or
> rather privately...
> Dear b-hebrew Moderators, what is your mind and/or advice? Could
> James's project or proposal be interesting for b-hebrew list members
> in a general way?
>
>
> Pere Porta
> (Barcelona, Catalonia, Spain)
>
> On 10 June 2010 09:55, James Christian <jc.bhebrew at googlemail.com> wrote:
>
>> Hi all,
>>
>> I don't how familiar everybody is with theories of generation
>> grammar. In a nutshell theories of generation grammar can be
>> summarised as follows.
>> There
>> is a set of words and rules which describe any given natural language.
>> Given
>> a correctly defined set of words and rules the generation grammar
>> will accept all grammatically correct utterances of that language and
>> reject ill formed utterances.
>>
>> Attempts to form a generation grammar for any given language
>> generally suffer from problems of over generation and under
>> generation. Over generation is when the grammar is poorly defined
>> and, as such, accepts ill formed utterances as well formed
>> utterances. Under generation is when the grammar rejects utterances which
are known to be well formed.
>>
>> Testing of generation grammars consists of having two different test
sets.
>> One test set of well formed utterances and the other of ill formed
>> utterances. The goal is that the grammar we define will accept the
>> utterances known to be well formed and reject the utterances known to
>> be poorly formed.
>>
>> If we were to extend this method to b-hebrew we already have a corpus
>> of utterances we can consider to be well formed. Defining a corpus of
>> ill formed phrases is a little more difficult as informants are lacking.
>>
>> I was toying with the idea of encouraging fans of b-hebrew to put
>> their theoretical knowledge into practice by playing with building a
>> generation grammar of b-hebrew using a simple linguist friendly
>> Prolog driven interface. I just wanted to get a quick idea of numbers
>> of who may be interested in participating in such an experiment.
>> Please let your students etc. know about the prospect and get them to
>> write to me either on or off list so that I can get an idea of
>> whether it would be worth putting this together.
>>
>> James Christian
>>
>> _______________________________________________
>> b-hebrew mailing list
>> b-hebrew at lists.ibiblio.org
>> http://lists.ibiblio.org/mailman/listinfo/b-hebrew
>>
>
>
>
_______________________________________________
b-hebrew mailing list
b-hebrew at lists.ibiblio.org
http://lists.ibiblio.org/mailman/listinfo/b-hebrew







More information about the b-hebrew mailing list