Generalized Composition

[Víctor M. Castel]

Hi,

I need to use Generalized Composition in a Micro-Grammar of Spanish.
To this purpose, I have been using a category for verbs which looks like this:

s|np|np|np|np

But OpenCCG (i.e. my specification of a micro-grammar of Spanish) does not seem to allow for the use of Generalized Composition.

Question 1: Is it because Generalized Composition is not implemented in the current version
of OpenCCG?

Question 2: Or is it because I need the following bracketing for Generalized Composition
to work?

(((s|np)|np)|np)|np

Question 3: If the answer to Q2 is "Yes", am I correct in saying that the current XML
Schema does not allow for the specification of categories with this kind of bracketing?

Thanks a lot.

Víctor M. Castel
CONICET y UNCuyo
Mendoza, Argentina

[Jason Baldridge]

OpenCCG actually does allow generalized composition, so there is probably something like a feature in your categories that is causing things not to work. Can you give an example of the categories you are trying to combine?

Also,the following categories are completely equivalent:

s|np|np|np|np

(((s|np)|np)|np)|np

An example where you need paretheses can be seen in the following non-equivalent pair

s/(s/np)

s/s/np

The first one looks for a function which is missing an np to produce an s. The second looks for an np and then for an s, and is equivalent to (s/s)/np.

You might find some of the tutorials available at UT Austin helpful:

http://comp.ling.utexas.edu/wiki/doku.php/openccg

Jason

[Víctor M. Castel]

Jason,

thanks a lot for your prompt reply.

Here's the problem details illustrated with an example.

I can get the derivation (1) by Forward Composition, but the signs in (2) fail to compose:

(1)

s$1/(s$1/np[acc])    s|np[nom]|np[dat]|np[acc]|np[obl]
_______________________________________________________>B

           s|np[nom]|np[dat]|np[obl]

(2)

s$1/(s$1/np[dat])    s|np[nom]|np[dat]|np[acc]|np[obl]
_______________________________________________________>B2*

The failure in (2) appears to be caused by the inapplicability of Generalized Composition.

Víctor

[Víctor M. Castel]

Jason,

thanks a lot for your prompt reply.

Here's the problem details illustrated with an example.

I can get the derivation (1) by Forward Composition, but the signs in (2) fail to compose:

(1)

s$1/(s$1/np[acc])    s|np[nom]|np[dat]|np[acc]|np[obl]
_______________________________________________________>B

           s|np[nom]|np[dat]|np[obl]

(2)

s$1/(s$1/np[dat])    s|np[nom]|np[dat]|np[acc]|np[obl]
_______________________________________________________>B2*

The failure in (2) appears to be caused by the inapplicability of Generalized Composition.

Víctor

[Jason Baldridge]

Hi Victor,

Though generalized composition is supported, it isn't when there is a dollar involved. Dollars and sets cause a lot of extra pain in the application of rules because they are essentially standing in for lots of real categories (if you want more details on this, I can go into this in another post). What this means is that there are some gaps in what the application of the rules will support when dollars are involved. One way you can get around this is to forgo the use of dollars and use more fully specified categories.

It would actually be very nice to have a complete overhaul of OpenCCG's unification code. I made a lot of shortcuts that actually help out in terms of making things faster, especially in rule application, but it also leads to certain gaps like this one. Unfortunately, I don't have the time to do this myself -- but I'd be happy to help advise someone interested in doing it.

Jason

[Víctor M. Castel]

Jason,

Yes, please, I would like to have more details on why "dollars and sets cause a lot of extra pain in the application of rules".

And I'd be happy to try and find a solution to this problem. So, I wait for you to advise me on the issue.

Víctor