Portability

[Stefan Müller]

Hi,

> How portable are analyses in these different systems?  Obviously if an
> analysis uses some feature that exists in one system but not the other,
> you're out of luck.  But if you stick to what's *conceptually* common,
> how easy would it be to take a grammar written in one *formalism* and
> port it to the other?

The question is what you consider to be conceptually common. What is
common between the systems is basically typed feature structures (with
some subtle differences in the implementation, which I consider minor,
but the ones who implemented the systems certainly do not). If you want
to do more in the LKB, you have to simulate this with typed feature
structures. Detmar et. al. also wrote a paper about this some years ago:

http://www.sfs.uni-tuebingen.de/~dm/papers/meurers-dekuthy-metcalf-03.html

As for porting: Dan's and my experience is that it is healthy to start
over from time to time. Large grammars tend to contain artifacts that
are only there because of the history of the development of the grammar.
So, if you know what your grammar did in one system it may be better to
try to implement something in the spirit of your old grammar in the new
system rather than porting it declaration by declaration.

Detmar/Kordula/Vanessa and Gerald did port LKB grammars to TRALE so they
may have more to say on this.

> Putting it differently, twenty or thirty years from now, when neither
> system runs on whatever computers and OSs we're using then, will you
> still be able to run one of these systems?

Well twenty years from now is still before retirement and Ann, Dan,
Stephan, Gerald, Frank, Martin, Kilian, Johannes and all the others will
still be hacking =;-).

The LKB is LISP-based (Allegro). There used to be open source variants,
maybe there are even new ports now. There are reimplementations of the
LKB in C. TRALE is Prolog-based. It runs under SICStus-Prolog, an SWI
port is almost ready for shipping. Since SWI is open source (and
C-based), I hope that it will still be available in 30 years from now.

Of course both systems come with a lot of additional machinery as for
instance Stephan Oepen's [incr TSDB()] (pronounced TSDB++, also known as
`The Fine System') and various debugging software that uses TCL/TK, Java
or wahtever. [incr TSDB()] relies on the parallel virtual machine (PVM)
and has to be tuned to the respective system so that it can communicate
with Lisp/Prolog and the PVM processes.

The [incr TSDB()] manual says on page 37 that Stephan will help with the
installation, provided there is a beach near your institution.

http://www.delph-in.net/itsdb/publications/manual.ps.gz

I do not know whether this will still be true in 30 years.

Independent of all this I think that developers should document their
grammars so that the knowledge and expertise is preserved for the
future. Then the respective grammars can be taken and be implemented in
whatever system is available then. Take an example: Pollard and Sag
1994. To write such a book is an enormous achievement. It took seven
years from Pollard Sag 1987 to Pollard Sag 1994. But according to Bob
Carpenter, it took just three weeks to implement the grammar in the
appendix of PS94.

@article{Mineur95a,
author = {Anne-Marie Mineur},
title  = {Interview with {Bob Carpenter}},
journal = {Ta!, the Dutch Students' Magazine for Computational
Linguistics},
volume = 3,
number = 1,
year = 1995
}

So if we have a precise description of seven years work with pencil and
paper or with the computer, we can reuse it and the achievement is there
independently of specific hardware.

But taking it one step further: Who knows whether our linguistic
theories will still look the same in thirty years. If not, the only
think that will be important will be the detailed discussion of the data
that was only possible because of the connection to some theory and
because of questions that arise during an implementation. On the other
hand: the HPSG toolbox (features, values, relations, inheritance) is so
general that it is difficult to imagine why the basic machinery should
be different in thirty years. Of course some analyses will change and
topics of interest will change too.

> (Apologies if this is answered in Nurit Melnik's article; our
> institution doesn't subscribe to Research on Language and Computation,
> and Springer wants its usual outrageous dollar a page for a PDF.)

They have to pay their web hoster ...


Best wishes

        Stefan

-- 
Stefan Müller       Tel: (+49) (+30) 838 52973
                    Fax: (+49) (030) 838 4 52973
Institut für Deutsche und Niederländische Philologie
Deutsche Grammatik
Habelschwerdter Allee 45
14 195 Berlin

http://hpsg.fu-berlin.de/~stefan/

http://hpsg.fu-berlin.de/~stefan/Babel/Interaktiv/