[pet] Several requests about the parameters for running PET

[Stephan Oepen]

hi david,

just briefly, on how to invoke cheap with the ERG:

> [...] (the command is cheap -mrs -verbose=3 -nsolutions english)

to enable characterization (i.e. character ranges as LNK values
in MRS EPs) with the ERG, you will need at least:

 -repp -cm

i.e. REPP-based string normalization and tokenization and chart
mapping (tokenization adjustments and lightweight NER); to get
the benefit of lightweight NER, i would also strongly recommend:

  -default-les=all

these together, should put <x:y> LNK annotation on each EP
in the MRSs, which is a first step towards working out where
the various pieces of semantics were introduced.

in principle, it is possible to annotate EPs with edge identifier
LNK values, i.e. a unique pointer into the derivation tree that
gave rise to this MRS.  currently, however, this functionality is
only available in the Lisp-based MRS code and [incr tsdb()].

to deal with unknown (out-of-vocabulary for the ERG lexicon)
words, i would further suggest you consider:

  -tagger

by default, the ERG is configured to call out to TnT, which is
bundled with the LOGON tree; look towards the bottom of
‘pet/common.set’ (‘tnt-command’ and ‘tnt-arguments’) to see
how TnT is invoked.  in principle, you should be able to plug
in another tagger, as long as it will mimic the TnT input and
output syntax (one token per line).

in your example above, there is no argument to ‘-nsolutions’,
which i suspect is just a typo?  in your setup, you probably
want just one parse, i.e. ‘-nsolutions 1’?

finally, for good measure, i would always impose a ceiling
for memory and time usage, e.g.

  -memlimit=1024 -timeout=60

which will allow it to use up to 2 gbyte of dynamic memory,
and up to one minute per input for parsing.

—i hope to have more to say about running cheap in server
mode, where i am afraid there is no ready-to-run solution
for your needs.  though possibly various stubs to build on.
your requirements are fairly similar to what is behind the
ERG on-line demonstrator.  what is your attitude towards
involving Lisp in your setup?

best wishes, oe