[Fwd: Re: [developers] processing of lexical rules]
Ann Copestake
Ann.Copestake at cl.cam.ac.uk
Thu Feb 17 14:45:57 CET 2005
Thanks for those comments. I guess I should say that I'm assuming we'll use
the LKB to try a few things out - I'd expect to keep the implementation as
general as possible and for developers to use the FSs to constrain application
ordering etc.
As far as clitics go - with the 2b option (i.e., morphology doesn't do
retokenisation), I think that clitics either have to be handled in the prior
tokenisation step (which may not be possible without building in a lot of
linguistic knowledge) or treated as something which brackets with the thing
they are tokenised as attaching to. But I don't know much about how people
treat clitics in computational systems - I may be missing an option.
BTW, as far a `generative grammarian' goes - personally I think that
`generative grammar' is lexicalised as a MWE and that `ian' is attaching to
that. Which is something that we currently allow (sort of) and I'd expect to
allow in a reimplementation. That is, `generative grammar' is an entry in the
lexicon, morphology operates to give
`generative' (ian-rule `grammar')
`grammar' is recognised as a rightmost bit of a MWE, and we end up with
something which is equivalent to (ian-rule `generative grammar')
So there is a way of allowing for something which is tokenised as two tokens
to correspond to a single lexeme, at least under certain conditions. I have
to work out exactly what the conditions are - but e.g. I don't think we could
allow prefixation on the first element where that licenced suffixation on the
second element.
Ann
More information about the developers
mailing list