[developers] LKB: Unification in lexicon

[Emily M. Bender]

Hi Bart,

My understanding is that that restriction is there for
a fairly fundamental reason:  Positing instances which
inherits from multiple types requires reasoning over
the type hierarchy (and possibly refining it, in the form
of adding glb types).  Instead, the LKB finalizes the
type hierarchy and then loads the instances (lexical
entries, lexical rules, syntax rules).

Emily

On Thu, Sep 25, 2008 at 7:54 AM, Bart Cramer <bart.cramer at gmail.com> wrote:
> Dear delph-in members,
>
> I have a question regarding the use of lexicons in the LKB system. I
> would like to try the following line:
>
> lex-verb-sein-1-fin-1-sg-pres-ind := tl-verb-sein-1 &
> infl-verb-fin-1-sg-pres-ind & [ ORTH "bin" ].
>
> As you can see, I handle the morphology in a somewhat sloppy way, by
> just enumerating the forms (and I have my reasons to do so). However,
> LKB does not seem to pick this up. When I use Debug -> Check lexicon,
> I get an error message:
>
> Unifications specified are invalid or do not unify.
> Structure for LEX-VERB-SEIN-1-FIN-1-SG-PRES-IND coud not be created.
>
> I am pretty sure that the unification is correct. When I put the
> following in a regular type file:
>
> wt-verb-sein-1-fin-1-sg-pres-ind := tl-verb-sein-1 &
> infl-verb-fin-1-sg-pres-ind.
>
> and the following in the lexicon:
>
> lex-verb-sein-1-fin-1-sg-pres-ind := wt-verb-sein-1-fin-1-sg-pres-ind
> & [ ORTH "bin" ].
>
> where wt stands for "word type", it perfectly parses. The problem with
> this solution, however, is that there is going to be a type for each
> word, which leads to very, very long start-up times (lower bound
> stuff...).
>
> Can anyone enlighten me why this restriction in the lexicon is there,
> and how I could circumvent it?
>
> Many thanks in advance,
>
> Bart.
>