XML-Based Interface to (External) Preprocessors

As part of the EU-funded Deep-Thought project, the LKB was interfaced to an external finite-state preprocessor, morphological analyzer, and tagger using an XML-based interface. This interface could serve to harmonize existing connections to external preprocessors (e.g. to ChaSen in the Japanese grammar) over time, and probably should be used as the role model for future integrations.

The Simple PreProcessor Protocol is documented on the LkbSppp page.

The MAF-based XML annotation scheme is documented on the LkbMaf page.

(Internal) Finite-State Preprocessing

To use the built-in finite-state preprocessor please a line such as the following in your grammar's script file:

  (read-preprocessor (lkb-pathname (parent-directory) "preprocessor.fsr"))

Once a preprocessor has been loaded, preprocess-sentence-string will automatically send input to the correct finite-state preprocessor. If no preprocessor has been loaded, you get the LKB's default built-in preprocessor (this throws away punctuation, modulo *bracketing-p*. and creates a sequence of tokens by splitting on space characters).

Characterization

Characterization support allows the deep processor to relate components of the linguistic analysis to the input text. To enable simple (character-point based) characterization support, set the following in your grammar's globals.lsp:

  (setf *characterize-p* t)

You must also ensure that CFROM/CTO features (of type lkb::*toptype*) are provided in the (R)MRS semantics component of your grammar. Eg. [from the ERG]

  relation := relation_min &
  [ PRED predsort,
    LBL handle,
    WLINK *list*,
    CFROM *top*,
    CTO *top* ].

These features are used to store and propogate pointers during linguistic processing. You can see these features in the semantic analysis:

The dog barks. 
<rmrs cfrom='-1' cto='-1'>
<label vid='1'/>
<ep cfrom='0' cto='14'><gpred>prpstn_m_rel</gpred><label vid='1'/><var sort='e' vid='2' tense='present'/></ep>
<ep cfrom='0' cto='3'><realpred lemma='the' pos='q'/><label vid='6'/><var sort='x' vid='9' pers='3' num='sg'/></ep>
<ep cfrom='4' cto='7'><realpred lemma='dog' pos='n' sense='1'/><label vid='10'/><var sort='x' vid='9' pers='3' num='sg'/></ep>
<ep cfrom='8' cto='14'><realpred lemma='bark' pos='v' sense='1'/><label vid='11'/><var sort='e' vid='2' tense='present'/></ep>
<rarg><rargname>MARG</rargname><label vid='1'/><var sort='h' vid='4'/></rarg>
<rarg><rargname>RSTR</rargname><label vid='6'/><var sort='h' vid='8'/></rarg>
<rarg><rargname>BODY</rargname><label vid='6'/><var sort='h' vid='7'/></rarg>
<rarg><rargname>ARG1</rargname><label vid='11'/><var sort='x' vid='9' pers='3' num='sg'/></rarg>
<hcons hreln='qeq'><hi><var sort='h' vid='4'/></hi><lo><label vid='11'/></lo></hcons>
<hcons hreln='qeq'><hi><var sort='h' vid='8'/></hi><lo><label vid='10'/></lo></hcons>
</rmrs>

For more sophisticated characterization support see LkbMaf.

LkbPreprocessing (last edited 2011-10-08 21:12:10 by localhost)

(The DELPH-IN infrastructure is hosted at the University of Oslo)