shallow parsing

[Peter G. Liu]

in our constraint-based chart parser I used an HPSG-similar grammar format
(which can be seem as a shallow version of it).

1. the shallow issue

It is shallow mainly in the following sense: It does not support the
extensive DAG embedding. It does not support the mechnism for long distance
dependence, no gap, no filler. Though the syntactic construction goes as
high as to S node and also forms trees of WH/YesNo questions, only phrase
level categories such as NP, PP, VP, ADJP, ADVP etc are used extensively (to
provide morphological and syntactic information for information extraction
in our case)


Basically the grammar format is a mixture of PATRII and HPSG. It is
HPSG-similar in the following snese: It handels complements and adjucts in
the way as normally done with the X-Bar schema. It supoorts the
head-principle and the principle of semantic contribution (it does not
support the subcategorisation principle, but there might be a easy way to
implement this principle to some extend).

The system has ca. 100 rules (part of them are lexical rules). As a shallow
parser, it has processeed ca. ten thousands of emails so far supporting the
information extraction.


2. Examples:

a) the modified DAG structure

the feature structure (a simplified HPSG format)

|phon:   _			|
|syn   |head:|cat:      _ |  |	|
|      |     |[feature]:_ |  |	|
|      |     |subtype:  _ |  |	|
|      |subcat:  _	     |	|
|				|
|sem   |pred :   _	  |	|
|      |modlist :_	  |	|
|      |arglist :_	  |	|


b) rule examples (for simplicity, we just take the string as a base
predicate such as book for BOOK(x))

# 100	VP	V	NP	PP

>V.syn.head.cat = V
>V.syn.subcat = %v2ppdir
>NP.syn.head.cat = N
>NP.syn.head.case = acc
>PP.syn.head.cat = PREP

>VP.syn.head < V.syn.head
>VP.sem.pred < {pred, V.phon}
>VP.sem.arglist < {theme, NP.phon}
>VP.sem.arglist < {destination, PP.sem.arglist}


#  100	S	VP

>VP.syn.head.cat = V
>VP.syn.head.tense = %present
>VP.syn.head.person = not!per3
>VP.syn.head.subtype = not!infinitive
>VP.syn.subcat = nil

>S.syn.head < VP.syn.head
>S.syn.head.person < per2
>S.syn.head.ctype < command
>S.sem < VP.sem
>S.sem.arglist < {agent, theMessageReceiver}


c) syntactic construction and semantic composition (output result)
____________________________________________
(0 - 5) S : 	VP .
Rule number: 32
Edge number: 249
AVM:

Category:	S
cat		V
num		_
person	per2
case		nom
aux		_
tense		present
subtype	not!base
s-type	command
subcat	[]
sem	(pred,      put)
arglist	(theme,     it)(destination,      your    commercial)(agent,
theMessageReceiver)
modlist
phon	[<S> , [<VP> , [<V> , [<V> , put]], [<NP> , [<PRON> , it]], [<PP> ,
[<PREP> , on], [<NP> , [<DET> , your], [<NBAR> , [<N> , commercial]]]]]]

____________________________________________
_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.