next up previous contents
Next: Summary Up: Lexical Organization Previous: Implementation

Generating grammars

We have used our tool to specify a grammar for English in order to produce the trees used in the current English XTAG grammar. We have also used our tool to generate a large grammar for Chinese. In designing these grammars, we have tried to specify the grammars to reflect the similarities and the differences between the languages. The major features of our specification of these two grammars26.3 are summarized in Table [*] and [*].
 
Table: Major features of English and Chinese grammars
  English Chinese
examples passive bei-construction
of LRRs dative-shift object fronting
  ergative ba-construction
examples wh-question topicalization
of transformation relativization relativization
blocks declarative argument-drop
# LRRs 6 12
# subcat blocks 34 24
# trans blocks 8 15
# subcat frames 43 23
# trees generated 638 280

 


 
Table: Comparison of the two grammars
  both grammars English Chinese
  causative long passive VO-inversion
LRRs short passive ergative ba-const
    dative-shift  
  topicalization    
trans blocks relativization gerund argument-drop
  declarative    
  NP/S subject   zero-subject
subcat blocks S/NP/PP object PL object preverbal object
  V predicate prep predicate  

 

By focusing on the specification of individual grammatical information, we have been able to generate nearly all of the trees from the tree families used in the current English grammar developed at Penn26.4. Our approach, has also exposed certain gaps in the Penn grammar. We are encouraged with the utility of our tool and the ease with which this large-scale grammar was developed. We are currently working on expanding the contents of subcategorization frame to include trees for other categories of words. For example, a frame which has no specifier and one NP complement and whose predicate is a preposition will correspond to PP P NP tree. We'll also introduce a modifier field and semantic features, so that the head features will propagate from modifiee to modified node, while non-head features from the predicate as the head of the modifier will be passed to the modified node.
next up previous contents
Next: Summary Up: Lexical Organization Previous: Implementation
XTAG Project
1998-09-14