Resemblances between MeaningText Theory and Functional Generative Description

1
Resemblances between Meaning-Text Theory and
Functional Generative Description

• Zdenek abokrtský
• Institute of Formal and Applied Linguistics
• Charles University, Prague

2
Functional Generative Description

• developed in Prague since mid 60s (Sgall,1967)
• sharing most of the peculiarities of the MTM
  (Bolshakov and Gebulkh,2000)
• multilevel character of the model
• orientation to synthesis
• distinguishing deep and surface syntactic
  representation
• accounting of communicative structure
• orientation to languages of a type different
  from English
• labeling syntactic relations between words
• keeping traditions and terminology of classical
  linguistics

3
Levels of representationin MTT and FGD
semantic
tectogrammatical
deep-syntactic
surface-syntactic
surface-syntactic
deep-morphological
morphological
surface-morphological
morphonological
deep-phonological
phonetic
surface-phonological
4
DSyntR vs. tectogrammatics

• in both
• skelet of the representation dependency tree
  (plus non-tree relations of co-reference)
• nodes semantically full lexemes
• inflectional meanings grammemes/grammatemes
• ficitious lexemes
• valency actants vs. circumstantials
• in DSyntR
• DSynt prosodic structure
• in TGTS
• semantically motivated inventory of dependency
  relations, so called functors (ACT, PAT, ADDR,
  ORIG, EFF, CAUS, DIR?, LOC, TWHEN, CAUS, BEN...)

5
Side remark re-inventing the DSyntR/TGTS in
PropBank

• 2002 annotated propositions only verbs and
  their arguments
• adding modifiers of event variables
• adding arguments of nouns
• adding discourse connectives

6
FGD implementationPrague Dependency Treebank

• long-term research project aimed at creating a
  syntactically annotated corpus based on the
  framework of FGD
• since 1995, inspired by Penn Treebank
• manually annotated Czech newspaper texts
• layered annotation scheme
• PDT 1.0 released in 2001 (distributed by LDC)
• PDT 2.0 to appear in 2006

7
Layered annotation scenarioof PDT 2.0

• 3 layers of annotation
• t-layer - tectogrammatical layer
• a-layer analytical layer
• m-layer morphological layer
• original text
• w-layer original sentence

8
m-layer sample
(Some contours of the problem seem to be clearer
after the resurgence by Havel's speech.)
9
a-layer sample
10
t-layer sample
11
Coordination in dependency trees in PDT

• physically still a tree structure, but tree edges
  do not always directly correspond to dependencies
• the real dependency and coordination relations
  can be (deterministically) derived by edge
  composition
• direct vs. effective parent/children

12
PDT 2.0 amount of the data
13
Summary

• FGD similar to MTT in several aspects
• PDT implementation of FGD framework on a large
  data