Lexicalized Tree-Adjoining Grammars (LTAG) - ad-teaching.infor...
Lexicalized Tree-Adjoining Grammars (LTAG) - ad-teaching.infor... Lexicalized Tree-Adjoining Grammars (LTAG) - ad-teaching.infor...
Why Lexicalization Syntactic structures associated with single words can be seen as more powerful POS-tags (“supertags”). Example (Transitive vs. intransitive verb) NP↓ S V eats VP NP↓ NP↓ S VP V sleeps (Finite) lexicalized grammars are finitely ambiguous. ⇒ The generated string languages are decidable. Lexicalization is useful for parsing since it allows us to drastically restrict the search space (as a preprocessing step). 12 / 52
Lexicalization Example (CFG which is not strongly lexicalizable with a TSG) Consider the following CFG: An intuitive approach to lexicalize it with a TSG might be: S → SS S → a S a S↓ S S S S S↓ But this is not a strong lexicalization because it cannot generate the following tree (which the CFG can generate): S a S a S S S S S Problem: In TSGs the distance between two nodes in the same initial tree cannot increase during derivation. a a a a 13 / 52
- Page 1 and 2: Lexicalized Tree-Adjoining Grammars
- Page 3 and 4: Context-Free Grammars Definition (C
- Page 5 and 6: Tree-Substitution Grammars Example
- Page 7 and 8: Outline 1 Why CFGs are not enough (
- Page 9 and 10: Cross-Serial Dependencies Example (
- Page 11: Lexicalization Example (Lexicalized
- Page 15 and 16: Outline 1 Why CFGs are not enough (
- Page 17 and 18: Initial Trees Definition (Initial T
- Page 19 and 20: Substitution Definition (Substituti
- Page 21 and 22: Adjunction Definition (Adjunction)
- Page 23 and 24: Adjunction Constraints Given TAG G
- Page 25 and 26: Lexicalization Example (strong lexi
- Page 27 and 28: Further Formal Properties of TAL Tr
- Page 29 and 30: TAG Parsing Parser: Given a string
- Page 31 and 32: Recognizing Adjunction But the algo
- Page 33 and 34: Dotted Tree We introduce the notion
- Page 35 and 36: Chart Items The algorithm stores in
- Page 37 and 38: Scan Operations Input string: c 1
- Page 39 and 40: Complete Operations 1 2 3 [γ, adr,
- Page 41 and 42: Recognizer Algorithm Algorithm (Rec
- Page 43 and 44: Complexity of Recognizer Given: n:
- Page 45 and 46: Recognizing Substitution Recognizer
- Page 47 and 48: LTAG-Spinal Parser LTAG-spinal: Rou
- Page 49 and 50: LTAG-Spinal Parser Graphical repres
- Page 51 and 52: Conclusion TAG: a grammar formalism
Why Lexicalization<br />
Syntactic structures associated with single words can be seen<br />
as more powerful POS-tags (“supertags”).<br />
Example (Transitive vs. intransitive verb)<br />
NP↓<br />
S<br />
V<br />
eats<br />
VP<br />
NP↓<br />
NP↓<br />
S<br />
VP<br />
V<br />
sleeps<br />
(Finite) lexicalized grammars are finitely ambiguous.<br />
⇒ The generated string languages are decidable.<br />
Lexicalization is useful for parsing since it allows us to<br />
drastically restrict the search space (as a preprocessing step).<br />
12 / 52