Mokusei: Language Understanding (cont’d)
Problem: Left recursive structure of Japanese requires look-ahead to resolve role of content words
- Nihon wa . . .
- Nihon no tenki wa . . .
- Nihon no Tokyo no tenki wa . . .
Solution: Use trace mechanism
- Parse each content word into structure labeled “object”
- Drop off “object” after next particle, which defines role and position in hierarchy
- Because Japanese is left-recursive, parse trees tend to be deep at the beginning and shallow at the end.
- A top-down parser has to propose many different theories for the first word, all but one of which is eventually pruned.
- The trace mechanism in TINA allows the parser to delay building the superstructure until later in the sentence, leading to tremendous savings in computation