| Date |
Sept 16, 2014 |
| Speaker |
Prof. Colin de la Higuera
|
| Title |
Learning Probabilistic Context-Free Grammars
|
| Abstract |
Grammatical inference concerns the question of learning an automaton or a grammar given information about a language. The learning of finite state machines is a difficult but well understood question. The one of learning context-free grammars is harder, with far less results. this is even more so the case when the grammars are probabilistic.
In this talk we will survey the main grammatical inference results, discuss some reasons (involving undecidable questions concerning the computation of distances between distributions generated by probabilistic context-free grammars (PCFGs) and present a new algorithm allowing to learn these, which is sufficiently robust to do so in the presence of small datasets but at the same time can scale up and learn grammars with many rules.
|
|
アブストラクト